Forest Fires

BEHAVIOR AND
ECOLOGICAL EFFECTS
This Page Intentionally Left Blank
 Forest
Fires
BEHAVIOR AND
ECOLOGICAL EFFECTS

Edited by
EDWARD A. JOHNSON
Department of Biological Sciences
and Kananaskis Field Stations
University of Calgary
Calgary, Alberta
Canada

KlYOKO MiYANISHI
Department of Geography
University ofGuelph
Guelph, Ontario
Canada

ACADEMIC PRESS
C^ A Harcourt Science and Technology Company

San Diego San Francisco New York Boston London Sydney Tokyo
Front cover picture: Boreal forest crown fire estimated to have a frontal
intensity of about 30,000 kW m " \ The fire is about 20 km away, burning
away from the camera. At least five convective updrafts can be identified
by their darker smoke indicative of incomplete combustion. The wind was
about 45 km h""\ Notice the suggestion of shearing of the convective up-
drafts. Tornadic fire whirls were observed immediately in front of this
advancing fire line. The fire eventually burned 48,000 ha.

This book is printed on acid-free paper.

Copyright © 2001 by ACADEMIC PRESS

All Rights Reserved.

No part of this publication may be reproduced or transmitted in any form or by any
means, electronic or mechanical, including photocopy, recording, or any information
storage and retrieval system, without permission in writing from the publisher.

Requests for permission to make copies of any part of the work should be mailed to:
Permissions Department, Harcourt Inc., 6277 Sea Harbor Drive,
Orlando, Florida 32887-6777

Academic Press
A Harcourt Science and Technology Company
525 B Street, Suite 1900, San Diego, California 92101-4495, USA
http ://www. academicpress. com

Academic Press
Harcourt Place, 32 Jamestown Road, London NWl 7BY, UK
http ://www. academicpress .com

Library of Congress Catalog Card Number: 00-110672

International Standard Book Number: 0-12-386660-x

PRINTED IN THE UNITED STATES OF AMERICA

00 01 02 03 04 05 SB 9 8 7 6 5 4 3 2 1
 CONTENTS

Contributors xi
Preface xiii
Acknowledgments xvii

1 Strengthening Fire Ecology's Roots

E. A. Johnson and K. Miyanishi
I. Introduction 1
II. Processes 3
III. Transfer Rates and Budgets 4
IV. Examples of Traditional vs. Proposed Approach 5
References 9

2 Flames
K. Saito
I. Introduction 12
II. Basic Aspects of Combustion in Forest Fires 13
III. Temperature, Velocity, Species Concentration, and Flame
Height 18
IV Premixed and Diffusion Flames 23
V. Extinction of Diffusion Flames 27
VI. Diffusion Flames and Scaling Analysis 29
VII. Spreading Flames 41
VIII. Structure of Flame Base 45
Vi Contents

IX. Conclusions 47
Notation 48
References 51

3 Combustion Chemistry and Smoke

D. Ward
I. Introduction 55
II. Fuel Chemistry and Combustion 57
III. Smoke Production 62
IV. Minimizing Smoke Production 72
V. Conclusions 74
References 75

4 Water Relations of Forest Fuels

Ralph M. Nelson, ]r.
I. Introduction 79
II. Forest Fuels 81
III. Fuel Moisture Relationships 96
IV. Moisture Content Estimation 137
Notation 139
Additional Reading 142
References 143

5 Wildland Fire Spread Models

R. O. Weber
I. Introduction 151
II. Head Fire Rate of Spread (Physical Principles and Their
Mathematical Embodiment) 152
III. Head Fire Rate of Spread: Australia 155
IV. Head Fire Rate of Spread: United States 156
V. Head Fire Rate of Spread: Canada 158
VI. Smoldering 158
VII. Whole Fire Modeling—Fire Shape 160
Notation 164
References 165

6 Wind-Aided Fire Spread

F. E. Fendell and M. F. Wolff
I. Introduction 171
II. Laboratory-Scale Setup 176
Contents Vll

III. Fire Spread Model 180

IV. Preliminary Testing of the Model 204
V. Test Results for the Effect of Wind Speed and Fuel Loading
on the Rate of Fire Spread 214
VI. Conclusions 217
Notation 218
Recommended Reading 221
References 221

7 Fire P l u m e s
G. N. Mercer and R. O. Weber
1. Introduction 225
11. Modeling Fire Temperature Maxima 229
111. Plumes above Fires in a Cross Wind 243
Notation 252
References 253

8 C o u p l i n g A t m o s p h e r i c a n d Fire M o d e l s
Mary Ann Jenkins, Terry Clark, and Janice Coen
1. Introduction 258
11. Vorticity Dynamics in a Fire 260
III. Coupling between Atmosphere and Fire 266
IV. The Elements of Fire Modeling 269
V Modeling the Atmosphere 269
VI. The Coupled Fire—Atmosphere Modeling Approach 272
Vll. Idealized Studies of Wildfire Behavior 277
Vlll. Infrared Observations of Fires 286
IX. Conclusions and Future Work 288
Appendix I. Circulation and Vorticity 290
Appendix II. Development of Vertical Rotation in a
Frictionless Fluid 292
Appendix III. Generation of Vertical Motion in Rotating
Convective Cells 297
Notation 299
References 301

9 Surface Energy Budget a n d Fuel Moisture

Kenneth E. Kunkel
1. Introduction 304
11. Evapotranspiration Processes and the Meteorological
Controlling Factors 305
Vlll Contents

III. Estimation of Potential Evapotranspiration Rates 317

IV. Functional Dependence of PET and AET 327
V. Characteristics of PET 329
VI. Near-Surface Environment 333
VII. Models of Land-Surface Interactions 336
VIII. Remote Sensing of the Surface Energy Budget 339
IX. Fire Weather Rating Systems 341
Notation 343
Suggested Reading List 346
References 346

10 Climate, Weather, and Area Burned

M. D. Flannigan and B. M. Wotton
I. Introduction 351
II. Weather and Area Burned—Synoptic Surface Features 352
III. Weather and Area Burned—Upper Air Features 354
IV. Teleconnections 359
V. Future Warming and Area Burned 365
VI. Summary 369
References 369

11 Lightning and Forest Fires

Don Latham and Earle Williams
I. Introduction 376
II. Lightning 376
III. Previous Studies of Lightning-Initiated Fire 380
IV. Interaction between Lightning and Fuels 389
V. How Ignition Occurs 391
VI. Ignition Experiments with Real Forest Fuels 402
VII. Generating Models for Operational Use 406
VIII. Smoke, Lightning, and Cloud Microphysics 410
IX. Global Implications of Lightning Ignition Characteristics 411
X. Conclusion 413
References 414

12 Statistical Inference for Historical Fire Frequency

Using the Spatial Mosaic
W.J. Reed
I. Introduction 419
11. Graphical Analysis 420
Contents IX

III. Statistical Inference with Prespecified Change Points 422

IV. The Efficiency of Sample vs. Map Data 427
V. Determining Epochs of Constant Fire Frequency 430
References 435

13 Duff Consumption
K. Miyanishi
I. Introduction 437
II. Characteristics of Duff 439
III. Empirical Studies of Duff Consumption 441
IV. Flaming Combustion 443
V. Smoldering Combustion and Pyrolysis 448
VI. Models of Smoldering Combustion 456
VII. Contribution of Smoldering Combustion Models to
Understanding of Duff Consumption 464
Notation 468
References 470

14 Fire Effects on Trees

M. B. Dickinson and E. A. Johnson
I. Introduction 477
II. Effects of Fire on the Tree Bole 483
III. Effects of Fire on Canopy Components 505
IV. Root Necrosis 514
V. Tree Mortality 515
VI. Discussion 517
Notation 518
Additional Readings 520
References 521

15 Forest Fire Management

David L. Martell
I. Introduction 527
II. The Relationship between Fire and Forest Land Management
Objectives 532
III. Assessing Fire Impacts 534
IV. Forest Fire Management Organizations 544
V. Level of Fire Protection Planning 572
 Contents

VI. Some Challenges 576

Further Reading 581
References 581

Index 585
 CONTRIBUTORS

Numbers in parentheses indicate the pages on which the authors' contributions begin.

TERRY L. CLARK (257), National Center for Atmospheric Research, Mesoscale

and Microscale Meteorology, Boulder, Colorado 80307 ([email protected])
JANICE COEN (257), National Center for Atmospheric Research, Meso-
scale and Microscale Meteorology; Boulder, Colorado 80307 (janicec@ncar
.ucar.edu)

MATTHEW B. DICKINSON (477), Kananaskis Field Stations, University of

Calgary, Calgary, Alberta T2N 1N4, Canada ([email protected])

FRANCIS E. FENDELL (171), Space and Technology Division, TRW Space and
Electronics Group, Redondo Beach, California 90278 ([email protected])
MICHAEL D. FLANNIGAN (351), Canadian Forest Service, Edmonton, Al-
berta T6H 3S5, Canada ([email protected])
MARY ANN JENKINS (257), Department of Earth and Atmospheric Science,
York University, Toronto, Ontario M3J 1P3, Canada ([email protected])
EDWARD A.JOHNSON (1,477), Department of Biological Sciences and Kana-
naskis Field Stations, University of Calgary, Calgary, Alberta T2N 1N4, Canada
(j ohnsone@ucalgary. ca)

KENNETH E. KUNKEL (303), Midwestern Climate Center, Illinois State Water

Survey, Champaign, Illinois 61820 ([email protected])

DON LATHAM (375), Intermountain Fire Sciences Laboratory, USDA Forest

Service, Rocky Mountain Research Station, Missoula, Montana 59802 (dlatham
@fs.fed.us)
Xii Contributors

DAVID L. MARTELL (527), Faculty of Forestry, University of Toronto, To-

ronto, Ontario M5S 3B3, Canada ([email protected])
GEOFFREY N. MERCER (225), School of Mathematics and Statistics, Univer-
sity College, University of New South Wales, Australian Defence Force Acad-
emy, Canberra 2600, ACT, Australia ([email protected])
KIYOKO MIYANISHI (1,437), Department of Geography, University of Guelph,
Guelph, Ontario NIG 2W1, Canada ([email protected])
RALPH M. NELSON, JR. (79), USDA Forest Service, NFSNC Supervisor's Of-
fice, Asheville, North Carolina 28801 ([email protected])
WILLIAM J. REED (419), Department of Mathematics and Statistics, University
of Victoria, Victoria, British Columbia V8W 3P4, Canada ([email protected])
KOZO SAITO (11), Department of Mechanical Engineering, University of Ken-
tucky, Lexington, Kentucky 40506 ([email protected])
DAROLD E. WARD (55), Fire Chemistry Research Work Unit, USDA Forest
Service, Intermountain Research Station, Missoula, Montana 59807 (pyroward
@aol.com)
RODNEY O. WEBER (151, 225), Department of Mathematics, University Col-
lege, University of New South Wales, Australian Defence Force Academy, Can-
berra 2600, ACT, Australia ([email protected])
EARLE R. WILLIAMS (375), Department of Earth, Atmospheric and Planetary
Sciences, Massachusetts Institute of Technology, Cambridge, Massachusetts
02139 ([email protected])
MICHAEL F. WOLFF (171), Space and Technology Division, TRW Space and
Electronics Group, Redondo Beach, California 90278 ([email protected])
B. MICHAEL WOTTON (351), Canadian Forest Service, Great Lakes Forestry
Centre, Sault Ste Marie, Ontario P6A 5M7, Canada ([email protected])
 PREFACE

Traditional approaches to the study of fire behavior and fire ecology (by forest-
ers and ecologists) have been strongly rooted in a descriptive tradition in which
fire behavior has been described in a very informal way, using loosely defined
terminology such as "hot/cool fires" or "flammability" with no dimensions or
units of measurement given. With this approach, the coupling of fire behavior
and fire effects on individuals, populations, or communities could be, at best,
qualitative or categorical. Furthermore, even when more quantitative measures
of fire behavior have been attempted, many studies have used less than effective
measures for attempting to understand the relationship between fire behavior
and fire effects (e.g., the use of color-changing temperature-recording paints to
correlate temperature with tree mortality). To effectively relate fire behavior to
ecological effects, it is necessary to show that such instruments or measures
are meaningfully related to the way that organisms are affected by fire. In other
words, such studies should incorporate some model of heat transfer to the or-
ganisms or particular living tissues (e.g., to serotinous cones or to the tree cam-
bium through bark).
In wildfires, variation in the response of the organism can result from varia-
tion in the appropriate measurements of fire behavior variables and variation in
the organism's physical characteristics which affect heat transfer processes from
the fire to living tissues. Many ecological studies cannot separate these compo-
nents of variation and thus cannot make useful statements about the causes. On
a different scale, studies of the spatial pattern of tree mortality within large
burns (i.e., unburned "islands") have often looked for explanations of these
patterns in differences in usual environmental factors such as topography, hy-
drology, and vegetation composition. While such differences may account for
some unburned patches, in many cases these unburned islands do not appear
XIV Preface

to be related to any of these factors. Again, an understanding of the physical

behavior of fire (e.g., coupling of fire and three-dimensional convection above
the fire) might provide a useful approach to understanding some of these un-
burned islands.
Since the 1950s, the body of literature on fire behavior has been growing in
journals of engineering, geophysics, meteorology, etc. Foresters and ecologists
have not used much of this literature on the physical aspects of fire behavior to
understand fire effects on individual organisms, populations, or communities.
Part of the reason may be that their interests lie in the ecological responses to
fire rather than in the physical processes of fire itself. Nevertheless, as indicated
above, an understanding of the ecological effects of fire will necessarily involve
some understanding of the processes involving the fire in order to ask appropri-
ate research questions and to measure meaningful fire behavior variables. The
purpose of this book is to provide an introduction to this literature and to con-
vince those working in fire ecology of the importance of understanding these
physical fire processes. The book is not meant to be a definitive coverage of the
topic of fire behavior and ecological effects; some important topics are not cov-
ered because of space limitations. Possible future editions of the book may in-
clude other aspects of fire behavior and ecological effects.
This book attempts to bridge the gap between elementary texts on fire for for-
esters, ecologists, and other environmental specialists (e.g.. Brown and Davis,
1973; Luke and McArthur, 1978; Pyne, 1984; Trabaud, 1989) and the technical
combustion and heat transfer literature (e.g., Journal of Fluid Mechanics, Sym-
posium on Combustion, Combustion and Flame, Combustion Science and Technol-
ogy). We assume that the reader has an elementary knowledge of physics, chem-
istry, and biology and an introduction to calculus. Derivations of equations are
shown when necessary for an understanding of the processes and relationships
between variables. All fields have conventions (i.e., terminology, symbols, ways
of presenting relationships) that are often difficult for outsiders to penetrate.
Also all fields tend to take for granted that the reader knows why certain things
are being done. For individuals outside the field, this is rarely the case. The con-
tributors come from a wide range of different fields and they have tried to ex-
plain the conventions and approaches used as they arise in each chapter.
Although the contributors were instructed in writing their chapters to keep
in mind the primarily non-physical-science background of the intended audi-
ence, we also wanted to provide enough detail and depth to allow the reader to
grow into the material. Thus, the book may be read on different levels. A first
reading of a chapter may provide some readers with a general intuitive grasp of
the topic and its relevance without necessarily a complete understanding of the
details and equations. We hope this first reading will entice the reader to delve
further into the topic and to use some of the suggested readings to develop an
understanding at a deeper level.
Preface XV

Development of this book was aided immeasurably by a workshop attended

by all of the contributors in which first drafts of each chapter were discussed in
terms of the general objective of the book, the intended audience for the book,
and overall coordination of the connections between the chapters. The work-
shop was held at and funded by the National Center for Ecological Analysis and
Synthesis in Santa Barbara. We also gratefully acknowledge Louis Trabaud and
Paul Zedler, who, as representatives of the intended audience, provided invalu-
able feedback to all of the contributing authors.

REFERENCES

Brown, A. A., and Davis, K. P. (1973). "Forest Fire: Control and Use," 2nd ed. McGraw-Hill, New
York.
Luke, R. H., and McArthur, A. G. (1978). "Bushfires in Australia." Australian Government Pub-
lishing Service, Canberra.
Pyne, S. J. (1984). "Introduction to Wildland Fire: Fire Management in the United States." W^iley,
New York.
Trabaud, L. (1989). "Les Feux de Forets: Mecanismes, Comportement et Environment." France-
Selection, Aubervilhers Cedex, France.

E. A. Johnson
K. Miyanishi
This Page Intentionally Left Blank
 ACKNOWLEDGMENTS

We thank the following people for reading and providing helpful comments on
individual chapters: Frank Albini, Kerry Anderson, George F. Carrier, Terry
L. Clark, Matthew B. Dickinson, Francis Fujioka, Robert Haight, Donald A.
Haines, Stan Heckman, Rodman R. Linn, Tilden Meyers, Ralph M. Nelson, Tom
Ohlemiller, Luis A. Oliveira, James Quintiere, Jon Regelbrugge, Jeffery S. Reid,
Walter Skinner, Louis Trabaud, Charles E. Van Wagner, Neil R. Viney, Richard
H. Waring, David R. Weise, Forman A. Williams, Zong-Liang Yang, and Paul H.
Zedler. We gratefully acknowledge the National Center for Ecological Analysis
and Synthesis in Santa Barbara, California, for providing funding and support
for a workshop that greatly facilitated collaboration and interaction among the
contributing authors. The Center is funded by NSF (Grant no. DEB-94-21535),
the University of California-Santa Barbara, the California Resources Agency,
and the California Environmental Protection Agency. We also acknowledge the
contributions of Louis Trabaud and Paul H. Zedler in reading and commenting
on all of the chapters as well as participating in the NCEAS workshop.
Marie Puddister in the Department of Geography, University of Guelph, did
an admirable job of converting all of our graphics (rough hand drawings, photo-
copied diagrams, spreadsheet charts, and graphics files in numerous formats)
into uniformly formatted figures. We greatly appreciate her skill and patience
in working with authors scattered across North America and Australia.
Finally, we thank our editors, Charles Crumly, Donna James, Danielle Cum-
mins, and Joanna Dinsmore, for their help in producing this book.
This Page Intentionally Left Blank
 CHAPTER

Strengthening Fire
Ecology's Roots
EDWARD A. J O H N S O N
Department of Biological Sciences, University of Calgary, Calgary, Alberta, Canada

KlYOKO MiYANISHI
Department of Geography, University of Guelph, Guelph, Ontario, Canada

I. Introduction
II. Processes
III. Transfer Rates and Budgets
IV. Examples of Traditional vs. Proposed Approach
References

I. INTRODUCTION
Research on the connection between wildfires and ecological systems goes back
to the early discovery that natural disturbances were a recurrent phenomenon
in ecosystems and, as such, required an understanding of their effects on eco-
system structure and function. However, connecting wildfires to ecological sys-
tems has proceeded slowly. This is probably because forestry and ecology, the
two fields primarily interested in wildfire effects on ecosystems, have been side-
tracked by their traditional approach to studying ecological systems. Foresters
are mostly interested in extinguishing or eliminating wildfires or in managing
burns to produce certain effects in the forest (e.g., reduced competition between
certain trees or creation of wildlife habitat). Ecologists have been interested in
how fires change the composition and structure of ecological systems. The ap-
proach that has been taken to investigate these issues has, in general, involved

Forest Fires
Copyright © 2001 by Academic Press. All rights of reproduction in any form reserved. 1
Z Johnson and Miyanishi

describing patterns of fire effects and correlating these to environmental factors.

This approach does not immediately lead to a concern with the mechanism of
interaction between fire processes and ecosystem processes. One could read the
ecological literature for a long time before finding any discussions that indicate
an appreciation that the processes of combustion and heat transfer lie at the heart
offireecology.
Fire ecology is a part of both environmental biophysics and ecology. Conse-
quently, it is concerned with combustion, transfer of heat, mass and momentum
in wildfires, heat transfer between the fire and the organism, and finally how
these physical processes affect ecological processes. In the following chapters,
we will not be able to deal with these concerns in any complete manner. Con-
sequently, we have chosen to accent certain topics which we feel ecologists and
foresters have not traditionally addressed and about which we believe they
should have more information. This means that most of the book is devoted to
presenting information on fire processes. The information on ecological effects
of fire is more limited than what most ecologists might hope to see. However,
this reflects the limitation of our understanding of combustion and heat, mass,
and momentum transfer. Particularly lacking are well accepted process-based
models for the connection of fire to ecological processes. This has been an on-
going problem noted by several authors over the last decades (Trabaud, 1989;
Johnson, 1992; Whelan, 1995; Bond and van Wilgen, 1996).
Readers may be asking at this point: As an ecologist/forester, do I really have
to understand all this about combustion and heat transfer? In fact, the pur-
pose of this book is to convince readers of the necessity for understanding the
relevant processes involved in the ecological effects of fire. In the absence of a
process-based approach, the discipline of fire ecology is left with a collection
of case studies, that is, site-specific, species-specific, and also often method-
specific correlations between some ecological response and traditionally used
fire variables. Thus, there is little justification for the selection of variables
and relationships and no way to distinguish between relationships that are
physically and ecologically based and those that are statistically significant but
spurious.
Central to the approach advocated in this book is that we must invest in a
basic understanding of wildfire materials and processes and their effects on eco-
logical (biological) materials and processes. Materials have properties and are
assembled into structures. Thus, in terms of heat transfer from a fire to the liv-
ing tissues of a tree and its effect on tree mortality, an important material prop-
erty of bark is its thermal diffusivity. The structure could be considered the
changing thickness of bark on different parts of the tree stem. Processes are
(natural) phenomena marked by a series of operations, actions, or mechanisms
which explain a particular result—in other words, how something works. Some
of the wildfire processes are ignition, extinction, flaming and glowing combus-
Chapter 1 Strengthening Fire Ecology's Roots 3

tion, rate of spread, and fuel drying. Some of the ecological processes affected
by fire processes are population dynamics (i.e., birth, death, immigration, and
emigration processes), nutrient cycling, and productivity.

II. PROCESSES
Briefly, this book is concerned with the following topics, most of which involve
processes. Combustion is considered primarily as the means of generating heat
and combustion products such as smoke (Chapter 3). Flames (Chapter 2) are
one of the principal forms of combustion, smoldering being the other. Flames
are also the main source of heat by which most wildfires spread. Fire spread
(Chapters 5 and 6) brings together the different processes in the fire and its en-
vironment to explain why it moves. The plume (Chapter 7) above the flame is,
as often as the flame, the cause of death of organisms. Since fuel moisture (Chap-
ters 4) plays such a central role in coupling fire behavior and other environ-
mental variables, it is given special treatment. Although the role of weather and
climate has always been recognized as important in wildfires, recent decades
have seen some significant advances in our understanding of all scales of inter-
action of wildfires and the weather. The coupling of a wildfire and the atmospheric
convection (Chapter 8) above it has only begun but may help us understand bet-
ter not only fire spread but also some phenomena such as unburned remnants
left by fires. Mesoscale meteorology (Chapter 9) has shown how fuel moisture is
affected by the energy exchange with the regional environment above the eco-
systems. Synoptic scale meteorology and climatology (Chapter 10) have the long-
est history in fire ecology, but understanding in recent decades has helped us
understand why wildfires occur only under limited kinds of synoptic condi-
tions and how these conditions persist because of midtropospheric circulation
patterns in the atmosphere. Lightning (Chapter 11) is the major natural cause
of wildfires; however, until the advent of lightning locating systems, there was
little understanding of the relationship between lightning and wildfire ignition.
Smoldering combustion (Chapter 13) is a form of combustion that occurs in ma-
terials with particular properties (e.g., the partially to well decomposed organic
matter (duff) that accumulates on top of the soil in some ecosystems). The pro-
cess of smoldering is particularly important in ecosystems in which duff removal
is necessary for plant regeneration. The heat transfer to plants (Chapter 14) uses
many of the processes described earlier as input for the transfer into the plant
to cause death of the whole plant or necrosis of parts. Many forest landscapes
are mosaics of different ages resulting from the overburning of past fires. Deter-
mining the frequency of fires from this kind of spatial data has proved to be a so-
phisticated statistical problem (Chapter 12). Fire management has undergone
changes in past decades from being primarily prevention and suppression to
4 Johnson and Miyanishi

more ecological and economic concerns of understanding and mimicking the

fire regime (Chapter 15).

III. TRANSFER RATES AND BUDGETS

The method for studying the preceding processes is generally by use of transfer
rates and budgets of heat, mass, and momentum. Heat transfer results from the
difference between objects in their average kinetic energies. The mechanisms of
heat transfer are radiation, conduction, convection, and latent heat. Radiation
is dependent on the temperature of the object. An object's temperature is con-
sidered to be proportional to its average kinetic energy. All objects above ab-
solute zero radiate to their surroundings. When fireplaces were an important
source of heat in homes, chairs had screens or wings to block the radiation from
the fireplace. Thus, radiative transfer depends on the view of the radiating ob-
ject. Conduction is heat transferred by the motion of adjacent molecules. Solid
objects feel warm or cold to us because of conductive transfer between them
and our skin. Convection is the mass transfer of heat in a fluid such as water or
air as illustrated by hot-air heating systems in homes. Latent heat is the heat ab-
sorbed or released when water changes state; for example, evaporation of water
absorbs 2450 J g~\ In their simplest form, these mechanisms consist of the dif-
ference in temperature between objects times some coefficient which measures
the resistance to transfer.
Mass transfer is the movement of materials and involves diffusion in which
different classes of particles have different velocities. Diffusion has two modes:
molecular and turbulent. Molecular diffusion is caused by the random motion
of molecules, while turbulent diffusion is disorderly movement, often in eddies.
Like heat, mass transfer is along a gradient but in this case a gradient of concen-
tration or density rather than temperature. As with heat transfer, diffusion is
modified by a resistance term.
Momentum transfer occurs because of resistance to flow in gases and liquids
generated by collisions among rapidly moving molecules. Rates of convective
heat transfer and of mass transfer are proportional to the viscosity or resistance
of the fluid in which they occur. Perhaps the most important issue in momen-
tum transfer discussed in the coming chapters will be boundary layers—the
relatively still layer of fluid near solid surfaces in which heat and mass transfer
are carried out by molecular diffusion or by boundary layer streaks.
The transfer rates are assembled into budgets which determine the input-
output or storage of heat, mass, or momentum based on the law of conservation
of energy, mass, and momentum. These budget equations form simple ways of
understanding the dynamics of the processes of interest. Ecologists should be
familiar with this approach since population dynamics (birth, death, and mi-
gration) are budgets which tell us how the numbers in populations come about.
Chapter 1 Strengthening Fire Ecology's Roots 5

We believe that the reason fire ecologists have not been recruited to the ap-
proaches given in the following chapters is the lack of tools to address process
questions. This explains why seemingly obvious (to physical scientists) ap-
proaches are not taken. Consequently, the approaches used in the following
chapters should be of as much interest as the contents.
Some tools are familiar to fire ecologists (e.g., careful measurements, lab and
field experiments), but others, such as dimensional analysis (but see Gurney
and Nesbit, 1998) and reasoning that seeks to interpret phenomena in light of
physical principles, are not. Dimensional analysis helps define what variables
should be incorporated and which ones are unnecessary. It also might provide
a method for assembling these variables into functional relationships. Experi-
ments (in the widest sense) supply the numerical constants and check the cor-
rectness against independent data. Biologists and foresters develop algebraic
equations from empirical data by statistical curve fitting, usually with little
physiological or biological justification. These functional relationships are
simply mathematical descriptions of the data. They often make no dimensional
sense. For example, what are the units of fire severity? Experiments are thus
used to test classifications of a phenomenon, not the processes involved.

IV. EXAMPLES OF TRADITIONAL

VS. PROPOSED APPROACH
Finally, perhaps a comparison between the traditional approach to a problem
in fire ecology and the approach advocated in this book would be useful. A tree
can be killed by a surface wildfire that has heated its base long enough to kill
the cambium beneath the bark. Clearly this process of killing the cambium is
affected by the temperature of the surface of the tree, how long this tempera-
ture is maintained, and how well the heat is transferred through the bark to the
cambium. The question usually asked by ecologists in this situation is: What is
the tolerance of different species to surface fires?
The traditional approach by ecologists in studying this fire tolerance has been
to heat the base of the tree and record differences in temperature at the cam-
bium (Uhl and Kauffman, 1990; Hengst and Dawson, 1994; Pinard and Huff-
man, 1997). The tree is heated with a torch, radiant heater, or a burning fuel-
soaked rope which is tied around the tree. An unshielded thermocouple is
exposed at the surface of the bark, and another is placed in the cambial layer
beneath the bark. The temperature course of both thermocouples is recorded
during heating.
The time-temperature curve of the outer bark surface (Figure 1) is compared
between species, with differences in maximum temperature used to "demon-
strate that external bark characteristics can affect heat absorption" (Uhl and
Kauffman, 1990). It is not clear how heat absorption could be measured by a
 Johnson and Miyanishi

500

O 400 H

0)
Q.
300 H
E

200

100

0 100 200 300 400 500

Time (s)
FIGURE 1 Temperature of bark surface during wick (experimental, tree surface) fires of Tetra-
gastris altissima, Jacaranda copaia, and Manilkara huheri at Vitoria Ranch near Paragominas, Para,
Brazil. From Uhl and Kauffman (1990), with permission.

thermocouple at the surface of the bark. Any differences in temperature at the

bark surface are probably a result of the experimental setup related to variation
in the wick fire and perhaps to flames from the burning bark if this is occurring.
The time-temperature curve of the thermocouple at the cambium surface (in-
side the bark) is usually related to bark thickness (Figure 2). Time-temperature
curves increase more slowly in general with thicker bark species. Notice in Fig-
ure 2 that Uhl and Kauffman (1990) use the term temperature flux when the
graph gives only temperature. Finally a relationship (Figure 3) is usually given
between the maximum cambium temperature and bark thickness (for all spe-
cies) . The equation is then used to calculate the bark thickness at which a tree's
cambium will exceed 60°C and cambium death is presumed to happen. Thus,
bark thickness is used to infer fire tolerance.
A number of important issues are not considered in such studies. For ex-
ample, heat transfer is never considered; only temperature is considered. There
is no consideration of which heat transfer processes are operating. The transient
nature of the heating is never incorporated into the results. Finally, the effect of
the material properties of the bark on heat transfer is not considered.
This approach is indeed unfortunate since a more coherent and rigorous ap-
proach is almost always given in the references cited by these studies. However,
Chapter 1 Strengthening Fire Ecology's Roots

Ecclinusa sp. (2.4 mm)

Jacaranda copaia (4.8 mm)
Pourouma guianesis (4.8 mm)
96 Inga alba (5.8 mm)
Metrodorea flavida (6.3 mm)
Lecythis idatimon (9.4 mm)
Cecropia sciadophylla (7.3 mm)
84 Lecythis lunda (11.4 mm)
O Manilkara huberi (11.3 mm)

g 72-
0)
Q.
E 60
(D
£

I 48
O

36-

24-

80 160 240 320 400

Time (s)
FIGURE 2 Mean temperature flux [sic] at the cambium surface during wick (experimental, tree
surface) fires for two individuals (between 20 and 30 cm dbh) in each of nine taxa: Edinusa sp.,
Jacaranda copaia, Metrodoreaflavida,Pourouma guianensis, Inga alha, Cecropia sciadophylla, Lecythis
idatimon, Lecythis lurida, and Manilkara huberi at Vitoria Ranch near Paragominas, Para, Brazil.
From Uhl and Kauffman (1990), with permission. Each species is labeled with the mean bark thick-
ness (mm).

the significance of the heat transfer model (transient heat flow in a semi-infinite
solid) given in Spalt and Reifsnyder (1962) does not seem to be recognized. The
model (described in detail in Chapter 14) is as follows. Assume that the flame
is heating primarily by conduction and that the surface (boundary layer) resis-
tance is minimal, considering the proximity of the flame. The fire suddenly in-
creases the surface temperature. The tree is further assumed to be large enough
in diameter so that heating from the opposite side does not affect the cambium
(i.e., the main conductive heat transfer is occurring perpendicular to the sur-
face of the bark). The transient heat flow can then be described by

- erfl (1)
r,
8 Johnson and Miyanishi

108

96

84
(D
Q.
72

60
E
cc
o
48
CO
(D
CL

36

8 12 16 20 24
Bark thickness (mm)

FIGURE 3 The relationship between bark thickness and peak temperature of the cambium dur-
ing wick fires of 30 individuals distributed among 15 species at Vitoria Ranch near Paragominas,
Para, Brazil. From Uhl and Kauffman (1990), with permission.

The left side of Eq. (1) gives the rise in temperature; the numerator gives the
difference between the temperature at which the cambium is killed (T) and the
flame temperature (Tf), and the denominator gives the difference between
the initial (before heating) ambient temperature (T^) and the flame temperature
(Tf). The temperature rise is thus given relative to the initial difference between
the flame and tree temperatures. Notice that the gradient of heat is defined by
this term. On the right side of the equation, erf is the error function, a value
which can be easily looked up in tables for the values in parentheses. The heat
transfer is directly proportional to the bark thickness (x) (i.e., the depth for
which the temperature gradient is being determined) and inversely proportional
to the square root of the thermal diffusivity (a) (i.e., how the bark material af-
fects the flow of heat), and the time it takes for the lethal temperature (T) to be
reached (r). Thus, if we solve for r, we should be able to see how either trees
of different a but same x or different x but same a have different fire tolerances.
Thermal diffusivity (a) contains the relevant bark characteristics such as bark
density, moisture, and conductivity that influence fire tolerance.
The model we have given is the simplest, and more complicated ones can be
formulated (e.g., Costa et al, 1991). However, even this simple model illustrates
an approach to the study of fire tolerance that is based on the process (heat con-
duction) by which the cambium is heated to lethal temperature. This model
also provides a rationale for the choice of variables and shows how the variables
interact.
chapter 1 Strengthening Fire Ecology's Roots ^

REFERENCES
Bond, W. J., and van Wilgen, B. W. (1996). "Fire and Plants." Chapman and Hall, London.
Costa, J. J., Oliveira, L. A., Viegas, D. X., and Neto, L. P. (1991). On the temperature distribution
inside a tree under fire conditions. Int. J. Wildl. Fire 1, 87-96.
Gurney, W. S. C , and Nisbet, R. M. (1998). "Ecological Dynamics." Oxford University Press,
New York.
Hengst, G. E., and Dawson, J. O. (1994). Bark properties and fire resistance of selected tree species
from the central hardwood region of North America. Can. J. For. Res. 24, 688-696.
Johnson, E. A. (1992). "Fire and Vegetation Dynamics: Studies from the North American Boreal
Forest." Cambridge University Press, Cambridge.
Pinard, M. A., and Huffman, J. (1997). Fire resistance and bark properties of trees in a seasonally
dry forest in eastern Bolivia. J. Trop. Fcol. 13, 727-740.
Spalt, K. W., and Reifsnyder, W. E. (1962). "Bark Characteristics and Fire Resistance: A Literature
Survey." Occasional Paper No. 103. USDA Forest Service, Southern Forest Experiment Station.
Trabaud, L. (1989). "Les Feux de Forets: Mecanismes, Comportement et Environment." France-
Selection, Aubervilliers Cedex, France.
Uhl, C , and Kauffman, J. B. (1990). Deforestation, fire susceptibility, and potential tree responses
to fire in the eastern Amazon. Ecology 71, 437-449.
Whelan, R.J. (1995). "The Ecology of Fire." Cambridge University Press, Cambridge.
This Page Intentionally Left Blank
 CHAPTER

Flames
K. SAITO
Department of Mechanical Engineering, University of Kentucky, Lexington, Kentucky

I. Introduction
II. Basic Aspects of Combustion in Forest Fires
A. Governing Equations
B. Adiabatic Flame Temperature and Soot Formation
III. Temperature, Velocity, Species Concentration, and Flame Height
A. Temperature Measurement
B. Velocity Measurement
C. Species Concentration Measurement
D. Flame Height Measurement
IV. Premixed and Diffusion Flames
A. The Premixed Flame and Its Burning Velocity
B. Ignition
C. Ignition Temperature
D. Flammability Limits
E. Minimum Ignition Energy
V. Extinction of Diffusion Flames
VI. Diffusion Flames and Scaling Analysis
A. Re Number-Controlled Laminar Diffusion Flames with Low Pe Number
B. Re Number-Controlled Turbulent Jet Diffusion Flames with High Pe Number
C. Re and Fr Number-Controlled Laminar Diffusion Flames with Low
Pe Number
D. Fr Number-Controlled Turbulent Diffusion Flames with High Pe Number
E. Scaling Laws
F. Wood Crib Fires
VII. Spreading Flames
A. Downward and Horizontal Flame Spread
B. Turbulent Upward Flame Spread
C. Flame Spread through Porous Fuel Beds
VIII. Structure of Flame Base
IX. Conclusions
Notation
References
Forest Fires
Copyright © 2001 by Academic Press. All rights of reproduction in any form reserved. II
12 K. Saito

I. INTRODUCTION
This chapter intends to help ecologists, forest fire researchers, and fire-fighting
strategists understand some of the fundamental aspects of combustion and
think scientifically about their forest fire problems. As with any discipline, the
best approach to understanding forest fires is to grasp the fundamentals—in
this case, the structure and behavior of flames. Explanation and discussion
are based on the physical and chemical aspects of combustion, with little em-
phasis on strict mathematical treatment of equations. The chapter addresses ba-
sic knowledge on the structure of diffusion flames and scaling laws, premixed
flames, ignition, diffusion flame extinguishment, spreading flames, and the
mechanism of diffusion flame anchoring. Some general background in com-
bustion research is introduced at the beginning.
Fire is the common name we give to high-temperature gaseous combus-
tion. This combustion can occur in open land or an enclosed environment. In
either case, regardless of the type of fire, two things are certain: heat is released,
and the fire may spread. Flame is the fundamental element of fire that produces
the heat and the combustion by-products (such as CO2, H2O, CO, and smoke)
by means of chemical reactions that occur between fuel and oxygen—called
combustion.
Modern combustion research is composed of four theoretical supporting
structures: thermodynamics, chemical kinetics, fluid mechanics, and transport
processes (Williams, 1992). Additionally, three basic tools (experiment, theory,
and computation) are available to help researchers approach combustion prob-
lems. Each tool has strengths and weaknesses.
Experiments have two roles: "insightful observation" and verification of as-
sumptions that theory and computational methods employ (Hirano and Saito,
1994). The term "insightful observation" means to see things with unbiased
minds using imagination and intuition. The second role of an experiment is
well explained elsewhere in the scientific literature, while the first role is rarely
emphasized, although it is important. In scientific research, the first and second
roles often play together. For example, a simple experiment designed to ver-
ify assumptions also offers additional information to researchers. W h e n a re-
searcher conducts an experiment, two different points of view need to be kept
in her or his mind: confirmation of assumptions and insightful observation.
When the researcher's mind is entirely occupied by the confirmation role only,
she or he may miss the opportunity for discoveries and inventions through
insightful observation. Our history proves that insightful observation is the
source of discoveries and inventions (Ferguson, 1993).
Computation, also referred to as computer experiments, can provide de-
tailed results and even a virtual reality by simulation. The combination of two
Chapter 2 Flames 13

or more of these tools will increase not only the accuracy of the results but also
the chance of getting correct answers, especially during the early stages of a re-
search and development project (Wilson, 1954). Computational methods not
only save time and energy of human calculation but also provide details of vir-
tual reality under ideal initial and boundary conditions, which may be difficult
to achieve by experiments.

II. BASIC ASPECTS OF COMBUSTION

IN FOREST FIRES
"Combustion is exothermic (heat releasing) chemical reactions in flow with
heat and mass transfer" (Williams, 1985; Linan and Williams, 1993). To be more
specific, we can divide combustion into two types, based on the kind of chemi-
cal reaction and the flow regime (laminar and turbulent): high speed and low
speed. Good examples of high-speed combustion are gas turbines, rocket mo-
tors, and internal combustion engines. Examples of low-speed combustion in-
clude forest fires and candle flames.
We can be more specific still by considering the type of flame in low-speed
combustion. Forest fires, for example, burn with what are called diffusion
flames. Diffusion flames can be maintained by a feedback loop; in the gas phase,
heat is released by exothermic combustion of the secondary gaseous fuels, while
in a condensed phase some of this heat is transferred back causing gasification
of the primary fuel. Gasification is an endothermic (heat-absorbing) process
that requires gaseous combustion in order to occur. This feedback of heat from
the gas phase flames to the condensed phase fuels is an essential mechanism for
sustaining the combustion process.

A. GOVERNING EQUATIONS

Governing equations or conservation equations can provide mathematically

strict relationships among various quantities. In combustion science, conserva-
tion equations are partial differential equations expressing conservation of mass,
momentum, energy, and chemical species. Derivation of conservation equa-
tions are detailed by Williams (1985) and Linan and Williams (1993) and are
beyond the scope of this chapter; only the concept of these equations together
with mathematical formulae for only mass and momentum conservation (be-
cause of their simple formulae) are explained.
By taking a volume element of flame fluid and assuming fluid velocity to be
14 K. Saito

much smaller than the speed of light, we can conceptualize the conservation
equations for mass, momentum, energy, and chemical species as follows.
Mass Conservation:
Rate of accumulation Rate at which mass flows
+ into the volume element = 0.
of mass
The mathematical expression is tu^/Q^t + V{QU) = 0, where m = ratio of flow
time to evolution time, Q = density over characteristic density, and u = veloc-
ity vector over characteristic velocity.
Momentum Conservation:
Rate of increase Inertia Pressure Body" Viscous
+ + +
of momentum _ _ force force _ force. force
The mathematical expressions is vjd{ Qu)/dt = — V( QUU) — Vp/M^ + F/Fr +
VTs/Re, where p = pressure over characteristic pressure, M = Mach number,
F = body force, Fr = Froude number. Re = Reynolds number, and Ts = shear
stress tensor.
Energy Conservation:
Thermal Chemical Kinetic
energy + energy J
energy J
Energy lost by conduction Work done on
+ -f = Constant
and radiation surroundings
Chemical Species Conservation:
Convection of the Diffusion of
Accumulation
chemical species out the species into
rate of given
of the volume element the volume
L chemical species
by fluid motion element
Production of the
chemical species
+ by chemical
reactions
The conservation (mass and energy) equations are often applied to the con-
densed phase.
When these conservation equations are solved under specified boundary
and initial conditions, they can provide predictions on changes of pressure, ve-
locity, temperature, and chemical species as a function of time and space. Linan
and Williams (1993) point out that no combustion problems require the full
Chapter 2 Flames 15

description of all the terms in the conservation equations and recommend the
simplification of combustion phenomena. For combustion of solids, which in-
cludes most forest fires, for example, the phenomenological derivation is more
satisfying than the first principle approach of using these complete conserva-
tion equations. Here, phenomenological derivation means to come up with em-
pirical equations using experimental data and physical principles, but not start-
ing from the basic governing equations.

B. ADIABATIC FLAME TEMPERATURE

AND SOOT FORMATION

The most desirable condition for combustion would be to reach the adiabatic
flame temperature, the ideal maximum temperature that a combustion system
can attain. In such a case, we would have no heat loss from the system. How-
ever, real systems don't reach this point because the actual combustion process
consists of finite-rate chemical reactions with heat loss that occurs in the forms
of convection, radiation, and conduction. All these losses contribute to lead-
ing the combustion system to incomplete combustion. Incomplete combustion
produces products of incomplete combustion (PICCs). Common examples of
PICCs include CO, NO^, SO^, dioxin, all kinds of intermediate hydrocarbons
( Q H ^ , where n and m = 1, 2, 3, . . .) and soot (solid particulates). The follow-
ing diagram describes the generic process of combustion:
[Fuel] + [Oxidizer] - > [FCC] + [PICC] + Q (1)
where PCC are the products of complete combustion. When complete combus-
tion occurs in an adiabatic system, no PICCs are formed, the heat of combus-
tion Q reaches a maximum, and the system achieves adiabatic flame tempera-
ture. For example, adiabatic flame temperature for a methane + air reaction
is 1875°C and that for most hydrocarbon + air systems falls between 1800°C
and 2200°C. The detailed procedure to calculate adiabatic flame temperature
is straightforward (e.g., Classman, 1996) using the JANAF Thermochemical
Tables (1985).
Flame temperature for an incomplete combustion system can also be calcu-
lated by assuming incomplete combustion products. Computer programs (e.g.,
NASA's Gordon and McBride and STANJAN, see Classman, 1996) are available
to calculate the flame temperature and equilibrium compositions. Equilibrium
compositions, defined as species concentrations at a specified temperature, can
be calculated using the equilibrium constant that is available in the JANAF
Table. For example, an arbitrary second-order reaction can be written as

A.-hBj,^Cp + Dp (2)
16 K. Saito

where AR and BR are reactants and Cp and Dp are products. A simple example is

CH4 + 2O2 -> CO2 + 2H2O (3)

The reaction rate describing how quickly a given reactant, A^, for example,
disappears and converts to the product can be written using the Arrhenius
approximation:

-d[A^]/dt = -k[Ap][Bp] - d[Cp]/dt = -Aexp{-E/RoT)[Ap][Bp] (4)

where [ ] means concentration, t is time, h is specific reaction rate constant,

A is preexponential factor, E is activation energy, RQ is universal gas constant,
and T is temperature. The values of k, E, and A have been studied for many dif-
ferent types of hydrocarbon-air combustion systems and are available in the lit-
erature (e.g., Glassman, 1996). For further information regarding chemical ki-
netics, see Benson (1960) and Glassman (1996).
For combustion efficiency, the ratio of the actual heat release to the maxi-
m u m ideal heat of combustion ranges from 50 to 95% (Pyne et al, 1996). The
combustion efficiency can be defined more specifically as the ratio of the actual
carbon contained in the emissions of carbon dioxide compared to that theoreti-
cally possible, assuming that all the carbon was released as carbon dioxide. Un-
der such combustion efficiency, many PICCs will be formed. As a result, for
such a system, the temperature drops a few hundred degrees below the adiabatic
flame temperature. However, there is a critical extinction temperature beyond
which flame won't exist. A universal value of extinction, approximately 1300°C
(Rashbash, 1962), was found for cellulosic fuels.
Soot is a typical PICC and has a large impact on the environment when it is
formed in the case of forest fires by degrading ambient air quality, impairing
visibility, worsening regional haze, and causing significant health problems
(Pyne et ah, 1996, and Chapter 3 in this book). Therefore, a brief explanation
of soot formation and properties will be given here. When soot is formed in a
high-temperature flame zone, it emits a luminous yellow (black body) radiation;
when it escapes from the flame, it can be seen as a cluster of small black par-
ticles. Smoke is a rather large cluster of soot particles (100 junn or larger) and is
usually large enough for human eyes to recognize, while no clear-cut distinction
is made between smoke and soot.
Scanning electron microscopy (SEM), transmission electron microscopy
(TEM), electron spectroscopy for chemical analysis (ESCA), and laser desorp-
tion mass spectrometry (LDMS) have been used to investigate the physical and
chemical structures of soot.
Soot can be collected directly from the hot flame zone by different direct
sampling techniques (Fristrom and Westenberg, 1985): (1) a filter paper to col-
lect soot through a vaccum line, (2) a small quartz needle and a stainless-steel
Chapter 2 Flames 17

mesh-screen to be inserted into a flame to deposit soot on them (Saito et ah,

1991), and (3) a thermophoretic sampUng probe to coUect soot (Dobbins and
Megaridis, 1987). Technique (1) is suitable for measuring an average amount
of soot produced in a relatively long time period (roughly an order of 10s or
greater). The two other techniques are suitable for detailed chemical analysis.
The residence time of technique (2) is roughly an order of a few seconds or
greater, and that of technique (3) is roughly 50 milliseconds. Technique (2) pro-
vides a layer of soot deposit to be analyzed with little effect of heterogeneous
catalysis on the soot and probe material. Technique (3) can provide individual
soot particles that are collected by inserting a thermophoretic probe directly
into a flame. Soot in the flame is attracted by copper and other metal grids by
thermophoresis.
Most of the soot studies were conducted for laboratory scale candlelike lam-
inar diffusion flames because they allow researchers easy access to laser optics
and other measurement systems that require well-controlled laboratory condi-
tions. These studies found that soot is actually a cluster consisting of small pri-
mary soot particles. The average diameter of the mature primary soot particle
produced by small laminar diffusion flames was identified by a scanning electron
microscope and a transmission electron microscope to be approximately 20 to
50 n m (1 n m = 10"^ m ) . The average molecular weight of soot is 10^ g/mole,
and the C/H ratio is 8 to 10 (Glassman, 1996). The mechanism of primary soot-
particle formation, however, is not well understood. Two competing mecha-
nisms that have been proposed recently include: (1) a gradual buildup of a
carbonaceous product via formation and interaction of polycyclic aromatic hy-
drocarbons (PAHs) or acetylene (C2H2) molecules through dehydrogenation
and polymerization processes, and (2) a direct condensation of liquidlike car-
bonaceous precursor materials to form a solidlike precursor soot particle as a
result of fuel pyrolysis reaction, polymerization, and dehydrogenation.
Regarding the soot and smoke formation mechanisms, there is a belief that,
because soot is formed in the gas phase during the process of fuel pyroly-
sis, difference in fuel type (either gas or liquid or solid) has little influence on
soot formation processes. This may be true for the physical shape of primary
soot particles which are spherical. However, a recent laser-desorption mass-
spectroscopy study (Majidi et al, 1999) showed that mature soot collected from
several different laminar hydrocarbon-air diffusion flames have different in-
depth chemical structures. This is interesting because, if it is proved true for a
wide range of fuels, soot could be used to identify its parent fuel, helping in fire
investigations. No LDMS data are available for soot collected from forest fires.
A few studies focused on smoke generated from liquid pool fires. Fallen
smoke particles were collected on the ground (Koseki et ah, 1999; Williams and
Gritzo, 1998) or sampled at a post flame region by an airborne-smoke-sampling
package (Mulholland et al, 1996). They used crude oil fires whose diameters
18 K. Saito

ranged from 0.1 to 20 m and sampled smoke particles generated by these fires;
primary smoke particles from these crude oil fires looked spherical, and the pri-
mary particle diameters varied between 20 and 200 nm depending on the sam-
pling location and diameter of the pool. These studies found a general trend: the
larger the fire, the larger the primary smoke particle, suggesting the observed
result may be attributable to the longer resident time.
The smoke generation rate from many different types of solid fuels was
tested using a laboratory scale apparatus. When small solid samples were ex-
posed under constant heat flux, they increased their temperature, released py-
rolysis products and eventually emitted smoke. The amount of smoke can be
measured by a light-scattering device or can be collected on a filter paper for
weight measurement. When a pilot ignitor is placed over the sample, the sample
can be ignited to study material flammability and flame spread characteristics
(Fernandez-Pello, 1995; Kashiwagi, 1994; Tewarson and Ogden, 1992; Babraus-
kas, 1988). The flame spread occurs in horizontal, upward, and downward di-
rections with and without an external (radiation) heat source. These kinds of
experimental apparatus require relatively small samples (e.g., 10 cm wide X
30 cm long for flame spread tests, Saito et ah, 1989), but some of them are de-
signed for large scale upward spread tests (Orloff et al, 1975; Delichatsios et ah,
1995). Effects of sample size and geometrical shape on ignition, flammability
and flame spread rate are not well understood (Long et ah, 1999). Caution
should be taken when the laboratory test results are applied to evaluate behav-
ior of forest fires that involves many different types, kinds and sizes of trees,
and bushes and duff in the forest.

III. TEMPERATURE, VELOCITY, SPECIES

CONCENTRATION, AND FLAME HEIGHT
Measurements of flame temperature, velocity, height, and chemical species con-
centrations are important for understanding the structure and characteristics
of forest fires. The first three quantities can be measured at a point and thus al-
lows determination of profiles. Height can be measured as the time-averaged
maximum height of the yellow flames since it is most relevant to radiant heat
released from the flame.
For the first three quantities, profile data can provide more information than
point data, but any (one-, two-, and three-dimensional) profile data require
more human effort and better (often more sophisticated) equipment than point
measurement. The profile data provide us with the relative change of the mea-
sured quantity (such as temperature), thus helping us better understand fire
phenomena.
Chapter 2 Flames 19

A. TEMPERATURE MEASUREMENT

Temperature measurement with thermcouples is the most commonly used

flame temperature measurement technique. It is an intrusive point-by-point
measurement and is affected by radiation (qrad)' convection (qcov)' ^^^ conduc-
tion (qcod) h^^t losses. These heat losses can be described in terms of heat flux:
^rad ~ ^bO-Tl{l - a ) ; q^ov ~ ^7(7^^ " ^ J ; ^cod ^ A(Tj, - Tj/I^-. Here, e^ = emis-
sivity of the thermocouple bead, a = absorption coefficient of the gas surround-
ing the thermocouple, a = Stefan-Boltzmann constant, Tj, = thermocouple
bead temperature, T^ = gas temperature surrounding the thermocouple bead,
Ij = a characteristic length related to the high-temperature gradient in the gas,
7] = heat transfer coefficient, and A = thermal conductivity of gas. To minimize
these heat losses, a thermocouple with a small bead and a small wire diameter
is advantageous. To minimize the conduction heat loss, the wire can be placed
along an isotherm in the flame. Radiation heat loss proportional to the fourth
power of temperature may become dominant above 1000°C. Soot deposit on
the thermocouple junction changes the thermocouple output reading due to
the change in emissivity of the junction. A sheathed thermocouple is available
to minimize the radiation effect. When a bare thermocouple is used, caution
should be taken for the measurement on sooty flames because the fire-generated
smoke and soot can coat the thermocouple bead quickly and change its emis-
sivity, increasing the radiation heat loss. An extrapolation to zero insertion time
can be used to eliminate the soot-coating effect on the thermocouple (Saito et al,
1986). When thermocouples are used for field tests, a lead wire that connects
the thermocouple and its detector should be as short as possible and lifted
above ground to avoid a short circuit due to an accidental step or contact with
water. The lead wire should be covered by a heat-insulating material to prevent
convective heat loss when significant convective heat loss is expected.
The infrared thermograph technique involves use of an infrared camera and
an image recording and analysis system capable of obtaining a two-dimensional
thermal image from a remote location. This technique was applied to measure
upward flame spread rate on wood slabs (Arakawa et al, 1993) and temperature
profiles in forest fires (Clark et al, 1998; 1999). Radke et al (1999) applied an
airborne imaging microwave radiometer to obtain a real-time airborne remote
sensing. These techniques can be very useful in identifying the hot spot or py-
rolyzing condensed-phase front which can be related to the active flame spread
front. Note that the infrared technique may not represent exactly the actual tem-
perature of the flame or condensed phase because the phase emissivities vary
from location to location and because absorption by smoke, HjO, and COj will
influence the infrared signal. Arakawa et al (1993) suggest the use of a band-
pass filter (10.6 ± 0.5 /xm) to eliminate the absorption effects of these gases.
20 K. Saito

A color video camera may be used to obtain relative flame temperature

change in forest fires because the flame color can be roughly related to the flame
temperature. The color video camera technique is simple and useful for seeing
relative change throughout the flame, but it is more qualitative than the infra-
red thermograph. Combination of the thermocouple, infrared thermograph,
and color video camera techniques may be most ideal for reseachers to assess
flame temperature correctly.
Other temperature measurement procedures that are mainly suitable for
laboratory-scale flames include the holographic interferometer (Ito and Kashi-
wagi, 1988), two-color pyrometer (Gaydonand Wolfhard, 1979), spectrum-line
reversal method (Gaydon and Wolfhard, 1979), and coherent antistoke Raman
spectroscopy (Demtroder, 1982). The two-color pyrometer employs two dif-
ferent wavelength band-pass filters to eliminate emissivity of the object. It is
suitable to measure from a remote location a local temperature of sooty-yellow
flames whose profiles are fairly uniform along the direction of measurement.
Typical examples are combustion spaces of internal combustion engines and
boilers. All other techniques are based on a laser optic system which requires a
dark room and accurate alignment. These systems are sensitive to environmen-
tal change and not suited to field measurements.

B. VELOCITY MEASUREMENT

A pitot tube velocity measurement is a commonly used velocity measurement

technique. It is an intrusive and point-by-point measurement (Sabersky et al,
1989). Pitot tubes can accurately measure velocity in the range of a few meters
per second to above 100 m/s, covering most of the wind velocity range in forest
fires. The pitot tube is relatively simple and can measure a one-dimensional ve-
locity component, but it also can be applied to a three-dimensional measure-
ment by changing the direction of the pitot tube head when the flow is at steady
state.
A simple velocity measurement technique with practical use is a video cam-
era. Recording the fire behavior in a video camera and reviewing it on a moni-
tor screen, researchers may be able to trace the trajectory of smoke and fire
brand particles moving in and over the fire area and calculate approximately
the local velocity as well as two dimensional velocity profiles. This technique
is qualititive at best because the smoke and particles may not exactly follow the
flow stream and also the flow may be complex and three-dimensional. Wisely
using the video image, however, researchers can gain valuable information on
the approximate flow velocity in relation to the overall fire behavior.
Other available velocity measurement techniques, all of which are laser-based
Chapter 2 Flames 21

methods and only suitable to laboratory experiments, include laser doppler ve-
locimetry (LDV) and particle image velocimetry (PIV) or laser sheet particle
tracking (LSPT). LDV is a point-by-point measurement, and PIV and LSPT are
capable of measuring two-dimensional, and possibly three-dimensional veloc-
ity profiles. All three of these techniques require seeding of small trace par-
ticles, laser optics, and a data acquisition system. An example of these tech-
niques applied to flame base structure is shown later.

C. SPECIES CONCENTRATION MEASUREMENT

Gas chromatography is the most commonly used method of analyzing the con-
centration of species, such as CO2, CO, H2O, N2, O2, and many different hydro-
carbons. A sampling probe collects sample gases by means of either a batch or
continuous flow system. The batch sampling requires transport of the sample
to a laboratory where the samples will be analyzed. The continuous flow sys-
tem offers an on-line analysis at the site. The continuous flow system is better
than the batch sampling system because the on-line method eliminates the pos-
sibility of sample contamination during transportation and storage. For field
experiments, portable gas analyzers can be used for the on-line measurement
of concentrations of some or all the following species: O2, N2, CO2, H2O, CO,
NO^, and other hydrocarbons. These instruments require frequent calibration.
Their sensitivity and response time should satisfy the required accuracy of each
experiment.
A quartz microsampling probe is commonly used for sampling of laboratory-
scale flames. It consists of a tapered quartz microprobe with a small sonic ori-
fice inlet which accomplishes rapid cooling and withdrawal of the sampling gas
by adiabatic expansion (Fristrom and Westenberg, 1985). During the measure-
ment of water concentration, the sampling line should be heated above the dew
point to prevent water condensation; this is called wet-base sampling. Another
method is dry-base sampling in which all water is removed before analysis. The
dry-base technique requires a water condensation unit in the sampling line. A
water-cooled stainless-steel sampling probe is often used to provide dry-base
sampling of large scale fires. When an uncooled probe is used for dry-base mea-
surement, a water condensation unit is needed to be sure all water is removed
before analysis. The sampling line that connects the probe and the batch (or the
probe and the analytical instrument) should be free from contamination by left-
over gas or air leaks. The commonly used sampling line materials are Teflon,
quartz, and stainless steel.
The previously mentioned sampling techniques are intrusive and provide
time-average species concentrations within a specific sampling volume. Physical
22 K. Saito

disturbances due to the probe may be negligible for forest fires, but they may
become important for small-scale laboratory flames. For futher information, see
Fristrom and Westenberg (1985).

D. FLAME HEIGHT MEASUREMENT

The purpose of flame height measurement is to estimate radiation intensities

from the flame, an important parameter for assessing hazards to personnel and
the rate of flame spread. The most widely accepted interpretation of flame height
is the height where the flame achieves the maximum temperature. Another defi-
nition that has more practical application to forest fires is the vertical distance
from the flame base to the yellow visible flame tip. These two definitions may
not always give the same flame height. For a small candle flame the maximum
temperature is near the visible yellow flame tip, while for a 1-m diameter crude
oil fire the maximum temperature is achieved at about two thirds of the time-
averaged visible flame height, probably because a fire-induced strong turbulent
air flow cools the upper portion of the 1-m diameter fire.
Here, two different methods of determining flame height will be introduced.
The first is to measure the visible flame tip, and the second is to identify the lo-
cation of the maximum flame temperature. The values obtained by these tech-
niques are valid to within ±10% uncertainty.
To measure the time-average visible yellow flame height, a motor-driven
35mm still camera or a motion color video camera can be used. Both are capable
of measuring temporal flame behavior from which time-averaged flame height
can be measured. The latter equipment has an advantage in accuracy because it
offers the measurement of flame height over the slow motion video screen. For
a smoke-covered flame, however, neither technique is applicable because the
flame tip is invisible. For these types of fires, an infrared camera can be used to
penetrate the smoke layer and measure the maximum flame temperature loca-
tion. Caution is required when the infrared camera is used to identify the maxi-
mum flame temperature and/or visible flame tip because emissivity of the flame
is unknown, and it varies with fuel and location in the flame.
Figure 1 is a simplified plot of dimensionless flame height (visible flame
height divided by a horizontal scale of fire) as a function of dimensional pa-
rameter measuring Froude number (the ratio of inertia force to buoyancy force)
(Williams, 1982). The error bars in the correlation are due to different mea-
surement techniques, different fuels, and different flame height definitions.
However, the correlation itself is reasonably high over the 15 orders of magni-
tude in Froude number and the 6 orders of magnitude in the ratio of flame
height to the fuel-bed dimension.
Chapter 2 Flames 23

Modified Froude Number 10^(g^/cm^s^)

FIGURE 1 Dimensionless flame height as a function of modified Froude number (Williams,
1982). Quantity / is a horizontal scale of the fire.

IV. PREMIXED AND DIFFUSION FLAMES

The idea of categorizing flames into two different types, premixed and diffusion
flames, and then studying each type thoroughly has proven to be successful.
Many studies have contributed to a thorough understanding of these two types
of flames. Based on these studies, much progress has been made in the field of
combustion (Williams, 1985; Linan and Williams, 1993; Glassman, 1996; Hi-
rano, 1986). Although diffusion flames can represent forest fires, the premixed
flame allows us to better understand the diffusion flame by comparing its struc-
ture with that of premixed flames. Thus, in this chapter, characteristics of both
diffusion and premixed flames are explained. However, note that these concepts
are by no means perfect and may not be applicable to all forest fire phenomena.
For example, the microstructure of a flame's leading edge can be interpreted by
using a different concept (Venkatesh et al, 1996).
A diffusion flame is defined as any flame in which the fuel and the oxidizer
are initially separated. Diffusion flames are represented by candle flames, match
flames, wood fires, and forest fires. A premixed flame is defined as any flame in
which the fuel and the oxidizer are initially mixed. A good example of a pre-
mixed flame is a uniform combustible mixture contained inside a tube that is
ignited by a spark creating a flame spreading along the tube (Williams, 1985).
The most significant difference between diffusion and premixed flames is that
in the former the fuel-oxidizer ratio varies throughout the flame, whereas in the
latter it is constant everywhere in the flame. Some flames possess both charac-
teristics, and they will be called partially premixed flames.
24 K. Saito

According to the flow regime, both diffusion and premixed flames can be
classified as laminar, turbulent, or transitional. Laminar flames have smooth,
steady flow characteristics in and around the flame, whereas turbulent flames
have irregular, disorganized flow characteristics in and around the flame. The
use of Reynolds number (the ratio of viscous force to the inertia force) forjudg-
ing flames to be either turbulent or laminar is not always straightforward be-
cause of uncertainty involved in determining characteristic parameters; a visual
observation is often good enough to make this judgment. Some flames may be
laminar in one section and turbulent in the other section (e.g., the flame shape
in a 1-m diameter crude oil fire is turbulent except at the flame base where the
flame anchors and flame shape is laminar). In the following discussion, pre-
mixed flames and four different types of diffusion flames will be explained.

A. T H E PREMIXED FLAME
AND ITS BURNING VELOCITY

One important characteristic of the premixed flame is the burning velocity. As-
suming the pressure p constant, and with the specific heat c^ of the mixture
taken as constant, the overall energy conservation shows that the energy per
unit mass added to the mixture by the combustion, HQ, is

Ho = S(T^ - To) (5)

where TQ and T^ are the initial and final temperature, respectively. This energy
can be determined by the rate of conversion of reactant mass to product mass.
Using Eq. (4), this conversion rate (mass per unit volume per unit time) can be
written as

w = AY^Y^l) exp(-E/RoT) (6)

where the subscripts F and O identify fuel and oxidizer, the mass fractions Yp
and YQ are proportional to the concentrations (the number of molecules per
unit volume) of the reactants, and m and n are the overall reaction orders with
respect to fuel and oxidizer, respectively. In terms of the thickness 8 of the
flame, the chemical energy released per unit area per unit time may be approxi-
mated as

H,w8 = A(T^ - To)/8 (7)

where A is the thermal conductivity.

Using Eqs. (5) and (7) and solving for 5, we obtain

8 = [A/(c,w)]i/^ (8)
Chapter 2 Flames 25

which shows that the flame thickness varies inversely with the square root of
the reaction rate. The mass of reactant converted per unit area per unit time is
PQVQ, where pQ is the initial density of the mixture and VQ is the laminar burn-
ing velocity. By conservation of fuel mass in the steady flow, we have
PoVo ^ w6 (9)
Using Eqs. (8) and (9), the laminar burning velocity takes the form
vo = (l/po)[(wA)/Cp]i'^ (10)
Equation (10) shows that the laminar burning velocity is proportional to the
square root of the ratio between the diffusivity, A/(CpPo), and the reaction time,
PQ/W. Equation (10) agrees with laminar flame spread experiments using a ver-
tical tube filled with a fuel and oxidizer mixture. When the mixture is ignited,
a steady laminar flame spread occurs from the top to the bottom. The flame
spread rate can be measured fairly accurately using a motor-driven camera or a
high-speed video camera or an array of thermocouples placed along the inner
surface of the tube.

B. IGNITION

Ignition can be achieved either by heating the system containing a combustible

mixture with thermal energy or by creating chain reactions with autocatalytic
reactions. Chemical reactions whose reaction rate is controlled by the concen-
tration of initially present reactants are called thermal, whereas reactions whose
rate is affected by the concentrations of the intermediate and final products are
called autocatalytic.
Thermal ignition can be achieved by supplying an external source of heat to
the system. An electric arc, a spark plug, or a pilot flame can be the external
heat source. The external heat source provides excess heat to the system and
raises its temperature continuously if the heat input exceeds the heat loss. At
some point, the temperature rise will become highly accelerative and a high
heat evolution will occur. This is called ignition.
Autocatalytic ignition can be achieved by increasing the rate of chain car-
rier generation over the termination reaction of chain carriers. The initiation
of chain carrier generation may be achieved by thermal energy. After the ini-
tiation reaction has started, the reaction may become self-accelerative even
after the external heat is removed. Then the system will ignite by satisfying the
self-sustaining condition. Lightning may be a good example of autocatalytic
ignition.
An important topic in forest fires is determining the condition under which
given combustible forest materials can ignite. There are two types of ignition—
26 K. Saito

spontaneous (unpiloted) ignition and piloted ignition. Unpiloted ignition can

be achieved by raising the temperature of the system with hot boundaries and
adiabatic compression. Piloted ignition can be achieved by adding heat to the
system with an external heat source. Ignition temperature (explained later) for
unpiloted ignition is higher than for piloted ignition. For example, ignition
temperature for piloted ignition of cellulose is approximately 350°C, whereas
that for unpiloted ignition is approximately 500°C (Williams, 1982).
Other important aspects of ignition related to forest fires include ignition
temperature, flammability limits, and minimum ignition energy.

C. IGNITION TEMPERATURE

Ignition temperature can be defined as the critical temperature that the con-
densed phase needs to achieve for burning to begin (Williams, 1985). The igni-
tion temperature is a useful criterion to assess the flammability of materials and
estimate flame spread rate. However, there are some variations among different
ignition conditions. For example, some researchers use a black body as the heat-
ing source, and others use a halogen or infrared lamp (Babrauskas, 1988).
Ignition tests conducted in an inert atmosphere under various external heat
flux conditions showed that the ignition temperature indeed varies as a func-
tion of external radiant heat flux. The higher the external heat flux, the higher
is the ignition temperature (Delichatsios and Saito, 1991). Thus, some caution
should be taken when laboratory ignition temperature data are applied to for-
est fire problems.

D. FLAMMABILITY LIMITS

If there is too little or too much fuel in a fuel-air mixture, ignition may not oc-
cur even if a pilot flame provides sufficient heat to the mixture. For example, a
propane + air mixture with fewer than 2.2 mole of propane or more than 9.5%
of propane can't ignite. The maximum and the minimum concentrations of fuel
within which flame can spread are called, respectively, rich and lean limits of
flammability. Flammability limits depend on pressure and temperature; such
data for gases are well established (Zabetakis, 1965). Flammability of solid fuels
can be linked to ignition of solids that release pyrolysis products under exter-
nal heating. These pyrolysis products will mix with air forming a flammable
mixture whose flammability can be tested. Thus, knowledge gained through
studies on flammability limits for gases can be applied to understand the flam-
mability of solids.
Chapter 2 Flames 27

E. MINIMUM IGNITION ENERGY

Williams (1985) offers a simple and practically useful definition of minimum

ignition energy. "Ignition will occur only if enough energy is added to the gas
to heat a slab about as thick as a steadily propagating adiabatic laminar flame
to the adiabatic flame temperature." Thus, the minimum ignition energy, H, can
be described by

H = {A8)po[c^{T^-To)] (11)
where A is the cross-sectional area of the slab, 8 is the adiabatic laminar flame
thickness, PQ is the initial density of the mixture, Cp is the average specific heat
at constant pressure, T^ is the adiabatic flame temperature, and TQ is the initial
temperature of the mixture. Using 8 ~ XIC^PQMQ (Williams, 1985), Eq. (11) can
be written as

HM = A(T^ - To)/vo (12)

where A is the average thermal conductivity of the gas and VQ is the laminar
burning velocity discussed earlier.

V. EXTINCTION OF DIFFUSION FLAMES

There are two main flame extinction criteria: the Damkohler extinction crite-
rion and the Zel'dovich extinction criterion, both of which were detailed by
Williams (1985) and Linan and Williams (1993). Here a summary of the Dam-
kohler extinction criterion is presented.
Extinction can be defined as a balance between a chemical time, t^, and a resi-
dence time, ty, in the gas-phase flames. A nondimensional parameter known as
Damkohler number. Da, is defined as the ratio of a flow (or diffusion) to chemi-
cal reaction times in the gas:
Da number = (A flow or air diffusion time)/(Chemical reaction time) = tjt^
(13)
Using the chemical time, t^ = AYpYQ exp{E/RQT), and the residence time, t^ =
l/u or l^/a, Eq. (13) can be written as

Da n u m b e r = (I/U)(AYFYO)~^ e x p ( - E / R o T )
(14)
or {lVa){AYpToy^ exp(-E/RoT)
where u is the characteristic flame velocity, I is the characteristic length, a is the
thermal diffusivity of the gas, and m = n = I denoting the first-order chemical
reaction.
28 K. Saito

Q.
E
B 'EX
E
E

'IG — •*- — — — — " — _ 3 # IGNITION

Damkohler Number
FIGURE 2 Schematic illustration of the dependency of the maximum temperature on the reduced
Damkohler number for diffusion flames. Here, for cellulose-based fuels, T^x (extinction tempera-
ture) = 1300 ± 100°C, TjG (ignition temperature) = 350 ± 50°C, and TQ (initial fuel tempera-
ture) = 20°C.

The maximum temperature can be taken as an important parameter to con-

trol extinction. Figure 2 shows temperature as a function of the Da number, a
well-known S-shaped curve. Variation of parameters that control fires causes
the Da number to change continuously. For example, if the Da number based
on TQ is increased beyond the critical Da number marked as ignition (Di^), then
a discontinuous change to a much higher temperature will occur (this passage
is shown in a double arrow). If the upper branch of the Da number is decreased
continuously to the critical Da number marked as extinction (D^x), then a dis-
continuous change to a lower temperature will occur (this passage is shown
in a single arrow). At the lower branch temperature, there is no combustion,
whereas at the upper branch temperature, active combustion is taking place.
The middle branch is unstable and not physically relevant. The Damkohler ex-
tinction criterion is useful in obtaining overall chemical kinetic parameters for
gas-phase combustion from experimental data.
There are eight different strategies to achieve extinction: isolate the fuel, iso-
late the oxidizer, cool condensed phase, cool the fuel, cool the gas, inhibit the
chemical reaction homogeneously, inhibit chemical reaction heterogeneously,
and blow the flame away (Williams, 1982).
The Zel'dovich extinction criterion, derived by assuming a large activation
energy asymptote (Linan and Williams, 1993), is beyond the scope of this book.
Readers who are interested in the Zel'dovich extinction criterion may consult
Linan and Williams (1993).
Chapter 2 Flames 29

VI. DIFFUSION FLAMES

AND SCALING ANALYSIS
Diffusion flames can be classified based on the forces that control the flow and
mixing processes. The major forces acting on the diffusion flames may be as-
sumed to include momentum, viscous, and buoyancy forces. Molecular diffu-
sion driven by concentration differences is also assumed to play an important
role in the mixing process. Based on these assumptions, scaling analysis (Wil-
liams, 1969; Emori and Schuring, 1977) can be conducted to categorize the
diffusion flames into four different types. Scaling laws are important in fire re-
search as described in the scaling law section later in this chapter. By defining
UQ = the fuel exit velocity, d = the burner port diameter, D = the binary diffu-
sion coefficient of a fuel into air, g = the acceleration due to gravity, and v =
the kinematic viscosity, the following three pi-numbers (dimensionless num-
bers) can be formed when a fire burns in calm air:
• Peclet (Pe) number = [Fuel mass transferred by momentum of incoming
fuel flow]/[Fuel mass transferred by concentration difference] = UQCI/D
• Reynolds (Re) number = [Momentum force of incoming fuel flow]/
[Viscous force of air and fuel] = UQCI/V
• Froude (Fr) number = [Momentum force of air induced by the flame]/
[Buoyancy force in the flame] = ul/dg
Here four different types of flames, each identified by different combinations
of the preceding three pi-numbers, will be discussed.

A. Re NUMBER-CONTROLLED LAMINAR
DIFFUSION FLAMES WITH L O W Pe NUMBER

Consider the case of a hypodermic needle held vertically and in which a hy-
drocarbon fuel constantly flows through the bottom end to the top. When the
fuel is ignited, a small laminar diffusion flame with flame height of a few milli-
meters is established on the port of the hypodermic needle. This flame is called
a microdiffusion flame. The existence of this type of flame was predicted by
Williams (1985) and proven experimentally by Ban et al. (1994). It is governed
by the balance between the momentum of the incoming fuel flow and the mo-
lecular diffusion (both in axial and radial directions) of the gaseous fuel and air.
The shape of a microdiffusion flame is spherical and different from the more
commonly observed candlelike diffusion flame.
The momentum and species equations are coupled, and a similarity solu-
tion has been obtained for the coupled system. Microdiffusion flame experi-
ments on CH4, C2H2, and C2H4 were conducted under ambient atmospheric
30 K. Saito

pressure conditions. Flame shape was photographically recorded and compared

well with the theoretical predictions. Figure 3 (see color insert) shows a photo-
graph and a schematic of an ethylene-air microdiffusion flame established on a
0.25-mm-diameter hypodermic needle (Ban et al, 1993).

B. Re NUMBER-CONTROLLED TURBULENT J E T
DIFFUSION FLAMES WITH H I G H Pe NUMBER

By increasing the incoming fuel velocity in the previously mentioned hypoder-

mic needle, the flame will become unstable, and at some point the flame can no
longer be sustained at the burner port. This is because the characteristic time
of the incoming fuel velocity is smaller than the time required for the flame
to complete the chemical reaction. By increasing the diameter of the burner, a
stable flame with a higher Re number can be achieved. For example, a hydrogen
jet diffusion flame can be stabilized on a L43-mm-diameter nozzle burner up
to Re = 4800, and a methane jet diffusion flame can be stabilized on a 9.45-mm-
diameter nozzle burner up to Re = 8440 (Takahashi et al, 1996). Figure 4 (see
color insert) shows a hydrogen-air turbulent jet diffusion flame stabilized on the
L43-mm-diameter burner (Takahashi et al, 1996). For turbulent jet diffusion
flames, the flame will be established as the result of the mixing of the incoming
fuel jet with quiescent air. The momentum force controls the fuel jet, and the
inertia and viscous forces of both the air and the fuel control the mixing pro-
cess. This type of flame (high Pe and Re numbers) is stabilized by a turbulent
shear mixing, but not by molecular diffusion (Takahashi et al, 1996; Venkatesh
etal, 1996).

C. Re AND Fr NUMBER-CONTROLLED LAMINAR

DIFFUSION FLAMES WITH L O W Pe NUMBER

Applying a relatively slow fuel flow rate (approximately a few centimeters per
second) to the burner with a diameter of a few millimeters to a few centimeters,
a candlelike flame can be established over the burner port. This is the regime
of buoyancy-controlled laminar diffusion flames. Buoyancy forces acting on the
hotter, less-dense gas in the region of the flame often distort the shape of the
flame sheet. Figure 5 shows a candlelike laminar methane-air diffusion flame
established on a 1.6-cm-diameter Pyrex burner. Faraday (1993, originally pub-
lished in 1861) was the pioneer to identify this type of diffusion flame as the
most basic type and conducted a series of noble experiments. After Faraday,
Chapter 2 Flames 31

thin blue
orange

yellow

thin yellow

thin blue
thick blue

air

A small laminar methane - air diffusion flame

FIGURE 5 A schematic of a small methane-air diffusion flame established on a 1.6-cm diameter
Pyrex tube (the average fuel flow rate at the burner port is 2.2 cm/s and the air flow rate is
9.3 cm/s).

many researchers have used this type of diffusion flame to further understand
their chemical, thermal, and fluid mechanical structures. These flames are also
ideal for optical and chemical species sampling measurements because they are
stable, steady, and large enough to obtain accurate measurements.
The structure of laminar diffusion flames in this category is also of current
interest for a number of reasons. There have been recent demands for under-
standing the mechanisms of turbulent flames. A promising approach to the
analysis of turbulent flames is to introduce conserved scalars (Linan and Wil-
liams, 1993) so that data on laminar flames can be used. Correlations of lami-
nar flame structures on the basis of mixture fractions allow turbulent flame
properties of interest to be calculated with relative simplicity. The mixture frac-
tion, Z, for the two-feed system (oxidizer and fuel) can be defined as

Z = {Y,- YaViYa " ^,2) (15)

Here Yj is the mass fraction of the i-chemical element; the subscripts 1 and 2,
respectively, refer to the composition in the fuel and the oxidizer flows. There
are indications that the correlations can be approximately applied even for soot
32 K. Saito

and chemical species that are known to be out of the chemical equilibrium
region (Koylu and Faeth, 1991). If the binary diffusion coefficients of all pairs
of species are approximately equal, then the concentrations of all species can
be uniquely related to Z through algebraic equations. If this condition is not
satisfied, then different diffusion rates can cause Z for different elements (N, O,
C, H) to take different values (Linan and Williams, 1993).
Here, structure of a laminar coflow methane-air diffusion flame will be in-
troduced because methane is the simplest saturated hydrocarbon, yet it can
represent the general nature of hydrocarbon-air combustion. A 20-mm-high
methane-air diffusion flame, established on a 16-mm-diameter Pyrex burner,
was probed by a microsampling probe, and 15 different steady chemical species
were separated; concentration of each of these 15 species was measured with a
gas chromatograph. Two-dimensional profiles of steady flame temperature were
measured by traversing a fine thermocouple. Figure 5 shows a schematic of the
laminar methane-air diffusion flame that was probed using the experimental
apparatus shown in Figure 6 (Saito et at, 1986). The main elements of the ap-
paratus include a quartz coflow burner with Pyrex chimney that can produce a
stable and steady laminar diffusion flame, a quartz microsampling probe that
can probe chemical species from the flame, fuel and air supply and control sys-

Exhaust

Control valve
Cathetometer \
jl, Pyrex chimney - ^

pn-off valve

FIGURE 6 A schematic of methane-air diffusion flame experimental apparatus (Saito et al, 1986).
Chapter 2 Flames 33

tern, a gas chroma tograph, and a ca the tome ter for flame height measurement.
The samphng hne that connects the quartz probe and gas chromatograph is
wrapped with a heating tape to prevent condensation of water and heavy mo-
lecular species.
The distinctive color differences shown in the Figure 5 flame can be ex-
plained as follows. A thick blue near the burner port is associated with chemi-
luminescence from exited CH and CO molecules. Above that thick blue, there
is a thin blue (chemiluminescence from CO) covering the entire flame. Inside
the thin blue, there is yellow-soot region and on the top of that yellow-soot re-
gion, a narrow orange-soot region is formed. The yellow- and orange-soot re-
gions are clearly separated. The two soot-color zones were found by Saito et al.
(1987). At the yellow region, young soot particles are formed, and the soot par-
ticle temperature is higher than the gas temperature because the young soot
particles can actively react with the gas molecules, causing a strong exothermic
reaction on the soot particle surface. However, at the orange-soot region, young
soot becomes mature and has lost its surface reactivity, bringing the soot par-
ticle and gas temperatures into equilibrium.
The centerline profiles of chemical species of the flame is shown in Figure 7.
The structure of the methane-air diffusion flame, based on results from careful
measurements such as Figure 7, is illustrated in Figure 8 (Linan and Williams,
1993). In Figure 8, Z^ is the stoichiometric mixture fraction, and the one-step
rate function, Eq. (6), is used. At the point, 2 = Z^, the fuel and oxidizer are
consumed at an infinite rate, releasing the maximum heat (and achieving the
maximum temperature) due to the complete combustion on the stoichiometric
fuel-oxidizer mixture. The mixture is fuel-lean for Z<Z^ and fuel-rich for Z >
Z^; for both cases, temperature drops from the maximum. At or around Z = Z^,
an active chemical reaction is taking place by balancing with diffusion of chem-
ical species. This region is called the reactive-diffusive zone. There are two sep-
arate zones in each side of the reactive-diffusive zone. One is the Z < Z^ side
where oxygen diffuses into the flame and the other is the Z.> Z^ side where fuel
diffuses into the flame. In these (convective-diffusive) zones, no chemical reac-
tion takes place and convection and diffusion terms control the structure.
Kinetic models have been proposed to calculate concentrations of species for
methane-air flames. These models often consist of more than 200 chemical re-
action steps, but simplifications are needed to understand the structure. A re-
duced kinetic mechanism has been proposed by carefully selecting only essen-
tial reactions. Williams proposed to use the minimum sets that can describe the
main feature of the flame and apply steady state approximations in this set to
obtain a reduced mechanism. Table 1 shows the proposed four-step mechanism
(Linan and Williams, 1993) consisting of fuel consumption, water-gas shift, re-
combination, and oxygen consumption and radical production. The Table 1 re-
sults were obtained from guesswork on 14 fundamental reaction steps offered
34 K. Saito

Thin blue

1.0r

0.9

0.8

0.7

CD
X 0.6
S^ u
0
Q. X^
0}
O o: 0.5
<=f
X
o
0.4

3r 0.3

2h 0.2

11- 0.1

OL
0.6 0.8
h/H
FIGURE 7 The centerline profiles of temperature and cliemical species for a laminar methane-
air diffusion flame established over an overventilated L6-cm-diameter coflow burner (Saito et al,
1986).

by Linan and Williams (1993) and applying steady state approximation for
the intermediaries, O, OH, HO2, CH3, CH2O, and CHO. The usefulness of this
four-step mechanism has been demonstrated. Even simpler three- or two-step
mechanisms were also proven to be sufficient to explain many aspects of flame
structure.
 Convective- Reactive- Convective-
diffusive diffusive diffusive
zone zone zone

Fo

FIGURE 8 Thermal a n d chemical structures of the methane-air diffusion flame plotted against
mixture fraction Z.

TABLE 1 Four-Step Mechanism for the Methane-Air Flame

Fuel consumption CH4 + H - > CH3 + H2

CH3 + O ^ CH2O + H
CH2O + H ^ CHO + H2
CHO + M ^ C O + H + M
H + O H ^ O + H2
H + H 2 0 ^ 0 H + H2
CH4 + 2H + H2O ^ CO + 4H2
Water-gas shift CO + OH < ^ CO2 + H
H + H2O 44> OH + H2
CO + H2O ^ CO2 + H2
Recombination O2 + H + M - ^ HO2 + M
OH + H 0 2 - ^ H 2 0 + O2
H + H2O - > OH + H2
2H + M ^ H 2 + M
Oxygen consumption and O2 + H < ^ OH + O
radical production O + H2 < ^ OH + H
OH + H2 < ^ H2O + H
OH + H2 < ^ H2O + H
O, + 3H, ^ 2H,0 + 2H
From Linan a n d Williams, 1993.
36 K. Saito

D. Fr NUMBER-CONTROLLED TURBULENT
DIFFUSION FLAMES WITH H I G H Pe NUMBER

w h e n the velocity of incoming fuel jet is relatively low compared to the veloc-
ity of hot gas accelerated in the turbulent flame zone by buoyancy, a buoyancy-
controlled turbulent flame will be obtained. Good examples of this type of flame
are turbulent liquid pool fires, which can be created by igniting a liquid fuel
placed in a relatively large diameter open top pan (say larger than 1 m for most
hydrocarbon fuels and a lesser diameter for alcohol fuels). W h e n the diameter
of the pan is less than a few centimeters, the flame shape will be laminar. W h e n
the diameter is approximately between a few centimeters and 1 m, the flame will
be in a transient regime between laminar and turbulent. W h e n the diameter is
larger than 1 m, the flame exhibits fully developed turbulence and often emits
black smoke; sometimes the entire flame will be covered by a thick dense smoke.
Figure 9a (see color insert) shows a diesel oil fire of a 15-m^ open-top con-
tainer, conducted in an open field with a wind of approximately 9 - 1 2 m/s; a
thick dense smoke covers almost the entire flame making the flame height mea-
surement difficult. An infrared thermograph technique, described in the tem-
perature measurement section, was applied to obtain a two-dimensional tem-
poral map of flame temperatures from a remote location (Figure 9b; see color
insert). Interestingly, the infrared image in Figure 9b shows the highest tem-
perature zone for such a large-scale pool fire to be near the flame base. A similar
result was found in a recent large-scale crude oil fire where an infrared (IR) ther-
mograph was used to measure temperature distributions of a smoke-covered
flame (Koseki et al, 1999). It is not clear whether the result is false due to the
change of emissivity of the flame or real. If it is real, it may be attributed to
the enhanced cooling effect by buoyancy-induced convection of outside air in
the upper portion of the flame where a large-scale puffing motion takes place.

E. SCALING LAWS

Forest fires spread nonuniformly and nonsteadily over nonhomogeneous ma-

terials. To understand complicated mechanisms of forest fires, it may be neces-
sary to design a laboratory-scale test without changing the essential chemistry
and physics governing the forest fires and measure specific aspects of forest fire
behavior under well-controlled laboratory conditions. Prediction models then
can be developed against these experimental data. Scale modeling is an experi-
mental technique to realize these needs. The purpose of scale modeling is well
described by Johnstone and Thring (1957): "commit your blunders on a small
scale and make your profits on a large scale." The role of scale modeling in the
Chapter 2 Flames 3 7

Study of forest fires is certainly important because many forest fire phenomena
are so large that full-scale experiments are difficult to conduct due to safety con-
ditions, economical and time constraints, as well as technical difficulty in con-
trolling the parameters. In scaling combustion and fire phenomena, the same
temperature for both the model and the full-scale may be required to make the
same chemical reactions for both.
The rules that researchers must follow in setting up and interpreting the re-
sults of scale model experiments are called scaling laws or model rules (Emori
and Schuring, 1977). To give a reliable forecast of full-scale phenomena, the de-
velopment of scaling laws is very important, and indeed the use of scale mod-
els can be justified only if they are able to provide information relevant to the
full-scale phenomena.
Governing equations can be applied to obtain strict scaling laws shown
below.
77i = 7T[ (16)

where TTJ (pi-numbers) represent dimensionless numbers involved in the full-

scale (or prototype) phenomena (i = 1, 2, 3 , . . .), and the prime represents the
scale model. Williams (1969) showed us that the strict governing equations
in combustion systems yield 29 pi-numbers. Thus, it is difficult to satisfy the
Eq. (16) requirement. For example, take TT^ as Reynolds number and TTJ as
Froude number. Then TT^ requires the characteristic velocity to be proportional
to (Length scale) ~\ whereas 772 requires the characteristic velocity to be pro-
portional to (Length scale) ^^^, resulting in a scaling conflict. However, if we can
find reasons to neglect one of them, then the scaling problem can be resolved.
Thus, the essential problem of scale modeling is to select only a few groups of
pi-numbers and discard others to realize the scaling laws.
Scaling laws were derived for the last type of flames, referred to as pool fires,
by employing the following assumptions (Emori and Saito, 1983): (1) the pres-
sure is atmospheric for both model and the full-scale fire, (2) the dominant heat
input to the fuel surface is given by radiation from the optically thick flame,
(3) conducted heat through the rim of the pan is negligibly small, (4) inertia
and buoyancy terms in the motion of gas and air are dominant compared with
viscous term, and (5) the flame pulsation phenomenon is governed by the in-
ertial and buoyant effect of combustion gas and air.
With the same ambient conditions for both the scale model and the full-scale
model and with the use of the same fuel under the same ambient conditions,
scaling laws were reduced to a functional relationship of the form
0{ui/u2, hid, df/ui, dg/ui EJipjqjVj)} = 0 (17)
where u^ is the lateral wind velocity, Uj is the buoyancy-induced velocity, h is
the visible flame height, d is the diameter of the liquid pool,/is the frequency
38 K. Saito

of the flame pulsation, g is acceleration due to gravity, E^ is the radiant heat flux
received by a radiometer at geometrically similar points, scaled according to the
size of the fuel bed, Vj is the rate of decrease of height of fuel surface by com-
bustion, <l> is an arbitrary function, pj is the density of the fuel, and qj is the heat
value per unit mass of liquid fuel.
It was found experimentally that the regression rate Vj does not depend on
the diameter of the fuel container. The burning time tj, = V ^ / ' where hj is the
depth of liquid fuel in the container, does not depend on the size of container
if the initial fuel thickness is the same. As a result, the condensed phase is con-
trolled by E^/{pjqjVj) in Eq. (17), whereas the flame behavior is governed by
four other pi-numbers, Ui/uj, h/d, df/ui, dglu\. This is the major difference be-
tween pool fires and crib fires (to be explained in the following section), where
the common group of pi-numbers governs both condensed and gas phases.
Equation (17) can be used to design scale models and determine specific ex-
perimental conditions for scale model experiments. Using h/d and dgl\^, for ex-
ample, the model experiments are to be performed with wind velocities pro-
portional to the square root of the fuel bed diameter. The scaling then predicts
variations of flame height and vertical velocity in the flame. Wind tunnel fire ex-
periments can be designed to verify these predictions (Emori and Saito, 1983).

F. W O O D CRIB FIRES

In contrast to the pool fires where the regression rate of the fuel is the same
in winds of different velocities, the burning rate of wood crib fires can easily be
increased by flowing air on it. Thus, scaling laws on crib fires employ the same
assumptions—(1), (3), (4), and (5)—as employed for the pool fires. Assump-
tion (2), however, is modified to account for energy transfer by buoyancy. Thus,
assumption (2) for crib fires includes both radiation and convection as the domi-
nant heat input. Figure 10 shows the configuration of a crib fire bed and a Japa-
nese cedar crib fire. The configuration of the crib promotes three-dimensional
turbulence so that the flow remains turbulent in cribs at smaller diameters than
in pool fires, suggesting that the geometrical configurations play an important
role in combustion.
With the use of the same crib materials and under the same ambient condi-
tions, the pi-numbers obey a functional relationship of the form

^ { u i / u 2 , h/I„ lju\ l,g/ui E,t/{p,l,q,)} = 0 (18)

In addition to the symbols defined for pool fires, here l^ is the length of the crib
stick, t is the characteristic time, "^ is an arbitrary function, p^ is the density of
the crib stick, and q^ is the heat value per unit mass of the crib material.
Chapter 2 Flames 39

b. T
mm m m m mm
mm mm m m

FIGURE 10 A Japanese cedar wood crib fire and crib bed configuration.

Experimental data are available to support the proposed scaling laws (Emori
and Saito, 1983). Figure 11 shows that the temporal changes of irradiance from
pool fires of 10 m (prototype) and 1.2 m (the scale model) in diameter become
the same at corresponding times after the burning. This is consistent with the
scaling law prediction, E^/(pjcjjVj), which yields EjVj = K^Vj and Vj = Vj for
40 K. Saito

Pan dia. (m) 1

^ ^ ^ 10
—— l"2 1

1000

y
r
V 1 *
500 /#
(
ii V
_l 1 1 ^ ^ 1 — 1

0 1 2 3 4
Burning time: t (min)
FIGURE 11 Irradiance measured at geometrically similar poims as a function of time after igni-
tion for hexane pool fires (Emori and Saito, 1983).

pool fires using the same fuel. For crib fires, the temporal change of total ir-
radiance, both irradiance and time, are scaled to the size of the crib with l^ =
90 cm ( = [3) according to the scaling laws IJuh, Icg/uj, and E^t/{pJcq^). Fig-
ure 12, temporal change of total irradiance from crib fires of three different
sizes, proves the scaling law predictions [Eq. (18)].
Byram and Nelson (1970) and Porscht (1975) confirmed the prediction from
df/ui and dg/uj that the pulsating frequency becomes inversely proportional to
the square root of characteristic length (e.g., diameter) of the fuel bed. Blinov
and Khudyakov (1957) and Emori and Saito (1983) showed that irradiance E^
and burning time t are independent of pool diameter. Emori and Saito verified
that the burning rate of a 1.2-m-diameter hexane pool fire is little influenced by
lateral wind of up to 10 m/s and that the flame inclination is the same if dg/uj
is kept constant.
The concept of pool and crib fires can be used to categorize the currently ex-
isting various types of fires. In forest and urban fires (Emori and Saito, 1982;
Soma and Saito, 1991), for example, the heat release rate and the velocity of air
is known to be respectively scaled by (Characteristic length)^^^ and (Character-
istic length)^^^. In enclosure fires (Quintiere, 1988), such as a fire in the room
of a house, the burning rate is known to depend on ventilation from openings,
and the Fr number can scale the velocity of air. These fires, therefore, can be
categorized as crib fires. On the other hand, a fire on a railroad passenger car
was best characterized as a combination of pool and wood crib fires (Emori and
Saito, 1983).
Chapter 2 Flames 41

Crib length: €p(m)

0.9
L 0.6 1
0.3

CO

03

•a
Lx'
k
^ ^ "

^ 1
o
CO

0
1
Scaled time: t x A^^U (min)
FIGURE 12 Scaled irradiance measured at a geometrically similar location as a function of scaled
time for crib fires (Emori and Saito, 1983). The / is characteristic length of crib bed, and I3 is the
crib length of 90 cm. The square root relationships are based on Fr number scaling described in
Eq. (18).

VII. SPREADING FLAMES

In forest and urban fires, both laminar and turbulent flame could spread either
in one-dimensional, two-dimensional, or three-dimensional fuel configurations
with steady or unsteady modes of spread. However, most of the fire spread
modes will be turbulent with strong or minor interactions with wind. Flame
spread direction can be either concurrent or opposed to the wind direction. Be-
cause of the difficulty associated with the unsteady nature of turbulence, most
of the detailed measurements on spreading fires are done for a two-dimensional
laminar flame with a steady state spreading rate over a one-dimensional or two-
dimensional fuel surface. Theoretical models developed range from simple one-
dimensional laminar steady state models to complex three-dimensional turbu-
lent spread models (Williams, 1976; Wichman, 1992). These models, derived
for and tested in well-controlled laboratory conditions, address the fundamen-
tal nature of flame spread.
The structure of spreading flames over condensed fuels, either liquid or solid,
has been studied both theoretically and experimentally. A difficulty, however,
lies in the transient measurement technique, which is far more difficult than
42 K. Saito

the Steady-State measurements because of the highly transient nature of flame

spread behavior. It has been a challenge to obtain the two-dimensional and
three-dimensional profiles of chemical species concentration, velocity, and tem-
perature for the spreading flames. Progress has been made recently by an inter-
national collaborative effort (Ito et at, 1999) to use simultaneously laser sheet
particle tracking (LSPT), single-wavelength (HI), or dual-wavelength (DWHI)
holographic interferometer and infrared thermograph techniques. This simul-
taneous measurement technique made transient two-dimensional measure-
ments of velocity, fuel concentration, and temperature for spreading flames pos-
sible. A dual-wavelength hologram was obtained for a spreading flame over
propanol that was placed in a Pyrex tray 30 cm long and 1 cm wide (Ito et al,
1999). Two different lasers—argon-ion and He-Ne—were used as the light
source. The dual-wavelength hologram contains two different data—tempera-
ture and fuel concentrations. For the temperature holograms, each colored line
represents an isothermal contour, whereas for the concentration holograms it
represents equal fuel density contours. From the second hologram, the fuel va-
por concentration profiles in the upstream from the spreading flame-front was
obtained.
Fire spread can occur if there is communication between the burning region
and the virgin fuel. With the definitions of v = the flame spread rate, q = the
net energy per unit area per second transported across the surface of fire in-
ception, and Ah = the difference in thermal enthalpy per unit mass between the
fuel at its ignition and the virgin fuel, the energy balance can be written as

PjvAh = q (19)

Flame spread can be classified based on its spreading direction (i.e., down-
ward, horizontal and upward). Using this classification, for example, flame
spread over a 15-degree angle up-slope hill can be treated as the combination
of upward and horizontal spread.
Many other parameters can influence the spread rate in forest fires. Those
parameters include the type of fuel (thermally thick or thermally thin solid fuel,
char-forming or non-char-forming, porous, liquid, etc.), the fuel conditions
(moisture contents, temperature, and aging), the size and geometry (discrete or
continuous) of the fuel bed, and the ambient (wind, temperature, and humidity)
conditions. Rothermel (1972) extended the flame spread equation (19) to for-
est fires by adding two important components, wind and fuel bed slope effects:

V = qlpjAh = lRf(l + $ . , + ^,)lpjCoQ,^ (20)

where Ij^ = reaction intensity, ^ = propagating flux ratio (the proportion of the
reaction intensity that heats adjacent fuel particles to ignition), ^^ = a dimen-
sionless multiplier that accounts for the effect of wind in increasing the propa-
Chapter 2 Flames 43

gating flux ratio, ^^ = a dimensionless multiplier that accounts for the effect of
slope in increasing the propagating flux ratio, (o = the effective heating num-
ber (the proportion of a fuel particle that is heated to ignition temperature at the
time flaming combustion starts), and Qig = ignition energy. Here Ah = o)Qig.
Further explanation of Eq. (20) is not offered here because it is available in
Rothermel (1972) and Pyne et al (1996).
Williams (1982) stresses that developing a general formula to predict the
spread rate including all the preceding parameters is not practical nor the best
avenue. He rather suggests developing the spread rate formula for special cases
and studying each case thoroughly. Taking his advice as a guide, four types of
fire spread that have application to forest fires are summarized here (Williams,
1976, 1982). Note that for downward and horizontal spread, radiation is be-
lieved to become important, whereas for upward spread both radiation and
convection become important.

A. DOWNWARD AND HORIZONTAL FLAME SPREAD

Types of forest fuel may be categorized as thermally thin or thermally thick. For
thermally thin fuels, temperature rises across the entire cross section of the vir-
gin fuel prior to arrival of the flame. For thermally thick fuels, the fuel interior
may not be heated appreciably prior to arrival of the flame, and the tempera-
ture won't rise throughout the entire cross section of the virgin fuel prior to ar-
rival of the flame. For downward and horizontal spread, the spread direction
is opposed to the fire-induced flow direction, resulting in a slow and steady
spread rate (~ 1 cm/s or less).
Downward and horizontal spread may be associated with an early stage of
flame spread in forest fires. A small circulation that was found in the gas phase
just ahead of the spreading flame's leading edge may play an important role in
providing heat from the flame to a virgin fuel surface (Ito et al, 1999). Gas-
phase heat conduction and radiation may also be important.
For forest fires, radiation from the flame may be a dominant heat transfer
mode of flame spread (Albini, 1986). Applying q = {£aT%) sin d/£ to Eq. (19),
Eq. (21) will be obtained for thermally thin fuels:

V = {EaT%) sin e/{€pjAh) (21)

where h is the visible flame height, 0 is the angle between the flame (usually the
vertical dimension) and the exposed virgin fuel surface (6 > 77/2), and € is fuel
bed thickness. When the fuel is thermally thick, it won't be heated throughout
its depth prior to ignition. Then, using thermal diffusivity a {= X/c^pj), the fuel
44 K. Saito

thickness that is heated by the flame, (ah/v)^^^, can be used instead of the fuel
thickness € in Eq. (21). Thus, for thermaUy thick fuels,

V = (saTyhsm^d/[apj{Ah)Y (22)

B. TURBULENT UPWARD FLAME SPREAD

Upward flame spread occurs with a large heat release rate created by rapid (~
1 m/s or more) and acceleratory spread modes because the flame comes in con-
tact with the fuel in the downstream direction effectively preheating the fuel
surface by convection and radiation. Upward spread rate is much faster than
downward and horizontal spread rates where buoyancy-induced flow has
either negative or no effect. Saito et a\. (1985) conducted a study of upward
flame spread over vertically suspended (PMMA, wood and particle board) wall
samples (0.3 m wide X 1.2 m high X 0.015 m thick) and formulated the spread
rate. In the forest, upward flame spread will occur when a tree trunk ignites
near the root and flame spreads along the trunk. Figure 13 shows a schematic
of upward flame spread over a vertical wall. The spread rate is given by
V- ^q\xj Xp)/[77Apj.Si^P ~ ^o) ] (23)

where Xj is the visible flame tip height, Xp is the pyrolysis flame tip height, T^ is
the ignition temperature, and TQ is the initial fuel temperature. For a wall burn-
ing in building fires, a typical value oi q = 2-2.5 W/cm^. The value q may be
considerably larger for forest fires where intense radiant heat is generated from
the surrounding burning trees.

Area where the major heat

transfer occurs in the form of
radiation and convection

Flame

FIGURE 13 A schematic of upward flame spread over a vertical wall (Saito et al., 1985).
Chapter 2 Flames 45

C. FLAME SPREAD THROUGH POROUS FUEL BEDS

Forests can be treated as porous fuel beds consisting of layers of dead leaves and
pine needles and thickets of trees. When these fuels are burned, flame spread
occurs through the porous media fuel beds. When radiation from the burning
fuel element dominates the heat transfer, Eq. (19) becomes
V = ea{Tj - n)/f^pj{Ah) (24)
where Tj is the burning fuel surface temperature, TQ is the initial fuel tempera-
ture, and fp is the porosity of the fuel beds. Progress has been made recently
using a two-dimensional numerical calculation model to predict the maximum
spotting distance in fires over porous media (McDonough et al, 1998). This
numerical model can accommodate variable geometrical shape of terrain and
wind speed. The model prediction was roughly verified against qualitative ob-
servations made during forest fires.
Wind-aided flame spread is important in forest fires because of the enhanc-
ing effect of winds on spread rate in running crown fires (Pyne et ah, 1996), fire-
brands and spotting fires (Albini, 1979), and fire whirls and fire storms (Emori
and Saito, 1982; Soma and Saito, 1991). Detailed explanations and discussions
appear in Chapters 5 and 6 in this book and are also available from Albini
(1981), Carrier et al (1991), Wolff et al (1991), and Pyne et al (1996), so no
further discussion is provided here.

VIII. STRUCTURE OF FLAME BASE

One of the important aspects of diffusion flames is the mechanism of flame an-
choring. A commonly raised question is why and how diffusion flames are an-
chored. Study of the anchoring mechanism of diffusion flames is important be-
cause of its connection to fire extinguishment, control of fire spread, and flame
instability. Here, two different mechanisms of flame anchoring will be explained:
one sustains Fr number controlled pool fires and the other sustains Re number-
controlled turbulent jet diffusion flames. Figure 14 shows an LSPT experimen-
tal apparatus applied to a small-scale pool fire. Flame-induced flow near and
around the flame leading edge where the pool fire is anchored was visualized
by the LSPT technique. Figure 15 shows a two-dimensional flow structure vi-
sualized by the LSPT technique for a 45-mm-diameter propanol pool fire (Ven-
kateshetal., 1996).
For the Re number-controlled turbulent jet diffusion flames, the mechanism
of anchoring depends, on the formation of shear-stress-related circulation zones.
Because of the shearing between the fuel jet, whose flow speed may be much
higher than the flame propagation speed, and the oxidizer, a stagnant circulation
46 K. Saito

Flame

^<i-^Z®
© Argon Laser © ® Particle Feeder
(D Cylindrical Lens @ Water-Cooled Stainless-Steel Pan
@ © Compressed Air © Liquid-Level Controller
@ ® Control Valve
FIGURE 14 Schematic of a laser sheet particle tracking system applied to a small-scale propanol
pool fire.

Visible flame
Flow boundary
/
/ A 14
1 m/s
X 12 ?

i 10 N

H 8
6

/
i 4
y •i 2

iili
22 20 18 16 14 12 10 8 0
r (mm)
FIGURE 15 A two-dimensional flow structure of the flame base obtained by the LSPT technique
shown in Figure 14. Center of the flame base is at r = 0.
Chapter 2 Flames 47

zone is developed where mixing occurs and the flame is allowed to anchor. For
pool fires, the shear stresses at the rim are lower by two orders of magnitude
compared to the jet diffusion flame, demonstrating the insignificance of shear-
stress-induced mixing and suggesting the presence of a molecular-diffusion-
mixing zone due to finite-rate chemistry. Nevertheless, both anchoring mecha-
nisms require the generation and sustained presence of a mixing zone, where
fuel and oxidizer can intermix. The mechanisms by which these two flames pro-
duce these zones are quite distinct.
The structure of the flame leading edge where the pool fire is anchored is
similar to that of a spreading flame. Because of finite-rate chemistry influences,
the flame sheet has a finite thickness. Because the diffusion flame is bounded
on either side by two premixed flames (Figure 8), the concept of the premixed
flame propagation speed (described earlier) can be applied. The direction of the
flame propagation is opposite to the tangential component of the air entrain-
ment vector, V^. If V^ is greater than the flame propagation speed near the base,
the flame will not be able to anchor or stabilize. The LSPT measurements, shown
in Figure 15, confirmed that V^ in the convective-air entrainment zone was lower
than the stoichiometric flame speeds for the anchored pool fire. For the flame
with Vf- higher than the stoichiometric flame speed, however, the flame was no
longer able to anchor the burner port resulting in lift-off or blow-off, which led
to extinction. A schematic of structure of diffusion flames shown in Figure 8
also provides additional explanation. The flame is able to spread when the oxi-
dizer and fuel mixture is in the reactive-diffusive zone; the spread rate is maxi-
m u m at Z = Z^ and becomes slower both at Z < Z^ and at Z > Z^.

IX. CONCLUSIONS
Topics introduced in this chapter cover a small portion of flame studies that
have been accomplished in the past. Some interpretations given here are the au-
thor's view and subject to further discussion. The intention of the author is to
discuss fundamentals of combustion and to stimulate readers to think scien-
tifically when they deal with forest fire problems. References are provided for
readers who are interested in advancing their understanding in flames and
combustion.

ACKNOWLEDGMENTS
Ralph Nelson read the manuscript and offered many valuable comments. This author is largely
indebted to Forman Williams and Alvin Gordon who taught principles of combustion and have
48 K. Saito

been the constant source of education. This work is sponsored both by Kentucky-NASA EPSCoR
program and Trinity Industrial Corporation through a general education gift.

NOTATION
ROMAN LETTERS

A pre-exponential factor s"^

A cross-sectional area of the slab m^
D binary diffusion coefficient of a fuel into air m^ s~^
E activation energy J mol"^
E^ radiant heat flux received by a radiometer of geo- W m~^
metrically similar points
F body force N
H minimum ignition energy J
Hg energy per unit mass added to the mixture by the J g~^
combustion
JR reaction intensity W m~^
PCC product of complete combustion
PICC product of incomplete combustion
Q heat of combustion J
Qig ignition energy J
T temperature K
T^ gas temperature surrounding the thermocouple K
Tj, thermocouple bead temperature K
T^ the initial temperature K
Tj burning fuel surface temperature K
Tp ignition temperature K
T^ the final temperature K
T5 shear stress tensor Pa
Vj the rate of decrease of height of fuel surface by m s~^
combustion
Vf the tangential component of the air entrainment m s~^
vector
Yp mass fraction of fuel
Chapter 2 Flames 49

Yj mass fraction of the i-chemical element

Yji mass fraction of the i-chemical species in the fuel
YQ mass fraction of the i-chemical species in the
oxidizer
YQ mass fraction of oxidizer
Z mixture fraction
Z^ stoichiometric mixture fraction
a absorption coefficient of gas surrounding the
thermocouple
Cp specific heat at constant pressure J K"^ mol~^
d burner port diameter or diameter of liquid pool m
-1
/ frequency of the flame pulsation
/p porosity of the fuel beds
g acceleration due to gravity m s"- 2
h visible flame height m
hj depth of liquid fuel m
Ah the difference in thermal enthalpy per unit mass Jg^
between the fuel at its ignition and the virgin fuel
k specific reaction rate constant s"^
I characteristic length m
€ fuel bed thickness m
^3 90 cm
l^ crib stick length m
Ij a characteristic length related to the high tempera- m
ture gradient in the gas
m, n reaction order
p pressure over characteristic pressure
q the net energy per unit area per second transported W cm'
across the surface of fire inception
q^ heat value per unit mass of the crib material Jg-^
q^od conduction heat loss Wm"-2
(^cov convection heat loss W m '-2
qj heat value per unit mass of liquid fuel Jg-^
qj-ad radiation heat loss
Wm~-2
t characteristic time
50 K. Saito

t^ chemical time s
ty residence time s
UQ fuel exit velocity m s~
V flame spread rate m s"
VQ laminar burning velocity m s"
Ui lateral wind velocity m s"
Uj buoyancy-induced velocity m s~
u velocity vector over characteristic velocity
w production rate (mass per unit volume per unit time) gm~
of chemical species by chemical reaction
Xf the visible flame tip height m
Xp the pyrolysis height m

GREEK LETTERS

an arbitrary function
a dimensionless multiplier that accounts for the effect
of wind in increasing the propagating flux ratio
a dimensionless multiplier that accounts for the effect
of slope in increasing the propagating flux ratio
an arbitrary function
a thermal diffusivity m^s-i
8 adiabatic laminar flame thickness m
Sb emissivity of the thermocouple bead
A thermal conductivity of gas Wm"^K
V kinematic viscosity m^s-i
77 pi-number
Pc density of the crib stick gm
PI density of fuel gm"
Po the initial density of the mixture gm"
Q density over characteristic density
ratio of low time to evolution time
CO the effect of heating number
Chapter 2 Flames 51

^ propagating flux ratio, the portion of the reaction

intensity that heats adjacent fuel particles to ignition
7] heat transfer coefficient W m~^ K~^

DiMENSIONLESS GROUPS

DjG critical Da number for ignition

DEX critical Da number for extinction
Fr Froude number
M Mach number
Pe Peclet number
Re Reynolds number

CONSTANTS

a Stefan-Boltzmann constant
Rn universal gas constant

REFERENCES
Albini, F. A. (1979). "Spot Fire Distance from Burning Trees—A Predictive Model." General Tech-
nical Report INT 56, Intermountain Forest and Range Experiment Station, Forest Service,
USDA, Ogden.
Albini, F. A. (1981). A model for the wind-blown flame from a line fire. Combustion and Flame 43,
155-174.
Albini, F. A. (1986). Wildland fire spread by radiation—A model including fuel cooling by natural
convection. Combustion Science and Technology 45, 101-113.
Arakawa, A., Saito, K., and Gruver, W. A. (1993). Automated infrared imaging temperature measure-
ment: With application to upward flame spread studies. Combustion and Flame 92, 222-230.
Babrauskas, V. (1988). "Handbook of Fire Protection Engineers." SFPE, Boston.
Ban, H., Venkatesh, S., and Saito, K. (1994). Convection-diffusion controlled micro-diffusion
flames. J. Heat Transfer 116, 954-959.
Benson, S. W. (1960). "The Foundations of Chemical Kinetics." McGraw-Hill, New York.
Blinov, V. 1., and Khudyakov, G. N. (1957). Certain laws governing diffusion flames in liquids.
Academiia, Nauk, USSR, Daklady 113, 1094-1098.
Byram, G. M., and Nelson, R. M.,Jr. (1970). The modeling of pulsating fires. Fire Tech. 6,102-110.
Carrier, G. F., Fendefl, F. E., and Wolff, M. F. (1991). Wind-aided fire spread across arrays of dis-
crete fuel elements. I. Theory. Combust. Sci. Tech. 75, 31-51.
Clark, T. L., Coen, J. L., Radke, C. L., Reeder, M., and Packham, D. (1998). Coupled atmosphere-
fire dynamics. In "Third International Conference of Forest Fire Research," pp. 67-82. Coim-
bra, Portugal.
52 K. Saito

Clark, T. L., Radke, L., Coen, J., and Middleton, D. (1999). Analysis of small-scale convective dy-
namics in a crown fire using infrared video camera imagery. J. Appl. Meteorol. 38, 1401-1420.
Delichatsios, M. A., and Saito, K. (1991). Upward fire spread: Key flammability properties, simi-
larity solutions and flammability indices. In "Fire Safety Science—Proceedings of Third Inter-
national Symposium" (G. Cox and B. Langford, Eds.), pp. 217-226. Elsevier Applied Science,
London.
Defichatsios, M. A., Delichatsios, M. M., Lougheed, G. D., Crampton, G. P., Qian, C , Ishida, H.,
and Saito, K. (1995). Effect of external radiant heat flux on upward fire spread: Measurements
on plywood and numerical predictions. In "Fire Safety Science—Proceedings of the Fourth In-
ternational Symposium" (T. Kashiwagi, Ed.), pp. 421-432. International Association for Fire
Safety Science, Washington.
Demtroder, W. (1982). "Laser Spectroscopy." Springer-Verlag, New York.
Dobbins, R. A., and Megaridis, C. M. (1987). Morphology of flame-generated soot as determined
by thermophoretic sampling. Langmuir 3, 254-259.
Emori, R. 1., and Schuring, D.J. (1977). "Scale Models in Engineering." Pergamon Press, London.
Emori, R. L, and Saito, K. (1982). Model experiments of hazardous forest fire whirl. Fire Tech. 18,
319-327.
Emori, R. 1., and Saito, K. (1983). A study of scaling laws in pool and crib fires. Combust. Sci. Tech.
31, 217-230.
Faraday, M. (1993). "The Chemical History of a Candle." Cherokee Pub., Atlanta.
Ferguson, E. S. (1993). "Engineering and the Mind's Eye." The MIT Press, Cambridge.
Fernandez-Pello, A. C. (1995). The solid phase. In "Combustion Fundamentals of Fire" (G. Cox,
Ed.), Chapter 2, pp. 31-100. Academic Press, San Diego.
Fristrom, R. M.,and Westenberg, A. A. (1985). "Flame Structure," 2nd ed. McGraw-Hill, New York.
Gaydon, A. G., and Wolfhard, H. G. (1979). "Flames: Their Structure, Radiation and Temperature,"
4th ed. Chapman Hall, London.
Glassman, I. (1996). "Combustion," 3rd ed. Academic Press, San Diego.
Hirano, T. (1986). "Combustion Physics." Kaibundo, Tokyo. (In Japanese.)
Hirano, T., and Saito, K. (1994). Flame spread phenomena: The role of observation in experiments.
Progr. Energy Combust. Sci. 20, 461-485.
Ito, A., and Kashiwagi, T. (1988). Characterization of flame spread over PMMA using holographic
interferometry: Sample orientation effects. Combustion and Flame 71, 189-204.
Ito, A., Konishi, T., Narumi, A., Tashtoush, G., Saito, K., and Cremers, C.J. (1999). The measure-
ment of transient two-dimensional profiles of velocity and fuel concentration over liquids.
J. Heat Transfer 121, 413-419.
JANAR (1985). "Thermochemical Tables," 3rd ed. (M. W. Chase, Jr., et al, Eds.). J. Physical and
Chemical Reference Data.
Johnstone, R. E., and Thring, M. W. (1957). "Pilot Plants, Models, and Scale-up Methods in Chemi-
cal Engineering." McGraw-Hill, New York.
Kashiwagi, T. (1994). Polymer combustion and flammability—Role of condensed phase. In
"Twenty-fifth International Symposium on Combustion," pp. 1423-1437. The Combustion
Institute, Pittsburgh.
Koseki, H., Iwata, Y., Natsume, Y., Takahashi, T., and Hirano, T. (1999). Tomakomai large scale
crude oil fire experiments. Fire Tech. 36, 24-38.
Koylu, U. O., and Faeth, G. M. (1991). Carbon monoxide and soot emission from liquid-fueled
buoyant turbulent diffusion flames. Combust. Flame 87, 61-76.
Linan, A., and Williams, F. A. (1993). "Fundamental Aspects of Combustion." Oxford University
Press, New York.
Long, R. T., Jr., Torero, J. L., Quintiere, J. G., and Fernandez-Pello, A. C. (2000). Scale and trans-
Chapter 2 Flames 53

port considerations on piloted ignition of PMMA. In "Fire Safety Science—Proceedings of Sixth

International Symposium." The International Association for Fire Safety Science, Washington.
Majidi, V., Saito, K. Gordon, A. S., and Williams, F. A. (1999). Laser-desorption time-of-flight
mass-spectrometer analysis of soot from various hydrocarbon fuels. Combust. Sci. Tech. 145,
37-56.
McDonough, J. M., Garzon, V. E., and Saito, K. (1998). "A Porous Medium Model of Turbulent
Wind-driven Wildland Fires." Department of Mechanical Engineering Internal Report, Univer-
sity of Kentucky, Lexington, KY.
Mulholland, G., Liggett, W, and Koseki, H. (1996). The effect of pool diameter on the properties
of smoke produced by crude oil fires. In "Twenty-sixth International Symposium on Combus-
tion," pp. 1445-1452. The Combustion Institute, Pittsburgh.
Orloff, L., de Ris, J. N., and Markstein, G. H. (1975). Upward turbulent flame spread and burning
of fuel surface. In "Fifteenth International Symposium on Combustion," pp. 183-192. The
Combustion Institute, Pittsburgh.
Porscht, R. (1975). Studies on characteristic fluctuations of the flame radiation. Comhust. Sci. Tech.
10,73-76.
Pyne, S. J., Andrews, P. L., and Laven, R. D. (1996). "Introduction to Wildland Fire," 2nd ed. John
Wiley & Sons, New York.
Quintiere, J. G. (1988). "Handbook of Fire Protection Engineers." SFPE, Boston.
Quintiere, J. G. (1998). "Principles of Fire Behavior." Delmar Pub., Albany, NY.
Radke, L. F., Clark, T. L., Coen, J. L., Walter, C , Riggan, R J., Brass, J., and Higgans, R. (1999). Air-
borne remote sensors look at wildland fires. In "Fourth International Airborne Remote Sensing
Conference and Exhibition/ 21st Canadian Symposium on Remote Sensing." Ottawa.
Rashbash, D. J. (1962). The extinction of fires by water sprays. Fire Res. Ahstr. Rev. 4, 28-53.
Rothermel, R. C. (1972). "Mathematical Model for Predicting Fire Spread in Wildland Fuels." Res.
Pap. INT-115, Intermountain Forest and Range Experiment Station, Forest Service, USDA,
Ogden.
Sabersky, R. H., Acosta, A. J., and Hauptman, E. G. (1989). "Fluid Flow," 3rd ed. Macmillan,
New York.
Saito, K., Quintiere, J. G., and Williams, F. A. (1985). Upward turbulent flame spread. In "Fire
Safety Science—Proceedings of the First International Symposium," pp. 75-86. Hemisphere
Pub., Washington.
Saito, K., Williams, F. A., and Gordon, A. S. (1986). Structure of laminar co-flow methane-air dif-
fusion flames. J. Heat Transfer 108, 640-648.
Saito, K., Gordon, A. S., and Williams, F. A. (1987). A study of two-color soot zone for small hy-
drocarbon diffusion flames. Combust. Sci. Tech. 51, 291-312.
Saito, K., Wichman, I. S., Quintiere, J. G., and Williams, F. A. (1989). Upward turbulent flame
spread on wood under external radiation. J. Heat Transfer 111, 438-445.
Saito, K., Gordon, A. S., Williams, F. A., and Stickle, F. A. (1991). A study of the early history of
soot formation in various hydrocarbon diffusion flames. Combust. Sci. Tech. 80, 103-119.
Soma, K., and Saito, K. (1991). Reconstruction of fire whirls using scale models. Combust. Flame
86, 269-284.
Takahashi, E, Vangsness, M. D., Durbin, M. D., and Schmoll, W J. (1996). Structure of turbulent
hydrogen jet diffusion flames with or without swirl. J. Heat Transfer 118, 877-884.
Tewarson, A., and Ogden, S. D. (1992). Fire behavior of polymethylmethacrylate. Combustion and
Flame 89, 237-259.
Venkatesh, S., Ito, A., Saito, K., and Wichman, I. S. (1996). Flame-base structure of smafl scale pool
fires. In "Twenty-Sixth International Symposium on Combustion," pp. 1437-1443. The Com-
bustion Institute, Pittsburgh.
54 K. Saito

Wichman, I. S. (1992). Theory of opposed-flow flame spread. Progr. Energy Combust. Sci. 18, 5 5 3 -
593.
Williams, F. A. (1969). Scaling mass fires. Fire Res. Ahstr. Rev. 11, 1-23.
Williams, F. A. (1976). Mechanisms of flame spread. In "Sixteenth International Symposium on
Combustion," pp. 1281-1294. The Combustion Institute, Pittsburgh.
Williams, F. A. (1982). Urban and wildland fire phenomenology. Progr. Energy Combust. Sci. 8,
317-354.
Williams, F. A. (1985). "Combustion Theory." Benjamin/Cummings, San Francisco.
Williams, F. A. (1992). The role of theory in combustion science. In "Twenty-Fourth International
Symposium on Combustion," pp. 1-17. The Combustion Institute, Pittsburgh.
Williams, J. M., and Gritzo, L. A. (1998). In situ sampling and transmission electron microscope
analysis of soot in the flame zone of large pool fires. In "Twenty-Seventh International Sympo-
sium on Combustion," pp. 2707-2714. The Combustion Institute, Pittsburgh.
Wilson, E. B., Jr. (1954). "An Introduction to Scientific Research." McGraw-HiU, New York.
Wolff, M. F., Carrier, G. F., and Fendell, F. E. (1991). Wind-aided fire spread across arrays of dis-
crete fuel elements. 11. Experiments. Combust. Sci. Tech. 77, 261-289.
Zabetakis, M. G. (1965). "Flammability Characteristics of Combustible Gases and Vapors." U.S.
Bureau of Mines, Bulletin #627, Washington.
 CHAPTER 3
Combustion Chemistry
and Smoke
D. WARD
Fire Sciences Laboratory, Rocky Mountain Research Station, USDA Forest Service,
Missoula, Montana

I. Introduction
II. Fuel Chemistry and Combustion
A. Fuel
B. Combustion Chemistry
C. Pyrolysis
D. Phases of Combustion
III. Smoke Production
A. Release of Carbon
B. Formation of Particles
C. Emission of Trace Gases
IV. Minimizing Smoke Production
V. Conclusions
References

I. INTRODUCTION

Fires burning in vegetative fuels, whether prescribed or wildfires, emit a com-

plex mixture of particles and gases into the atmosphere. The diversity in com-
position of combustion products results from wide ranges in fuel types, fuel
chemistry, and fire behavior. This chapter discusses the state of knowledge con-
cerning the effect of these variables on the characteristics of smoke. In addition,
the chemical and physical characteristics of the vegetation affect the rate of com-
bustion and influence the overall fire behavior.
The need to bring fire back into ecosystems that, historically, have been de-
pendent on fire has created a great deal of concern over the smoke released and
the effect of the smoke on air quality values in both urban and nonurban set-
tings. This has resulted in a need to develop methods for estimating the rate of
release of smoke and for methods of burning to minimize smoke production
56 D. Ward

consistent with achieving goals of the managers. Air Quahty Regulations have
been the main drivers for improving guidelines needed for reducing smoke
emissions and managing the dispersion of smoke in the atmosphere. Many of
the principles discussed in detail in this chapter have direct application in re-
ducing smoke from wildland fires or understanding the toxicity of the smoke
to humans.
Relatively little biomass is burned in the United States in comparison to the
rest of the world. The total biomass consumed globally is currently estimated
to be on the order of 5 - 8 Pg (1 Pg equals 10^^ g) per year (Crutzen and Andreae,
1990). In comparison, 1988 was one of the most extreme "fire years" in recent
history in the United States with almost 2 million ha of land burned by wild-
fires, including the large fires of Yellowstone National Park. If we consider the
average fuel consumption per ha to be 45 t (1 metric tonne is 10^ g), the total
fuel consumed by wildfires in the United States in 1988 would have been —90 Tg
(1 Tg is 10^^ g). In addition, Ward et a\. (1993) estimated that prescribed fires
for 1989 consumed about 40 Tg of fuel in the United States. The total of ~ 130 Tg
of fuel consumed per year is —2% of the total burned globally. It is interesting
that more than 80% of all biomass burning takes place in tropical countries
( H a o a n d L i u , 1994).
The discussion in this chapter will focus on the chemistry of natural eco-
system fuels important from an emissions production standpoint; then com-
bustion processes will be discussed; and finally a general discussion of the
chemistry of smoke from fires in natural vegetation fuels will be presented. Pro-
cess level mechanistic approaches have not been as useful for explaining the
range and mix of emissions resulting from the burning of biomass in the open
environment as has proven beneficial for studying the mix of emissions from
the combustion of hydrocarbon and coal (Glassman, 1977). For wildland fires,
the chemistry of fuels is very complex, the distribution of fuel elements is ran-
dom but highly variable, and weather influences are nonlinear and exception-
ally difficult to quantify with moisture gradients occurring throughout the fuel
complex. Outstanding use has been made of controlled environment combus-
tion laboratory experiments to simulate field conditions (Yokelson et at, 1996).
Correlation and linear regression modeling techniques have been used almost
exclusively in establishing the influence of independent variables on smoke
production. Generally, approaches used in studying smoke emissions can be
broken into four levels of investigation based on scale and control of variables.
These are:

1. Highly controlled microcombustion or pyrolysis devices (Ward, 1979;

Clements and McMahon, 1980; Susott, 1980),
2. Laboratory experiments (Lobert et al, 1991; Reinhardt and Ward, 1995;
Yokelson et al., 1997),
Chapter 3 Combustion Chemistry and Smoke 57

3. Field-based tower experiments, generally used for studying smoke

emissions from small plots (Ward and Hardy, 1991; Ward et al, 1992;
Hoffaetal., 1999), and
4. Airborne studies of full-scale fires burning under ambient atmospheric
conditions (Radke et al, 1991; Nance et al, 1993; Reid et al, 1998;
Yokelson et al., 1999).

The characterizations of fuels and resulting emissions discussed in the fol-

lowing sections draws heavily from all four scales of research.

II. FUEL CHEMISTRY AND COMBUSTION

A. FUEL

Plant material consists of polymeric organic compounds, generally described by

the chemical formula C^HgO^ (Byram, 1959), although we know that this gen-
eralized chemical formula varies depending on whether we are interested in the
carbon content of grass or wood or material that is undergoing decay. Regard-
less, plant tissue is approximately 50% carbon, 44% oxygen, and 5% hydrogen
by weight. The content of most wood varies between 4 1 % and 53% cellulose, be-
tween 15% and 25% hemicellulose, and between 16% and 3 3 % lignin (Brow-
ning, 1963). The lignin content of wood is much higher (up to 65%) in decay-
ing (punky) wood, in which the cell wall polysaccharides are partially removed
by biological degradation.
Dead plant material accumulates on the forest floor or in lakes and bogs. This
material is rich in nutrients and is transformed by microorganisms into humus
in which individual plant parts are no longer identifiable. In the process, sulfur
and nitrogen accumulate in the organic matter because many of the microor-
ganisms require these elements for growth and reproduction. In addition, tree
needles and leaves generally have a higher composition of nitrogen and sulfur
than woody stems and limbs. Allen (1974) reported sulfur content values be-
tween 0.08% and 0.5% for plant material and 0.03% and 0.4% for organic soils.
Nitrogen is one of the most dominant of the macronutrients and is of primary
concern because of the large number of nitrogen-based compounds produced
when biomass is burned. Nitrogen makes up as much as 0.2% of the older wood
of some species, 1% of needles of some pines, and up to 2.7% of fallen hard-
wood leaves (Clements and McMahon, 1980). We will discuss the nitrogenous
products of combustion in Section III.C under the topic of emissions of trace
gases (Ward et al, 1996).
58 D. Ward

B. COMBUSTION CHEMISTRY

The generalized representation of biomass (or fuel) does not explain the diver-
sity of compounds produced in smoke as a result of burning the material. Com^
bustion efficiency is a term used to describe the overall conversion of carbon to
carbon dioxide (Ward and Hardy, 1991). Byram (1959) illustrated the case for
complete combustion of plant fuel with the following chemically balanced oxi-
dation reaction:

4C6H9O4 + 25O2 + [O.322MH2O + 94.ON2] ...

^ 1 8 H 2 0 + 24CO2 + [O.322MH2O + 94.ON2] + 11.6 X 10^ J ^^
Moisture in the fuel and water and nitrogen in the air are shown as bracketed
quantities because they do not take part in the combustion reaction. It should
be noted that the combustion of plant material releases the moisture contained
with the fuel shown in brackets for this case and the water of combustion which
is 18 kg moles of water for each 4 kg moles of plant material [(18*18 g)/(4 * 145)
= 0.559]. For each kg of plant material (dry weight basis), 0.559 kg of water is
released. Therefore, if one considers the moisture content of the fuel, it is read-
ily seen that the rate of release of water is equal to the water of combustion plus
the contained moisture (0.559 + 0.01 M). It should further be recognized from
Eq. (1) that 4 kg moles of plant material release 11.6 X 10^ J of heat energy. If
we divide this heat release by the molecular weight and number of kg-mol of
fuel, the high heat of combustion can be derived as 20,000 kj/kg. The high heat
of combustion is defined as the heat released from a unit mass of fuel without
accounting for the heat required to vaporize the moisture of combustion or
moisture content of the fuel (see Byram (1959) for a more complete treatment).
The combustion of plant material from open fires is seldom if ever 100% effi-
cient; hence, products of incomplete combustion are produced that are of a ma-
jor concern from an air pollution standpoint. Carbon monoxide is the major
carbonaceous product of incomplete combustion, whereas carbon dioxide and
water are the major products of complete combustion. Carbon dioxide contains
as much as 95% of the carbon released in smoke from biomass fires where the
combustion efficiency is high (e.g., grassland fires). On the other hand, the sum
of carbon in other compounds of incomplete combustion makes up less than 5%
(Figure 1). For fires of lower combustion efficiency, the percentage of carbon re-
leased as carbon dioxide would be lower and the products of incomplete com-
bustion higher than shown in Figure 1. Minor constituents such as nitrogen,
phosphorus, and sulfur affect the mix of pollutants generated by burning plant
material. Other factors are important as well and contribute to the diversity
of combustion products. For example, most plant material contains classes of
compounds known as extractables, consisting of aliphatic and aromatic hydro-
Chapter 3 Combustion Chemistry and Smoke 59

88.0%
80-

60-
C
<D
O 40-

20-
9.5%
1.5% 0.5% 0.5%
0- . '' ^ 'J ' ' 1 1 '
CO CO, CH,
Emission component

FIGURE 1 The average distribution of carbon between the primary products of combustion for
broadcast burns of logging slash in the Pacific Northwest in the United States.

carbons, alcohols, aldehydes, gums, and sugars. These extractables, as a group,

have a higher heat value than cellulose, lignin, and hemicellulosic substances.
Other extractables from forest fuels contain a complex mixture of terpenes, fats,
waxes, and oils. Not all of the extractable substances have low boiling points.
For example, some of the extractives were shown by Shafizadeh et at. (1977) to
evaporate or pyrolyze only at temperatures above 300°C.

C. PYROLYSIS

Pyrolysis is defined as the chemical breakdown of solid fuel under the influ-
ence of heat and usually in an oxygen-deficient environment. The pyrolytic
decomposition of cellulose is generally believed to follow one of two paths
dependent on whether the pyrolysis is occurring under high-temperature or
low-temperature conditions. Usually under low-temperature conditions (200-
280°C), the cellulose undergoes dehydration with the evolution of char, H2O,
CO2, CO, and other compounds. Under higher temperature conditions (280-
340°C), the pyrolysis proceeds in the production of levoglucosan, a volatile fuel
that supports a gas-phase flame (Kilzer and Broido, 1965).
Thermal gravimetric analysis methods have been used by Susott (1980) and
Susott et at. (1979) to evaluate the evolution of pyrolysis gases from solid fuels
as a function of temperature during heating. The evolution of pyrolysis gas us-
ing this technique exhibits a spectrum reflecting the thermal stability of the fuel
components as shown in Figures 2 and 3. Fach component released can have a
60 D. Ward

0.24

0.20
o E

= _ 0.16H
_cg CO

o Bo 0.12

0.08 H

0.04 H

100 200 300 400

Temperature (°C)
FIGURE 2 Different classes of forest fuels tend to volatilize at different rates as a function of
temperature. The rates of volatilization were determined using thermal gravimetric measurement
techniques.

different molecular weight and chemical form which can have significant im-
plications regarding the formation of emissions. These materials pass through
the flame structure or are released directly into the atmosphere. Oxidation may
or may not occur at the solid fuel interface. For woody biomass, most research-
ers have studied the decomposition of cellulose and hemicellulose. Little is
known about the thermal decomposition of bark, of which suberin—a poly-
mer of long-chain hydroxy fatty acids esterified to phenolic acids—is one sub-
stance. The degradation products in the 425°C peak form a sizable fraction, but
the chemical content may be different than for the woody carbohydrate prod-
ucts (Figure 2).

D. PHASES OF COMBUSTION

Preheating, flaming, smoldering, and glowing combustion compete for available

fuel (Figure 3) and are markedly different phenomena that contribute to the di-
versity of combustion products. The fuel characteristics (including arrange-
ment, distribution by size class, moisture, and chemistry) dominate in affecting
the duration of flaming and smoldering combustion phases and combustion effi-
ciency. Open combustion occurs through a diffusion flame process in which the
fuel from the interior of the flame (oxygen-deficient area) diffuses outward, and
the oxygen from the free-air diffuses inward. This results in a narrow envelope
Chapter 3 Combustion Chemistry and Smoke 61

Unburned fuel

Glowing or_ Flaming • Preheating

smoldering

FIGURE 3 Diffusion flame model of pyrolysis and vaporization of volatiles during the preheating,
flaming, glowing, and smoldering phases of combustion. The more highly volatile oleoresins are
evaporated from the surface fuels in advance of the flame front for an advancing fire moving from
left to right. The diagram above is a cross-section through a line of fire with the line of fire extend-
ing through the paper. There are several features describing the thermal environment and the ef-
fect of changing heat load on the early evaporation and later pyrolysis of fuel:
1. Evaporation of the highly volatile compounds. No condensation or polymerization is expected
for the compounds released that are not in close proximity to the flame envelope.
lla. Evaporation of the highly volatile compounds increases in this zone and the potential for poly-
mers to form increases as bond rupturing takes place for the terpenes.
lib. Abundant evaporation of oleoresins occurs with partial oxidation due to the intense thermal
environment and extreme bond rupturing resulting from the high-intensity radiant energy
source.
III. Introduction of evaporated oleoresins into a high-temperature oxygen-deficient environment
with considerable oxidation occurring as molecules diffuse through the flame envelope. Those
not undergoing complete oxidation may be fragmented into ethylene units and/or free radicals.
IV. Very high-temperature environment with oxygen depletion occurring as the carbon fragments
reach the tip of the flame envelope. The amount of compound oxidation is dependent on the
depth of the flame envelope and the amount of ventilation (turbulence). As the flames become
taller, the heat feedback to the solid fuels becomes less, and the radiant energy loss becomes
greater. The result is that particle formation/polymerization reactions may be increased be-
cause of the loss of heat within the fuel-rich zone.
V. Recombination takes place with the formation of compounds not found during the evapora-
tion phase or inside the flame envelope due to pyrolysis. Aromatic hydrocarbon molecules are
synthesized during this phase of transport.
VI. Products of pyrolysis and glowing combustion are transported across the flame surface.
VII. Transport of products of pyrolysis and glowing combustion completely miss the flame enve-
lope and enter the atmosphere with no additional oxidation.
62 D. Ward

around the fuel-rich zone where the oxygen and fuel are mixed in stoichio-
metric proportions to produce rapid oxidation chemical reactions which result
in the visible emission of light called a flame. The chemical reactions also cause
the rapid liberation of heat which feeds back to the interior of the flame enve-
lope and ahead of the flame envelope (Figure 3). This causes further vaporiza-
tion of fuels with low vapor pressures and pyrolysis of solid fuels.
Initial heating of unburned fuel releases the more volatile components by
distillation which then leads to pyrolysis and fragmentation of polymers and
the release of oxygenated organic compounds (Yokelson et al, 1996). Flaming
is initiated when the fuel-to-oxygen mixture reaches flammable proportions and
there is a source of ignition. Flaming and smoldering combustion are reasonably
distinct combustion processes that involve different chemical reactions and are
quite different in appearance (see also Chapters 2 and 13 in this book). Flaming
combustion dominates during the startup phase, with the fine fuels and surface
materials supplying the volatile fuel required for the rapid oxidation reactions
to be sustained in a flaming environment. The heat from the flame structure
and the diffusion and turbulent mixing of oxygen at the surface of the solid
fuel promote generation of the heat required to sustain the pyrolysis processes.
Early in the flaming phase, the more volatile hydrocarbons are vaporized from
the fuels. Later the cellulosic and lignin-containing cellular materials decom-
pose through pyrolysis. These processes produce the fuel gases that sustain the
visible flame structure (Figure 4).
Once carbon and ash begin to build up on the solid fuel surfaces, the py-
rolytic reactions no longer produce sufficient fuel gases to maintain the flame
envelope. For combustion to continue, oxygen must diffuse to the surface of the
fuel. Diffusion of oxygen and the availability of oxygen at the fuel surface are
enhanced through turbulence in the combustion zone and through premixing
by oxygen transport at ground level. This allows oxidation to take place at the
solid fuel surface and provides for heat evolution and heat feedback to acceler-
ate the pyrolytic reactions and volatilization of the fuel gases from the solid
fuel. The process ultimately leads to the production of charcoal for which the
only combustion occurring is of the glowing type—a surface reaction of oxy-
gen with carbon.

III. SMOKE PRODUCTION

The smoldering combustion phase produces large amounts of particulate matter
and carbon monoxide. Fires of very low intensity (those in which the flaming
combustion phase is barely sustained) produce proportionately higher amounts
of particulate matter. Heading fires generally are associated with conditions that
produce two to three times as many particles (by weight) as backing fires. The
Chapter 3 Combustion Chemistry and Smoke 63

| - Flaming Combustion -|—Preheating \ B Concentration

Reaction Oxygen Gasified fuel

Envelope /
Particulate matter

Carbon dioxide

Gasified fuel

Horizontal distance
FIGURE 4 Relative concentrations of basic combustion gases with horizontal distance, A-A, and
height, B-B, cross-sectioning a typical diffusion flame resulting from the oxidation of pyrolyzed
forest fuels.

formation of particles results primarily from two processes: (1) the agglomera-
tion of condensed hydrocarbon and tar materials and (2) mechanical processes
which entrain fragments of vegetation and ash.

A. RELEASE OF CARBON

when vegetation is burned, carbon is released in the form of carbon dioxide,

carbon monoxide, hydrocarbons, particulate matter, and other substances in
decreasing abundance (Figure 1). A carbon mass-balance procedure is often
employed by researchers in developing emission factors for different fuel con-
64 D. Ward

ditions (Ward and Hardy, 1984). An emission factor is defined as the mass of a
specific combustion product released per unit dry mass of fuel consumed and
usually is expressed in pounds per ton or grams per kilogram. Emission factors
are highly variable, but generally a large part of the variance can be explained
through using a measurement of combustion efficiency for the independent
variable. The combustion efficiency is a measure of the overall oxidation capac-
ity for the combustion of fuel under a given set of weather and fuel conditions.
It is a ratio of the mass of carbon released in the form of carbon dioxide to the
mass of carbon in the original fuel and ranges from 0.98 for flaming combus-
tion of fully cured grass vegetation with virtually no smoldering combustion to
0.75 for 100% smoldering combustion of deep duff. To simplify the calculation
of combustion efficiency, Ward et al. (1996) adopted the use of modified com-
bustion efficiency (MCE), or the ratio of the carbon released as carbon dioxide
to the sum of the carbon released in the form of carbon dioxide and carbon
monoxide. MCE is illustrated in Figure 5 for a number of different fuel types
and is often used to predict emission factors for different compounds and par-
ticulate matter (Ward and Hardy, 1991; Ward et al, 1996). For example. Ward
et al. (1996) showed that MCE can be predicted for savanna fuels by using the

Fuel type
FIGURE 5 Range in combustion efficiency (the ratio of carbon released as COj to the sum of car-
bon released as COj and CO) for fires burning in seven different fuel types as follows: DF, Douglas
Fir; HW, Hardwood; LP, ponderosa pine and lodgepole pine (long-needled conifers); MC, Mixed
Conifer; TP, Tractor Piles; CP, Crane Piles; PILE, Combined TP and CR
Chapter 3 Combustion Chemistry and Smoke 65

ratio of the mass of standing grass to the total mass of standing grass and htter
using:
MCE = 0.85 4- 0.111 Grass/(Litter + Grass)^^^ R^ = 0.68 (2)
The model was developed for African savanna conditions of completely cured
grass (no green grass). When the grass is lying flat (i.e., grass = 0) the predicted
MCE is 0.85. When the grass is standing and there is no litter, the MCE is
predicted to be 0.96. A summary of algorithms for predicting emission factors
for several vegetation types, including the African savanna ecosystem, will be
provided.

B. FORMATION OF PARTICLES

Wildland fires are a complex form of the diffusion flame process where pyroly-
sis of solid fuels produce vapors that interdiffuse with oxygen from the atmo-
sphere. Turbulent mixing of fuel and oxygen is important, and as the turbu-
lence increases, the flame characteristics and the chemical processes occurring
in the flame zone change. Although this has never been completely substanti-
ated, it is believed that there is a level of fire intensity where MCE reaches its
highest level (perhaps 0.96 to 0.98 for many fuel types) (Ward, 1983) and par-
ticulate matter production is lowest (Nelson and Ward, 1980). For very high-
intensity fires, some of the pyrolyzed fuel may no longer pass through an active
oxidation zone. At times, even in lower intensity fires, pockets of unburned,
partially oxidized gaseous fuels escape the combustion zone or undergo delayed
ignition. The influence of flame turbulence on combustion efficiency is not fully
understood; however, as the intensity of the fire increases and the zone of com-
plete mixing of gaseous fuel and oxygen moves farther from the solid fuel, com-
bustion efficiency is believed to decrease and the production of pollutants to
increase. Because of the increased depth and height of the flame zone, head-
ing fires and area fires create an extended reducing environment in which con-
tinued pyrolysis and synthesis of hydrocarbon gases and fragmented particles
can occur under conditions of reduced oxygen content. In addition, heat re-
radiated from the particles to the atmosphere can slow down the reactions as the
unburned gases and particles are convected away from the active combustion
zone. Combustion of the particles requires prolonged exposure at high tempera-
tures (>800°C) in a zone with near ambient concentrations of oxygen (Glass-
man, 1977). Greater premixing seems to reduce the production of fine particles
(< 2.5 jjim. diameter).
Fuel chemistry plays an important role in the formation and release of par-
ticles to the atmosphere. For example, it is well known that fuels high in oleo-
resins smoke profusely when burned. The pitch has a high terpene content; the
66 D. Ward

terpenes are similar to aromatic compounds in structure (five carbon atom

building blocks instead of six carbon atoms such as occurs in benzene and some
other highly aromatic compounds). These highly volatile compounds have a
low oxygen content and during pyrolysis produce ethylene units which tend to
polymerize and form long-chain compounds which, in turn, condense into tar-
like substances that coat most particles from fires. Ward (1979) found that emis-
sion factors for particulate matter are inversely related to the oxygen content of
the fuel molecules and suggested that cellulose and hemicellulose may produce
less particulate matter when burned (due to their high oxygen content) than the
very low oxygen content oleoresins.
Smoke particles have been measured using sophisticated instruments rang-
ing from Differential Mobility Particle Sizer (DMPS) to methods for sizing par-
ticles based on their aerodynamic properties. These instruments have been used
onboard aircraft to cover the broad range of particle sizes from 0.01 to over
43 fJim (Radke et al, 1991; Reid and Hobbs, 1998). The results suggest a very
pronounced number concentration peak at a diameter of 0.15 /xm. The volume
distribution, when one assumes a unit mass density for smoke particles, repre-
sents the mass distribution and exhibits a bimodal distribution with peaks at
about 0.3 /xm and greater than 43 /xm (Figure 6).
Emission factors have been measured for a number of different combinations
of forest fuel and weather influences (Lobert et al, 1991; Ward and Hardy, 1991;
Nance et al, 1993; Andreae et al, 1996; Ward et al, 1996; Ferek et al, 1998;

c
o

Particle diameter, D (|im)

FIGURE 6 Bimodal distribution of particles on a volume basis. The percentage of particles larger
than 10 fim typically increases for the flaming phase of combustion due to the turbulence gener-
ated in the combustion zone and the lofting of ash and partially burned biomass fragments.
Chapter 3 Combustion Chemistry and Smoke 67

40 60 80 100 120 140 160

Rate of heat release (kW/m")
FIGURE 7 Higher intensity fires entrain additional large particles relative to fine particles.

Hoffa et al, 1999; Yokelson et al, 1999). Empirical data suggest that emission
factors for particulate matter range from 4 to more than 40 g/kg for particles less
than 10 fim in diameter (EFPMIO); for particles without regard to size (EFPM),
the range may be 20% larger, and emission factors for particles less than 2.5 /jun
diameter (EFPM2.5) are typically 10% smaller than EFPMIO. However, the dif-
ference between EFPM and EFPM2.5 is highly dependent on fire intensity as
illustrated by Figure 7. For savanna grassland ecosystems, Ward et al. (1996)
found that EFPM2.5 can be predicted from the following equation:
EFPM2.5 - 87.65 - 88.51 MCE R' = 0.56 (3)
The content of particulate matter varies between flaming and smoldering
combustion. Between 1 and 10% of a particle's mass may consist of trace ele-
ments of potassium, chlorine, sulfur, phosphorus, and sodium. On a percentage
basis, the mass of trace elements contained in particles from smoldering com-
bustion is 10-20% of that from the flaming phase. Figure 8 for logging slash
fires in the Pacific Northwest illustrates this dramatic difference between flam-
ing and smoldering combustion (Ward and Hardy, 1989).
Emissions of graphitic and organic carbon are especially important because
of the increased absorption of light by smoke particles that are high in graphitic
carbon content. Absorption of light is due primarily to the black carbon con-
tent of the particles (Patterson et al, 1986). For logging slash fires of the Pacific
Northwest, emission factors for graphitic carbon ranged from 0.46 to 1.18 g/kg
of fuel consumed. For fuel beds burned under laboratory conditions, the emis-
sion factors ranged to 5.4 g/kg of fuel consumed. Though graphitic or black
carbon is produced proportional to the intensity of the fire, it is generally true
68 D. Ward

2.80

2.40 j j l j l Flaming
EZH Smoldering
§ 2.00

1.60

i 1-20

B 0.80
c
o
O
0.40

0.00 Qi r-f-ir-, r^ r^Pp, , n n-^rb^rn

Na Al Si P S CI K Ca Ti V Cr Mn Fe Ni Cu Zn Br Ag Cd Sn Pb

Elements
FIGURE 8 Difference in percent composition of particles for flaming and smoldering combustion
reflecting the difference in ash content of the fine particles.

that emission factors for PM2.5 are inversely proportional to higher intensity
burns. The organic fraction of the particles smaller than 2.5 /nm is as much as
50 - 7 0 % of the mass of the particulate matter for the smoldering phase, but can
be lower for the flaming phase emissions.
Reid et al. (1998) conducted a very intensive study of "young" smoke (less
than 2 hours old) in comparison to "aged" smoke (from 2 to 4 days old) from
biomass fires in the very humid Brazilian Amazon region. Among other find-
ings, they hypothesized that hydrocarbons of fewer than 11 carbon atoms were
depleted over time and converted to CO2, CO, and reactive molecular species
and likely removed through dry deposition and/or by conversion to particulate
matter. Although somewhat contradictory, the argument concludes that indi-
vidual particle mass increased over time, and it was estimated that 2 0 - 4 5 % of
the mass concentration of the particles was due to the condensation of organics.
It is interesting to note that most of the active "new" fires are lit between 1:00
and 3:00 P.M. local time and active flaming is nearly complete in 3 to 4 hours
with smoldering combustion then dominating. Smoldering combustion contin-
ues for several hours to several days (Ward et at, 1992). The organic content of
particles produced during smoldering combustion of biomass is approximately
20% higher than the organic content of particles produced during the flaming
combustion phase (Ward and Hardy, 1988) (likely dominating the emissions
for the "young" fire plumes of Reid et al, 1998). There are still many issues to
be resolved before it can be conclusively demonstrated that condensation of hy-
Chapter 3 Combustion Chemistry and Smoke 69

drocarbons occurs at a higher rate than volatiUzation in contributing to the

mass of particulate matter once the smoke is more than a few minutes old.

C. EMISSION OF TRACE GASES

In this section, fuel chemistry and combustion efficiency effects are discussed
as they affect the production of trace gases from combustion sources. In par-
ticular, studies are reviewed that have examined the production of gases con-
taining nitrogen, sulfur, and chlorine. Other gases produced, such as methane,
nonmethane hydrocarbons (NMHC) and oxygenated hydrocarbons, also are
discussed in this section.

1. Nitrogen Gases
Generally, temperatures within flame structures of vegetation fires do not ex-
ceed 1000°C, which suggests that molecular nitrogen gas (N2) from the atmo-
sphere is not dissociated to combine with free radicals within the combustion
zone to form oxides of nitrogen (NO^). Several studies have reported the pro-
duction of NO^ from the burning of biomass.
Evans et at. (1977) measured NO^ in forest fire plumes in Australia. Their
findings showed NO^ concentrations as high as 0.024 ppm. The conclusion was
that NO^ is produced proportional to fire intensity. The same relationship has
not been found for fires in the United States.
Results of Clements and McMahon (1980) suggest that the production of
NO^ strongly increases with the nitrogen content of the fuel burned. Conver-
sion of fuel-bound nitrogen to NO^^ can occur readily in oxygen-depleted air.
Simple linear regression equations were developed with coefficients of deter-
mination greater than 0.70 for predicting the production of NO, NO^, NO2, as
functions of the nitrogen content of the fuel. The quantity of fuel nitrogen con-
verted to NO^ was found to range from 6.1 to 41.7% for wood and organic soil,
respectively.
Proximate analysis for the nitrogen content is an important measurement to
make in interpreting differences between NO^ emissions for different ecosys-
tems. The emissions of N2O from combustion of forest fuels showed a small
variance for two extreme fuel types—the boreal forest of Canada and the chap-
arral fuels of California. However, for some measurements of NO^ the differ-
ence between boreal and chaparral NO^ emissions was quite large (Hegg et al,
1989).
Hegg et al (1989) measured emission factors for NH3 of 0.1 to 2.0, for NO^
of 0.81 to 8.9, and for N2O of 0.16 to 0.41 g/kg of fuel consumed. Nitrogen loss
from the biological site was significant. In addition, a fraction of the nitrogen is
contained within the particulate matter. NH4 is contained within the particles
and may contribute as much as 0.05 g/kg of fuel consumed.
70 D. Ward

Volatilization and release of nitrogen to the atmosphere can be significant on

a total mass basis per unit area burned. Yokelson et al. (1996) demonstrated a
relationship between the NH3/NO^ ratio and MCE where NO^ dominates the
emissions during flaming combustion and NH3 dominates during smoldering
combustion as follows:

NH3/NO^ = 14(1 - MCE) R^ = 0.96 (4)

The total of NH3 and NO^ accounts for approximately 50% of the fuel nitrogen.
Therefore, if the nitrogen content of the fuel is known and the MCE for the
combustion is known, then the emissions of NH3 and NO^ can be estimated.
Crutzen et al. (1985) found an average ratio of N2O to COj of 1.5 X 10"^ for
tropical biomass burning. If one sums the carbon released as CO, CO2, CH4,
and NMHC and assumes an emission rate of 1% for particulate matter (Ward
and Hardy, 1991), the conversion of biomass to COj is 77.8% efficient. This
compares favorably with average rates of combustion efficiency for logging slash
fires of 82.8% in the western United States (Ward and Hardy, 1984). The emis-
sion factor for N2O was computed from Crutzen et al (1985), and the result
is 0.22 to 0.44 g NjO per kilogram of fuel consumed. The emission factor for
burning of tropical biomass is quite similar to that for boreal, chaparral, and
coniferous biomass in North America as measured by Hegg et al. (1989). Ozone
is not a byproduct of biomass combustion but forms as a product of secondary
chemical reactions once the combustion products enter the atmosphere.

2. Sulfur Emissions (Carbonyl Sulfide)

Along with nitrogen, sulfur is one of the essential nutrients required in the
synthesis of plant amino acids and other physiologically important substances
(Grier, 1975; Tiedemann, 1987). Hence, the volatilization and loss of these im-
portant nutrients is of extreme interest in sustaining the productivity of eco-
systems. Nitrogen can be replaced through symbiotic N-fixation, whereas sul-
fur is replenished mainly through atmospheric deposition. Very little work has
been done in identifying the form of the sulfur- or nitrogen-containing emis-
sions released during the combustion of biomass fuels.
Crutzen et al. (1985) measured carbonyl sulfide (COS) emission ratios to
CO2 of approximately 4 to 8 X 10~^ (5.71 to 11.42 mg/kg of fuel consumed).
Ward et al. (1982) measured emission factors for COS of 0.18 to 2.36 mg/kg
of fuel consumed from controlled experiments in a combustion laboratory-
burning hood facility. The experiments were for fuels of varying sulfur content
ranging from 0.55% (by weight) for organic soil to a low of 0.065% for pine
needles. Ward et al. (1982) calculated emission factors from Crutzen et al.
(1979) of 16.2 and 35.3 mg/kg of fuel consumed for fires in the Rocky Moun-
tain area of the United States. The results of Ward et al. (1982) can be used to
Chapter 3 Combustion Chemistry and Smoke 71

predict emission factors for COS as a function of the rate of heat release per unit
area (JR, kW/m^) as follows: EF^os "^ 0.732 — 0.0065 IR where the coefficient
of determination is 0.99. This would suggest that the production of COS is in-
dependent of the sulfur content of the fuel—a result that probably is not valid.
However, it may be concluded that COS production is extremely sensitive to the
thermal environment reflected through the apparent dependence on J^^. Other
sulfur-containing compounds were measured as well including H2S, (CH3)2S,
CS2, CH3SSCH3, and other u n k n o w n mercaptan compounds. The sulfur quan-
tified made up less than 0.25% of the total sulfur released during the combus-
tion experiments. The balance of the sulfur may be partially explained by mea-
surements of Yokelson et at. (1996) who estimate the molar ratio of SO2 emitted
to that of CO2 as (7.3 ! 3.0) X 10""^. As with nitrogen compounds, the sulfur
content of the fuel is highly variable and becomes the dominant characteristic
in estimating the release of sulfur compounds.

3. Methyl Chloride

Methyl chloride has been suggested as a natural tracer unique to the combus-
tion of biomass fuels (Khalil et al, 1985) and seems to work well for apportion-
ing the impact of residential wood combustion. For open burning. Ward (1986)
found that methyl chloride is produced in much greater quantities in the smol-
dering combustion phase than in the flaming phase. Thus, unless the actual ra-
tio of flaming to smoldering is known for smoke impacting a site, the appor-
tionment may be in error.
Ward (1986) found an inverse relation between chlorine content of fine par-
ticles and the amount of methyl chloride released and that emission factors for
methyl chloride are inversely proportional to the rate of heat release. For chap-
arral fuels, methyl chloride emission factors ranged from 16 to 47 mg/kg of fuel
consumed (Ward and Hardy, 1989). Reinhardt and Ward (1995) found that
most, if not all, of the methyl chloride is produced from the smoldering com-
bustion process.

4. Carbon Monoxide
Carbon monoxide is the second most abundant carbon-containing gas released
during the combustion of biomass (Figure 1). Combustion efficiency and MCE
are nearly perfectly correlated with the ratio of the production of carbon mon-
oxide relative to carbon dioxide. Ward and Hardy (1991) found particulate
matter concentration to be strongly correlated with carbon monoxide concen-
tration (r = 0.89). Reinhardt and Ward (1995) found the concentration of for-
maldehyde to be correlated with the concentration of carbon monoxide (r =
0.93). Generally, emission factors for carbon monoxide on a mass basis are ten
72 D. Ward

times greater than for the fine particle fraction. Emission factors for carbon
monoxide range from 40 to over 300 g/kg of fuel consumed. Ward et al. (1996)
found for fires in savanna grassland ecosystems that the following relationship
can be used for computing emission factors for carbon monoxide (EFCO):

EFCO = 1834(1 - MCE) * 0.64 (5)

5. Methane and Nonmethane Hydrocarbons

Methane is produced in much larger quantities during the smoldering com-
bustion phase than in the flaming phase (Ward and Hardy, 1991). Emission fac-
tors are about two to three times greater for the smoldering phase than for the
flaming phase. For one study, however, they ranged from 5.7 to 19.4 g/kg of
fuel consumed for smoldering combustion and the flaming phase emission fac-
tors ranged from 1 to 4.2 g/kg. The volatile hydrocarbons are well correlated
with methane, and emission factors for the hydrocarbons are well correlated
with MCE. Ward et al (1992) and Hao and Ward (1993) demonstrated that lin-
ear relationships between CH4 emission factors and MCE are quite different for
forest and savanna fires. In addition to the different slopes of the linear regres-
sion models, typical MCE values were found to be 0.84 for slash fires in primary
forests of Brazil, 0.90 for second-growth deforestation burns in the same area,
and 0.94 for cerrado savanna ecosystems of Brazil. Ward et al. (1996) found for
fires in savanna ecosystems that the following equations can be used for calcu-
lating emission factors for CH4 and NMHC:

EFCH4 = 60.76 - 62.41 MCE R^ = 0.84 (6)

EFNMHC = 45.5 - 45.8 MCE R^ = 0.96 (7)

IV. MINIMIZING SMOKE PRODUCTION

One way to reduce the amount of air pollutants released to the atmosphere is
to adjust the fire return interval to minimize the total smoke production. Here,
an example is presented using information for a pure grassland ecosystem in
Africa (Ward et al, 1996). The decision to be made is whether it should be
burned annually or biennially or a fire return interval of 3 or more years should
be used to minimize the release of pollutants to the atmosphere. To assist in
making this decision, we can apply the equations presented in the previous sec-
tions to first of all estimate the MCE using Eq. (2) which in turn can be used to
estimate the emission factors needed for evaluating the emissions of PM2.5,
CO, CH4, and NMHC. We first need to know something about the rate of lit-
ter accumulation. For our Africa example, Ward et al (1996) assumed that the
Chapter 3 Combustion Chemistry and Smoke 73

area would be burned biennially and that half of the first year's standing grass
decomposes naturally and the other half becomes incorporated into the litter
layer during the 2-year period (i.e., L = 0.5G in the second year). The ratio un-
der this set of circumstances for the standing grass to the sum of standing grass
plus the litter would be 0.67. Even though the total amount of fuel consumed
is less, the MCE declines as a result of the increase in litter so that the emission
factors are larger. The total products of incomplete combustion are shown in
Table 1 to be more than the amount that would be produced were the area
burned annually. The amount of nonmethane hydrocarbons and PM2.5 is 8 and
12% greater and the CH4 released is 60% greater than would occur with annual
burning. Other examples could be developed based on higher or lower decom-
position rates for the litter and longer periods between burns.

TABLE 1 Example of Application of Algorithms for Assessing

the Optimal Burning Interval for Minimizing the Release of
Pollutants to the Atmosphere from the Use of Prescribed Fire
in a Pure Grassland Ecosystem

1 year 2 years

Available fuel 1.000 1.500

G/{G + L) 1.000 0.667
MCE 0.961 0.947
EFCO2 1762 1736
EFCO 45.8 62.6
EFCH4 0.8 1.7
EFNMHC 1.5 2.1
EFPM2.5 2.6 3.9
Emissions
CO2 1762 2604
CO 45.8 93.8
CH4 0.78 2.51
NMHC 1.49 3.21
PM2.5 2.59 5.79
Ratios to one year
CO2 1.00 1.48
CO 1.00 2.05
CH4 1.00 3.21
NMHC 1.00 2.16
PM2.5 1.00 2.23
The total emissions relative to burning every year
CO2 0.74
CO 1.02
CH4 1.60
NMHC 1.08
PM2.5 1.12

After Ward etd., 1996.

 74 D. Ward

TABLE 2 Algorithms and Ratios Developed for Logging Slash of the Pacific Northwest in the United
States and Used for Computing Emission Factors (g of emission per kg of dry fuel consumed)

Algorithm Ratio model Flaming Smoldering Units

MCE 0.92 0.75

Carbon dioxide MCE * 1834 1687 1376 g/kg
Carbon monoxide (1 - MCE) * 1834 * 0.64 94 293 g/kg
Methane 42.7 - MCE * 43.2 3.0 10.3 g/kg
Nonmethane 0.760 + EFCH4 * 0.616 2.6 7.1 g/kg
hydrocarbons
PM2.5 67.4 - MCE * 66.8 5.9 17.3 g/kg
PMIO 1.18*EFPM2.5 7.0 20.4 g/kg
PM 93.3 - MCE * 90.5 10 25 g/kg
Benzene 0.00592 * EFCO 0.56 1.74 g/kg
Benzo[a]pyrene 13 * EFPM 131 331 Mg/kg

After Ward eta/., 1993.

Ward et al. (1993) used similar algorithms developed for vegetation types in
the United States to estimate the total criteria pollutant and air toxic emissions
released from the prescribed use of fire nationally for the year 1989. The sys-
tem they employed identified the fuel consumed by National Fire Danger Rat-
ing Fuel Model types and assigned a ratio of flaming to smoldering for each
type. Then emission factors were computed for flaming and smoldering based
on using an MCE of 0.92 for flaming and 0.75 for smoldering. Table 2 lists the
algorithms and the emission factors for flaming and smoldering combustion
used in this study. It immediately follows that a weighted emission factor can
be computed by knowing the amount of fuel consumed during flaming versus
smoldering combustion. For example, for 100% flaming a valid emission factor
for particulate matter for particles less than 2.5 /mm diameter would be 5.9 g of
PM2.5/kg of fuel consumed on an oven dry weight basis. On the other hand,
for a fuel complex where we expect about 30% of the fuel to be consumed dur-
ing flaming and 70% during smoldering, we would weight the emission factor
by multiplying the flaming emission factor by 30% and the smoldering emission
factor by 70%, add the two together, and derive a weighted emission factor of
13.9 g of PM2.5/kg of fuel consumed.

V. CONCLUSIONS
The combustion of biomass in the open environment is a source of pollution
and greenhouse gases to the atmosphere. Incomplete combustion, primarily
from the smoldering phase, results in a very complex mixture of gaseous and
C h a p t e r 3 C o m b u s t i o n C h e m i s t r y and Smoke 75

particulate matter called smoke. The mixture is somewhat dependent on the el-
emental composition, size distribution, and moisture content of the fuel par-
ticles at the time of burning. The environmental conditions prevailing at the
time of the fire and the type of fire (whether heading, backing, flanking, or an
area fire) are important as well. Combustion efficiency can be calculated and/or
estimated for determining the magnitude of emission factors for different com-
pounds and particulate matter for different combinations of fuel and fire. The
outcome of these calculations can be used for assessing management strategies
in developing environmental impact assessments. By knowing fuel loading,
available fuel, and ratio of fuel consumed by phase of combustion, the amount
and rate of release of smoke can be estimated.

REFERENCES
Allen, S. E. (1974). "Chemical Analysis of Ecological Materials." Wiley, New York.
Andreae, M. O., Andreae, T. V/., Annegarn, H., Beer, J., Cachier, H., Canut, P., Elbert, W., Maen-
haut, W., Salma, I., Wienhold, F. G., and Zenker, T. (1996). Airborne studies of aerosol emis-
sions from savanna fires in southern Africa. 2. Aerosol chemical composition, J. Geophys. Res.
103,32,119-32,128.
Browning, B. L. (1963). "Methods of Wood Chemistry. Vol I." Wiley, New York.
Byram, G. M. (1959). Combustion of forest fuels. In "Forest Fire Control and Use" (K. P. Davis,
Ed.), pp. 1 5 5 - 1 8 2 . McGraw Hill, New York.
Clements, H. B., and McMahon, C. K. (1980). Nitrogen oxides from b u r n i n g forest fuels examined
by thermogravimetry and evolved gas analysis. Thermochim. Acta 3 5 , 1 3 3 - 1 3 9 .
Crutzen, P. J., and Andreae, M. O. (1990). Biomass b u r n i n g in the tropics: Impact on atmospheric
chemistry and biogeochemical cycles. Science 2 5 0 , 1 6 6 9 - 1 6 7 8 .
Crutzen, P. J., Delaney, A. C , Greenberg, J., et al. (1985). Tropospheric chemical composition mea-
surements in Brazil during the dry season. J. Atmos. Chem. 2, 2 3 3 - 2 5 6 .
Crutzen, P. J., Heidt, L. E., Krasnec, J. P., Pollock, W H., and Seller, W (1979). Biomass b u r n i n g as
a source of the atmospheric gases CO, Uj, N2O, NO, CH3CI, and COS. Nature (Lond) 2 8 2 , 2 5 3 -
256.
Evans, L. F., Weeks, I. A., Eccleston, A. J., and Packham, D. R. (1977). Photochemical ozone in
smoke from prescribed burning of forests. Environ. Sci. Tech. 1 1 , 8 9 6 - 9 0 0 .
Ferek, R. J., Reid, J. S., Hobbs, P V., Blake, D. R., and Liousse, C. (1998). Emission factors of hy-
drocarbons, halocarbons, trace gases and particles from biomass b u r n i n g in Brazil. J. Geophys.
Res. 1 0 3 , 3 2 , 1 0 7 - 3 2 , 1 1 8 .
Classman, 1. (1977). "Combustion." Academic Press, New York.
Grier, C. (1975). Wildfire effects on nutrient distribution and leaching in a coniferous ecosystem.
Can. J. Soil Sci. 5 , 5 9 9 - 6 0 7 .
Hao, W. M., and Liu, M. H. (1994). Spatial and temporal distribution of tropical biomass burning.
Global Biogeochem. Cycles 8, 4 9 5 - 5 0 3 .
Hao, W. M., and Ward, D. E. (1993). Methane production from global biomass burning, J. Geophys.
Res. 9 8 , 2 0 , 6 5 7 - 2 0 , 6 6 1 .
Hegg, D. A., Radke, L. F., Hobbs, P. V., et al. (1989). Emissions of some biomass fires. In "Proceed-
ings of the 1989 National Air and Waste Management Association Meeting," Paper No. 089-
025-003. Air and Waste Management Association, Pittsburgh.
Hoffa, E. A., Ward, D. E., Hao, W M., Susott, R. A., and Wakimoto, R. H. (1999). Seasonality of
76 D. Ward

carbon emissions from biomass b u r n i n g in a Zambian savanna. J. Geophys Res. 104, 1 3 , 8 4 1 -

13,853.
Khalil, M. A. K., Edgerton, S. A., and Rasmussen, R. A. (1985). Gaseous tracers for sources of re-
gional scale pollution. J. Air Pollut. Control Assoc. 3 5 , 8 3 8 - 8 4 0 .
Kilzer, F. J., and Broido, A. (1965). Speculations on the nature of cellulose pyrolysis, Pyrodynamics
2, 1 5 1 - 1 6 3 .
Lobert, J. M., Scharffe, D. H., Hao, W. M., Kuhlbusch, T. A., Seuwen, R., Warneck, R, and Crutzen,
R J. (1991). Experimental evaluation of biomass burning emissions: Nitrogen and carbon con-
taining compounds. In "Global Biomass Burning: Atmospheric, Climatic, and Biospheric Impli-
cations" (J. S. Levine, Ed.), pp. 2 8 9 - 3 0 4 . MIT Press, Cambridge.
Nance, J. D., Hobbs, R V., Radke, L. F., and Ward, D. E. (1993). Airborne measurements of gases
and particles from an Alaskan wildfire. J. Geophys Res. 9 8 , 14,873-14,882.
Nelson, R. M., Jr., and Ward, D. E. (1980). Backfire particulate emissions and Byram's fire inten-
sity. Research Note SE-290. USDA Forest Service, Southeastern Forest Experiment Station,
Asheville.
Patterson, E. M., McMahon, C. K., and Ward, D. E. (1986). Absorption properties and graphitic
carbon emission factors of forest fire aerosols. Geophys. Res. Lett. 13, 1 2 9 - 1 3 2 .
Radke, L. R, Hegg, D. A., Hobbs, R V., Nance, J. H., Lyons, J. H., Laursen, K. K., Weiss, R. E., Rig-
gan, P. J., and Ward, D. E. (1991). Particulate and trace gas emissions from large biomass fires
in North America. In "Global Biomass Burning: Atmospheric, Climatic, and Biospheric Impli-
cations" (J. S. Levine, Ed.), pp. 2 0 9 - 2 2 4 . MIT Press, Cambridge.
Reid, J. S., and Hobbs, P. V. (1998). Physical and optical properties of young smoke from individ-
ual biomass fires in Brazil. J. Geophys. Res. 103, 3 2 , 0 1 3 - 3 2 , 0 3 0 .
Reid, J. S., Hobbs, R V., Ferek, R. J., Blake, D. R., Martins, J. V., Dunlap, M. R., and Liousse, C.
(1998). Physical, chemical, and optical properties of regional hazes dominated by smoke in Bra-
zil. J. Geophys. Res. 103, 3 2 , 0 5 9 - 3 2 , 0 8 0 .
Reinhardt, T. E., and Ward, D. E. (1995). Factors affecting methyl chloride emissions from b u r n -
ing forest biomass. Environ. Sci. Tech. 2 9 , 8 2 5 - 8 3 2 .
Shafizadeh, R, Chin, P. S., and DeGroot, W F. (1977). Effective heat content of green forest fuels.
For. Sci. 2 3 , 8 1 - 8 9 .
Susott, R. A. (1980). Thermal behavior of conifer needle extractives. For. Sci. 26, 3 4 7 - 3 6 0 .
Susott, R. A., Shafizahdeh, E , and Aanerud, T. W (1979). A quantitative thermal analysis technique
for combustible gas detection. J. Fire Flamm. 10, 9 4 - 1 0 4 .
Tiedemann, A. R. (1987). Combustion losses of sulfur from forest foliage and litter. For. Sci. 3 3 ,
216-223.
Ward, D. E. (1979). "Particulate Matter and Aromatic Hydrocarbon Emissions from the Controlled
Combustion of Alpha Pinene," PhD dissertation. University of Washington, Seattle.
Ward, D. E. (1983). Particulate matter emissions for fires in the palmetto-gallberry fuel type For.
Sci. 29, 7 6 1 - 7 7 0 .
Ward, D. E. (1986). Characteristic emissions of smoke from prescribed fires for source apportion-
ment. In "Proceedings of the Annual Meeting of the Pacific Northwest International Section Air
Pollution Control Association."
Ward, D. E., and Hardy, C. C. (1984). Advances in the characterization and control of emissions
from prescribed fires. In "Proceedings of the 78th Annual Meeting of the Air Pollution Control
Association," Paper No 84-363. Air Pollution Control Association, Pittsburgh.
Ward, D. E., and Hardy, C. C. (1988). Organic and elemental profiles for smoke from prescribed
fires. In "Receptor Models in Air Resources Management" (J. G. Watson, Ed.), pp. 2 9 9 - 3 2 1 .
Air and Waste Management Association, Pittsburgh.
Ward, D. E., and Hardy, C. C. (1991). Smoke emissions from wildland fires. Environ. Int. 17, 1 1 7 -
134.
Chapter 3 Combustion Chemistry and Smoke 77

Ward, D. E., Hao, W. H., Susott, R. A., Babbitt, R. E., Shea, R. W, Kauffman, J. B., and Justice, C. O.
(1996). Effect of fuel composition on combustion efficiency and emission factors for African sa-
vanna ecosystems. J. Geophys. Res. 101, 23,569-23,576.
Ward, D. E., Hao, W H., Susott, R. A., Kauffman, J. B., Babbitt, R. E., Cummings, D. L., Dias, B.,
Holben, B. N., Kaufman, Y. J., Rasmussen, R. A., and Setzer, A. W (1992). Smoke and fire char-
acteristics for cerrado and deforestation burns in Brazil: BASE-B experiment. J. Geophys. Res.
97, 14,601-14,619.
Ward, D. E., McMahon, C. K., and Adams, D. F. (1982). Laboratory measurements of carbonyl
sulfide and total sulfur emissions from open burning of forest biomass. In "Proceedings of the
75th Annual Meeting of the Air Pollution Control Association." Air Pollution Control Associa-
tion, Pittsburgh.
Ward, D. E., Peterson, J., and Hao, W. H. (1993). An inventory of particulate matter and air toxic
emissions from prescribed fires in the USA for 1989. In "Proceedings of the 1993 Air and Waste
Management Association Meeting." Air and Waste Management Association, Pittsburgh.
Yokelson, R. J., Goode, J. G., Ward, D. E., Susott, R. A., Babbitt, R. E., Wade, D. D., Bertschi, I.,
Griffith, D. W T., and Hao, W. M. (1999). Emissions of formaldehyde, acetic acid, methanol,
and other trace gases from biomass fires in North Carolina measured by airborne Fourier trans-
form infrared spectroscopy. J. Geophys. Res. 104, 30,109-30,125.
Yokelson, R. J., Griffith, D. W T., and Ward, D. E. (1996). Open-path Fourier transform infrared
studies of large-scale laboratory biomass fires. J. Geophys. Res. 101, 21,067-21,080.
Yokelson, R. J., Susott, R., Ward, D. E., Reardon, J., and Griffith, D. W T. (1997). Emissions from
smoldering combustion of biomass measured by open-path Fourier transform infrared spec-
troscopy;. Geophys. Res. 102, 18,865-18,877.
This Page Intentionally Left Blank
 CHAPTER 4
Water Relations
of Forest Fuels
RALPH M . NELSON, JR.
USDA Forest Service, Rocky Mountain Research Station, Fire Sciences Laboratory,
Missoula, Montana

L Introduction
IL Forest Fuels
A. Fuel Classification
B. Fuel Burning Rates and Moisture Content
C. Fuel Characteristics Affecting Moisture Content
IIL Fuel Moisture Relationships
A. Water Potential
B. Conservation Equations
C. Live Fuel Moisture
D. Dead Fuel Moisture
IV. Moisture Content Estimation
Notation
Additional Reading
References

I. INTRODUCTION

Earlier chapters have described forest fire behavior and effects, the general na-
ture of forest fire flames, and the processes of combustion and smoke produc-
tion in forest fuels. It was pointed out that moisture in these fuels acts to retard
the rate of combustion. Among the fire behavior factors affected are the pre-
heating and ignition of unburned fuels, rate of fire spread (or fire grov^th), rate
of energy release, and production of smoke by burning and smoldering fuel. If
we are to improve our understanding of these and other aspects of fire behavior,
we must be able to quantify fuel moisture content within reasonable bounds.
Moisture content, expressed as a fraction, is the mass of water held by unit mass

79
80 Ralph M.Nelson, Jr.

of ovendry fuel and is determined primarily by fuel type and weather. It also may
be expressed as a percentage of the fuel ovendry weight by multiplying by 100%.
For our purposes, fuels will be considered as living or dead and, within these
broad categories, as individual particles or a collection of particles making up
a fuel complex, or stratum. In many cases, of course, a fuel complex is composed
of a mixture of live and dead particles from various fuel types whose moisture
contents may vary over a wide range. The open grasslands of the southwestern
United States exemplify a single fuel stratum, whereas the palmetto-gallberry-
southern pine fuels of northern Florida exhibit at least three strata—a mat of
pine needles in varying stages of decay intermixed with various living grasses
and forbs, an intermediate layer of shrubs, and a pine overstory. The trunks of
standing individual trees greater than 5 -10 cm in diameter usually are scorched
or charred by fire but do not burn and are not considered part of the fuel com-
plex. In the conifer forests of the western United States and Canada, needle lad-
ders (sometimes referred to as needle drape) often extend from the canopy to
the ground so that fire can climb into the tree crowns. Similar behavior in some
other conifer species such as spruce is caused by branches extending to the
ground. When the crown fuels are drier than normal and winds are strong,
these transitions from surface fire to crown fire occur with a large increase in
energy release rate. The fire appears to acquire a violent character, and its in-
creased energy often renders useless any attempt to control it.
The present chapter begins with a review of the influence of moisture on the
combustion of forest fuels and how fuel characteristics determine the moisture
content level of these fuels. It summarizes current understanding of the amount
of water these fuels can hold, gain and loss of this water, and how the govern-
ing processes have been described mathematically. Only fuels associated with or
originating from vascular plants (grasses, shrubs, and trees as opposed to mosses
and worts that lack internal structure for transporting water) are considered.
The discussion of live fuels emphasizes the physiological aspects of water trans-
fer, but only a few studies related to mechanisms of water transport in these
fuels are discussed. For example, the effects of photosynthesis, respiration, and
growth on water potential and water movement are not described. Brief men-
tion is made of soil water transport which, itself, has been a prominent topic in
many soil physics and hydrology texts. In the case of dead fuels, a description
of what the author believes are the more relevant studies is given. Only a few of
the published studies of moisture content change in wood and forest fuels have
dealt with diurnal change; a glimpse at prediction models for diurnal moisture
variation is provided. Near the end of the chapter, several methods of measur-
ing the moisture content of live and dead forest fuels are briefly discussed, and
the reader is referred to works related to the fuel moisture aspects of current fire
behavior and fire danger rating systems. Information on the general topic of
forest fuel moisture relationships may be found in Luke and McArthur (1978),
Chandler et al (1983), Pyne (1984), and Pyne et al (1996).
Chapter 4 Water Relations of Forest Fuels 01

II. FOREST FUELS

The behavior of a spreading fire is determined by factors such as weather, to-
pography, fuel quantity, and fuel moisture content. In practice, the burning
characteristics of individual forest and wildland fuel particles are difficult to de-
scribe, but the influence of moisture content, particle physical properties, and
particle arrangement on the burning characteristics of vegetation layers or lit-
ter and duff is even more complex and not well understood. This section dis-
cusses the effects of fuel moisture content on fuel burning rates and the fuel
characteristics related to moisture content change.

A. FUEL CLASSIFICATION

Because forest fires usually are categorized according to the location of the
uppermost fuel stratum through which they burn (ground, surface, or crown
fires), it makes sense to think of fuel classification in the same way. Thus fuels
may be classified as ground fuels, surface fuels, or crown fuels. Ground fuels
consist of the highly decomposed organic material in contact with the inor-
ganic layer and include duff, roots, peat, and rotten wood or bark coming from
downed twigs and branches. Next to the ground fuel is a layer of surface fuels
consisting of recently fallen and partially decomposed tree leaves (and/or coni-
fer needles), fallen twigs, bark and branches, live or dead grasses, forbs, and
shrubs less than L8 m tall (Davis, 1959). In moist climates, the fuels above
mineral soil have three components: the recently cast litter layer, the partially
decomposed/ermentation layer, and the well-decomposed humus layer. Stocks
(1970) refers to duff as all material above the upper surface of the mineral soil.
A distinction between litter and duff is made in this chapter, however, with "lit-
ter" referring to the litter and fermentation layers and "duff" referring to the
humus layer. The characteristics of duff layers are discussed further in Chap-
ter 13 in this book. Dead ground and surface fuels contain both free and bound
liquid water and are sensitive to precipitation and to changes in atmospheric
relative humidity and temperature. Crown fuels include the canopies of most
conifers and the chaparral and pocosin shrub types typical of the southwestern
and southeastern United States. The latter two fuel types, which some investi-
gators would categorize as surface fuels, grow to a height of 5 m or more. The
moisture content of these live fuels is determined primarily by environmental
and physiological factors. The canopies of deciduous species also are crown fu-
els, but fires rarely spread through these strata because the crowns are relatively
sparse and have higher moisture contents than conifers. Exceptions occur under
dry, windy conditions and when the fuels contain highly volatile and flam-
mable substances, as in various species of Eucalyptus.
82 Ralph M.Nelson, Jr.

B. FUEL BURNING RATES

AND MOISTURE CONTENT

The effect of fuel moisture on fire behavior is to slow the rate of burning, or rate
of fuel consumption. For a spreading line of fire, the average burning rate can
be computed from the mass of fuel consumed per unit area of ground divided
by the time required to consume fuel on the area. Thus high fuel moisture con-
tent retards the rate of fuel consumption per unit of burning area (kg m~^ s~^)
by decreasing the mass of fuel consumed and increasing the particle burning
time, or particle residence time (the time during which flame resides on individ-
ual particles in the fuel layer combustion zone). In addition, it increases the fuel
preheating time. With respect to the thermal history of unit mass of fuel, the
various phases of combustion include the processes of preheating, volatiliza-
tion, charring, smoldering, and glowing; these are described in Chapters 3 and
13 in this book.
Consider first the fuel preheating time (or ignition time). The heat required
for the onset and completion of volatilization of the fuel (volatilization begins
at about 200°C and is assumed complete when the fuel temperature reaches
400°C) is called the heat of ignition Qj (kj kg"^) by Wilson (1990) and is com-
puted from the defining equation

Qj = Qj^ MQM (1)

where Qj (kJ kg~^) is the heat required to raise unit mass of dry fuel from ambi-
ent temperature to 400°C, Q^ (kJ kg~^) is the energy to heat unit mass of water
to 100°C and then vaporize it, and M is the fractional moisture content (a fuel
particle or layer average expressed on an ovendry weight basis). Thus, an in-
crease in M increases the amount of heat required to raise the temperature of
unit mass of fuel from ambient temperature to 200°C and increases the preheat-
ing time. Various investigators have observed increases in ignition time with in-
creasing M in dead and live fuels (Pons, 1950; Xanthopoulos and Wakimoto,
1993). The analytical studies of Albini and Reinhardt (1995) suggest that the
increase in ignition time is due to increases in particle thermal conductivity and
volumetric heat capacity with increasing M. A different aspect of ignition deals
with what happens when firebrands, either short or long range, are blown or
dropped into unburned fuels. Again, preheating and ignition are slowed by fuel
moisture content, and the heat demand may be estimated by Eq. (1). Blackmarr
(1972) has shown that the probability of ignition in slash pine litter is a strong
function of moisture content and firebrand characteristics.
A second effect of fuel moisture on the burning rate involves a decrease in
fuel consumption owing to the interaction among various fire behavior charac-
teristics. The available fuel loading, defined here as the mass of fuel consumed
Chapter 4 Water Relations of Forest Fuels 83

per unit area of ground during flaming combustion, depends in a complicated

way on moisture content, flame temperature, and the mass fractions of volatiles
and char produced during combustion of unit mass of original fuel. The fuel
chemical composition and rate of thermal decomposition are two factors affect-
ing the relative amounts of volatiles and char produced which, in turn, deter-
mine the heat of combustion and the fraction of fuel available for flaming com-
bustion. These relationships are discussed in detail by Albini (1980) and Susott
(1982). The presence of moisture also causes a reduction in flame temperature
because some of the heat generated in combustion is used to heat the inert water
vapor in the products of combustion and because oxygen in the air is diluted by
water vapor leaving the heated fuel (Byram, 1959). The reduced temperature
retards the rate of decomposition and tends to drive the combustion process
toward production of char rather than toward production of high-temperature
volatiles. Results from burns oi Eucalyptus leaves in a flow calorimeter (Pompe
and Vines, 1966) suggest that water in fuel promotes smoke formation, reduces
the heat of combustion, lowers the rate of temperature rise of air flowing in the
calorimeter, and leads to much less intense burns. The effect of moisture on
fuel consumption was estimated by Van Wagner (1972a) who used data from
experimental fires spreading in jack pine, red pine, and white pine stands to de-
velop a semiempirical relationship showing that duff consumption decreases as
M increases. His equation for the weight of duff consumed, W ( k g m ~ ^ ) , is
given by

W = 0.941(1.418 - M)/(0.1774 + M) (2)

suggesting that consumption in these fuels approaches zero when M approaches

1.418. The origin of Eq. (2) is discussed in Chapter 13 in this book. In this
model, downward transfer of heat by radiation in the fuel layer combustion
zone drives off moisture and then raises the temperature of the dry duff to the
ignition point. Closure of the model requires a decrease in the emissivity of
the flaming front with increasing M, suggesting that reduced radiative trans-
fer to unburned fuel constitutes yet another means of slowing the rate of fuel
consumption. King (1973) discusses the reduction of flame emissivity due to
presence of moisture in terms of a reduction in soot concentration and an in-
crease in carbon monoxide concentration according to the water gas reaction
in which water vapor and carbon (soot) combine to form carbon monoxide and
hydrogen.
The third effect of fuel moisture on the rate of burning is an increase in the
fuel particle residence time. Albini and Reinhardt (1995) have shown with ex-
perimental and theoretical studies that, to the extent that flaming combustion
is fueled by a simple sublimation process, the characteristic particle burning
time is proportional to Qj from Eq. (1). Because this demand for heat is satisfied
84 Ralph M.Nelson, Jr.

by radiative and convective heat transfer within the combustion zone, moisture
also will increase the burning time by reducing radiation to the particles (King,
1973). It is noted that the effects of M on particle ignition time and particle
burning time are sometimes difficult to separate because a fraction of M is lost
in preheating and the remainder is lost during particle consumption by flaming
combustion. Research has not yet clarified how moisture change takes place
while a fuel layer burns, but the relative amounts of water lost during preheat-
ing and flaming combustion seemingly would depend on particle size, shape,
and arrangement and on the initial value of M (i.e., on the amounts of liquid
and/or adsorbed water present).

C. FUEL CHARACTERISTICS AFFECTING

MOISTURE CONTENT

If the effects of weather are disregarded, the most significant factors affecting
the amount of water held and transported in woody and vegetative particles are
chemical composition, internal structure, and physical properties. In the case
of a fuel bed, several layer characteristics may be significant also. The amount
of moisture held in the cell walls of fuel particles is related to composition and
crystalline structure of the walls, whereas the liquid water held in the cell cavi-
ties is determined by the larger scale capillary structure.

1. Chemical Composition
The chemical constitution of wood is similar to that of foliage. Stamm and Har-
ris (1953) give the cell wall composition of wood on a percentage of dry weight
basis as cellulose (40-55%), hemicellulose (15-25%), lignin (15-30%), and
extraneous and extractive matter (2-15%). Extractives include various organic
compounds such as resins, sugars, and fatty acids that may be soluble in either
alcohol, water, or organic solvents (e.g., xylene or ether). The only mineral con-
stituent they cite is ash with a content between 0.1 and 4%. Thus, the holocellu-
lose content (cellulose plus hemicellulose) ranges from about 55 to 80%. When
ranked according to their hygroscopicity (or affinity for water), these com-
ponents are ordered as hemicellulose, cellulose, lignin, and extractives. On the
other hand, foliage contains the same components but in different amounts.
According to Susott (1980), the range in composition of old and new needles
of three western conifers is holocellulose (35-44%), lignin (18-19%), and ex-
tractives (37-47%). The higher percentage of extractives in needles than in
wood tends to reduce water takeup and rates of moisture exchange (Anderson,
Chapter 4 Water Relations of Forest Fuels o5

1990a, 1990b). Yet, the experimental data of Anderson (1990b) tend to confirm
the resuUs of earlier investigators (Dunlap, 1932; Blackmarr, 1971), showing
that forest fuels are slightly more hygroscopic than wood under identical con-
ditions of air temperature and relative humidity. This could be the case if the
tendency toward smaller moisture contents due to smaller hemicellulose con-
tent of the fuels is more than offset by the fact that the hygroscopicity of hemi-
cellulose exceeds that of wood cellulose by 50% (Browning, 1963); alternatively,
the extractives in some conifer needles and deciduous leaves may be more hy-
groscopic than wood holocellulose. As litter fuel weathers over time, water-
soluble extractives are leached from the surface and interior of the particles,
and the holocellulose content is reduced due to consumption by microorgan-
isms. This loss of dry mass can be approximated as an exponential decrease
with time (Olson, 1963).

2. Internal Structure
The instantaneous level of moisture content in forest fuels is strongly influ-
enced by the internal structure of the material and by whether the fuel is alive
or dead. The factors controlling water movement in living plants are osmotic
forces due to intercellular differences in plant water concentration and capillary
tension forces created by transpirational demand at the external surfaces of the
leaves (the term "leaves" includes conifer needles also). This demand for water
must be satisfied primarily through absorption by the roots. In dead fuel par-
ticles, the most significant factors are two transport properties: (1) the perme-
ability of the fuel to liquid water (bulk flow in cell cavities by capillarity), and
(2) the moisture diffusivity (molecular flow in cell walls by bound water diffu-
sion or in air spaces by diffusion of water vapor). An additional factor in litter
and duff layers is liquid water drainage due to gravitational forces.

a. Live Fuels
The cells in living fuels contain protoplasm, cell wall material, and one or
more vacuoles. The living protoplasm is largely made up of water, protein, var-
ious dissolved organic compounds and salts (cytoplasm), and a nucleus that
controls inheritance and the metabolic activity of the cell through the genes it
contains. The vacuole is located within the cytoplasm and is filled with cell sap
(about 98% water plus various dissolved compounds). The watery solution
surrounded by the nonliving primary cell wall is called the protoplast The wall
itself is mainly composed of cellulose, hemicellulose, and pectin; the water it
contains is referred to as the apoplast As the cell matures, the vacuole enlarges
and eventually occupies most of the cell (Figure 1). Water movement in the cell
is accomplished primarily by osmosis, for which the thin outer layer of cyto-
86 Ralph M. Nelson, Jr.

Air

4.0 to 6.5 nm
pore diameter
Plasmalemma
Xylem Wall
microfibrils
surrounded
by otiner
polymers

Air/water
meniscus

FIGURE 1 Mature cell wall showing the xylem through which water flows and the protoplast
in the vacuole. The enlarged inset shows the apoplast, including air-water interfaces between the
polymer microfibrils. The plasmalemma is pressed against the wall from the inside and pulled in
the same direction by tension from the outside. This is the source of plant turgor. Adapted from
Kramer and Boyer (1995).

plasm in contact with the cell wall interior, the plasmalemma, functions as the
semipermeable membrane. In general, any given cell is surrounded by other
cells or by a solution containing various solutes. Hence cell A with higher sol-
ute concentration than cell B will have a lower water concentration than cell B
so that water will move from B to A by osmosis. Cell A will become turgid
and cause the cytoplasm to press against the cell wall. This is the origin of "tur-
gor pressure." In the opposite sense, when water leaves cell B, the cytoplasm
shrinks from the cell wall, and if carried to the extreme, cell B dies. In addition
to osmosis, liquid imbibition (or absorption) within the cell wall occurs in live
or dead cells. The influence of the fine structure of the wall on this process is
discussed in the next section.
In living fuels, liquid soil water absorbed by the roots moves by capillary flow
through the water-conducting xylem of the stem, into the leaves, and then into
the atmosphere through the leaves as water vapor. The processes of liquid ab-
sorption and vertical ascent are highly significant because they can limit the rate
of water supply to the leaves by reducing plant turgor and causing stomatal
closure. Figure 2A shows the structure of a deciduous leaf and pathways for
Chapter 4 Water Relations of Forest Fuels 87

--„ Cuticle
Lc Upper epidermis
Sclerenchyma

Palisade cell
Xylem
Spongy mesophyll cell
Phloem
Sclerenchyma
Guard cell

B. Resin duct
Transfusion tissue
Endodermis
Phloem
Xylem
Guard cell
Stoma
Chlorenchyma

Cuticle
Epidermis
Hypodermal sclerenchyma

FIGURE 2 Structure of leaves from a deciduous tree and a conifer. Water vapor loss is primarily
through the internal air spaces and guard cells, with a small amount of vapor diffusing directly
through the epidermis and cuticle: (A) species not identified, (B) eastern white pine. From Kramer
and Kozlowski (1979) and Kozlowski and Pallardy (1997), respectively.

liquid transfer. Food, water, and nutrients are supplied to and removed from
the leaf through veins (not shown) that penetrate the mesophyll (palisade and
spongy parenchyma) cells. These cells, in which photosynthesis occurs, are sur-
rounded by a network of small intercellular spaces; other spaces just behind the
lower epidermis are relatively large. These air spaces create a large amount of
internal surface from which evaporation can occur. The vapor then diffuses to-
ward and through the guard cells surrounding the stoma and into the atmo-
sphere. This process is referred to as stomatal transpiration. A small amount of
vapor also exits the leaf through the epidermal layers but is reduced by surface
cuticle—cuticular transpiration. The stomates are found on both sides of the
leaves but generally are more numerous on the underside. Conifers, with the
exception of pines, also produce needles in which the mesophyll consists of
88 Ralph M.Nelson, Jr.

palisade and spongy parenchyma cells. The pine needle cross section in Fig-
ure 2B shows sclerenchyma cells which provide mechanical support and chlor-
enchyma cells that contain chloroplasts. The transfusion tissue is involved in
solute transport.
If the supply of water is not a limiting factor, the two major sources of re-
sistance to evaporation are associated with the leaf and the atmosphere. The
former resistance primarily is due to covering of the leaf surfaces by a relatively
waterproof layer of cutin and wax (cuticle) and to stomatal resistance related
to size of the aperture. In pine species, wax also is present in stomatal cham-
bers. Aperture size of the stomates strongly depends on plant stresses and en-
vironmental factors such as sunlight, air temperature, and relative humidity.
Atmospheric resistance is due to the leaf boundary layer, a thin region adjacent
to the surfaces in which the factors affecting transport of vapor into the atmo-
sphere are changing from leaf values to atmospheric values. This resistance, de-
termined by particle size and shape and by wind speed, decreases as air flow
over the particle increases. More information on water transport as related to the
internal structure of living fuels may be found in texts such as Slatyer (1967),
Kramer and Boyer (1995), and Kozlowski and Pallardy (1997).

b. Dead Fuels
The amount of moisture held in the cell walls of live or dead fuels (as dis-
cussed earlier in connection with cell wall imbibition) is related to the fine-
scale structure of the walls. Approximately 50 individual cellulose chains are
held together in parallel chance groupings by hydrogen bonds and various other
types of secondary bonding to form a crystalline region, or crystallite. These
elements are thought to be about 0.06 /im long—much shorter than a cellu-
lose chain—due to the frequent occurrence of amorphous regions between the
crystallites. The amorphous regions are regions in which the chains are highly
disordered. Thus individual chains pass through several crystalline and amor-
phous regions but are linked at only a few points in the amorphous regions so
that most of the active hydroxyl groups are available to take on water molecules.
This water then forces the chains farther apart and swells the material. Because
water is adsorbed throughout the amorphous regions but only on the surfaces
of the crystallites, the water-holding ability of the cellulose varies directly with
the proportion of the cellulose that is amorphous. The crystallites, with their
associated amorphous regions, become further aggregated into fibrils that are
embedded within the lignin-hemicellulose matrix materials to make up the var-
ious layers of the cell wall. In these fibrous materials, a secondary wall (the S^,
S2, and 53 layers) composed mainly of cellulose is found inside the primary wall
and provides structural integrity (Figure 3A). In wood, the 52 layer makes up
60-80% of the cell wall, thus controlling shrinkage, swelling, and other physi-
chapter 4 Water Relations of Forest Fuels 89

A fi^l>^ B

FIGURE 3 Structural features of a softwood. (A) View of a mature tracheid in which ML is the
middle lamella that cements tracheids together, P is the primary wall, S^, S2, and S3 are secondary
walls, and W is the "warty" layer of unknown composition. From Cote (1967). (B) Pathways for
continuous liquid or vapor flow through the large white cell cavities and bordered pits and for in-
termittent bound water/water vapor flow through cavities and the darker cell walls (the middle
lamella between tracheids is not clear). FromStamm (1946). (C) Three kinds of pit pairs—a,simple
pit with the dark middle lamella, cross-hatched primary wall, and dotted secondary wall; b, bor-
dered pit with the torus midway in the pit membrane; c, half-bordered pit with no torus. Repro-
duced with permission of The McGraw-Hill Companies from Brown, H. P., Panshin, A. J., and For-
saith, C. C. (1949). "Textbook of Wood Technology, Vol. 1," 1st ed. McGraw-Hifl, New York.

cal properties (Siau, 1995). References on details of the fine-scale structure of

wood and cellulose are Browning (1963) and Panshin and de Zeeuw (1980).
The fractional moisture content M of dead fuel particles can reach maximum
values between 2.5 and 3, depending on the wood or foliage particle specific
gravity (Stamm, 1964). Both types of fuel exhibit sudden changes in moisture
transport mechanisms and in electrical, mechanical, and thermal properties
when M is in the range 0.21 to 0.35. These changes are associated with a value
of M called the fiber saturation point that may be estimated for each fuel; the
variability in its value can be due to differences in fuel temperature, mechani-
cal stress, specific gravity, or chemical composition. The fiber saturation point,
Mjsp, has been defined in various ways. It is most often defined as that M value
obtained when isothermal equilibrium data for M are plotted as a function of
90 Ralph M.Nelson, Jr.

increasing fractional relative humidity, H, and then extrapolated from H = 0.95

to H = 1 (Stamm, 1964). This procedure estimates the M value at which the
fuel cell walls are saturated but no condensed water exists in the capillary struc-
ture. The method is an idealization, however, because there is no abrupt tran-
sition from a "saturated cell wall" region to a "liquid water" region; instead, the
transition occurs over a range in H from about 0.9 to 0.995. Above the latter
value, the fuel becomes increasingly saturated. The figure generally used for
Mjjp by fire researchers is 0.30, independent of fuel type or state.
Gain or loss of moisture by dead fuel particles above Mj^^ occurs by move-
ment of liquid water through the fuel capillary structure in response to surface
tension forces. The specific permeability (m^) of a porous material is a measure
of the ease with which fluids can move in bulk through the material in response
to a pressure gradient in the fluid. In woody and vegetative particles, perme-
ability to water depends on the internal structure of the material and extent of
its swelling. Because forest fuels usually exchange moisture with the atmosphere
at temperatures below 60°C, vapor pressure differences are not extremely large,
and moisture transfer by gas flow often is considered negligible in comparison
with that due to liquid flow. In softwoods, passage of liquid water is through
fiber cavities in series with tiny holes in the pit membrane called pit membrane
pores; on the other hand, continuous liquid movement through the permanent
cell wall capillaries contributes little to the overall flow (Stamm, 1964). Fig-
ure 3B is a sketch of the gross capillary structure of a softwood and paths for
liquid transport through fiber cavities and pit membrane pores. Darkened areas
represent a network of lignocelluosic strands (or fibrils) intertwined to form the
cell wall, whereas the white areas represent fiber (or tracheid) cavities and pit
chambers within the bordered pits. Figure 3C shows three types of pit pairs
found in softwoods and hardwoods. Bordered and half-bordered pits are com-
mon in softwoods, whereas simple pits occur in hardwoods. The enlarged re-
gion in the center of the pit membrane is called the torus; the pit becomes aspi-
rated when the torus is sealed against a border by surface tension forces created
in drying (Hart and Thomas, 1967). The pits also can become impervious to
water because of encrustation by extractives.
The internal structure and liquid permeability of hardwoods is more com-
plex and more variable than that of softwoods. The primary conducting ele-
ments, or vessels, are interspersed among tracheids and other fibers and inter-
connected by simple pits. Resistance to liquid flow is offered by the pits and by
internal growths, or tyloses, lodged in the vessels of most species. In the pits,
resistance is due to encrustation rather than aspiration because pits in hard-
woods have no torus in the pit membrane. Furthermore, the openings in the pit
membrane are about 10 times smaller in hardwoods than in softwoods, but not
so small as to prevent water flow. One of the most permeable woods is red oak
(no tyloses) with a longitudinal permeability of about 2 X 10~^° m^. The sap-
wood and heartwood of many softwoods and hardwoods exhibit longitudinal
Chapter 4 Water Relations of Forest Fuels 91
permeabilities ranging from 10"^^ to 10~^^ m^. Comstock (1970) measured lon-
gitudinal/tangential permeability ratios in softwoods ranging from about 500 to
8 X 10"^. Siau (1995) refers to an unpublished study of eight hardwoods by Com-
stock reporting longitudinal/tangential permeability ratios between 3 X 10"^
and 4 X 10^. Additional information concerning flow pathways and the rela-
tion of internal structure to fluid flow in wood may be found in Stamm (1964)
and Siau (1995).
The flow of liquid water in layers of deciduous or coniferous litter and duff
is determined by the amount of moisture retained on and within the particles
and in the interstices between particles. In a series of laboratory experiments,
Stocks (1970) applied simulated rainfall amounts ranging from 12.7 to 50.8 mm
to air-dry ponderosa pine duff layers 7.6 cm thick at a rate of about 25.4 mm
h~\ The results showed that the percentage of the applied rainfall retained by
the duff decreased from 20 to 11% as rainfall amount increased. When the lab-
oratory experiment was repeated using moist duff, water retention was halved
and the percentages decreased from 10 to 5% as rainfall increased. Stocks (1970)
also reported field and laboratory experiments suggesting that the moisture con-
tent of the duff in a thin sublayer adjacent to underlying mineral soil is smaller
than in the sublayers just above it (Figure 4). These results are consistent with

2.0 inches rain

% Moisture content
50 100 150 200

20
Time (days)
FIGURE 4 Laboratory drying rates for five separate sublayers within a 7.6-cm duff layer (pri-
marily ponderosa pine) after 50.8 mm (2 inches) of simulated rain. "Layer V is at the top of the
duff layer. (A) Moisture content versus time after rain. (B) The vertical distribution of moisture con-
tent for specific times after rain. Adapted from Stocks (1970).
92 Ralph M. Nelson, Jr.

the idea that the amount of rainfall retained by the layer depends on a balance
among capillary forces across the duff-soil interface, capillary forces among the
fuel particles, gravitational forces acting over the vertical depth of the layer, and
attractive forces drawing water into the particles. Structural details of the layer
affect this balance by determining the specific permeability of the duff. Fosberg
(1977) measured the permeability of several western conifer duff layers to air
and reported values ranging from about 10"^ to 5 X 10"^^ m^. These values are
orders of magnitude larger than those for wood, reflecting the much larger sizes
of the conducting pores.
Moisture diffusivity (m^ s"^) is the property that governs exchange of mois-
ture in dead fuel particles below Mj^^. Water movement is due to combined
diffusion involving a gradient in bound water concentration across the cell
walls and a gradient in partial vapor pressure through the fiber cavities and pit
membrane pores. Stamm (1964) states that bound water diffusion differences
between softwoods and hardwoods are small, so diffusion of vapor through
the continuous void structure is the major source of variability among species.
Stamm (1964) and Choong (1965) suggest, however, that in softwoods the per-
centage of the transverse moisture diffusion attributable to vapor diffusion is
22% or less at temperatures below 50°C. Thus structural differences between
softwoods and hardwoods normally cause only small differences in diffusional
flow. Though the circumstances under which diffusion operates in dead foliage
particles are not well known, equilibrium and dynamic moisture relationships
in these fuels are well represented in form, if not in magnitude, by research re-
sults for wood over a large range in M. This is verified to some extent in data
reported by Van Wagner (1979) on drying of jack pine litter that are similar to
those from studies on wood and paper (Nelson, 1969). Other data demonstrate,
however, that diffusivities differ for particles of wood and foliage. Values of the
integral moisture diffusivity (the average value for the particle or layer), D^^, ob-
served by Linton (1962) for Eucalyptus ohliqua wood ranged from 1.4 to 42 X
10~^^ m^ s~^ for twigs 6 mm in diameter and from 3.5 to 45 X 10"^^ m^ s~^ for
twigs 3 mm in diameter. Linton also found that individual leaves from Euca-
lyptus ohliqua and Eucalyptus radiata exhibited D^y values ranging from 0.24
to 11 X 10"^^ m^ s"\ Anderson (1990a) reported a comprehensive study in
which grasses, hardwood leaves, conifer needles, and square softwood sticks
initially in equilibrium at 26.7°C and H of 0.9 were subjected to a rapid drop
in H to 0.2 at which they equilibrated and then were exposed to another step
change in H back to 0.9. Weathered foliage fuels were studied also. Results
showed that D^y of the wood ranged from about 0.1 to 2 X 10~^^ m^s~\ ingood
agreement with measurements of transverse diffusion in softwoods and hard-
woods for which D^^ ranged from 0.07 to 1.7 X 10~^° m^ s~^ (Stamm, 1960;
Simpson, 1993a). On the other hand, Anderson's foliage diffusivities (for re-
cently cast fuel) ranged from about 10~^^ to 10~^^ m^ s~^ These values are in
Chapter 4 Water Relations of Forest Fuels 93

approximate agreement with the observations of Linton (1962), but even An-
derson's values for weathered fuels (10~^^ to 10"^^ m^ s~^) are smaller than
those for wood by two orders of magnitude.

3. Physical Properties
The differences in moisture diffusivities of dead wood and foliage just described
are due, at least in part, to differences in mass density, particle size and shape,
and extractive content. Moreover, the effects of weathering on moisture rela-
tionships operate primarily through these variables. The density of wood de-
pends on its specific gravity and moisture content, ranging from about 300 to
800 kg m~^ (Simpson, 1993b); values for foliage fuels fall well within this range
(Anderson, 1990a). The effect of density on diffusion has not been identified
in experimental studies. Though he acknowledged the lack of conclusive data,
Hart (1964) argued that two samples of wood drying in response to identical
gradients in M will exhibit diffusivities inversely proportional to their densities.
Particle density also is affected by water-soluble extractives that, because of their
bulking action, leave the fiber cell wall in an expanded state when drying from
the green condition takes place. For foliage, this effect may be overshadowed by
the flow reduction due to wax on particle surfaces. Van Wagner (1969a) pre-
sents data showing that response time r (s) (the time required for accomplish-
ment of 1 — 1/e, or 63.2%, of the total moisture content change due to a step
change from a constant initial value to a constant final value) decreased by fac-
tors averaging about 5.5 and 2.5 for red pine needles and aspen leaves that had
been pretreated with xylene to remove the waxes and resins. This result implies
that fuel particles with surface wax removed by weathering will gain or lose
moisture about three to six times faster than unweathered particles whose sur-
face wax is intact.
The effects of particle size are best obtained from the theory of diffusion in
solids (Crank, 1975). The diffusion theory, discussed briefly in Section 11I.D.2,
indicates that the theoretical effect of minimum particle dimension d (m) on
Day is zero because the local diffusivity, D, is treated as a constant and therefore
equals D^^. Dimension d usually is the radius of a cylinder (or sphere) or the
half-thickness of a slab. It is well known, however, that D^^ for wood is not con-
stant, but a function of moisture fraction M and fuel temperature Tj.
The measurement of D^^ is further confounded by the possible occurrence of
two nondiffusional mechanisms that can be mistakenly interpreted as diffusion.
The first of these involves the slowing of sorption [this term is used when refer-
ence is made to the combined processes of adsorption and desorption (see Sec-
tion III.D. 1) or when no distinction between the two processes is necessary] due
to molecular rearrangements associated with relaxation of shrinkage or swell-
ing stresses. These rearrangements are more pronounced during adsorption in
94 Ralph M.Nelson, Jr.

thin particles (< 3 mm thick) when fractional humidity H exceeds 0.5; they
make new adsorption sites available at a slower rate than the initial diffusion rate
(Christensen, 1965). The second mechanism is moisture transfer at the solid
surface. The theoretical influence of this process is expressed in terms of the di-
mensionless mass transfer Biot number, Bi—the ratio of resistance to moisture
transfer at the surface to that within the particle. This number is discussed fur-
ther in Section III.D.2. Byram (1963, unpublished) expressed the results from
classical theory in terms of response time and found r proportional to d^ when
Bi approaches infinity. For smaller Bi, approaching zero, r was predicted as pro-
portional to d. Nelson (1969) tested Byram's predictions with laboratory drying
experiments on cellulosic materials exposed to fixed step changes in H at con-
stant temperature. Two values of r were observed for most runs. The r versus
d^ relationship was approximated for sawdust layers (r depended on d^^ rather
than cJ^), square wooden sticks, and thin slabs of paper (during the later stage
of drying) for initial values of M well above Mj^^. The approximate d^ depen-
dence observed in all fuels studied suggests that diffusivity D^^ was nearly inde-
pendent of d in these tests. Results for the paper slabs are presented in Figure 5.
The data show that the predicted r versus d result was observed in approximate
form (r depended on d^-^^ rather than d) during the early stage of drying when
the resistance to transfer at the surface must have exceeded that within the slabs.
For the two thinnest slabs, the tendency for r to become independent of d indi-
cates that the drying rate depended entirely on the external conditions. In other
work, the effect of thickness on moisture change in Eucalyptus regnans sapwood
specimens was studied by King and Linton (1963) who found that r increased
with increasing d, but not according to d^ as would be expected from theory.
Taken together, the data suggest that the predicted effects of d on r will be ap-
proximately correct for forest fuels and that D^^ is nearly independent of d in
the diffusion-controlled range when external conditions are constant.
Particle shape theoretically affects moisture exchange by means of differ-
ences in constants derived in analytical solutions of the equations describing
diffusion in the various particles. In practice, these effects due to shape operate
partly through Biot number Bi and partly through the particle surface-to-volume
ratio a (m~^) that influences the exchange of heat and mass between the par-
ticle and the surrounding fluid. Larger values of the ratio result in larger ex-
change rates because of increasing surface area per unit of particle volume. This
ratio is one of the factors defining the openness of fuel layers which, in turn,
may determine the rate-controlling dimension during moisture exchange. A
second controlling factor, the dimensionless fuelhedpacking ratio, /3, represents
the volume of fuel particles within unit volume of fuel bed. An interesting ques-
tion not yet answered by research is the following: if a fuel layer of constant
packing ratio j8, vertical depth 8 (m), and uniformly distributed cylindrical par-
ticles of radius d (or constant cr) is exposed to a step change in H at constant T,
Chapter 4 Water Relations of Forest Fuels 95
320_

160

80

B 40

— 20

10

2.5 ^ 1 \ 1 1 —

0.01 0.02 0.04 0.08 0.16 0.32

r (centimeter)
FIGURE 5 Logarithmic plot of response time versus total thickness of paper slabs drying at
27.2°C and 36% relative humidity, where r^ refers to early drying when surface effects were pres-
ent but not dominant and T2 to the later stages when internal diffusion was controlling the process.
For small values of slab thickness, response time tends to become independent of thickness due to
surface control. From Nelson (1969).

is its rate of moisture exchange controlled by 8 or dl Though the answer to this

question partly depends on the imposed boundary conditions, in general one
would expect that moisture gain or loss in tightly packed beds is controlled by
6, and that as jS decreases and the beds become more open, the rates of mois-
ture exchange would approach the rate for a single particle. The limited data of
Anderson et at (1978) for beds of ponderosa pine needles suggest this is so. In
later work, Anderson (1990a) compared the diffusivities of layers with differ-
ent degrees of packing with those of widely spaced needles drying in similar
environmental conditions and found that D^^ decreased as the fuel beds became
deeper or more dense. Using the method of nonlinear least squares, Anderson
also developed equations for predicting the effects of fuel particle and fuel bed
variables. He produced four groups of equations—each group distinguished by
96 Ralph M.Nelson, Jr.

the hardness of wax on the fuel surfaces or by the amount of surface cuticle.
Within each group, equations were developed for recently cast and weathered
fuels undergoing adsorption and desorption. All equations showed that re-
sponse times were proportional to the quantity (J^J8^5'^, where exponents a, b,
and c differed among fuel species and sorption conditions. Anderson stated that
increasing /3 and 8 when a is constant decreases D^^ and increases r, and that
additional study is needed.

III. FUEL MOISTURE RELATIONSHIPS

The moisture content of living fuels involves free water and water vapor in plant
void spaces and in the cell walls and surrounding tissues. For example, an equi-
librium relationship between water content and water potential describes water
retention in plant tissue, bulk capillary flow takes place in the stems of woody
plants, and water exits the interior of leaves by vapor diffusion (Kramer and
Boyer, 1995). Similar phenomena take place in dead fuels. Approaches for de-
scribing moisture held at equilibrium in hygroscopic porous solids range from
empirical and semiempirical models to rigorous theories based on the physical
chemistry of solutions or statistical thermodynamics. Water movement in po-
rous bodies, including dead forest fuels, can be described mathematically in
two ways. In the first, water transport is assumed to obey the conservation
equations of engineering mechanics (Spolek and Plumb, 1980). The second ap-
proach involves the procedures of nonequilibrium thermodynamics (Katchal-
sky and Curran, 1965). The underlying foundation of this method rests in the
conservation equations of mass and energy and in the thermodynamic laws but
involves additional assumptions about the conservation of momentum and de-
viations from equilibrium. The primary driving force for mass flow is the spa-
tial gradient of the partial molal Gibbs free energy; this idea has been accepted
by soil and plant water researchers for many years and now is drawing the atten-
tion of wood-drying researchers (Cloutier and Fortin, 1991, 1993; Siau, 1995;
Zhang and Peralta, 1999). The rapidly developing thermodynamic methods,
which have been little used in forest fuel moisture research, are not discussed
further because of space limitations.

A. WATER POTENTIAL

Researchers studying water relationships in soils and plants as separate ther-

modynamic systems have long recognized that water tends to move in these
materials from regions of high Gibbs free energy to regions of lower free energy.
This is consistent with the thermodynamic concept of equilibrium which tells
Chapter 4 Water Relations of Forest Fuels 97

us that two bodies (or two regions within a single body) are in equiUbrium only
when the Gibbs free energy difference between them is zero. Water moves across
the soil-atmosphere, soil-plant, and plant-atmosphere interfaces in response
to free energy gradients, so viewing the three separate domains as one unified
thermodynamic system is a logical research approach. This single-system con-
cept has been labeled SPAC to refer to the "soil-plant-atmosphere continuum"
and has been accepted by many workers in the two subject matter areas.
The energy status of water in living or dead plants, soils, or the atmosphere
measures the potential energy of the water in its current state when tempera-
ture, pressure, and chemical composition are the only variables. This energy
is the partial molal Gibbs free energy, or chemical potential, ii (J mol~^). The
standard reference state in both soil and plant water research, denoted by /XQ,
is taken as pure liquid water at atmospheric pressure and at the temperature of
water in the material. The reference location is an arbitrarily chosen elevation.
Because water is essentially incompressible (except at very small moisture frac-
tions), its density (kg m~^) and molecular weight (kg mol"^) can be used to ex-
press [JL in units of pressure, J m~^. Thus water potential ijj is defined as

where p^, M^, and V^ are, respectively, the density, molecular weight, and mo-
lar volume of liquid water and {JL and IJLQ are chemical potentials of water in the
plant and in the reference state. For plant water in the reference state, ij/ = 0.
The water potential ip (J m~^), applicable in both soils and plants, is written as

(A = fAp + (A^ + (A, + (Ag (4)

where if/^ is the pressure potential owing to plant turgor (hydrostatic forces in
excess of those due to atmospheric pressure acting on cell walls and internal
membranes), J/^^ is the matric potential caused by capillary or binding forces at
internal surfaces of cell walls, if/^ is the osmotic potential due to a reduction in
water vapor pressure by solutes and surface attraction, and ip^ is the gravita-
tional potential which originates from the difference in gravitational force act-
ing on water at a given location and at the reference location. In general, ijj^ is
taken to be negligible in the water of living plants, and if/p is considered to be
zero in soil water. Cowan (1965) and Philip (1966) were among the early work-
ers who attempted to describe water flow in living plants with physical coeffi-
cients like diffusivity and conductivity, but within the continuum framework.
Lemon et at. (1971) and Riha and Campbell (1985) are other modelers who have
used a similar approach. References summarizing the SPAC concept and re-
search on water transport in plants are Kramer and Boyer (1995) and Kozlow-
ski and Pallardy (1997).
98 Ralph M.Nelson, Jr.

B. CONSERVATION EQUATIONS

Though the conservation laws are equally applicable to live fuels, they are dis-
cussed here in the context of dead fuels because their application to water trans-
port problems in these fuels is relatively simple. The solid fuel and its associated
moisture comprise four separate phases: the dry solid, free liquid water, bound
liquid water, and a gas phase consisting of a mixture of dry air and water vapor.
Equations expressing the conservation of mass, momentum, and energy may
be written for each phase. The intricate chemical and structural features of fuel
particles or layers require the moisture exchange modeler to consider carefully
the elementary volume selected as the basis for writing equations governing
the process. Though a set of differential equations can be written to describe
changes within each phase in the system, it is not possible to know which set
to assign at each point in the medium. One approach to this problem is to de-
fine an elementary volume within the particle just large enough to contain all
phases and satisfy the definitions of all system variables. Then a set of volume-
averaged equations may be written that should be valid at all locations and the
fuel-water system may be treated as a continuous medium (Spolek and Plumb,
1980). This procedure is known as local volume averaging and has been used
by several researchers to describe heat and mass transfer in wood. Perre et al.
(1990) pointed out that moisture transport in wood at temperatures between
60 and 120°C is best considered a low- to moderate-temperature process taking
place with variable total pressure. On the other hand, an assumption of constant
total pressure may be reasonable for open duff layers or for moderately thin fuel
particles because within-particle temperatures, even in intense sunshine, sel-
dom exceed 60°C, and the internal pressure does not differ greatly from atmo-
spheric. Perre et al. also assume thermal equilibrium among all phases—an
assumption that is commonly used because of the mathematical simplification
it provides.
In the present discussion, we consider the unaveraged equations (equations
for each phase) that would describe heat, mass, and momentum transfer in an
elementary volume of fuel particle. Suppose that shrinking and swelling of the
particle are ignored and gravitational body forces, kinetic energy changes, and
the rate of doing work on the water are negligible. Then the equation that bal-
ances the transport of phase i in direction x (m) in a plane body (a slab, for ex-
ample) may be written as
d{p,(l>,)ldt = -d{p,(j),v,)ldx -w,= -d{],)/dx - w, (5)
where Pj (kg m~^), (/)i, v^ (m s~^), and Wj (kg m~^ s"^) are the density, volume
fraction, velocity, and evaporation rate of phase i—all on a unit volume of par-
ticle basis—referred to a coordinate system on the stationary particle. The mass
fluxes Ji (kg m~^ s~^) take forms that depend on the transport mechanism for
phase i. Subscript i (i = s, w, b, g) denotes the solid, free liquid, bound liquid.
Chapter 4 Water Relations of Forest Fuels 99

or gas phase, respectively. The dry air component of the gas phase, indicated by
subscript a, can be described with its own mass conservation equation given by

d{PaCt>a)'dt = - a ( p A V a ) / ^ ^ = " ^ ( J a ) / ^ ^ (6)

where w^ = 0. In addition, the diffusion flux of air can be written as

J , = -p,sD,d[ln{Pj - P,)]/dx = - p , 8 D , a [ l n ( P j ] / a x (7)

where sD^ (m^ s~^) is the diffusivity of water vapor in air (or air in water vapor)
corrected for presence of the solid (s is the void fraction) and total gas pressure
P j depends on the partial pressures of air and vapor, P^ and P^, according to

PT = Pa^ Pv (8)
Equality of temperatures T^ and T^ of the two gases and applicability of the ideal
gas law are assumed.
For moisture change in forest fuels, it is usually not necessary to include
equations for momentum conservation explicitly because flow rates are slow
and fuel temperatures rarely exceed 60°C. Though the transport relationships
describing Jj (Pick's law and Darcy's law) may be derived from approximate ex-
pressions of momentum conservation, their origin is empirical because of their
discovery during 19th century experiments. The equations for Jj, where i = w,
b, V (v denotes the water vapor component of the gas), may be written as

J , = -{Kj7j^)d{P^)/dx = -{Kjiq^)d{Pr - Pj/ax (9)

A = -psmD^d[\n{Pj,)]/dx (10)

J, = -{Kjr]g)d{Pj)ldx - p,8D,a[ln(P,)]/ax (11)

where K^ (m^), V^^ (m^ s~^), and P^ (J m~^ or kg m"^ s"^) in Eq. (9) are the
specific permeability of the medium to liquid water and kinematic viscosity and
pressure of the liquid phase. This equation is Darcy's law describing movement
of liquid water in an unsaturated two-phase system by capillary forces. Capillary
pressure P^ (J m~^) is the pressure difference at a gas-liquid interface given by

Pc = (Pg - Pw) = {PT - Pw) = 2y/r (12)

where y is the surface tension of the liquid and r is the mean radius of the in-
terface (Siau, 1995). Equation (10) originates in the equation of motion for the
adsorbed phase (Babbitt, 1950; Nelson, 1986b) and describes bound water dif-
fusion in the cell walls. Quantities p^ (kg m~^), Dj, (m^ s~^), m, and Pj, (J m~^)
denote the density of the dry particle, bound water diffusivity, local moisture
content fraction, and spreading pressure of water in the cell wall. The two-
dimensional spreading pressure of a film adsorbed on the internal surface of a
solid is analogous to the pressure of a three-dimensional gas in that adsorbed
water molecules j u m p from one sorption site on the surface to another because
of differences in film surface tension. Terms on the right side of Eq. (11) account
100 Ralph M. Nelson, Jr.

for the capillary flow and diffusion of water vapor in the medium. Quantities
Kg (m^), T7g (m^ s~^), p^ (kg m~^), and sD^ (m^ s~^) refer to the specific perme-
ability of the medium to gas flow, kinematic viscosity of the gas, density of the
vapor, and corrected vapor diffusivity. Equations (9)-(11) may be substituted
into Eq. (5) to obtain the within-phase equations of moisture transport. When
i = w in Eq. (5), Wj = w^; when i = h, w^ = w^,; when i = g, Wj = Wg = w^ =
— (w^ + Wj,) and the interphase sink/source terms sum to zero.
The conservation of energy within phase i may be written as
p^^qdydt + cJidydx + WjAi = -dq^/dx = d{}i^dTjdx)ldx (13)
where q (J kg~^ K"^) and Tj (K) are the constant-pressure specific heat and
temperature of phase i, Aj (J kg~^) is the heat of evaporation/condensation of
phase i, k^ (J m~^ s~^ K~^) is the thermal conductivity, and q^ (J m"^ s~^) is
the heat flux that accounts for all mechanisms of heat transfer to the phase. In
Eq. (13), it is assumed that conduction is the only means of heat transfer, and
use is made of another empirical relationship—the Eourier law of heat conduc-
tion. Equation (13) must be summed over all phases (assuming local thermal
equilibrium) because all are capable of transferring heat. It is noted that J^ =
^s ^ '^a~ K~ K~ ^\ in addition, Aj, = (A^ + A^), where A^ (J kg~^) is
the differential heat of sorption. A discussion of general wood-water thermo-
dynamic relationships, including the differential heat of sorption, is given by
Skaar (1988).
Consistent with their application and the desired level of complexity,
Eqs. (5)-(13), or a subset of them, may be solved analytically or numerically in
conjunction with fixed or variable boundary conditions involving one or more
of the following variables: moisture content, temperature, relative humidity,
wind speed, solar radiation, and net longwave radiation from various sources.
Use of these equations requires still more information that includes expressions
for void fraction, transport properties, thermal properties, and thermodynamic
relationships. For multiple phases, relative amounts of space occupied by free
water and water vapor are needed. Examples of the incorporation of this kind
of information into models describing multiphase transport through volume
averaging are the studies of Spolek and Plumb (1980), Perre et al. (1990), and
Fernandez and Howell (1997).

C. LIVE FUEL MOISTURE

The understanding of water retention and movement in plants and in the soil
supporting their growth is basic to development of models for predicting the
moisture content of live surface and crown fuels. These models would help for-
est fire management and control personnel quantify the behavior of crown fires
and anticipate the transition from surface fire to crown fire. Testing of a crown
Chapter 4 Water Relations of Forest Fuels 101

fire spread model that uses fuel moisture content as an externally supplied in-
put parameter is underway (Alexander et al, 1998).

1. Moisture Characteristic Curves

The equilibrium relationship between plant tissue water content and water po-
tential at constant temperature is often referred to as the moisture character-
istic curve by plant water researchers, and similar terminology is used in soil
water studies. It is determined, in effect, by a balance between water potential
of the solution in the vacuole of the cell (protoplast) and water in the cell walls
(apoplast) and provides a mechanism by which liquid water moves into and
through plants (see Figure 1). Though the cell wall is permeable to water and
solutes, the membrane (or plasmalemma) surrounding the wall's inner edge is
impermeable to solutes. Thus water moves into or out of the vacuole according
to the difference in potential across the membrane. These differences, however,
are usually so small that local equilibrium may be assumed (i.e., il/ya = il^cw
where subscripts va and cw denote the vacuole and cell wall). Techniques are
available to measure the components of if/^^ and i/^cw separately—ma trie poten-
tial of the cell wall water J/^^^), pressure (or turgor) potential of the solution
in the vacuole, and the osmotic potential of both regions (Kramer and Boyer,
1995). The relationship between if/^^^-^ and relative water content, RWC, is shown
in Figure 6 for a Taxus branch suspended in a pressure chamber (McGilvary
and Barnett, 1988; Kramer and Boyer, 1995). Plant water researchers express if/
in terms of bars or MPa rather than J m~^, so their terminology is used here
(1 bar = 0.1 MPa = 10^ J m~^). The osmotic component in the cell wall, ifj^f^a)
in Figure 6, contributes in a minor way to the total potential of the wall, which
is the sum of the two components displayed. Quantity RWC is the water con-
tent of a material expressed as a percentage of the water content when the ma-
terial is fully turgid (in equilibrium with pure water). This turgid state is a more
convenient reference point than the dry state because it is experimentally re-
producible and not as subject to error due to plant growth. Thus RWC is com-
puted from

RWC = [(W, - W^)/{W, - Wa)]100% (14)

where W^.W^, and W^ are the material weights in the current, turgid, and oven-
dry (dried at 100°C) states. Gardner and Ehlig (1965) are among the research-
ers who have studied the water potential-RWC relationship. They measured
RWC, if/, if/^, and ipp of cotton, pepper, sunflower, and birdsfoot trefoil leaves
and found a linear relation between osmotic potential ij/^ and RWC. The total po-
tential (/^ was not linearly related to RWC because of a change in the modulus
of elasticity of the cell wall when the turgor potential ij/p approximated +2 bars.
This change in elasticity was associated with the early stages of wilting.
102 Ralph M. Nelson, Jr.

Cut surface
Adjustable seal

Wet filter
paper on wall

Baffle L-Disk
r False
^ ' f bottom
—--

Gas inlet—• o o o

B.

80 90 100
RWC (%)
FIGURE 6 Determination of the moisture characteristic curve. (A) Pressure chamber in which
incoming gas increases the pressure around a Taxus branch and forces xylem sap through the cut
surface. The pressure is adjusted to maintain zero flow at the surface and provides an estimate of
the internal tension on the apoplast. (B) Pressure chamber measurements show that water poten-
tial decreases as relative water content decreases and that a greater pull is being exerted by the leaves
on water in the xylem as water content decreases. Osmotic water potential of the cell wall is given
by *As(a)- From Kramer and Boyer (1995).

2. Water Transport in the SPAC

The mechanisms governing transport of water through living fuels may be de-
scribed by several methods. One is empirical and often used to fill knowledge
gaps with information based on experimental data. In the second method, it is
assumed that the water-conducting system is analogous to an electrical circuit
Chapter 4 Water Relations of Forest Fuels 103

and requires evaluation of flow resistances in series and/or parallel connections.

A third method involves solution of the differential equations governing flow
in the various s o i l - p l a n t - a t m o s p h e r e domains and uses conductivities and dif-
fusivities rather than resistances. The domains most conducive to analysis with
differential equations are those describing soil water flow and uptake by roots.
Modelers often describe flow in the whole plant using some combination of
these three methods (Federer, 1979; Riha and Campbell, 1985).

a. The Ohm's Law Analogy

The driving force for water movement in transpiring plants originates in the
leaves and is transmitted to the roots through the cell sap. Transpiration re-
duces the leaf water potential, and this disturbance is propagated through the
xylem of the leaves and stem to the roots where the reduced potential induces
water flow toward the roots. This process occurs when water content of the soil
is high or the transpirational demand is low. On the other hand, when the soil
water content is low or the transpirational demand is great, leaf water potential
falls to such a low value that the leaves lose their turgor and the stomates tend
to close. Thus, the leaf resistance is greatly increased, the rate of transpiration
is reduced, and further loss of water is controlled by the plant and soil rather
than the atmospheric conditions (Cowan, 1965). In the SPAC method of analy-
sis, the overall movement of water may be analyzed using an analog of Ohm's
law which states that the flow of current in an electrical circuit is proportional
to the potential difference across the circuit. The electrical analog shown in
Figure 7 is a simplified representation of the paths taken by water from its en-
try via the water table to its exit through the plant leaves. The xylem portion of
the circuit applies to the complete vascular system—xylem in the roots, stem,
and leaves. The overall process may be crudely approximated with four sub-
processes: water flow from soil to roots, from roots to stem, from stem to leaves,
and from leaves to the atmosphere. Figure 8 illustrates the relative magnitudes
of change in water potential for each component of the SPAC. Because the
change in potential from root to leaf is only a few bars, the stem sometimes is
omitted from considerations of overall flow.
The SPAC is nonstationary in character because of diurnal and seasonal vari-
ation and because of periods of hydration and desiccation. To eliminate some
of this complexity, plant water researchers often assume a steady state rate of
transport. Though this represents considerable simplification, investigators can
work with a series of steady states instead. The steady state assumption implies
that a disturbance of water potential in the roots (or leaves) is propagated in-
stantaneously throughout the plant to the leaves (or roots) and that internal
changes in water content during the process are small in comparison with the
magnitude of the disturbance (Cowan, 1965). Jarvis (1975) and Waring and
Running (1976) argue, however, that in plants with water sources of sufficient
104 Ralph M. Nelson, Jr.

Cuticle

FIGURE 7 Electrical analog showing soil, root, vascular, leaf, and atmospheric flow resistances
in the SPAC. Resistances r^;, r^, and r^ correspond to r^^, r^, and r^ in Eq. (15); the overall resistance
due to r^„, r^, and rj,| is represented by r| in Eq. (15). The shaded area at the right indicates relative
magnitudes of water potential for the various components. From Rose (1966).

size in the stem and branches and with radial (or horizontal) pathways to the
stem xylem of low enough resistance, sufficient water may be withdrawn from
these regions of storage to satisfy the transpirational demand for considerable
periods. Jarvis points out that little is known about resistances to flow between
these storage regions and the xylem and about the moisture characteristic of
the cambium, phloem, or roots. Nevertheless, if steady state is assumed, stem
resistance is omitted, and soil moisture is not limiting, the transpiration rate
Q (m s~^) equals the flow rate in all plant domains and
Q = i^so - ^rVirso + r,) = ((A, - (A,)/(r, + r,) = {iff, - (Aa)/(n + r j
(15)
where r (bar-s m^) symbols denote resistances and subscripts so, r, I, and a re-
fer to soil, root, leaf, and atmosphere. The last term in this equation is a rough
approximation during the day because its value is influenced by temperature
Chapter 4 Water Relations of Forest Fuels 105

-500 H

-300

-100

1 r
B p P E
soil ' root stem . leaf atmosphere
FIGURE 8 The distribution of water potential throughout a woody plant (not to scale). Curves
1 and 2 indicate extremely wet soil; curves 2 and 4 indicate a high atmospheric demand. The largest
change in potential occurs in the leaf. From Hillel (1980b).

changes associated with solar radiation and periods of cloudiness. Moreover,

the term often is written with differences in vapor pressure or vapor concen-
tration replacing the water potential difference (Kramer and Boyer, 1995).
It is recognized by plant water researchers that the assumptions commonly
associated with the SPAC approach, namely steady state flow and constant flow
resistances, make this kind of analysis a crude approximation at best. For ex-
ample, total resistances often must be evaluated by dividing the potential differ-
ence by transpiration rate Q. They are continuously changing, however, because
transport can be controlled either by the soil, plant, or atmosphere. Numerous
authors have studied the relative importance of r^^ and r^, and differences of
opinion continue to exist. Results from these studies are briefly summarized
by Kramer and Boyer (1995) who state "evidence seems to indicate that over
a considerable range of soil water content and with average root density, root
resistance exceeds soil resistance to water movement." Leaf resistance also is
complicated and includes components due to liquid flow in the petiole, veins,
and mesophyll cells as well as vapor diffusion through the cuticle, substomatal
106 Ralph M. Nelson, Jr.

cavities, and stomates (Kreith and Sellers, 1975). Root and leaf resistances from
several studies are reported by Slatyer (1967) and Cowan and Milthorpe (1968).

b. Soil-Root Models
Movement of soil water may be described by the Richards equation which is
based on mass continuity and momentum conservation as expressed by Darcy's
law. Derivations of the Richards equation for saturated and unsaturated flow
may be found in soil physics texts such as Hillel (1980a). If the influence of os-
motic potential on soil water flow is negligible and the system is isothermal, the
Richards equation for vertical flow in unsaturated soils may be written as
do/dt = Q(ah/at) = d[Ki,{dh/dz) + K^ysz (16)
where 6 is the volumetric soil water content, t (s) is time, h (m) is the pressure
head applicable to flow caused by capillary and attractive forces, C^ (ni~^) is
the specific water capacity given by {36/dh), K^^ (m s~^) is the hydraulic conduc-
tivity, and z (m) is the vertical distance from the reference location. Head h is
given by P^^Pw^^ where P^ (J m~^ or kg m~^ s~^) is the hydrostatic pressure, p^
(1000 kg m~^) is the density of water, and g (9.8 m s~^) is the gravitational ac-
celeration. For horizontal flow, z must be interpreted as a horizontal distance,
and the second term in the brackets is set to zero. In the case of saturated flow,
conductivity K,^ may be regarded as the hydraulic conductivity of the saturated
soil (a constant), and the derivative of the second term in brackets vanishes.
Various empirical and theoretical relationships expressing conductivity KJ^ in
terms of 6 or matric potential if/^ and soil texture have been reported (Scholl,
1976; Jury et al, 1991; Alessi et al, 1992). Equation (16) may be expressed in
terms of if/^ (J kg"^) by substituting the relationships

^m = PJpw = gh and p^Kj, = gK^ (17)

into Eq. (16) to obtain
p^XdOldt) = p^C^idijjJdt) = d[K^{diljJdz) + gK^]/dz (18)
where C^ (s^ m~^) is the slope of the soil-moisture characteristic curve, or
specific water capacity (dOldilj^), and conductivity K^ (kg-s m~^) differs from
K/^. Equations (16) and (18) contain no source or sink terms to account for loss
or gain of soil water by the roots. Riha and Campbell (1985) added a sink term
to the right side of Eq. (18) to account for absorption of such water. Their term
was of the form

U = PA^SO - ^rVirso + rr) (19)

where U is the root sink strength (kilogram of water absorbed per second per
cubic meter of soil) and p, the root density (meter of root length per cubic me-
ter of soil). Potentials are in units of J kg~^ and resistances in m^ kg~^ s ~ \
Chapter 4 Water Relations of Forest Fuels 107

The uptake of soil water by roots may be modeled by two differing ap-
proaches in which the roots are treated as uniform cylinders of length L. The
first method considers a single root of uniform radius surrounded by a cylindri-
cal sheath of soil. W h e n solute effects are neglected and equilibrium is achieved
between liquid in the cell vacuole and that in the cell wall, radial transport of
water into the root may be described by

difjjdt = D^{d^il/Jdr^) + {Dj,/r){dilfjdr) (20)

where D,^ (m^ s~^) is the hydraulic diffusivity of the root tissue, r (m) is radial
distance in the root, and potential i/^^ has the units J m~^ (Hillel, 1980b). The
relationship between K^ of Eq. (16) and Dj, of Eq. (20) is

Q D , = K, (21)

whereas Eqs. (16)-(18) may be used to write

Q = gC„, (22)
When the soil moisture contains a significant amount of solutes, differences in
osmotic and matric potential between the sheath of soil surrounding the root
and the root itself must be considered (Hillel, 1980a; Flowers and Yeo, 1992).
The second approach is to consider the entire root system as a moisture sink
which penetrates the soil to a known depth. This root zone depth may be sub-
divided into layers that contain different root masses per unit volume of soil.
The major fault of this method is that it is based on gross averages of root den-
sity and water potential over the root zone.

c. Leaf-Atmosphere Models
The influence of the atmosphere on water movement in plants can be evalu-
ated from the rate of evaporation from a free water film covering the leaf surface.
This rate is referred to as the potential rate of evaporation, QQ (kg m~^ s"^), and
may be estimated from the simple "Ohm's law" equation

Qo = {Psat - Pva)l^a (^3)

where p^^^ (kg m~^) is the concentration (or density) of water vapor at the leaf
surface under saturated conditions and p^a the vapor concentration in the am-
bient air. Further discussion of Qo is given in Chapter 9 in this book. Resistance
offered by the air to vapor diffusion from the leaf surface is r^ (s m"^) and can
be estimated by two methods. The first method, for continuous areas of more
or less uniform vegetation, utilizes the logarithmic wind profile (see Chapter 9
in this book) and therefore considers such factors as vegetation height, zero
plane displacement, and roughness length (Monteith, 1963a; Rutter, 1968). The
second approach, for single leaves, makes use of boundary layer concepts and
108 Ralph M. Nelson, Jr.

depends on leaf size and shape and on wind speed (Kreith and Sellers, 1975).
When evaporation takes place from the interior leaf surfaces, the process is re-
ferred to as transpiration. In this case, the resistance to vapor diffusion occur-
ring within the leaf, rj, should be accounted for and r^ in Eq. (23) replaced by
i^a + n) • Resistance rj includes cuticular resistance which is in parallel with the
resistance to vapor diffusion. This diffusion, or stomatal, resistance is made up
of resistances originating in the mesophyll cell walls, intercellular spaces, and
stomatal pores linked in series. Schonherr (1976) studied the diffusion of water
through cuticular membranes and found that the water permeability of the cu-
ticle is completely determined by waxes rather than cutin in the cuticular layer.
Cuticular resistance is 10 to 100 times greater than the stomatal resistance and
generally can be ignored in calculations of rj (Kozlowski and Pallardy, 1997).
The term evapotranspiration refers to loss of water from the soil due to the
combined water loss from plants by transpiration and from the soil surface by
evaporation (see Chapter 9). W h e n all surfaces are moist, evapotranspiration
proceeds at the potential rate and depends primarily on meteorological factors.
In this case, the potential evapotranspiration, QQ, may be calculated by at least
two different methods. The first is outlined by Kreith and Sellers (1975) and is
applicable to evaporation from large areas (an ecosystem or field). Their ap-
proach requires that temperatures of the evaporating surfaces be known. The
second method utilizes the Penman equation—an equation which accounts for
the effect of solar radiation on the evaporation from a water surface in such a
way that knowledge of surface temperatures is not required. Various forms of
the equation are widely used in agricultural applications to compute QQ. Mon-
teith et al. (1965) and Monteith (1980) modified the Penman equation to de-
scribe potential evaporation from whole plants or crops over an extended area.
The modified model, referred to as the Penman-Monteith equation, is discussed
in Chapter 9.
According to Monteith (1980), there are only two ways to compute the ac-
tual evapotranspiration rate, Q. One method is to utilize complex models within
the SPAC framework (Lemon et al., 1971; Waring and Running, 1976); the sec-
ond is to apply models such as the Penman-Monteith equation to the entire
plant canopy. This application, which assumes that stomatal resistance is uni-
form throughout the canopy and that temperature, humidity, and momentum
profiles are similar within and above the canopy, was questioned by Philip
(1966) and Tanner (1968). Subsequent research has shown that the Penman-
Monteith equation is useful when applied to well-developed crop canopies but
underpredicts transpiration from sparse canopies when soils are dry and over-
predicts it when soils are wet. Refinements in the equation by agricultural re-
searchers have led to compartment schemes in which evaporation from the soil
surface and transpiration from the crop canopy are treated separately (Shuttle-
worth and Wallace, 1985; Choudhury and Monteith, 1988). This approach has
Chapter 4 Water Relations of Forest Fuels 109
produced better agreement with experimental measurements of transpiration
because of improved modeling of soil heat and vapor fluxes (Wallace et al,
1990; Tourula and Heikinheimo, 1998).
When the soil moisture or soil water potential falls below some critical value,
the evapotranspiration rate falls below the potential value. Lopushinsky and
Klock (1974) found that transpiration rates in several conifer seedlings began
to decrease when the soil water potential fell below —2 bars (Figure 9). The
transpiration rates of several pines were slower than those of fir and spruce by
more than a factor of two, explaining the ability of pines to survive on fairly dry
sites and the need for fir and spruce to occupy more moist sites. Lopushinsky
and Klock attributed the decrease in transpiration rate to stomatal closure and
noted that such closure takes place in all species when the leaf water potential
ranges from about —14 to —25 bars. Brown (1977) studied the wilting process
in two range grasses under controlled environmental conditions. He measured

1 r
0 -2 -4 -6 -8-10-12-14-16-18-20 0 -2 -4 -6 8 -10-12 -14-16-18-20
Soil water potential (bars)
FIGURE 9 Transpiration rate as a function of soil water potential for seedlings of various coni-
fers. Curves at the lower right represent regression equations through the experimental data. From
Lopushinsky and Klock (1974). Reprinted from Forest Science (vol. 20, no. 2, p. 183) published
by the Society of American Foresters, 5400 Grosvenor Lane, Bethesda, MD 20814-2198. Not for
further reproduction.
no Ralph M. Nelson, Jr.

c
0)
o
Q.
9?
13
CO
w
Q)

lethal
threshold

protoplast collapse
FIGURE 10 Relationship between pressure (or turgor) potential and leaf water potential for two
range plants. From Brown (1977).

soil and leaf water potentials during a drying process in which the soil water
potential decreased from 0 to about —65 bars. In the early stages of drying, the
pressure potential (the difference between the leaf water and osmotic potentials)
decreased to zero—a value labeled by Brown as the physiological wilting point
(Figure 10). As drying continued, the pressure potential decreased to a mini-
mum of about —5 bars (the point oi protoplast collapse) and then increased to-
ward zero where severe wilting occurred at a leaf water potential between —70
and —80 bars; Brown labeled this point (or range) the lethal threshold and asso-
ciated it with the collapse of all plant cells. Brown (1977) stated that other un-
published research by him indicates the measurement techniques he used are
equally applicable to trees, shrubs, forbs, and grasses growing in arid environ-
ments and that similar experiments can provide quantitative guidelines for
evaluating the adaptability of range plants to dry conditions.

3. Canopy Moisture Content

The literature of forest physiology contains much information on evaporation
and transpiration rates but is almost devoid of attempts to model the water
content of understory and overstory canopies. It is likely that within a given
fuel type there exist critical combinations of canopy moisture content, relative
weights of live and dead fuel, and ambient wind speed which will permit the
spread of fire in tree crowns, understory shrubs, or open grassland. The ability
Chapter 4 Water Relations of Forest Fuels i l l

to predict canopy moisture content seasonally would be a help in understand-

ing and anticipating the initiation and spread of fire in these fuel types.
Changes in the moisture content of live fuels are poorly understood—partly
because plant physiologists have studied the response of plants to environ-
mental stress in terms of water potential rather than moisture content (Brown
et al, 1989). These authors also state that seasonal variation in water content
is more closely related to phenological changes such as leaf growth and matu-
ration than to environmental factors. In earlier work, Kozlowski and Clausen
(1965) pointed out that changes in plant tissue moisture content can result
from changes in actual water content, tissue dry weight, or both. Their pheno-
logical study of leaves from several forest trees during the 1963 growing season
showed that the moisture content of all deciduous leaves studied decreased un-
til mid-June, after which time the rate of decrease slowed and became depen-
dent on species. This result was attributed to increases in dry weight rather than
changes in the mass of water present. On the other hand, moisture content of
zero-age conifer needles decreased continuously, whereas in 1-year-old needles
it increased until about mid-July and then became relatively constant. In this
case, the dry weight changes were attributed to carbohydrate translocation into
new needles and out of the older ones. Little (1970) conducted a year-long study
of water, sugar, starch, and crude fat content in juvenile balsam fir needles. In
contrast with the results of Kozlowski and Clausen (1965), he found that sea-
sonal fluctuations in moisture content generally were caused by changes in ac-
tual water content rather than by dry weight changes. Little pointed out, how-
ever, that changes in moisture content due to dry weight changes increase with
increasing age, and that increased carbohydrates may not have accounted for
the total change in dry weight of his study—as they might have in the matur-
ing balsam fir needles of the Kozlowski and Clausen study. Research on Engel-
mann spruce needles conducted by Gary (1971) showed that moisture contents
were determined primarily by dry weight changes and that differences between
needles from north and south slopes were small. Chrosciewicz (1986) sampled
the foliar moisture content of several Canadian conifers in central Alberta. His
sampling procedure was designed to minimize variation due to the diurnal
cycle and sample location within the canopy. Figure 11 shows some of his re-
sults for new and old (needles 1,2, and 3+ years old) foliage measured during
1974. The flushing of new foliage coincides with the moisture content mini-
mum in the old foliage. Moisture content of the new foliage approaches that of
the old in September.
In a study extending from January to October 1971, Hough (1973) mea-
sured extractive content, phosphorus content, and moisture content of old and
new needles to determine their effects on the flammability of natural sand pine
stands. A sharp rise in moisture content of new needles up to 10 months in age
coincided with the initiation of new growth, whereas ether-soluble extractives
112 Ralph M. Nelson, Jr.

480-

460-

440-

420-

400-

380_

360-

340-

320-

300-

280_
B
New foliage
o
o 260-
0)

240-
220-
200-
180-

160-
140-
120-

100-

80-
Mar. I Apr. I May I June I July I Aug. I Sept.
60-
1I I\ I\ I\ I\ In I I\ i'\ InI i'\ Ir M ' I r i ' I I
1 11 21 31 10 20 3010 20 30 9 19 29 9 19 29 8 18 28 7 17
1974
FIGURE 11 Seasonal moisture content variation in new and combined old foliage (1, 2, and
3+ years old) of four Canadian species. Right-pointing arrows indicate the start of flushing of new
foliage. With the arrows as reference points, the species from top to bottom of the figure are white
spruce, black spruce, balsam fir, and jack pine. From Chrosciewicz (1986).

of new and old (9-18 months in age) needles peaked in late February and then
decreased until June. Phosphorus content of new needles decreased until March
and then peaked in late May, while that of the old needles decreased until April
and then slowly increased. Hough concluded that both weather and fuel vari-
ables affect fire occurrence in stands of sand pine in Florida and that crown fires
Chapter 4 Water Relations of Forest Fuels 113
would be most likely to occur in late February and early March. He stated that
the severity, or intensity, of wildfires is not reduced until the new needles, with
their high moisture content and low extractive content, make up most of the
crown.
One naturally would expect understory vegetation to show patterns of varia-
tion in moisture, mineral, and extractive content similar to those in tree crowns
(Hough, 1973), and the results of numerous experimental studies have shown
this to be so. Fire researchers have documented the moisture content of chapar-
ral and brush fuels in southern California for many years. Countryman (1974),
for example, reported field data on the variation in moisture content of man-
zanita and snowbrush (Figure 12). New growth in both species typically begins
in early June with moisture contents of about 200% of the fuel ovendry weight.
The moisture content of old foliage increases more slowly, and both species
reach a minimum level in September. Countryman stated that the percent of
total fuel made up of living material usually ranges from 65 to 85% and that
fire behavior in the chaparral types depends on both live fuel moisture content
and relative amount of dead fuel present. According to Blackmarr and Flanner
(1975), the understory shrubs growing on organic soil in the pocosin areas of
eastern North Carolina cause a severe fire hazard during periods of drought.
These authors sampled the moisture content of six shrub species native to the
area for a 22-month period. Their 1964 and 1965 data for deciduous and ever-
green species showed that all parts of the evergreen foliage were consistently
drier (ranging from 250 to 100% during the summer) than those of the decidu-
ous species (350 to 150%), tending to make them more flammable. In earlier
work, Blackmarr and Flanner (1968) measured the diurnal moisture content
variation in four North Carolina pocosin species over a 36-hour period dur-
ing June 1965 and found differences ranging from about 25 to 40% moisture
content.

220 A New manzanita

\
^ 200
"c 180 New snowbrush •,
0)
c 160
o
o 140
1 Old snowbrush
CO 120
o
^ 100
80
60-4 \ \ \ \ \ 1 \
April May June July August September October
FIGURE 12 General trend of seasonal moisture content variation in two chaparral fuels in south-
ern California. From Countryman (1974).
114 Ralph M. Nelson, Jr.

The moisture content of live fuels in open grasslands generally follows pat-
terns similar to those of the understory shrub and crown fuels discussed earlier,
but it also differs in some respects. These patterns are determined primarily by
weather. The annual grasses usually have shallow root systems and therefore
depend strongly on moisture availability near the soil surface for their growth.
In the spring, these green grasses can attain fractional moisture content values
larger than 3.5. As hotter weather approaches, the growth of annuals (roughly
from May to July in the United States) is shortened by the absence of rainfall.
The plants mature, produce seed, and then begin to cure (or dry). Their rate of
curing depends on soil moisture and rainfall, resulting in curing times that may
vary from three to eight weeks. As the plants flower and produce seed, the color
of their various parts turns from green to yellow (or other colors such as purple,
depending on species). According to Luke and McArthur (1978), this process
begins in the annuals of Australia when the moisture content fraction has de-
creased to about 1.75. As the moisture fraction drops below 1.0, yellowing has
affected about 30% of the plant surface, and the grass is said to be 30% cured; at
a moisture fraction of about 0.5, the grass is 60% cured and has reached a state
of dryness such that it cannot recover its moisture regardless of the amount of
additional rainfall. At moisture fractions below 0.3, annuals are close to fully
cured and become dead fuel now subject to changes in atmospheric relative hu-
midity and temperature. The perennial grasses exhibit a similar, but longer,
curing period (from May to September in the United States) because they have
root systems that penetrate more deeply into the soil than those of annuals; this
makes them less sensitive to short-term weather and surface soil moisture. In
warm and humid climates, some parts of perennials cure and die whereas other
parts remain alive but in a dormant state (Schroeder and Buck, 1970). In the
United States, Brown (1977) reported water potential relationships he observed
during the wilting of two range grasses (see Section Ill.C.l.c); Brown et al.
(1989) measured the moisture content of perennial grasses under aspen stands
in Wyoming. The latter authors noted a change in color of the green vegetation
at a moisture fraction of about 1.0 (corresponding to 30% curing of Australian
annuals) and found that these grasses green up and increase in moisture con-
tent in response to rainfalls greater than 6 mm. Similarly, Luke and McArthur
(1978) reported a delay in the curing of Australian perennials due to 30 mm of
rain. The moisture content of grassland fuels in Australia has been related to the
Keetch-Byram Drought Index (see Section IV for several references on drought).
Luke and McArthur (1978) state that Australian annuals are fully cured when
the Keetch-Byram index exceeds 200 and that perennials are cured when the
index is close to 600. In studies conducted in the United States (northeastern
California), Olson (1980) observed that the foliar moisture content of blue-
bunch wheatgrass (a perennial) decreased exponentially as the Keetch-Byram
index increased.
While the need for models that predict moisture content change in the cano-
Chapter 4 Water Relations of Forest Fuels 115

pies of grass or understory/overstory vegetation is generally recognized, the

complexity of the task may have precluded all attempts to develop a reasonably
complete physics- and physiology-based model (Tunstall, 1991). The required
model should include compartments treating such topics as soil water transport
to roots and directly to the atmosphere, water takeup by roots, water rise in the
main stem, growth of roots, stem, and foliage as determined by photosynthesis
and respiration, storage and translocation of water and food, water potential in-
teractions, water vapor diffusion through leaves and into the atmosphere, and
senescence (the process of wilting and dying). A practical solution to the prob-
lem may be achievable if we can eventually identify and model a few key con-
trolling processes (Howard, 1978); alternatively, perhaps we can successfully
modify existing models of growth and transpiration in conifers and understory
vegetation to obtain reasonable predictions of canopy moisture content (Run-
ning, 1978, 1984a, 1984b).

D. DEAD FUEL MOISTURE

The moisture content of dead forest fuels in the hygroscopic range is of consid-
erable interest to fire managers. The equilibrium moisture content is useful in
laboratory studies in which classical diffusion theory describes moisture change
driven by step changes in the fuel's environment. More recently, equilibrium
concepts have been combined with diffusion theory to develop new methods
for estimating the moisture transport properties of fuels under field conditions.

1. Equilibrium Moisture Content

The fractional equilibrium moisture content M^ represents the constant value of
M attained by dead forest fuels when they are exposed for an extended period
in air of constant relative humidity and temperature and in which changes due
to variability in wind, solar radiation, and barometric pressure are small. In lab-
oratory experiments, equilibrium values of fuel moisture content depend on
whether that state is attained following adsorption or desorption because of
various mechanical and environmental factors that result in sorption hystere-
sis. The effects of hysteresis are not apparent in field studies because of a con-
tinuously changing environment. Models that describe the equilibrium state in
fuels and soils over the entire range in relative humidity eventually will simplify
the description of capillary and sorption processes in composite layers of soil,
duff, and litter.

a. Adsorption and Desorption

W h e n wood and foliage reach equilibrium at various values of H and con-
stant T, the resulting M^ values, plotted versus H, form a sigmoid curve charac-
116 Ralph M. Nelson, Jr.

• 55° Cn
0.30-^ 0.30
•
•
40°
25°
C
C
Adsorption 1
A 10° C- it

i
0 55° Cn A/1
O.25J D 40° c
Desorption
/O i 0.25
0 25° C
A 10° C-

0.20^ — 0.20 ^
A/ ^' O

,6: / 7
0.15^ /a //y — 0.15
o

0.10- \- 0.10

0.05- ^ ^^^y^y^^y^^^^ V- 0.05

^
n ^ 0
u 1
"1 ^ ^0.4 ^ ^ 0.7r V~~\
0 0.1 0.2 0.3 0.5 0.6 0.8 0.9 1.0
Relative vapor pressure
FIGURE 13 Sorption data for klinki pine wood at four temperatures. Note crossing of the 40°C
adsorption isotherm over the 23°C curve at a relative vapor pressure of about 0.95. From Kelsey
(1957). Reproduced by permission of CSIRO Australia.

teristic of most natural materials. If the successive H values increase, M^ in-

creases, and the process is referred to as adsorption; when the H values decrease,
desorption is occurring. The resulting curves are called adsorption and desorp-
tion isotherms, and the combined processes are often referred to as sorption.
Figure 13 illustrates these isotherms for klinki pine wood at temperatures from
10 to 55°C. The region in H from 0 to about 0.2 is considered one in which
most of the water in the wood is held by hydrogen bonding in a single layer. The
slightly curvilinear region from 0.2 to about 0.9 involves water held in multiple
layers within the cell walls but possibly also includes a small amount of capil-
lary condensed water in the finest cell wall capillaries as H approaches 0.9.
Chapter 4 Water Relations of Forest Fuels 117

Most of this water adds its volume to that of the cell walls and the composite
is best thought of as a solid solution (water dissolved in cell wall substance).
Because of swelling, the internal area available for adsorption due to forming
cellulose-water bonds or water bridges between hydroxyl groups is about 1000
times greater than the area that would be available if water did not swell wood.
True capillary condensation takes place for H between 0.9 and 0.95, but these
values are rarely exceeded in commonly used experimental methods of mea-
surement. Specialized measurements extending H to 0.999+ have been made
on wood and wood pulp using the tension plate, pressure plate, pressure mem-
brane, or nonsolvent water methods (Cloutier and Fortin, 1991; Zhang and
Peralta, 1999).
The effect of previous history on sorption for a given temperature is evident
in Figure 13 as the difference in M^ values for adsorption and desorption at con-
stant H, referred to as sorption hysteresis. The two most generally accepted ex-
planations of hysteresis are based on the availability of sorption sites in the
material and on the rheological properties of wood. In the sorption site theory
(Urqhart, 1960), it is supposed that during the initial desorption from the green
condition, all available hydroxyl groups will have water molecules attached to
them. As these molecules leave the sites and shrinkage occurs, many of the sites
become mutually satisfied through formation of cellulose-cellulose bonds and
a certain amount of rearrangement of the cellulose takes place; some unsatis-
fied hydroxyl groups remain, however, even when H is zero. Now as H is in-
creased during adsorption, water molecules are adsorbed on these remaining
sites, and nearly all sites will have one or more attached molecules, reducing
the attraction for additional molecules. Thus a greater H is required during ad-
sorption to attach the same number of water molecules (or achieve the same M^
value) as in desorption. This mechanistic theory of Urqhart (1960) remains, for
many researchers, the most reasonable explanation of hysteresis.
Research has not yet shown whether Urqhart's theory conflicts with the
thermodynamic arguments proposed by Barkas (1949) to explain hysteresis in
wood based on rheological properties, but it seems likely that the two explana-
tions will prove to be compatible. The rheological theory states that hysteresis
appears because gels such as wood are not perfectly elastic and hydrostatic
stresses caused by resistance to shrinking or swelling of the cell wall introduce
Mg values differing from the stress-free (or perfectly elastic) value. On the one
hand, fiber direction differences in the cell wall (Figure 3A) induce tension into
wood undergoing desorption, causing a larger M^ than if the wood were stress
free. On the other hand, interfibrillar bonding causes wood that is swelling dur-
ing adsorption to be in compression, resulting in a smaller M^ than if the wood
were stress free. Barkas (1949) shows that the area enclosed by a constant-
temperature hysteresis loop (as in Figure 13) represents the net work loss dur-
ing the entire sorption cycle. For spruce wood, Barkas calculates that this area
118 Ralph M. Nelson, Jr.

equals 17% of the total work of desorption. According to Skaar (1988), this
means that the ratio of M^ values for adsorption and desorption in spruce should
come to 0.83 (1 — 0.17 = 0.83) when averaged over the entire range in H. This
overall hysteresis ratio averages about 0.83 for both softwoods and hardwoods
(Skaar, 1988). For woody fuels, the hysteresis ratio for specific values of H gen-
erally ranges from 0.75 to 0.85 over the entire range in H from 0 to 1 (Stamm,
1964). These discrete ratios for some foliage fuels are more sensitive to H, as in-
dicated by values ranging from 0.61 to 0.86 (Nelson, 1984).
The immediate effect of temperature on water vapor sorption by forest fuels
is a reduced value of M^ as T increases. The reduction is due to the strong de-
pendence of saturation vapor pressure on T and the resulting tendency for
more water molecules to leave their sorption sites as T increases but H remains
constant. This reversible effect is illustrated in the desorption data of Figure 13.
Van Wagner (1972b) reported a decrease in M^ averaging about 0.024 per 10°C
increase for desorption in aspen and pine leaf litter and about 0.018 in aspen
wood splints. On the other hand, King and Linton (1963) and Anderson et at.
(1978) reported a decrease of about 0.01 per 10°C increase for sorption in as-
sorted natural fuels and ponderosa pine needles, respectively. At high values of
T and H, a second temperature effect, an irreversible reduction in hygroscopic-
ity, occurs when exposure times are long. The wood becomes discolored, and
the most hygroscopic component, the hemicellulose, decomposes first followed
by the pure cellulose. A still different effect at the higher values of H is that the
adsorption isotherms of wood and cellulosic materials at temperatures exceed-
ing about 40°C may cross those for lower temperatures. The effect is present in
Figure 13 in which the 40°C adsorption isotherm crosses the 25°C isotherm at
about H = 0.95. Though the 55°C isotherm does not extend to sufficiently high
H values to indicate crossing at that temperature, the data suggest that crossing
can occur in once-dried wood at about 40°C. Unfortunately, there exist no ex-
tensive laboratory data illustrating crossover in wood. The phenomenon has
been observed in American Uplands cotton (Urquhart, 1960) in which the 70
and 90°C isotherms crossed those for lower temperatures when H exceeded
0.85. Both Urquhart (1960) and Stamm (1964) attribute the crossover effect
in cotton to plasticization of the cellulose fibers; the resulting molecular re-
arrangement relieves drying stresses introduced when the cotton was first dried,
making new sites available for adsorption. Figure 13 also shows that for H = 0.5
the magnitude of hysteresis in klinki pine wood decreases by a factor of nearly
2 as T increases from 10 to 55°C; Skaar (1988) refers to German research indi-
cating that sorption hysteresis in European spruce wood disappears between 75
and 100°C. In a study of water vapor sorption by conifer and hardwood litter.
Van Wagner (1972b) noted that hysteresis decreased by a factor of about 2.5 at
"medium" values of H when T increased from 16 to 49°C.
Sorption of water vapor by forest fuels in the field is strongly affected by en-
Chapter 4 Water Relations of Forest Fuels 119

vironmental variables, and even the finest fuels rarely achieve equilibrium with
the continuously changing H and T values of the ambient air. The effects of H
and T on sorption are modified, however, by wind and by daytime solar radia-
tion and nighttime longwave radiation to the sky. These factors cause H and T
in the air immediately adjacent to the fuel surface to differ from their ambient
values. Thus M^ for a given fuel is a continuously changing value that would be
obtained if the fuel were exposed for a hypothetically infinite time to these dif-
ferent values of "fuel" temperature and fractional relative humidity, Tj and Hj.
One way to estimate M^ is to assume that Tj and Hj are the same as T and H. In
this case, approximating Tj with T might be acceptably accurate only if the fuel
were beneath a dense forest canopy, and even then, perhaps only at night. The
second, and more appropriate, way to obtain M^ under field conditions should
utilize measurements or models to evaluate Tj and Hj. Byram and Jemison
(1943) collected data on fuel temperature, air temperature and humidity, wind,
and fuel moisture content in an artificial sun apparatus in which fuels (hard-
wood leaf litter or basswood slats) were exposed to simulated solar radiation
and wind generated by light bulbs and fans. The data were used to develop con-
stants in a mathematical expression describing the effects of radiation and con-
vection on the temperature difference {Tj — T). Byram and Jemison then used
this difference to compute the ratio Hj/H by assuming that vapor pressures in
the fuel and air were equal. Values of M^ were determined by applying Tj and
HJ to tabulated equilibrium sorption data for wood. These figures were in good
agreement with measured M^, substantially verifying the substitution of Tj and
HJ for T and H. Byram and Jemison also studied experimentally the effect of
wind and solar radiation on M^ obtained following the drying of basswood slats
containing free water. They found that M^ for irradiated slats was larger when
ventilated than when wind was absent and attributed the difference to higher
HJ and smaller Tj values caused by the cooling action of the wind. Van Wagner
(1969b) repeated the field experiments of Byram and Jemison in a laboratory
setting to test the form of their equation for Tj. He observed similar effects of
wind and radiation on Tj for jack pine and aspen litter fuels, but the mathemati-
cal form of his equation was slightly different. Viney (1991) critiqued the Byram
and Jemison (1943) approach and noted omission of the effects of longwave ra-
diation and vertical gradients in vapor pressure on Tj and Hj.

b. Sorption Models
The study of sorption systems involving various solid-gas interactions has
led to identification of five types of sorption (Stamm, 1964). Type 1 sorption oc-
curs on all solids in a single molecular layer at temperatures above the critical
temperature of the gas. Types II and III take place in swelling solids and exhibit
sorption in multiple layers. In Type II, large quantities of heat are liberated due
120 Ralph M. Nelson, Jr.

to Strong interaction between the solid and the gas; in Type III, this interaction
is weak. Types IV and V are special cases of Types II and III and describe sorp-
tion in solids with very fine pores that become filled before the saturation vapor
pressure is reached. Researchers in the fields of wood products, textile manu-
facturing, leather processing, and food technology have found that water vapor
sorption isotherms for their respective materials follow Type II sorption. One
of the first isotherms derived for Type II adsorption was the Brunauer-Emmett-
Teller (BET) isotherm—a model that did not represent water takeup in wood
very well but spawned other theories that describe water takeup over the entire
range in H. Prior to the BET theory, however, the Bradley isotherm had been
developed on the basis of reduced dipole attraction between successive layers
of adsorbed water vapor molecules. Anderson and McCarthy (1963) derived an
equation similar in form to the Bradley isotherm. Their semiempirical theory
was based on experimental data that suggested the natural logarithm of the
partial molal enthalpy of adsorption is linearly related to M^; it successfully de-
scribed water vapor sorption data in silk, nylon, and cotton fibers over a range
in H from about 0.1 to 0.85. A detailed discussion of sorption theories is be-
yond the scope of this chapter; the reader may consult the summary in Skaar
(1988).
Despite its apparent success, the Anderson-McCarthy sorption equation
must be regarded as approximate because it uses the highly improbable as-
sumption that the entropy change of the adsorbed water always is zero. In an
alternative approach, Nelson (1984) reported that for sorption in forest fuels
the natural logarithm of the change in Gibbs free energy. In AG, is linearly re-
lated to Mg over most of the hygroscopic range according to
In AG = A + BM, = In AGa[l - (M,/M,,,)] (24)
where AG^ (J kg"^) is the limiting value of AG as M^ and H approach zero, and
A and B are constants. Quantity M^^^ is an estimate of the fiber saturation point
Mjsp under desorption conditions; for adsorption, M^^^ approximates the aMj^^
product, where a is the hysteresis ratio. The decrease in Gibbs free energy
when pure liquid water is adsorbed by forest fuels is defined as the difference
between the Gibbs free energy of water vapor in equilibrium with pure liquid
water (the reference state) and the energy of water in the adsorbed state. Thus
at constant T,
rpo rp
AG = GQ- G = V,dp = - {RT/M^)dp/p = -{RT/M^) In H (25)
^P ^Po

where GQ refers to the reference state, V^ is the specific volume of water vapor
at ambient temperature T, R is the universal gas constant (8.32 J mol~^ K"^),
and M^ is the molecular weight of water (0.018 kg mol"^). Vapor pressures in
Chapter 4 Water Relations of Forest Fuels 1Z1

the adsorbed and reference states are p and PQ, respectively, and the vapor is
assumed to be an ideal gas. The sorption isotherm may be written by combin-
ing Eqs. (24) and (25) to give

M, = {ln[(RT/M,)(-ln H)] - A}/B (26)

in which B is a negative number. Clearly, as H approaches 0, M^ approaches

negative infinity, and as H approaches 1, M^ approaches infinity. Increasingly
erroneous values of M^ are predicted as H decreases below about 0.1 or increases
beyond 0.85. The model in Eq. (26) was used to describe water vapor sorption
in forest fuels, including hardwood leaves, pine needles, and wiregrass (Nelson,
1984). The maximum difference between measured and modeled M^ values was
about 0.02.
In a comprehensive study, Anderson (1990b) utilized Eq. (26) to describe
sorption in recently cast and weathered (over one winter) litter collected from
sites in northern Idaho and western Montana. The twofold purpose of the re-
search was (1) to test applicability of the model for forest fuels and (2) to de-
termine whether the model would aid in separation of the litter types studied
into groups with similar M^ values. A total of 94 runs involving adsorption and
desorption in fresh and weathered fuels and spanning a temperature range from
278 to 322 K was made. Anderson found that a minimum of 89% of the varia-
tion in In AG of Eq. (24) was explained by variation in M^. He identified four
distinct fuel groups for recently cast fuels on the basis of ±0.02-wide bands
within the M^ data; there were only a few species changes within the basic groups
for the weathered litter. In order of increasing Mg, the four groups are referred
to as grasses, spruce and fir needles, pine and cedar needles, and aspen and larch
foliage. Encouraged by results of the study, Anderson stated that only a few fuel
groups were needed to describe changes in M^ for a large assortment of fuel lit-
ters. The constants A and B of Eq. (26), documented in his Table 2, provide a
wealth of information for researchers interested in modeling equilibrium rela-
tionships or water transport below Mj^^ in foliage fuels.

c. Complete Sorption Isotherms for Wood and Soil

The isotherms of the previous section involve sorption of water below Mj^p
and, therefore, are strictly applicable for H < 0.9. Soil and plant researchers
have been interested in the region H > 0.9 for many years, but only a few wood
and paper researchers have reported on sorption in this region since the mid-
1960s. What is referred to here as a complete isotherm involves a range in M^
from the ovendry to water-saturated condition. Only when wood is totally sat-
urated will H= I and water potential if/ = 0. For abiotic materials such as wood
122 Ralph M. Nelson, Jr.

and soil, ijj must be interpreted as the sum of the osmotic and matric potentials
because both of them cause a reduction of the vapor pressure of liquid water.
Moreover, they are not easily separated experimentally (Cloutier and Fortin,
1991). Thus Eqs. (3) and (25) may be combined to give
^ = p^ifjL - ix,)lM^ = -p^AG = (p^RTlM^) In H (27)
so that water potential is negative, as it should be if the vapor pressure is re-
duced below that of the reference state. Capillary water within and between
fuel or soil particles is assumed to form hemispherical menisci in which the liq-
uid-gas interface separates liquid pressure P^ and gas pressure Pg. These pres-
sures are related to the principal radii of curvature of the air-water interface,
Ti and Tj, and the surface tension of water, y, by the equation
K-P,= - y [ ( l / r i ) + (l/r^)] - -2y/r = ^ (28)
when Vi = Vj = r and the wetting angle of water on the solid material is as-
sumed to equal zero. Figure 14 shows data relating M^ to if/ for western hem-
lock specimens 10 mm thick in the longitudinal direction at T = 21°C. The
data at high M^ values were obtained in 1979 by Fortin (cited by Cloutier and
Fortin, 1991) with the tension plate and pressure plate methods; the dashed
lines represent data obtained with saturated salt solutions in the hygroscopic

240-

^ 200 H Desorption
Absorption
c
0
co 160-
o
I 120-

80-

40

-10' -10' -10' -10' -10' -10^ -10° -10^

Water potential \|/ (J kg')
FIGURE 14 Moisture content of western hemlock sapwood as a function of water potential over
the range from the nearly ovendry to saturated condition. Temperature is 21°C. From Cloutier and
Fortin (1991).
Chapter 4 Water Relations of Forest Fuels 123

region using Eq. (27). For this figure, ip has units J kg~\ but these are easily
changed to J m~^ because the density of water is 1000 kg m~^. The hysteresis
prominent in the data of Figure 14 is attributed by Clou tier and Fortin (1991)
to the "ink-bottle effect" in which water gain or loss is dependent on the dis-
tribution of capillary sizes and shapes. In wood, evaporation to a given water
content is determined by radii of the smaller channels (pit membrane pores con-
necting cell cavities), whereas condensation approaching the same water con-
tent depends on the radii of the larger pores (the cell cavities). Thus Eqs. (27)
and (28) indicate that both if/ and H are smaller for desorption than absorption
at constant water content, in qualitative agreement with Figure 14.
In earlier years, soil water equilibrium relationships were expressed in terms
of matric suction (a pressure difference across a differentially permeable mem-
brane) and moisture content. Campbell et al. (1993) have simplified the presen-
tation of such data by developing a complete isotherm similar in form to Eq. (26).
The isotherm of Campbell et at, given by their Eq. (4), may be written (using
symbols of this chapter for water activity and equilibrium water content) as
H = exp{-(M,/RT) exp[(l - MJM{) ln(-^^)]} (29)
where M^ is the extrapolated value of M^ when ijj = —\] kg~^ and ^^ is the wa-
ter potential of ovendry soil, evaluated as the constant —10^ J kg~\ Because
the quantities —ifj^ and M^ of Campbell et al. are equivalent to AGQ and M^^t of
Eq. (24), it is easy to show using Eqs. (27) that Eqs. (24) and (29) are identi-
cal in form. In addition to simplifying the mathematical modeling of soil dry-
ing, Eq. (29) may be used with Eq. (26) to characterize moisture equilibrium
between mineral soil and the adjacent duff and litter layers. This possibility was
pointed out by Siau (1995).

2. Laboratory Studies
The transport of moisture through a fuel particle is determined by the relative
magnitudes of internal and external transfer of water as determined by particle
properties and ambient conditions. However, in highly impermeable wood or
in a composite fuel layer (litter, duff, and mineral soil, for example), internal
evaporation and/or infiltration (gravitational flow of liquid water) also may take
place. These processes initially were studied with diffusion theory, but advances
in describing the flow of liquid water in wood and granular materials by capil-
larity now provide more realistic analyses of moisture transport.

a. Transport Mechanisms
The mechanisms by which water moves through the interior of dead forest
fuel particles and layers are capillarity, diffusion, and infiltration. The latter pro-
cess is described in soil physics texts such as Hillel (1980b). When gravitational
124 Ralph M. Nelson, Jr.

forces have ceased to operate, liquid water is brought to the particle or layer
surface where it evaporates into the atmosphere; water in vapor form either
condenses on a cell wall or continues to diffuse until it passes into the environ-
mental air. Spolek and Plumb (1980) stated that the permeability of wood to
the gas phase can be neglected because many of the bordered pits become aspi-
rated by surface tension forces associated with liquid flow. Thus for materials
with fine capillaries, Darcy's law [Eq. (9)] is used to compute flows of liquid,
and the flux of vapor [Eq. (11)] may be regarded as due to diffusion only. In an-
alyzing liquid flow, Spolek and Plumb (1981) defined the local liquid satura-
tion, 5, as
S = {m- mj,p)/{m^^ - mj,p) (30)
where m denotes local values of the moisture fraction and m^^ is the maximum
possible m calculated from the wood density (Stamm, 1964). Clearly, 5 can
range from 0 to 1 when m > rrij^p. If the fuel temperature is below 60°C and the
total pressure Pj in Eq. (9) is considered constant, the liquid water flux J^ may
be expressed in terms of S or m by establishing a relationship between capillary
pressure P^ [seeEq. (12)] andS. Spolek and Plumb (1980) reported a set of equa-
tions for this purpose based on the capillary flow model of Comstock (1970).
In later work, Spolek and Plumb (1981) presented the empirical equation
P, = 12405"^'^ (31)
where P^ has the units J m~^. These authors implied that local values of 5 can
replace the volume-averaged values they measured at several positions within
each of their wood samples, so 5 is used in Eq. (31). When this equation is com-
bined with Eq. (9) and Pj is assumed constant, the flux of liquid becomes
J, = -756{Kjr]^)S-''\dS/dx) (32)
where J^ is on a volume of whole wood basis. Combination of the relation
Px,4>w =" P/(^ ~ ^/sp) = Pfi^max ~ ^/sp)^ wlth Eqs. (5) and (32) leads to the
liquid water transport equation
Pjim^ax - mj,p)dS/dt = 756{Kjr]^)d[S--''\dS/dx)]/dx - w^ (33)
where pj is density of the dry fuel.
The diffusion mechanism involves bound water and water vapor. Local dif-
fusion of bound water when m < mj^p is due primarily to surface diffusion—a
process in which water molecules hop from one sorption site to another in two
dimensions. This diffusion is driven by a gradient in spreading pressure, P^, and
is described by Eq. (10). Practical use of Eq. (10), however, requires a known
relationship between m and P^ (e.g.. Nelson, 1986a). In addition, water vapor
traverses the cell cavities by diffusion. The ideal gas law may be used to express
Eq. (11) (with the diffusion term only) in terms of a gradient in vapor pressure.
Chapter 4 Water Relations of Forest Fuels 125

Thus combined bound water and water vapor diffusion consists primarily of a
series of steps involving evaporation from a cell wall, vapor diffusion across the
adjacent cell cavity, condensation on the opposite wall, and bound water diffu-
sion through the wall. Stamm and Nelson (1961) found from combined diffu-
sion calculations based on an electrical conductance analogy utilizing premea-
sured wood structural features that 86% of the total flow in softwoods with a
specific gravity of 0.4 drying at 50°C is due to this mechanism. Much of the re-
maining flow is continuous vapor diffusion through the cavities and pit mem-
brane pores. Rather than solve separate mass conservation equations for bound
water and water vapor, other modelers (Choong, 1965; Gong and Plumb, 1994)
have used a combined diffusivity in place of Dj, in Eq. (10). When this modified
equation is substituted into Eq. (5), the diffusivity in the resulting diffusion
equation becomes a complex function of fuel internal structure, air (or fuel)
temperature, and local moisture fraction m. Numerically based prediction mod-
els combining capillary and diffusive mechanisms with energy conservation and
volume averaging have been used to simulate the drying of wood (Spolek and
Plumb, 1980; Perre et al, 1990; Gong and Plumb, 1994).
In some types of fuel, evaporation takes place within the material rather than
on the surface. This retreating-surface mechanism occurs in thick layers of po-
rous materials (Nissan and Kaye, 1957) and in wood in which the free water is
relatively immobile due to little or no continuous void structure (Stamm, 1964).
Nissan et al (1959) analyzed this phenomenon in thick textiles wound on bob-
bins that were dried in a wind tunnel at about 50°C. The temperature-time data
for wool in Figure 15 indicate a temperature plateau at 30°C, about 5°C greater
than the wet-bulb temperature. This increase is due to a balance between heat
arriving at the internal surface of evaporation and heat used to evaporate free
water and produce outward vapor diffusion. Nissan et al called this intermedi-
ate temperature the pseudo-wet-hulb temperature and described its calculation.
The two major assumptions are that (1) temperature and vapor concentration
gradients in the outer dry layer are similar and (2) all incoming heat is used to
evaporate free water at the retreating free water surface. The heat balance equa-
tion derived by Nissan et al is
fe(0 - 0 b) = {^l^v)pwh\wh{ppwh - P)
(34)
= 0.00216(8D,),,,Ap,,[(p^,,/Tp,j,) - (p/T)]
where subscript py^h denotes evaluation at the pseudo-wet-bulb temperature,
absence of a subscript indicates ambient air values, k (J m~^s~^K~^) is thermal
conductivity of the material, 0 and T are Celsius and Kelvin temperatures, p
(kg m~^) is vapor concentration, A (J kg~^) is the latent heat of evaporation,
and p (J m~^) the vapor pressure. With this equation and a generalization of
Eq. (43), it can be shown that response time r should vary directly with d^ and
126 Ralph M. Nelson, Jr.

O 45

"co
CD
Q.
E

300 400 600

Time (min.)
FIGURE 15 Temperature variation with time at various depths for a thick porous wool bobbin
dried at 50.9°C and an airspeed of 4.35 m s~\ Point A corresponds to the end of the constant-rate
period; the pseudo-wet-bulb temperature, indicated by points B, C, and D, is about 30°C; point E
represents the time at which the bobbin is essentially dry. Thermocouple depths in inches are in-
dicated in the lower right corner. From Nissan and Kaye, reprinted by permission from Nature 180:
142-143, copyright 1957, Macmillan Magazines Ltd.

T/P,^i, but inversely with (1 — H). Here, d is layer thickness, P^^t is the satura-
tion vapor pressure at T, and H is the humidity fraction. Interestingly, approxi-
mations to these relationships were observed for layers of sawdust (Nelson,
1969) and jack pine litter (Van Wagner, 1979) that were losing water only
through their upper surfaces. These results were further verified in unpub-
lished studies by the writer with measured temperature distributions similar to
those in Figure 15 and with predicted drying rates and plateau temperatures
close to measured values for sawdust layers drying in air at temperatures from
14 to 32°C and humidity fractions from 0.25 to 0.79.

h. Data Analysis
The drying of solids under fixed external conditions (constant M^) typically
is divided into a constant-rate and two falling-rate periods (Van Wagner, 1979).
In the constant-rate period, drying depends on external factors but is indepen-
dent of the nature of the solid. This period is seldom seen in forest fuels except
for short periods of time—when the exposed surface is covered with a water
Chapter 4 Water Relations of Forest Fuels 127
film. The decrease in M with time t is linear. When this free water has disap-
peared, or the rate of water supply from the fuel interior becomes smaller than
the surface evaporation rate, the first falling-rate period is triggered as the sur-
face begins to dry. Moisture loss depends on both internal and external factors;
in most fuels, the decrease in M — M^ with time approximates an exponential
decay. A second falling-rate period often occurs when all free water has disap-
peared and the fuel temperature begins to rise. This second period, while differ-
ing in mechanism from the first, also may exhibit an exponential decay, but one
that usually is preceded by a time of transition from the first exponential period
to the second. This type of response was observed by Anderson (1990a) in ex-
periments on drying and wetting of forest foliage below Mj^p and is shown in
the semilogarithmic plot of Figure 16. Similar results were reported by Nelson
(1969), Anderson et al (1978), and Van Wagner (1979). The fraction of evap-
orable moisture remaining in the fuel at any time, denoted by E, is given by
E = {M- M,)/(Mi, - MJ (35)
where M is the fuel particle (or layer) average moisture fraction and Mi„ is the
initial value. Note that E in Eq. (35) corresponds to (1 — E) in the graph of An-
derson (1990a) in Figure 16.

Douglas-fir needles

160
Time since start (H)
FIGURE 16 Variation in the fractional amount of evaporable moisture remaining as a function of
drying time for several forest fuels. The quantity (1 — E) in this figure corresponds to E in Eq. (35).
The four fuels exhibit two or more drying stages. Pine sticks dry quickly when compared with pine
and fir needle layers. From Anderson (1990a).
128 Ralph M. Nelson, Jr.

Laboratory drying (or wetting) results such as those in Figure 16 may be an-
alyzed by means of classical diffusion theory (Crank, 1975). The theory assumes
that the fuel is homogeneous and isotropic, drying is isothermal, and diffusiv-
ity D is constant. The equation for diffusion of moisture in a plane slab of thick-
ness 2d (moisture exchange from both surfaces) is
dm/dt = D{d^m/d^) (36)
and may be solved subject to the convective boundary condition
—D{dm/dx)^ = hj^m^ — m^) (37)
where subscript s indicates a surface value, h^ (m s~^) is the surface emission (or
convective mass transfer) coefficient, and m^ = M^, a constant. Though Eq. (36)
strictly applies to movement below fiber saturation, the equation often is ap-
plied when free water is present for convenience. Such an analysis may fortu-
itously describe the moisture change in cases when the interior free water is
highly mobile but drying is controlled by diffusion in the drier layers. Solving
Eq. (36) subject to Eq. (37) leads to an infinite series solution that may be in-
tegrated from x= —dto X = d and divided by 2d to obtain E in the form
oo

E = 2Bi^ 2 {exp{-l3lFo)/[l3M + Bf + Bi)]} (38)

n=l

where Bi = h^d/D is the mass transfer Biot number, Fo = Dt/d^ is the mass trans-
fer Fourier number, and (3^ is the nth positive root of the equation jB Tan (3 =
Bi. When Fo exceeds 0.1, the series converges rapidly and an approximate so-
lution is found by retaining only the first term for n = 1. If the ratio of surface
resistance to internal flow resistance is small, Bi effectively approaches infinity
and Pi approaches ir/l. The moisture change is driven by a significant gradient
in m. Thus E is written as
E = (8/77^) exp(-7r^Dt/4d^) = K exp(-t/T) (39)
where constant K is sometimes assumed equal to unity and the response time is
T = 4dV77^D (40)
On the other hand, when the surface resistance is relatively large and the gra-
dient in m is small, Bi approaches zero and fil approaches Bi. In this case, E cor-
responding to n = 1 becomes
E = exp(-BiFo) = exp(h^t/d) = exp(-t/T) (41)
and T is given by
T = d/K (42)
Equations (40) and (42) display the r-versus-d dependence described in Sec-
tion II.C.3. Similar analyses are possible for other fuel shapes—the cylinder,
Chapter 4 Water Relations of Forest Fuels 129

sphere, or square rod—and form the basis of the response time theory initially
proposed by Byram (1963, unpublished). This analysis provides a convenient
framework with which to interpret experimental data, but D is not a constant in
fibrous materials and should be considered an integral diffusivity over the range
of moisture content change. Procedures for calculating the bound water and
water vapor diffusivities separately in terms of m and T (or Tj) and then com-
bining them into an overall diffusivity are described by Nelson (2000).
The method used to evaluate integral diffusivity D^^ from experimental data
depends to some extent on the type and amount of data available. Crank (1975)
shows that a semiinfinite solid dries or wets in proportion to the square root of
time t and that plots of E versus t^'^ according to Eq. (38) for constant Bi are
effectively linear during roughly the first half of the exchange process. Thus
when data are plentiful for short times, D^^ for the slab may be found from a dif-
ferent solution of Eq. (36) written as
D^, = ird^E^/M = TTdVl"^ (43)
where s is the slope of the linear portion of a plot of E versus t^^^ (Crank, 1975).
This equation is based on three assumptions: (1) the slab dries or wets as a semi-
infinite solid early in the process of moisture change, (2) shrinkage/swelling is
negligible, (3) the surfaces instantaneously attain the equilibrium value, M^.
McNamara and Hart (1971) suggest using Eq. (43) in the range 0.5 < E < 1.
When data are available for more widely spaced time intervals or when E < 0.5,
Eq. (39) may be used to obtain an estimate of D^^ by solving for D.
For a single experiment, the effect of surface transfer on the moisture ex-
change process cannot be separated from that due to internal diffusion because
the mass transfer correlations applicable to saturated wood surfaces are not
valid for unsaturated surfaces (Plumb et al, 1985; Siau and Avramidis, 1996).
The effects can be separated, however, with a procedure that requires two or
more experimental runs with samples of differing thickness (Choong and Skaar,
1972). A summary of the results of various investigators utilizing this procedure
reports a 100-fold variation in mass transfer coefficient h^ (Siau and Avrami-
dis, 1996). A theory for predicting h^ in the hygroscopic range was reported
by Moren et al (1992), and an identical relation was developed by Siau and
Avramidis (1996).

3. Field Studies
Linton (1962) used the theory of diffusion in solids to describe moisture change
in fuel particles in terms of a constant diffusivity and evaluated this quantity
with laboratory experiments. Byram (1963, unpublished) also applied theory to
the problem, but his timelag concept is applicable only to step changes between
fixed initial and final values of average moisture fraction M. The oscillating
130 Ralph M. Nelson, Jr.

moisture content data obtained under field conditions differ from laboratory
data which tend to decrease exponentially. Linton (1962) and Viney and Hat-
ton (1989) discussed a different timelag that is useful in field studies. Their
timelag is concerned with the lapse in time between the continuously varying
equilibrium moisture content, M^ (defined as the M value determined by as-
suming equilibrium with the diurnally changing fuel temperature and relative
humidity), and the actual M value for the fuel. Thus Viney and Hatton (1989)
and Viney and Catchpole (1991) recommended that "timelag" be used only to
denote the lag of M behind Mg and that "response time" be applied to the time
needed for 63% of the total change between fixed moisture contents. Recog-
nizing this distinction, Anderson (1990a) used "response time" to characterize
the moisture exchange rates of fuels in his laboratory tests. The terminology of
Viney and Catchpole is adopted in this chapter and is recommended for all fu-
ture studies of fuel moisture relationships to minimize confusion.

a. Weather Factors
In addition to continuous variation in air temperature and relative humidity,
other variables influencing moisture change in forest fuels include wind speed
and direction, solar and longwave radiation, precipitation, canopy interception
of rainfall and sunshine, filtration through fuel layers, deposition of dew, con-
tact with fog, and source/sink effects of underlying soil. The models of Byram
and Jemison (1943) describing fuel temperature and relative humidity fraction
in terms of ambient values, solar radiation, and wind speed were briefly sum-
marized in Section III.D.l. Using the iterative Newton method, Viney (1992a)
presented coupled theoretical models for soil surface temperature and humid-
ity fraction based on a surface energy balance and complex expressions for re-
sistance to vapor transport. Viney compared predictions of Tj and Hj based on
his surface balance models and on the Byram-Jemison models with values mea-
sured at the under surfaces of Eucalyptus globulus leaves and found that his
models provided more accurate predictions of Tj and Hj than the simpler mod-
els. The mean absolute error in Tj predicted by Viney (1992a) was 0.52 K; the
error for predicted Hj was 0.051.
Byram and Jemison (1943) pointed out the influence of wind on drying rate
and final M^ values for forest fuels below Mj^^. Their data suggested that the
cooling action of wind on the drying rate of irradiated fuels is greater than the
tendency of wind to increase the evaporation rate. The data showed that desorp-
tion rates are slowed and M^ is increased, whereas both adsorption rates and M^
values are increased by the action of wind. Other data showed that the wind ef-
fect is reversed in the absence of radiation—i.e., drying rates for nonventilated
fuels are slower than for ventilated fuels. Van Wagner (1979) conducted labo-
ratory tests on the drying of jack pine litter at 27°C and 30% relative humidity
Chapter 4 Water Relations of Forest Fuels 131
200n

o 10CH
LU
I

n—\—r
01 2 3 4 5 6 7 8 9 10 11 12 1 3 1 4 1 5
Time (h)
FIGURE 17 Change of free moisture content (that above the equilibrium moisture content, EMC)
of jack pine litter with time at four wind speeds ranging from 0 to 9.62 km hr"^ (2.7 m s"^). Even
at the high moisture contents implied here, the effect of wind is evident only during the first hour
of drying. From Van Wagner (1979).

in wind but without solar radiation; the resuhs in Figure 17 show that the ht-
ter (which initially contained considerable free water) dried in two stages when
the wind speed exceeded about 0.75 km hr"^ (0.2 m s"^). These results indi-
cate that differences in drying rate are established during the first hour of dry-
ing; though this first hour constitutes only a small fraction of the total drying
time, a significant fraction of the moisture content change takes place during
this period. For times exceeding one hour, the drying rate is roughly indepen-
dent of wind speed.
The rise in average moisture content fraction, AM, due to rainfall has been
studied by Stocks (1970) for duff (see Section II.C.2) and by Fosberg (1972) for
wood cylinders. In his theoretical approach, Fosberg used numerical and ana-
lytical solutions of the diffusion equation to argue that water uptake by wood
cylinders is limited by the rate at which the wood can adsorb water and that this
limiting rate is of the order of 1 mm day~^ He also stated that the theory pre-
dicts realistic rates of adsorption and desorption associated with precipitation
and postprecipitation drying and that M increases exponentially with increas-
ing duration of rainfall. In the empirical approach of Van Wagner (1987), AM
132 Ralph M. Nelson, Jr.

is expressed in terms of the initial value of M and the net rainfall (observed
rainfall amount in the open less a correction of 0.5 mm for canopy intercep-
tion). These and other treatments of AM due to rainfall have been critiqued by
Viney (1991). Different aspects of rainfall are concerned with interception by
the canopy (if present as understory or overstory) and infiltration through the
litter and underlying soil but are not discussed here. These topics are treated by
Helvey (1967) and Hillel (1980b).
Nocturnal condensation of water on plant surfaces can be caused by deposi-
tion of water from the atmosphere or by distillation, the upward transport of
water vapor from the soil by turbulent diffusion (Monteith, 1963b). In a study
of the effect of condensation on leaf litter moisture content, Viney and Hat-
ton (1990) measured M for single Eucalyptus globulus leaves in contact with the
ground at two sites in southeastern Australia. They developed a theory of con-
densation into which measured fluxes of net all-wave radiation, soil heat, and
sensible and latent heat at the leaf surfaces were introduced to obtain a pre-
dicted AM (the change in average moisture fraction attributable to condensa-
tion). Good agreement between observed and predicted values is evident in
Figure 18, but M is underpredicted by the model when fog is present. Viney and
Hatton (1990) state that the model tends to predict a zero AM during fog epi-
sodes when, in fact, the leaves may have taken on moisture through direct con-

100
26-27 May 1988
• Observed
C3^ 80 Predicted

60-^

40

20

1500 1800 2100 2400 0300 0600 0900

Time (LST)
FIGURE 18 Comparison of predicted moisture contents of Eucalyptus leaves with those observed
overnight on 26 May 1988 during experiments in eastern Australia. Moisture content change due
to condensation of water vapor from the atmosphere and soil is predicted well, but the effects of
fog apparently are not explained by the model. Reprinted from Agricultural and Forest Meteorology,
Volume 51, Viney, N. R., and Hatton, T. J., Modelling the effect of condensation on the moisture con-
tent of forest litter, pages 51-62. Copyright 1990, with permission from Elsevier Science.
Chapter 4 Water Relations of Forest Fuels 133

tact with fog droplets. The measured values of M during condensation also may
be sensitive to fuel shape and orientation which could cause reductions in ab-
sorbed and surface-held water due to runoff (Viney, 1991).
Though the effects of soil moisture have been neglected in many studies of
dead fuel moisture, at least two studies have presented evidence that the former
affects the latter. Hatton et al. (1988) used a randomized block design to study
the response of leaves of several species of Eucalyptus to wetted (20% dry soil
weight or less) and untreated dry (8% dry weight or less) soil over a period of
48 hours. Within each cleared 8 X 8 m plot, leaves were arranged in a 2 X 1 X
0.16 m mound and also were uniformly scattered over the ground for full ex-
posure to the weather. The leaves were held in place by nylon netting. Hatton
et al concluded that soil moisture influenced fuel moisture because the scat-
tered leaves approached moisture contents of 100%, well beyond values attain-
able by vapor transport alone (dew was observed). In addition, leaves on top of
the mound on wetted soil were about twice as moist as corresponding leaves on
the dry soil. This difference was explained in terms of more water available for
distillation from the wetter soil. In a different study conducted in aspen stands
of central Alberta, Rothwell et al. (1991) measured the water potential and vol-
umetric water content (in %) at locations 2 - 3 cm above and below the transi-
tion zone between the litter layer and mineral soil. The authors concluded, on
the basis that water flows from higher to lower levels of water potential, that
soil water affects fuel availability and fire behavior in stands of aspen because
upward capillary flow from the soil kept the litter moist in their experiments.

b. Prediction Models
Viney and Hatton (1989) measured the moisture content of dead leaves, bark,
and twigs of Eucalyptus macrorhyncha and Eucalyptus rossii in contact with the
ground for the purpose of assessing models for predicting the moisture content
of fine fuels. They reported significant differences in equilibrium moisture frac-
tion Mg within the three fuel types of the study and stated that submodels for
dew and distillation should be included in models to be used in open or lightly
foliated areas. Viney and Catchpole (1991) and Viney (1992a, 1992b) devel-
oped a method for estimating fuel moisture response time and timelag (the lag
of fuel moisture fraction M behind M j from field observations. The basis of
their method is shown in Figure 19 in which it is assumed that both M and M^
follow a 24-hour sinusoidal cycle. In present notation, the equations describ-
ing such data are
M(0 =M,, + a Sm[(o{t - (pj] (44)
M,(t) =E^, + A Sin[(o{t - cp,)] (45)
where M^^, E^^, a, and A are constants obtained from least squares regression
analysis, t is time (h), cp^ and cp^ (h) are phase lags relative to midnight, and ct;
134 Ralph M. Nelson, Jr.

Time (h)

FIGURE 19 Idealized sinusoidal variation in equilibrium (dashed curve) and fuel (solid curve)
moisture contents with time under field conditions. The average values of E and M shown in this
figure are represented as E^^ and M^^, respectively, in Eqs. (44) and (45). The timelag (pj is the dif-
ference in phase for the two curves. From Viney (1992a, 1992b).

is the angular frequency given by (o = TT/II hr~\ Hourly observations of M and

Mg over a period of several days are filtered with a time series decomposition
algorithm to eliminate long-term trends and random fluctuations and to obtain
(p^ and (p^ from the two sets of data. As Figure 19 shows, the difference (p^ — cp^
is the timelag (or phase lag) of the fuel, (pj. The final step is to determine re-
sponse time T from
T = (1/w) Tan(a)(^j) (46)

The derivation of Eq. (46) is based on three additional assumptions: (1) the re-
sistance to internal moisture transport is small compared to that at the fuel sur-
face, (2) T is constant during the diurnal cycle, and (3) time since midnight of
the day prior to the first day of the measurement period (t = 0) is large com-
pared with the fuel's response time. These assumptions are reasonable for fine
or poorly ventilated fuels in which moisture is exchanged with little or no gra-
dient in moisture content. An alternative to evaluating cpj with the sinusoidal
assumptions is to represent the two data sets with cubic spline interpolations
and then choose the value of cpj that maximizes correlation between the two
sets. In a further modification of these methods, Catchpole et al. (in press) com-
bined spline interpolation with the sorption isotherm in Eq. (26) to find esti-
mates of Mg and r without assuming sinusoidal variation in M and M^. Accord-
ing to Catchpole et al, tests of the model with data collected in Tasmanian
buttongrass moorland and Western Australian mallee shrubland have given
Chapter 4 Water Relations of Forest Fuels 135

them "some confidence in the robustness of the method and the appropriate-
ness of the model." The methods described here should be used only for fuels
with short (3 h or less) response times and would apply during periods of sorp-
tion-controlled moisture exchange rather than periods of rainfall or evapora-
tion and condensation.
In larger fuels or fuels that dry in the presence of moisture content gradients,
internal resistance exceeds that at the surface. Viney (1992a, 1992b) pointed
out that such conditions require solution of the diffusion equation appropriate
for the type of fuel particle under consideration. To evaluate the constant mois-
ture diffusivity, D, for different particles, he solved the diffusion equations ap-
plicable to the semiinfinite solid (representing a layer of leaf or needle litter in
contact with the ground), the slab (a flat leaf), the cylinder (a twig or needle),
and the long square rod (a fuel moisture indicator stick)—all subject to sinu-
soidal surface boundary conditions. A parameter common to all four solutions
is the dimensionless number,fed,where k is given by
k = {OJ/IDY'^ (47)
and d is the depth of interest for the semiinfinite solid, thickness of the slab, di-
ameter of the cylinder, or side length of the square rod. Though solutions for
the average moisture fraction were derived analytically, their complexity pre-
cluded evaluation of hd, and hence D. Numerical representations of cpj for the
four fuel arrangements are shown in Figure 20. If the timelag is evaluated from

FIGURE 20 Relationship between the fuel timelag and dimensionless parameter kd for the four
fuel shapes indicated. Quantity d represents the depth for which the timelag is desired in a semi-
infinite solid, thickness for a slab, diameter for a cylinder, and side length for a square rod. From
Viney (1992a, 1992b).
136 Ralph M. Nelson, Jr.

field data and d is known, then k may be determined from Figure 20 and used
in Eq. (47) to compute D. Viney (1992a, 1992b) compared his calculated D val-
ues for leaf and twig data from the Viney and Hatton (1989) study with those
reported by Linton (1962) and found good agreement for leaves and 6 mm twigs;
his D for 3 mm twigs slightly exceeded Linton's values. In a different compari-
son, Viney's D for a lodgepole pine litter layer was within the range of corre-
sponding values measured by Anderson (1990a).
Several investigators have developed diurnal prediction models from field
data. Fosberg (1975) derived a model for conifer litter and duff from mass con-
servation equations for water vapor and the first law of thermodynamics. The
physical properties of litter and duff horizons were taken from measurements
on ponderosa pine and lodgepole pine layers or from the literature. Fosberg
adapted his model to field conditions by introducing time-dependent bound-
ary conditions consisting of step changes of arbitrary duration and magnitude.
Though he did not compare his computed M-versus-t curves with data from
laboratory or field experiments, Fosberg's predicted response times based on
layer properties were close to several values in the literature. A predictive model
of moisture content and evaporative loss in the litter of a mixed deciduous for-
est was reported by Moore and Swank (1975). Weather data requirements in-
cluded daily totals for sunlight and rainfall and daytime means for atmospheric
temperature and relative humidity. The model contains three hydrologic com-
partments—atmosphere, litter, and soil. In addition to evaporation from litter
to the atmosphere and liquid drainage from litter to the soil, litterfall and litter
dry weight (or decomposition) are modeled. Though they are acknowledged
as important at times, atmospheric condensation and upward flow of moisture
from the soil are not modeled. Independent data for validating model predic-
tions during a 7-day period in the summer and an 11-day period during the
winter were collected at the Coweeta Hydrologic Laboratory in western North
Carolina. Predicted moisture contents usually were within the error limits of
the experimental values. Simulated evaporation agreed to within 13% of the
measured evaporation but was more accurate for the first two days after rain
than under drier conditions. Moisture content comparisons during an addi-
tional 80-day verification study gave similar results. A model for computing fine
fuel moisture content during the diurnal cycle was developed by Van Wagner
(1987) using jack pine and lodgepole pine data; it is based on the assumption
that wetting and drying by vapor exchange with the atmosphere are exponen-
tial, whereas wetting due to rainfall depends on rainfall amount and yesterday's
fine fuel moisture content. In a different study. Nelson (2000) used diurnal
weather and moisture content data from two field sites in the eastern United
States to guide the formulation of a numerical model for computing hourly
changes in the moisture content and temperature of 1.27-cm-diameter ponder-
osa pine dowels. In the model, water moves by capillarity and combined bound
Chapter 4 Water Relations of Forest Fuels 137
0.4

Jun 11 13 14 15 16
Calendar date
FIGURE 21 Comparison between predicted and observed moisture contents for 10-h fuel indi-
cator sticks under field conditions in western Montana during June 1996. Rainfall occurred prior
to midnight on June 15 and 16 and at about 1800 on June 17. Solid lines represent predictions;
circles denote observations. From Nelson (2000).

water/water vapor diffusion. Either precipitation, evaporation, condensation,

or sorption determines the surface boundary condition; water takeup due to
fog is not considered. The model accepts data from remote automatic weather
stations (RAWS). Input consists of hourly readings of air temperature and rela-
tive humidity, solar radiation flux, and rainfall amount; changes in the four vari-
ables from one reading to the next are assumed to be linear. Independent obser-
vations for testing the model were compared with predicted values; predictions
generally agreed with observations to within the accuracy of the experimental
moisture content data, but rates of response appear to be slightly slower than
the observed rates (Figure 21).

IV. MOISTURE CONTENT ESTIMATION

Aside from prediction models, other means of estimating the moisture content
of live and dead fuels include weighing, correlation with Advanced Very High
Resolution Radiometer (AVHRR) data from NOAA satellites, resistance meters,
and artificial fuel sticks available with RAWS. Moisture content calculations re-
quire a sample ovendry weight. Drying in the usual convection oven is slow,
and the fuel is subject to loss of volatiles if dried too long at a high temperature.
138 Ralph M. Nelson, Jr.

In the literature, drying temperatures for wood and foliage range from 60 to
105°C. Perhaps the safest procedure for foliage is to dry at 100°C until a con-
stant weight is reached. This method, of course, requires frequent weighings to
prevent loss of volatiles due to prolonged drying. King and Linton (1963) con-
cluded that materials ovendried prior to testing adsorb water about half as fast
as those that have been dried only to a low moisture content. Thus experimen-
tal error may be reduced by determining ovendry weights after the measure-
ments are made. To avoid the slowness and labor intensiveness of this method,
instruments composed of a small balance and oven (or electric heater) have
been developed, but this method is suitable only for small samples. Commer-
cially available kitchen microwave ovens were tested by Norum and Fischer
(1980). Wet duff (up to 230% moisture content) and 0 to 0.64-cm-diameter
branchwood (up to 90%) samples weighing about 40 g were ovendried in
24 minutes or less. Live fuel moisture content can be estimated from satellites
with the Normalized Difference Vegetation Index (NDVI) which operates by
sensing red and near infrared radiation reflected from the surfaces of green or
curing vegetation. Burgan et al. (1991) showed that the NDVI is useful for cal-
culating overall site moisture content, whereas Pinol et al. (1998) developed a
rapid method for measuring the moisture content of individual plants. Chladil
and Nunez (1995) used AVHRR to follow curing of grasslands in Tasmania and
concluded that the NDVI gives good results for fuel and soil moisture contents
but is not a suitable standalone system for fire managers. Electrical resistance
meters are useful for estimating the moisture content of wood in the hygro-
scopic range but are sensitive to contact resistance and presence of a mois-
ture gradient (Skaar, 1988). Other factors mentioned by Clark and Roberts
(1982) include wood species, density, and grain angle; they used a minimum of
20 samples for calibrating their meter to the species utilized in their study. The
moisture content of the 10-h fuel moisture indicator sticks used in the U.S. Na-
tional Fire Danger Rating System (Bradshaw et al, 1983) may be estimated with
an artificial fuel moisture and fuel temperature stick that can be supplied with
RAWS installations. The stick is equipped with sensors to measure its relative
humidity and temperature, from which a moisture content is derived.
Most fire suppression and control organizations maintain a system with
which they assess the moisture content of the live and dead fuels within their
areas of jurisdiction. For operational applications, equations for estimating fuel
moisture content are found in fire danger rating and fire behavior prediction
systems around the world. A brief description of how moisture content is evalu-
ated in several of the fire danger rating systems is presented in Chapter 9 in
this book. In addition, operational models for prediction of dead fuel moisture
content in the hygroscopic range have been discussed and compared by Viney
(1991, 1992a). The reader is referred to Van Wagner (1987) and to the Forestry
Canada Fire Danger Group (1992) for details of the Canadian systems, to Brad-
shaw et al (1983) and Rothermel et al (1986) for the American systems, and
Chapter 4 Water Relations of Forest Fuels 139

to Noble et al. (1980) and Sneeuwjagt and Peet (1985) for the Australian sys-
tems. Another aspect of fire danger involves water loss from deep soils and large
fuels during extended dry periods. As before, the reader is referred to pertinent
papers on the subject. Discussions of drought and drought indexes are given in
key papers by Van Wagner (1987), Keetch and Byram (1968) and Alexander
(1992), and Mount (1972) for the Canadian, American, and Australian sys-
tems, respectively.

NOTATION

ROMAN LETTERS

C specific water capacity m " \ s^ m"^

c constant-pressure specific heat J kg"^ K~^
D local moisture diffusivity m^ s"^
D^y particle or layer average moisture diffusivity m^ s~^
Dh hydraulic diffusivity m^ s~^
d particle radius or half-thickness, layer depth m
E fraction of evaporable moisture
e base of Naperian logarithms
G Gibbs free energy J mol"^
g gravitational acceleration 9.8 m s~^
H ambient air relative humidity fraction
h hydrostatic pressure head m
h^ surface emission coefficient ms~^
J mass flux kg m"^ s~^
K specific permeability m^
k parameter in Eq. (47), thermal conductivity ni~\ J m"^ s"^ K~
L root length m
In Naperian logarithm
M particle or layer average moisture content
fraction
M^ molecular weight of water 0.018 kg moP^
Ml M value when i// = — 1 J kg~^
m local moisture content fraction
P fluid pressure, saturation vapor pressure J m"^
140 Ralph M. Nelson, Jr.

h bound water spreading pressure Jm-^

Pc capillary pressure Jm"^
V gas phase partial pressure Jm-3
Q transpiration or evapotranspiration rate tn s~'

9/ demand to heat dry fuel from ambient kjkg-i

to 400°C
Qu heat required to evaporate water at 100°C kj kg-^
QT heat of ignition kj kg-^
a heat flux Jm-^s"^
RWC relative water content
r radial distance, flow resistance, mean m; s m ~ \
radius of the gas-liquid interface bar-sni~\
m^ k g ' ^ s~^ m
n.rj radii of curvature of a gas-liquid interface m
S liquid water saturation
s slope of E versus t^^^ plot s-'^'
T ambient air temperature K
t time s, hr
U root sink strength kg m~^ s"^
V specific volume m^kg-i
V velocity m s~^
W plant tissue weight, duff weight consumed kg, kg m"^
by fire
w rate of evaporation kg m~^ s~^
X coordinate in direction of moisture transport m
z vertical distance from reference elevation m

GREEK LETTERS

a hysteresis ratio
P fuel layer packing ratio
jS„ nth root of pTm/S = Bi
y surface tension of water J ni"^
A finite change in associated variable
8 fuel layer depth m
Chapter 4 Water Relations of Forest Fuels 141

E fractional void volume

V kinematic viscosity m^s-i
0 temperature °C
6 volumetric soil water content
K hydraulic conductivity m s ~ \ kg-s m~^
A latent heat of evaporation/condensation at T Jkg-i
Aa differential heat of sorption Jkg-i
f^ chemical potential Jmor'
P mass density kgm"^
Pr root density m-2
a particle surface-to-volume ratio m-i
T moisture response time s, hr
<P sinusoidal phase lag hr
<i> phase volume fraction
^ water potential Jm-^Jkg-l
0) angular frequency 77/12 hr"'

DiMENSIONLESS GROUPS

Bi mass transfer Biot number

Fo mass transfer Fourier number

CONSTANTS

R universal gas constant 8.32 J mol~^ K~

SUBSCRIPTS

a atmospheric air, dry air component of the gas phase

av particle or layer average
h bound water phase
c current state
cw cell wall
d dry state
142 Ralph M. Nelson, Jr.

e equilibrium state
f fuel
fsp fiber saturation point
g gravitational potential, gas phase
h hydraulic head
i phase designator
in initial state
I leaf
m ma trie potential, moisture
n(d) apoplast matric potential
max maximum value
0 reference state
P pressure potential
pwh pseudo-wet-bulb value
r root
s solution (or osmotic) potential, solid phase, surface
s(a) apoplast solution (or osmotic) potential
so soil
sat saturated state
T total value
t turgid state
V water vapor component of the gas phase
va vacuole, vapor in ambient air
w liquid water phase

ADDITIONAL READING
Brown, R. W. (1995). The water relations of range plants: adaptations to water deficits. In "Wildland
Plants: Physiological Ecology and Developmental Morphology" (D.J. Bedunah and R. E. Sose-
bee, Eds.), pp. 291-413. Society for Range Management, Denver.
Marshall, M. J., Holmes, J. W., and Rose, C. W. (1996). "Soil Physics," 3rd ed. Cambridge Univer-
sity Press, Cambridge.
Smith, J. A. C, and Griffiths, H. (1993). "Water Deficits: Plant Responses from Cell to Community."
BIOS Scientific Publishers, Oxford.
Waring, R. H., and Running, S. W. (1998). "Forest Ecosystems: Analysis at Multiple Scales," 2nd
ed.. Chapter 2. Academic Press, San Diego.
Chapter 4 Water Relations of Forest Fuels 143

REFERENCES
Albini, F. A. (1980). "Thermochemical Properties of Flame Gases from Fine Wildland Fuels." Re-
search Paper INT-243. USDA Forest Service, Intermountain Forest and Range Experiment Sta-
tion, Ogden.
Albini, F. A., and Reinhardt, E. D. (1995). Modeling ignition and burning rate of large woody nat-
ural fuels. Int. J. WildlandFire 5, 81-91.
Alessi, S., Prunty, L., and Schuh, W. M. (1992). Infiltration simulations among five hydraulic prop-
erty models. Soil Sci. Soc. Am. J. 56, 675-682.
Alexander, M. E. (1992). The Keetch-Byram drought index: A corrigendum. Bull. Am. Meteorol.
Soc. 73, 61.
Alexander, M. E., Stocks, B. J., Wotton, B. M., and Lanoville, R. A. (1998). An example of multi-
faceted wildland fire research: The International Crown Fire Modelling Experiment. In "Pro-
ceedings, III International Conference on Forest Fire Research/14th Conference on Fire and
Forest Meteorology," pp. 83-112. University of Coimbra.
Anderson, H. E. (1990a). Moisture diffusivity and response time in fine forest fuels. Can. J. For. Res.
20,315-325.
Anderson, H. E. (1990b). "Predicting Equilibrium Moisture Content of Some Fohar Forest Litter
in the Northern Rocky Mountains." Research Paper INT-429. USDA Forest Service, Intermoun-
tain Research Station, Ogden.
Anderson, H. E., Schuette, R. D., and Mutch, R. W. (1978). "Timelag and Equilibrium Moisture
Content of Ponderosa Pine Needles." Research Paper INT-202. USDA Forest Service, Inter-
mountain Forest and Range Experiment Station, Ogden.
Anderson, N. T., and McCarthy, J. L. (1963). Two-parameter isotherm equation for fiber-water sys-
tems. Ind. Eng. Chem. Process Des. Develop. 2, 103-105.
Babbitt, J. D. (1950). On the differential equations of diffusion. Can. J. Res. A-28, 449-474.
Barkas, W.W. (1949). "The Swelling of Wood Under Stress." Department of Scientific and Indus-
trial Research, Forest Products Research, HSMO, London.
Blackmarr, W H. (1971). "Equilibrium Moisture Content of Common Fine Fuels Found in South-
eastern Forests." Research Paper SE-74. USDA Forest Service, Southeastern Forest Experiment
Station, Asheville.
Blackmarr, W H. (1972). "Moisture Content Influences Ignitability of Slash Pine Litter." Research
Paper SE-173. USDA Forest Service, Southeastern Forest Experiment Station, Asheville.
Blackmarr, W H., and Flanner, W B. (1968). "Seasonal and Diurnal Variation in Moisture Content
of Six Species of Pocosin Shrubs." Research Paper SE-33. USDA Forest Service, Southeastern
Forest Experiment Station, Asheville.
Blackmarr, W H., and Flanner, W B. (1975). "Moisture Variation in Selected Pocosin Shrubs of
Eastern North Carolina." Research Paper SE-124. USDA Forest Service, Southeastern Forest
Experiment Station, Asheville.
Bradshaw, L. S., Deeming, J. D., Burgan, R. E., and Cohen, J. D. (1983). "The 1978 National Fire-
Danger Rating System: Technical Documentation." General Technical Report INT-169. USDA
Forest Service, Intermountain Forest and Range Experiment Station, Ogden.
Brown, H. P., Panshin, A. J., and Forsaith, C. C. (1949). "Textbook of Wood Technology, Vol. 1,"
1st ed. McGraw-Hill, New York.
Brown, J. K., Booth, G. D., and Simmerman, D. G. (1989). Seasonal change in live fuel moisture of
understory plants in western U.S. aspen. In "Proceedings of the Tenth Conference on Fire and
Forest Meteorology" (D. C. Maclver, H. Auld, and R. Whitewood, Eds.), pp. 406-412. Forestry
Canada, Environment Canada, Ottawa.
Brown, R. W (1977). Water relations of range plants. In "Rangeland Plant Physiology" (R. E. Sose-
bee, Ed.), Range Science Series 4, pp. 97-140. Society for Range Management, Denver.
144 Ralph M. Nelson, Jr.

Browning, B. L. (1963). "The Chemistry of Wood." J o h n Wiley, New York.

Burgan, R. E., Hartford, R. A., Eidenshink, J. C , and Werth, L. F. (1991). Estimation of vegetation
greenness and site moisture using AVHRR data. In "Proceedings of the Eleventh Conference on
Fire and Forest Meteorology" (R L. Andrews and D. F. Potts, Eds.), pp. 1 7 - 2 4 . Society of Ameri-
can Foresters, Bethesda.
Byram, G. M. (1959). Combustion of forest fuels. In "Forest Fire: Control and Use" (K. P. Davis,
Ed.), pp. 6 1 - 8 9 . McGraw-Hill, New York.
Byram, G. M. (1963). "An Analysis of the Drying Process in Forest Fuel Material." Unpublished
report. USDA Forest Service, Fire Sciences Laboratory, Rocky Mountain Research Station,
Missoula.
Byram, G. M., a n d j e m i s o n , G. M. (1943). Solar radiation and forest fuel moisture. J. Agric. Res. 6 7 ,
149-176.
Campbell, G. S., Jungbauer, J. D., Jr., Shiozawa, S., and Hungerford, R. D. (1993). A one-parameter
equation for water sorption isotherms of soils. Soil Sci. 156, 3 0 2 - 3 0 5 .
Catchpole, E. A., Catchpole, W. R., Viney, N. R., McCaw, W. L., and Marsden-Smedley, J. B. Mod-
elling fuel response time and equilibrium moisture content from field data. Accepted for p u b -
lication. Int. J. Wildland Fire.
Chandler, C , Cheney, P., Thomas, P., Trabaud, L., and Williams, D. (1983). "Fire in Forestry: Vol. I."
John Wiley, New York.
Chladil, M. A., and Nunez, M. (1995). Assessing grassland moisture and biomass in Tasmania—
The application of remote sensing and empirical models for a cloudy environment. Int. J. Wild-
land Fire 5, 1 6 5 - 1 7 1 .
Choong, E. T. (1965). Diffusion coefficients of softwoods by steady-state and theoretical methods.
For. Prod. J. 1 5 , 2 1 - 2 7 .
Choong, E. T., and Skaar, C. (1972). Diffusivity and surface emissivity in wood drying. Wood Fiber
4,80-86.
Choudhury, B. J., and Monteith, J. L. (1988). A four-layer model for the heat budget of homoge-
neous land surfaces. Quart.]. Roy. Meteorol. Soc. 114, 3 7 3 - 3 9 8 .
Christensen, G. N. (1965). The rate of sorption of water vapor by thin materials. In "Humidity
and Moisture: Measurement and Control in Science and Industry" (A. Wexler, Ed.), Vol. 4,
pp. 2 7 9 - 2 9 3 . Reinhold, New York.
Chrosciewicz, Z. (1986). Foliar moisture content variations in four coniferous tree species of cen-
tral Alberta. Can.]. For. Res. 16, 1 5 7 - 1 6 2 .
Clark, B., and Roberts, F. (1982). A belt weather kit accessory for measuring woody fuel moisture.
Fire Manage. Notes 4 3 , 2 5 - 2 6 .
Cloutier, A., and Fortin, Y. (1991). Moisture content—water potential relationship of wood from
saturated to dry conditions. Wood Sci. Tech. 2 5 , 2 6 3 - 2 8 0 .
Cloutier, A., and Fortin, Y. (1993). A model of moisture movement in w o o d based on water
potential and the determination of the effective water conductivity. Wood Sci. Tech. 2 7 , 9 5 -
114.
Comstock, G. L. (1970). Directional permeability of softwoods. Wood Fiber 1, 2 8 3 - 2 8 9 .
Cote, W A., Jr. (1967). "Wood Ultrastructure." University of Washington Press, Seattle.
Countryman, C. (1974). Moisture in living fuels affects fire behavior. Fire Manage. Notes 3 5 , 1 0 - 1 3 .
Cowan, I. R. (1965). Transport of water in the soil-plant-atmosphere system. J. Appl. Ecol. 2, 2 2 1 -
239.
Cowan, 1. R., and Milthorpe, F. L. (1968). Plant factors influencing the water status of plant tissues.
In "Water Deficits and Plant Growth" (T. T. Kozlowski, Ed.), Vol. 1, pp. 1 3 7 - 1 9 3 . Academic
Press, New York.
Crank, J. (1975). "The Mathematics of Diffusion," 2nd ed. Clarendon Press, Oxford.
Davis, K. R (1959). "Forest Fire: Control and Use." McGraw-Hill, New York.
Dunlap, M. E. (1932). The drying rate of hardwood-forest leaves. J. For. 30, 4 2 1 - 4 2 3 .
Chapter 4 Water Relations of Forest Fuels 145

Federer, C. A. (1979). A soil-plant-atmosphere model for transpiration and availability of soil wa-
ter. Water Resources Res. 15, 555-562.
Fernandez, M. L., and Howell, J. R. (1997). Convective drying model of southern pine. Drying Tech.
15,2343-2375.
Flowers, T. J., and Yeo, A. R. (1992). "Solute Transport in Plants." Blackie Academic & Professional,
Chapman & Hall, Glasgow.
Fons, W. L. (1950). Heating and ignition of small wood cylinders. Ind. Engr. Chem. 42, 2130-2133.
Forestry Canada Fire Danger Group. (1992). "Development and Structure of the Canadian Forest
Fire Behavior Prediction System." Information Report ST-X-3. Forestry Canada, Science and
Sustainable Development Directorate, Ottawa.
Fosberg, M. A. (1972). Theory of precipitation effects on dead cylindrical fuels. For. Sci. 18, 9 8 -
108.
Fosberg, M. A. (1975). "Heat and Water Vapor Flux in Conifer Forest Litter and Duff: A Theoreti-
cal Model." Research Paper RM-152. USDA Forest Service, Rocky Mountain Forest and Range
Experiment Station, Fort Collins.
Fosberg, M. A. (1977). "Heat and Water Vapor Transport Properties in Conifer Duff and Humus."
Research Paper RM-195. USDA Forest Service, Rocky Mountain Forest and Range Experiment
Station, Fort Collins.
Gardner, W R., and Ehlig, C. F. (1965). Physical aspects of the internal water relations of plant
leaves. Plant Physiol 40, 705-710.
Gary, H. L. (1971). Seasonal and diurnal changes in moisture contents and water deficits of Engel-
mann spruce needles. Bot. Gaz. 132, 327-332.
Gong, L., and Plumb, O. A. (1994). The effect of heterogeneity on wood drying, part 1: Model
development and predictions. Drying Tech. 12, 1983-2001.
Hart, C. A. (1964). Principles of moisture movement in wood. For. Prod. J. 14, 207-214.
Hart, C. A., and Thomas, R. J. (1967). Mechanism of bordered pit aspiration as caused by capillar-
ity. For. Prod. J. 17, 61-68.
Hatton, T. J., Viney, N. R., Catchpole, E. A., and de Mestre, N.J. (1988). The influence of soil mois-
ture on Eucalyptus leaf litter moisture. For. Sci. 34, 292-301.
Helvy, J. D. (1967). Interception by eastern white pine. Water Resources Res. 3, 723-729.
Hillel, D. (1980a). "Fundamentals of Soil Physics." Academic Press, New York.
Hillel, D. (1980b). "Applications of Soil Physics." Academic Press, New York.
Hough, W. A. (1973). "Fuel and Weather Influence Wildfires in Sand Pine Forests." Research Pa-
per SE-106. USDA Forest Service, Southeastern Forest Experiment Station, Asheville.
Howard, E. A., III. (1978). A simple model for estimating the moisture content of living vegetation
as potential wildfire fuel. In "Preprint Volume, Fifth National Conference on Fire and Forest
Meteorology," pp. 20-23. American Meteorological Society, Boston.
Jarvis, P. G. (1975). Water transfer in plants. In "Heat and Mass Transfer in the Biosphere. Part I.
Transfer Processes in the Plant Environment" (D. A. de Vries and N. H. Afgan, Eds.), pp. 3 6 9 -
394. Scripta Book Company, Washington, DC.
Jury, W A., Gardner, W R., and Gardner, W. H. (1991). "Soil Physics," 5th ed. Academic Press, San
Diego.
Katchalsky, A., and Curran, P. F. (1965). "Nonequilibrium Thermodynamics in Biophysics." Har-
vard University Press, Cambridge.
Keetch, J. J., and Byram, G. M. (1968). "A Drought Index for Forest Fire Control." Research Paper
SE-38. USDA Forest Service, Southeastern Forest Experiment Station, Asheville.
Kelsey, K. E. (1957). The sorption of water vapour by wood. Aust.J. Appl. Sci. 8, 42-54.
King, A. R., and Linton, M. (1963). Moisture variation in forest fuels: the rate of response to cli-
mate change. Aust.J. Appl. Sci. 14, 38-49.
King, N. R. (1973). The influence of water vapour on the emission spectra of flames. Combust. Sci.
Tech. 6, 247-256.
146 Ralph M. Nelson, Jr.

Kozlowski, T. T., and Clausen, J. J. (1965). Changes in moisture contents and dry weights of buds
and leaves of forest trees. Bot. Gaz. 126, 20-26.
Kozlowski, T. T., and Pallardy, S. G. (1997). "Physiology of Woody Plants," 2nd ed. Academic
Press, San Diego.
Kramer, P. J., and Boyer, J. S. (1995). "Water Relations of Plants and Soils." Academic Press,
New York.
Kramer, P. J., and Kozlowski, T. T. (1979). "Physiology of Woody Plants." Academic Press, Orlando.
Kreith, F., and Sellers, W D. (1975). General principles of natural evaporation. In "Heat and Mass
Transfer in the Biosphere. Part I. Transfer Processes in the Plant Environment" (D. A. de Vries
and N. H. Afgan, Eds.), pp. 207-227. Scripta Book Company, Washington, DC.
Lemon, E., Stewart, D. W, and Shawcroft, R. W (1971). The sun's work in a cornfield. Science 174,
371-378.
Linton, M. (1962). "Report on Moisture Variation in Forest Fuels—Prediction of Moisture Con-
tent." Commonwealth Scientific and Industrial Research Organization, Division of Physical
Chemistry, Melbourne.
Little, C. H. A. (1970). Seasonal changes in carbohydrate and moisture content in needles of bal-
sam fir (Abies hahamed). Can.]. Bot. 48, 2021-2028.
Lopushinsky, W, and Klock, G. O. (1974). Transpiration of conifer seedlings in relation to soil
water potential. For. Sci. 20, 181-186.
Luke, R. H., and McArthur, A. G. (1978). "Bushfires in Australia." Australian Government Publish-
ing Service, Canberra.
McGilvary, J. M., and Barnett, J. P. (1988). Increasing speed, accuracy, and safety of pressure cham-
ber determinations of plant moisture stress. Tree Planters' Notes 39, 3 - 4 .
McNamara, W. S., and Hart, C. A. (1971). An analysis of interval and average diffusion coefficients
for unsteady-state movement of moisture in wood. Wood Sci. 4, 37-45.
Monteith, J. L. (1963a). Gas exchange in plant communities. In "Environmental Control of Plant
Growth" (L. T. Evans, Ed.), pp. 95-112. Academic Press, New York.
Monteith, J. L. (1963b). Dew: Facts and fallacies. In "The Water Relations of Plants" (A. J. Rutter
and F. H. Whitehead, Eds.), pp. 37-56. John Wiley, New York.
Monteith, J. L. (1980). The development and extension of Penman's evaporation formula. In "Ap-
plications of Soil Physics" (D. Hillel, Ed.), pp. 247-253. Academic Press, New York.
Monteith, J. L., Szeicz, G., and Waggoner, P. E. (1965). The measurement and control of stomatal
resistance in the field. J. Appl. Ecol. 2, 345-355.
Moore, A., and Swank, W T. (1975). A model of water content and evaporation for hardwood
leaf litter. In "Mineral Cycling in Southeastern Ecosystems," (F. G. Howell, J. B. Gentry, and
M. H. Smith, Eds.), pp. 58-69. ERDA Symposium Series, Technical Information Center, Office
of Public Affairs, U.S. Energy Research and Development Administration, Oak Ridge.
Moren, T., Salin, J.-G., and Soderstrom, O. (1992). Determination of the surface emission factors
in wood sorption experiments. In "Understanding the Wood Drying Process: A Synthesis of
Theory and Practice," (M. Vanek, Ed.), pp. 69-73. 3rd lUFRO International Wood Drying Con-
ference, Boku, Vienna.
Mount, A. B. (1972). "The Derivation and Testing of a Soil Dryness Index Using Run-off Data." Bul-
letin No. 4. Tasmania Forestry Commission, Hobart.
Nelson, R. M., Jr. (1969). "Some Factors Affecting the Moisture Timelags of Woody Materials." Re-
search Paper SE-44. USDA Forest Service, Southeastern Forest Experiment Station, Asheville.
Nelson, R. M., Jr. (1984). A method for describing equilibrium moisture content of forest fuels.
Can.]. For. Res. 14, 597-600.
Nelson, R. M., Jr. (1986a). Diffusion of bound water in wood. Part 1: The driving force. Wood Sci.
Tech. 20, 125-135.
Nelson, R. M., Jr. (1986b). Diffusion of bound water in wood. Part 2: A model for isothermal dif-
fusion. Wood Sci. Tech. 20, 235-251.
Chapter 4 Water Relations of Forest Fuels 147

Nelson, R. M., Jr. (2000). Prediction of diurnal change in 10-hour fuel stick moisture content. Can.
J. For. Res. 30, 1071-1087.
Nissan, A. H., and Kaye, W. G. (1957). Mechanism of drying of thick porous bodies during the
falling-rate period. Nature 180, 142-143.
Nissan, A. H., Kaye, W. G., and Bell, J. R. (1959). Mechanism of drying thick porous bodies during
the falling rate period. 1. The pseudo-wet-bulb temperature. AIChEJ. 5, 103-110.
Noble, I. R., Bary, G. A. V., and Gill, A. M. (1980). McArthur's fire-danger meters expressed as equa-
tions. Aust.J. Ecol. 5, 201-203.
Norum, R. A., and Fischer, W. C. (1980). "Determining the Moisture Content of Some Dead For-
est Fuels Using a Microwave Oven." Research Note INT-277. USDA Forest Service, Intermoun-
tain Forest and Range Experiment Station, Ogden.
Olson, C. M. (1980). An evaluation of the Keetch-Byram drought index as a predictor of foliar
moisture content in a chaparral community. In "Proceedings of the Sixth Conference on Fire
and Forest Meteorology," pp. 241-245. Society of American Foresters, Bethesda.
Olson, J. S. (1963). Energy storage and the balance of producers and decomposers in ecological sys-
tems. Ecology 44, 322-331.
Panshin, A. J., and de Zeeuw, C. (1980). "Textbook of Wood Technology, Vol. 1," 4th ed. McGraw-
Hill, New York.
Perre, P., Fohr, J. P., and Arnaud, G. (1990). A model of drying applied to softwoods: The effect of
gaseous pressure below the boiling point. In "Drying '89" (A. S. Mujumdar and M. Roques,
Eds.), pp. 91-98. Hemisphere Publishing Corporation, New York.
Philip, J. R. (1966). Plant water relations: some physical aspects. Ann. Rev. Plant Physiol. 17, 2 4 5 -
268.
Piiiol, J., Filella, 1., Ogaya, R., and Penuelas, J. (1998). Ground-based spectroradiometric estima-
tion of live fine fuel moisture of Mediterranean plants. Agric. For. Meteorol. 90, 173-186.
Plumb, O. A., Spolek, G. A., and Olmstead, B. A. (1985). Heat and mass transfer in wood during
drying. Int. J. Heat Mass Transfer 28, 1669-1678.
Pompe, A., and Vines, R. G. (1966). The influence of moisture on the combustion of leaves. Aust.
For. 30, 231-241.
Pyne, S. J. (1984). "Introduction to Wildland Fire." John Wfley, New York.
Pyne, S. J., Andrews, P. L., and Laven, R. D. (1996). "Introduction to Wildland Fire," 2nd ed. John
Wiley, New York.
Riha, S. J., and CampbeU, G. S. (1985). Estimating water fluxes in Douglas-fir plantations. Can. J.
For. Res. 15, 701-707.
Rose, C. W (1966). "Agricultural Physics." Pergamon Press, Oxford.
Rothermel, R. C , Wilson, R. A., Jr., Morris, G. A., and Sackett, S. S. (1986). "Modeling Moisture
Content of Fine Dead Wfldland Fuels: Input to the BEHAVE Fire Prediction System." Research
Paper lNT-359. USDA Forest Service, Intermountain Research Station, Ogden.
Rothwell, R. L., Woodard, P. M., and Samran, S. (1991). The effect of soil water on aspen litter mois-
ture content. In "Proceedings of the Eleventh Conference on Fire and Forest Meteorology"
(P. L. Andrews and D. F. Potts, Eds.), pp. 117-123. Society of American Foresters, Bethesda.
Running, S. W (1978). A process oriented model for live fuel moisture. In "Preprint Volume, Fifth
Joint Conference on Fire and Forest Meteorology", pp. 24-28. Society of American Foresters,
Bethesda and American Meteorological Society, Boston.
Running, S. W (1984a). "Documentation and Preliminary Vahdation of H20TRANS and DAY-
TRANS, Two Models for Predicting Transpiration and Water Stress in Western Coniferous
Forests." Research Paper RM-252. USDA Forest Service, Rocky Mountain Forest and Range Ex-
periment Station, Fort Collins.
Running, S. W (1984b). Microclimate control of forest productivity: analysis by computer simu-
lation of annual photosynthesis/transpiration balance in different environments. Agric. For.
Meteorol. 32, 267-288.
148 Ralph M. Nelson, Jr.

Rutter, A.J. (1968). Water consumption by forests. In "Water Deficits and Plant Growth" (T. T. Koz-
lowski, Ed.), Vol. 2, pp. 23-84. Academic Press, New York.
Scholl, D. G. (1976). Soil moisture flux and evaporation determined from soil hydraulic properties
in a chaparral stand. Soil Sci. Soc. Am. J. 40, 14-17.
Schonherr, J. (1976). Water permeability of isolated cuticular membranes: the effect of cuticular
waxes on diffusion of water. Planta 131, 159-164.
Schroeder, M. J., and Buck, C. C. (1970). "Fire Weather . . . A Guide for Application of Meteoro-
logical Information to Forest Fire Control Operations." Agriculture Handbook No. 360. USDA
Forest Service, Washington, DC.
Shuttleworth, W J., and Wallace, J. S. (1985). Evaporation from sparse crops—an energy combi-
nation theory. Quart.]. Roy. Meteorol. Soc. I l l , 839-855.
Siau, J. F. (1995). "Wood: Influence of Moisture on Physical Properties." Department of Wood Sci-
ence and Forest Products, Virginia Polytechnic Institute and State University, Blacksburg.
Siau, J. R, and Avramidis, S. (1996). The surface emission coefficient of wood. Wood Fiber Sci. 28,
178-185.
Simpson, W T. (1993a). Determination and use of moisture diffusion coefficient to characterize
drying of northern red oak (Quercus rubra). Wood Sci. Tech. 27, 409-420.
Simpson, W T. (1993b). "Specific Gravity, Moisture Content, and Density Relationship for Wood."
General Technical Report FPL-GTR-76. USDA Forest Service, Forest Products Laboratory,
Madison.
Skaar, C. (1988). "Wood-Water Relations." Springer-Verlag, Berlin.
Slatyer, R. O. (1967). "Plant-Water Relationships." Academic Press, New York.
Sneeuwjagt, R. J., and Peet, G. B. (1985). "Forest Fire Behaviour Tables for Western Australia."
Western Australian Department of Conservation and Land Management, Perth.
Spolek, G. A., and Plumb, O. A. (1980). A numerical model of heat and mass transport in wood
during drying. In "Drying, 1980," pp. 84-92. Hemisphere Publishing Corporation, New York.
Spolek, G. A., and Plumb, O. A. (1981). Capillary pressure in softwoods. Wood Sci. Tech. 15, 189-
199.
Stamm, A. J. (1946). "Passage of Liquids, Vapors and Dissolved Materials through Softwoods."
Technical Bulletin No. 929, USDA, Washington, DC.
Stamm, A. J. (1960). Combined bound-water and water-vapor diffusion into Sitka spruce. For.
Prod. J. 10,644-648.
Stamm, A.J. (1964). "Wood and Cellulose Science." Ronald Press, New York.
Stamm, A. J., and Harris, E. E. (1953). "Chemical Processing of Wood." Chemical Publishing Com-
pany, New York.
Stamm, A. J., and Nelson, R. M., Jr. (1961). Comparison between measured and theoretical drying
diffusion coefficients for southern pine. For. Prod. J. 11, 536-543.
Stocks, B.J. (1970). "Moisture in the Forest Floor—its Distribution and Movement." Publication
No. 1271. Department of Fisheries and Forestry, Canadian Forestry Service, Ottawa.
Susott, R. A. (1980). Thermal behavior of conifer needle extractives. For. Sci. 26, 347-360.
Susott, R. A. (1982). Characterization of the thermal properties of forest fuels by combustible gas
analysis. For. Sci. 28, 404-420.
Tanner, C. B. (1968). Evaporation of water from plants and soil. In "Water Deficits and Plant
Growth" (T. T. Kozlowski, Ed.), pp. 73-106. Academic Press, New York.
Tourula, T., and Heikinheimo, M. (1998). Modelling evapotranspiration from a barley field over the
growing season. Agric. For. Meteorol. 91, 237-250.
Tunstall, B. (1991). Live fuel water content. In "Proceedings, Conference on Bushfire Modelling
and Fire Danger Rating Systems" pp. 127-136. CSIRO Division of Forestry, Yarralumla.
Urqhart, A. R. (1960). Sorption isotherms. In "Moisture in Textiles" (J. W S. Hearle and R. H. Pe-
ters, Eds.), pp. 14-32. Interscience, New York.
Chapter 4 Water Relations of Forest Fuels 149

Van Wagner, C. E. (1969a). Drying rates of some fine forest fuels. Fire Contr. Notes 30, 5, 7, 12.
Van Wagner, C. E. (1969b). "Combined Effect of Sun and Wind on Surface Temperature of Litter."
Information Report PS-X-10. Canadian Forest Service, Petawawa Forest Experiment Station,
Chalk River.
Van Wagner, C. E. (1972a). Duff consumption by fire in eastern pine stands. Can. J. For. Res. 2,
34-39.
Van Wagner, C. E. (1972b). "Equilibrium Moisture Contents of Some Fine Forest Fuels in Eastern
Canada." Information Report FS-X-36. Canadian Forest Service, Petawawa Forest Experiment
Station, Chalk River.
Van Wagner, C. E. (1979). A laboratory study of weather effects on the drying rate of jack pine fit-
ter. Can. J. For. Res. 9, 267-275.
Van Wagner, C. E. (1987). "Development and Structure of the Canadian Forest Fire Weather In-
dex System." Forestry Technical Report 25. Canadian Forest Service, Ottawa.
Viney, N. R. (1991). A review of fine fuel moisture modelling. Int. J. Wildland Fire 1, 215-234.
Viney, N. R. (1992a). "Modelling Fine Fuel Moisture Content." PhD dissertation. Department of
Mathematics, University College, University of New South Wales, Sydney.
Viney, N. R. (1992b). Moisture diffusivity in forest fuels. Int. J. Wildland Fire 2, 161-168.
Viney, N. R., and Catchpole, E. A. (1991). Estimating fuel moisture response times from field ob-
servations. Int.J. WildlandFire 1, 211-214.
Viney, N. R., and Hatton, T. J. (1989). Assessment of existing fine fuel moisture models applied to
Eucalyptus htter. Aust. For. 52, 82-93.
Viney, N. R., and Hatton, T. J. (1990). Modelling the effect of condensation on the moisture con-
tent of forest litter. Agric. For. Meteorol. 51, 51-62.
Wallace, J. S., Roberts, J. M., and Sivakumar, M. V. K. (1990). The estimation of transpiration from
sparse dryland millet using stomatal conductance and vegetation area indices. Agric. For.
Meteorol. 5 1 , 3 5 - 4 9 .
Waring, R. H., and Running, S. W (1976). Water uptake, storage and transpiration by conifers: A
physiological model. In "Water and Plant Life, Problems and Modern Approaches" (O. L. Lange,
L. Kappen, and E.-D. Schulze, Eds.), pp. 189-202. Springer-Verlag, Berlin.
Wilson, R. A., Jr. (1990). "Reexamination of Rothermel's Fire Spread Equations in No-wind and
No-slope Conditions." Research Paper INT-434. USDA Forest Service, Intermountain Research
Station, Ogden.
Xanthopoulos, G., and Wakimoto, R. H. (1993). A time to ignition-temperature-moisture rela-
tionship for branches of three western conifers. Can. J. For. Res. 23, 253-258.
Zhang, J., and Peralta, P. N. (1999). Moisture content-water potential characteristic curves for red
oak and loblolly pine. Wood Fiber 31, 360-369.
This Page Intentionally Left Blank
 CHAPTER

Wildland Fire
Spread Models
R. O . WEBER
School of Mathematics and Statistics, University College UNSW, ADFA Canberra, Australia

1. Introduction
II. Head Fire Rate of Spread (Physical Principles and Their
Mathematical Embodiment)
III. Head Fire Rate of Spread: Australia
IV. Head Fire Rate of Spread: United States
V. Head Fire Rate of Spread: Canada
VI. Smoldering
VII. Whole Fire Modeling—Fire Shape
A. Ellipse Model of Fire Growth
B. Elliptical Shape and Wind Speed
C. Elliptical Shape and Fire Intensity
D. Other Fire Spread Issues
Notation
References

I. INTRODUCTION
The spread of individual wildland fires directly affects plant communities on a
short timescale. This is distinct from fire regimes over many years which help to
define vegetation mosaics. It is this fairly immediate impact which has prompted
a great deal of fire modeling activity over many decades. Naturally the desire
to predict the spread of wildland fires to aid in containment and management
of resources during particular fire incidents has also been a strong motivational
factor.
In this chapter, we will survey spread models and try to give the reader an
understanding of the main ideas behind the modeling as well as an overview of
the different approaches and the relative merits of the various fire spread mod-
els. The emphasis will be on the explanation of relevant physical principles and

Forest Fires
Copyright © 2001 by Academic Press. All rights of reproduction in any form reserved. 151
152 R. O.Weber

the way in which they are encapsulated into the mathematical models. In some
cases, it will be possible to completely define the model; in others, only a sketch
will be possible, and the reader will need to consult the references for a com-
plete understanding.
One topic that we will not cover is the influence of firebrands on fire prop-
agation. This is an important topic which has only been moderately researched
so far (see Lee and Hellman, 1970; several USDA Forest Service Internal Re-
search Papers by Albini and a recent ANU Ph.D. thesis still under examination)
and deserves further study. Another topic of great interest is crown fires; both
their initiation and their propagation. This is an active area of ongoing research
(e.g., Grishin, 1984; Albini and Stocks, 1986) and was recently briefly reviewed
by Albini (1996). The fire modeling that is described in the present chapter is
essential background for anyone wishing to study crown fires, firebrands, etc.,
but these topics will not be discussed here.

II. HEAD FIRE RATE OF SPREAD

(PHYSICAL PRINCIPLES AND THEIR
MATHEMATICAL EMBODIMENT)
One of the first observables of a landscape fire that would occur to most people
is: "How fast did it move?" By this they usually mean the speed at which the
fastest section of the fireline spreads into unburnt fuels. It was considered early
in this century that the principle of energy conservation, expressed mathemati-
cally, should provide the fundamental tool for modeling and predicting the rate
at which the head of a landscape fire will spread. Among the earliest published
papers on this topic are Pons (1946), Bruce ct a\. (1961), Emmons (1964), Em-
mons and Shen (1971), Anderson (1968, 1969) Hottel et al (1964, 1971), Mc-
Carter and Broido (1965), McArthur (1966), Rothermel and Anderson (1966),
Albini (1967), Berlad et al (1971), Thomas (1967, 1971), van Wagner (1967),
and Rothermel (1972).
Later studies include Albini (1976a, 1976b, 1981, 1985, 1986), Beer (1991a,
1991b, 1993,1995), Catchpole (1987), Cekirge (1978), DeMestreetaL (1989),
Dorrer (1984), Duarte (1997a), Frandsen and Andrews (1979), Fujii et al
(1980), Gill et al (1995), Green et al (1985, 1990), Grishin (1984), Konev and
Sukhinin (1977), Larini et al (1998), Pagni and Peterson (1973), Steward
(1974), Telisin (1974), Weber (1989, 1990), Weber et al (1995), and Williams
(1982). Literature surveys of the history and the mathematical models them-
selves are available in papers by Catchpole and De Mestre (1986) and Weber
(1991, also translated for the Chinese Journal of Mechanics and Practice).
For ecological aspects see Brown (1981, 1982), da Silva (1990), Gill (1981),
Green (1983a, 1983b), Greig-Smith (1979), Klukas (1973), McArthur and
Chapter 5 Wildland Fire Spread Models 153

Cheney (1966), Minnich (1983), Noble and Slayter (1980), Omi and Kala-
bokidis (1998), Rothermel and Deeming (1980), and Schmidt and Wakimoto
(1987). For good general introductions, see Luke and McArthur (1978) and
Chandler et al. (1983).
Fire spread is determined by the rate at which heat is transferred from the
burning matter to unburnt matter. The way to describe this mathematically is
with a precise statement of the rate at which the heat energy is accumulated in
unburnt matter as a balance between energy inputs and energy outputs. To do
this, we begin with the energy conservation expressed in a "word equation" as

J Rate of accumulation 1 J Rate at which 1 J Rate at which

t of heat energy J [ energy flows in J [energy flows out

This is to be applied individually to small elements of the unburnt fuel, with

the fuel being considered to be a collection of many such small elements. W h e n
coupled with some ignition criteria, it then allows one to determine how quickly
successive elements of fuel ignite and consequently the rate of fire spread. An
algebraic relation can be obtained from this approach (Williams, 1977, 1982,
1985); namely.

pyQig = q with Qig = \ CpdT

This can be rearranged to give (what some people refer to as the fundamental
equation for) the rate of spread

V = —^
pQu
or in words

/ R a t e ofl {Heat flux from the active combustion^}

i spread J {Heat required for fuel ignition}

Early purely thermal models (as described in Weber, 1991) embodied the idea
of many small elements in a differential equation for the temperature. An ex-
ample of this style of modeling can be found in the equations in De Mestre et al.
(1989), which then needs to be solved by integration. In fact, the solution of
these models usually proceeds by assuming a constant rate of spread for the fire
and changing frame of reference to a coordinate "moving with the fire." It is in
this new coordinate system that the differential equation can be integrated and
boundary conditions applied to complete the determination of the arbitrary
constants of integration and the rate of spread. The essential requirement for

^Equates with Anderson and Rothermel's "Propagating Flux"; see Section III.
154 R.O.Weber

accurate prediction of the rate of spread is a good description of the heat trans-
fer processes from the fire to the fuel. This is what determines the value for the
parameter A in the following example. The use of books on heat transfer, such
as Gebhart (1971) or Ozisik (1973), can assist in exploring all of the options
for radiative, convective, and conductive heat transfer. The fire spread papers
cited in this chapter have explored many of the possibilities, with limited suc-
cess when comparisons are made with experimental results.

Example

Energy Conservation Equation

T(x - Vt, t) = Ti
Ignition Interface Condition

T ( x - ^ c x ) , 0 = T,
Ambient Condition

Step 1 Introduce a moving reference frame

X = X- Vt:

dT dT
Then — = -V^
dt dX

Hence -pc"/— = Ae""^ - ISJT

and T(X = 0,t) = T^

Step 2 Integrate this differential equation to obtain the temperature profile:

VpCp{T{X, t) - TJ
rX
JS^^lS^Xlpc^ I ^-aX-{\Sjpc;)X ^^

-aX
Ae
IS,
a H
Chapter 5 Wildland Fire Spread Models 155

Step 3 Use the condition r ( X = 0, t) = Tj to find the speed:

V =
L o^pCp{Ti - Tj

Notes

1. The symbol V has been used for the fire spread velocity. Often R or ROS
are used in the wildland fire literature because they conjure up rate of spread.
Although any definition is equally appropriate in principle, certain symbols can
cause confusion when moving between research fields. As an obvious example,
R is often used to denote a radius in geometry problems.
2. Simple models, such as the preceding example which can be solved ex-
actly, can provide valuable understanding of processes and the effect of variables
such as ignition temperature on measurables such as spread rate. Other similar
examples, such as the lag time for fire build up, can be found in the literature
(Weber, 1989). They may also form the basis for more complete simulation of
a fire incident as will be shown later in this chapter, and they can also be used
as the input to other models such as mesoscale meteorological models.
3. The prediction for rate of spread in this model can be seen to fit precisely
with the word equation for rate of spread given at the beginning of this chapter.
The heat flux from the flames, moderated by heat losses, is in the numerator,
and the heat of ignition is in the denominator.
4. This sort of model only considers heat transfer aspects of fire spread. It
does not model the combustion process. Rather, it assumes that the radiative
output from the flame can be independently determined or estimated. More so-
phisticated models which include the combustion process modeled by some
simplified chemical kinetic scheme, such as Grishin et at (1983), Larini et a\.
(1998), and Weber et al. (1997), require sophisticated numerical methods for
their solution and are well beyond the scope of this chapter.

III. HEAD FIRE RATE OF SPREAD: AUSTRALIA

The most widely used rate of spread models in Australia are the McArthur
models for grassland fires and forest fires (McArthur, 1966; Noble et al., 1980).
These models make no attempt to include any physical mechanisms for fire
spread. Rather, they are a purely statistical description of test fires and as such
can be very successful in predicting observations in circumstances similar to
the test fires. The McArthur models were developed and tested in dry grassland
and forest litter in South Eastern Australia during dry winter months. Initially,
156 R. O.Weber

the results were used to make meters, a circular slide rule which required the
user to select appropriate values for degree of curing, wind velocity, and so on.
The meter could then be used to obtain a fire danger index and a fire spread
rate. Noble et al. (1980) derived mathematical relationships to represent the ac-
tion of the circular slide rules. For example, the equations for the Mark 4 grass-
land fire danger meter are

¥ = 1 e x p [ - 2 3 . 6 + 5.01 In Q + 0.0281 T^ - 0.226 H]'^ + 0.663 V\^]

V = 0.036 F

where Q is the degree of curing in % , T^ is the air temperature in °C, H^ is the

relative humidity in % , JJIQ is the wind velocity in m/s measured at a height of
10 m, F is the fire danger index, and V is the fire spread rate in m/s.
This type of model is easy to use and has been successful when used to pre-
dict fire spread in conditions similar to those during the test fires; for example,
low-intensity fuel reduction burning. However, other fuel types such as heath-
lands and markedly different environmental conditions are beyond the range of
variables for which these models were constructed.

IV. HEAD FIRE RATE OF SPREAD:

UNITED STATES
The model most widely used in the United States was named after R. C.
Rothermel (1972) who provided a practical implementation of the empirical
model of Frandsen (1971). This used the principle of conservation of energy to
write down an equation for rate of spread, but it did not distinguish between
different modes of heat transfer. Rather, laboratory experiments, covering a
range of environmental and fuel variation, were used to empirically determine
the propagating flux c|, as a function of reaction intensity 1^:

a = - e x p r ( 0 . 7 9 2 + 3.760o-^^^)(i8 + 0.1)1
^ (192 + 7.8940-) ^^^ ^^^ ^^
where a is the fine particle surface area to volume ratio in cm"^ and (3 is the
fraction of fuel bed volume occupied by solid fuel (called the packing ratio and
also equivalent to one minus the void volume fraction). The reaction intensity
is related to the rate of fuel mass loss per unit area, dw/dt, according to

dw
h = -H-
T
dt
Chapter 5 Wildland Fire Spread Models 157

where H is the heat of combustion of the fuel. By substituting into the energy
balance q = pQigV and rearranging, it is possible to obtain an expression for the
spread rate
(Hdw/dt)
^ = ^ /.\-. rexp[(0.792 + 3.760o-^^^)(i8 + 0.1)1
P%{192 + 7.8940-) ^^^ ^^^ ^^
This equation highlights the important fuel bed parameters of p, a, and /3. The
empirical nature is reflected in the functional forms found through laboratory
experimentation.
Wind and slope effects upon the rate of fire spread are also included in the
Rothermel model by multiplying the formula for the spread rate V by

-Ew
rB,,/ P
VPo
05 = 5.275 j8"^^tan^^

where 6 is the angle of the slope measured from the horizontal, B^, C^, and E^
are parameters involving cr, and /3o in the optimum packing ratio. There seems
to be no obvious physical reasons for the functional forms used for </)^ and 05,
or indeed for the way in which the spread rate formula is modified to accommo-
date (|)^^ and 05. It is essentially an empirical model which works well in many
situations. A careful examination of the parameters (Beer, 1991a) showed that
the exponent to which the wind speed is raised could exceed unity (see Table 1),
meaning that for sufficiently strong winds, the model would predict that the fire
spreads faster than the wind moves. This is not a failure of the model. Rather,
it is a consequence of the inappropriate choice for parameters, required in cases
where the model was never validated. The ad hoc solution has been to place
upper limits on the values of 0;^.

TABLE 1 Representative Values for C^, B^, Ew

a (cm ^) ^w Cm ^w

285 3.38 1.46 0.03

154 2.43 2.52 0.13
69 1.57 3.97 0.34
40 1.17 4.77 0.46
4 n/a n/a n/a

Following Beer (1991a).

158 R. O.Weber

Up until the Yellowstone fires of 1988, it was considered that the Rothermel
model contained sufficient variability to accommodate most wildland fire situa-
tions. However, there has been a concerted effort over the last decade to more
completely understand the limitations of the existing Rothermel model and to
make improvements which will ultimately result in a new version. Not with-
standing these reservations, the Rothermel model is generally a great success.
It has been incorporated into a complete operational management tool and, as
such, forms an integral part of the BEHAVE software—available on the World
Wide Web at www.fire.org.

V. HEAD FIRE RATE OF SPREAD: CANADA

The Canadian Forest Service have conducted measurements and field experi-
ments over a 25-year period to compile the Canadian Forest Fire Behaviour
Prediction System, now available in various book forms and electronically.
This provides a systematic method for including vegetation, topography, and
weather variables appropriate to Canadian ecosystems and assessing wildland
fire behaviour potential. It consists of mathematical formulae developed em-
pirically and it is usually presented in tabular form or through a computer in-
terface. With any model such as this, we can expect the predictions to be quite
accurate provided our study area is reasonably represented by one of the choices
available.

VI. SMOLDERING
Smoldering combustion is essentially the slow exothermic consumption of fu-
els (particularly organic matter in the case of wildland fires) under limited oxy-
gen (anaerobic) conditions. This occurs quite widely in the world, from the coal
seams in tropical Kalimantan (Indonesia) which cause occasional surface fires
(Goldammer, 1990), to peat bogs in cool temperate areas of the world.
Large litter layers, such as those found in the boreal forests of North Amer-
ica, can become quite dry and support smoldering combustion, particularly
in the so-called duff layer (see Chapter 13 in this book for more on smolder-
ing duff).
The prediction of smoldering rate of spread is in principle much easier than
for flaming combustion as the heat transfer aspect is greatly simplified with con-
duction of heat, the only mechanism of significance (Ohlemiller, 1985, 1988;
Jones, 1993; Drysdale, 1999).
Chapter 5 Wildland Fire Spread Models 159

However, the kinetics of smoldering combustion and hence the rate of re-
lease of heat under various circumstances is not well understood. This is par-
ticularly the case when there are small traces of minerals, such as in duff layers.
For this reason, smoldering models are usually based upon the simple energy
balance as introduced earlier in this chapter, and a fixed temperature is chosen
for the smoldering interface.
One method for estimating the rate at which heat is released in a smolder-
ing front is to consider the length scale of the front, namely, the distance over
which the temperature rises from ambient, T^, to the smoldering temperature,
Tj^. If this distance is called x and if we also assume that the conductive heat
transfer has reached a steady state, then the flux of heat per unit cross-sectional
area is
T - T
q = k
X

We can use this in our basic combustion front propagation equation to obtain
the speed

X T - T
V
pc^ij, - T j xpc^ \T,-T,

Given that T^^ = Tj, for cellulosic materials the thermal diffusivity is k/pCp =
10~'^ m^ s~\ and x = I mm, we can estimate
V = 10~^ms~^ (i.e., 0.1 mm/s)
The heat flux would have been

q = 0.01 X — = 3 X 10^ Js"^ m"^

^ 154 ^
compared to the heat of ignition
pCp{T, - T j = 73 X 1370 X 300 = 3 X lO^m"^
To put this into a familiar context, we note that a typical cigarette has a little
less than 60 mm of tobacco. Hence, it will smolder for approximately 10 min-
utes if left alone in suitable circumstances.
The transition from smoldering to flaming combustion has been and con-
tinues to be an active area of research (Drysdale, 1999). It has been established
beyond doubt that the main factor in the transition are limits on the supply of
oxygen.
160 R. O.Weber

VII. WHOLE FIRE MODELING—FIRE SHAPE

So far, the fire spread models have only considered head fire rate of spread. Nat-
urally, a whole fire consists of more than merely the fastest moving front, and
the whole fire has an impact upon the landscape and particularly the vegeta-
tion. As one might expect, the intensity will vary as one traverses the fire
perimeter just as the rate of spread also varies. For this reason, it has long been
of interest to consider fire shape models and also to investigate mechanisms or
algorithms for whole fire growth.
Some of the earliest whole fire modeling was documented by Van Wagner
(1969), Kourtz and O'Regan (1971) and Anderson et al (1982); although
Curry and Pons (1938) and Peet (1967) had considerably earlier made selected
observations on fire shape. Given homogeneous fuel and weather conditions
and assuming a constant moderate wind, a fire growing from a point ignition
source will evolve to a shape which we would describe as "elliptic." This is not
to say that the fire shape is exactly elliptical (or some other similar mathemati-
cal curve) but that a reasonable description, in terms of accuracy and relative
ease of calculating significant features, is an ellipse growing with time [but see
also McAlpine and Wotton (1993)]. In the next section, this whole fire model
will be described in detail, but first we should make it clear that there are many
different approaches to whole fire modeling, and we will survey several of these
in the remainder of this chapter (including cellular automata and percolatier
approaches), but this is a growing (pun intended!) area of fire research for both
management training and conservation or ecological purposes.

A. ELLIPSE MODEL OF FIRE GROWTH

Following Anderson et al. (1982), consider first a homogeneous fuel on level

terrain with no wind, so that the conditions may be summarized as being iso-
tropic. In this case, we expect (on average) that the fire will appear as a circular
front, growing in time, and we can describe this mathematically by the para-
metric equations
X = at cos X
(1)
y = at sin x
where x and y are the coordinates in the plane of a point on the front of the fire,
a is the rate of spread (uniform across the whole front in this idealised, isotro-
pic situation), t is the time elapsed since ignition (or some other convenient
temperal reference time such as the time at which the fire is first observed in
which case t would increase from some nonzero initial value to) and ;^ is a pa-
Chapter 5 Wildland Fire Spread Models 161

rameter which can be interpreted as an angular coordinate determining the

location of the front at angles between 0° and 360° from the x-axis. That these
parametric equations describe a circle can be seen by rearranging Eq. (1) and
using the trigonometric identity sin^ x "*" ^^^ A' ~ 1 to give a standard form
for the equation of a circle in the x, y plane; namely x^ + y^ = Oi'^. We note
that this model has the radius of the circle increasing linearly with time at a rate
determined by a, the rate of spread, and a, in turn, is determined by fuel type,
temperature, and moisture content.
Having established the nomenclature for describing a growing circular fire
in isotropic conditions, it is possible to modify the parametric equations to in-
clude the effect of a constant wind and yield an elliptical fire front. This is done
by writing the coordinates x, y of any point on the fire front as
X = at(/cos ;\f + g)
(2)
y = at{h sin x)
Clearly for the no wind case, / = h = 1 and g = 0; however, a windspeed of \J
will change the values of/, g, and h. The dependence upon wind speed, which
we may indicate by writing /(U), G(U), and h(l/), needs to be determined in
some other way, either empirically by fitting to the existing fire data or possibly
from physical arguments. Note that Alexander (1985) was able to determine
suitable values for a,/, g, and h for particular fires. Next, we note that Eq. (2) de-
scribe an ellipse with semiaxes at/and a\h, and whose center is moving (as the
fire grows) in the x-direction with a speed ag. This can most easily be seen by
again using the trigonometric identity cos^ x + sin^ X ^ ^ ^^^ rearranging
Eq. (2) to give

+ — = 1
ajt ) \ahtj
Anderson et a\. (1982) derived this elliptical model for a growing fire and were
then able to show that it concurred with a "modified" Huygen's principle (fa-
miliar from the ray propagation theory of light). This was a useful observation
for the subsequent computer implementation of this spread model but need not
concern us for the immediate purposes of this chapter.

B. ELLIPTICAL SHAPE AND W I N D SPEED

The parameters in the elliptical model can be combined to give the heading,
flanking, and backing rate of spread as follows:
162 R.O.Weber

In Figure 11 of Anderson et at (1982), graphs of time series for the forward rate
of spread, the lateral rate of spread, and the wind speed are all shown for a grass
fire conducted by the CSIRO in Australia. This and other similar plots are not
conclusive in relation to the wind speed dependence of/, g, and h. However, it
is encouraging to observe that variation in the ratio of forward to lateral rate of
spread, which we can write as

^(/+g) , /+g
ah h
and plot against wind speed U (as in Figure 12 of Anderson et al., 1982), can
be accounted for by wind speed variation alone.

C. ELLIPTICAL SHAPE AND FIRE INTENSITY

An additional benefit of the elliptical model is that it presents a useful and rea-
sonably simple method for extending the Byram index of head fire intensity to
a measure of intensity which changes around the perimeter of the fire. The in-
dex presented by Byram (1959) is written mathematically as
1 = HwV
where H is the heat of combustion of fuel, w is the mass of fuel consumed per
unit area, and V is the heading rate of spread of the fire. In the elliptical model
of the whole fire, we would identify V = a{f + g). Hence, we could write in-
tensity at the head, flank, and rear as
H w a{j + g), Hw ah, Hw a{f - g)
Alternatively, following Catchpole et al. (1982), we may prefer to determine
an expression for the intensity at any point on the perimeter. For this purpose,
one needs to introduce the angle of the normal to any point on the perimeter
of the ellipse if/. The result of this is an intensity equation^

J = H w a(g cos (A + Vf cos^ (// + h^ sin^ if/)

Catchpole et al. (1982) then demonstrate how this intensity varies around the
perimeter for two cases idealized by
a. Medium wind: / = 2, g = 1.8, h = 1
b. Strong wind: / "^ 4, g ^ 3.8, h = I

^Note that the angle if/ has value 0 at the head of the ellipse and values ir/l and rr at the flank
and rear, respectively. In fact, — TT < i// < TT.
Chapter 5 Wildland Fire Spread Models 163

In addition to using the elliptical model to determine the intensity at any point
on the perimeter at any moment in time, it is also possible to determine a total
fire flux by integrating the intensity around the entire perimeter.
Catchpole et al. (1982) performed this integration for the ellipse model to
find that the total fire flux is
ITTHW a^fht
This quantity is a measurement (in units of joules per second or watts) of the
energy released by the total fire per unit time. Note that this quantity increases
linearly with time as does the fire perimeter.
In principle, the elliptical model allows for a comprehensive analysis of the
entire progression of a fire, from its ignition and will allow estimates of fire in-
tensity to be made at all points in space and time. Using the temperature model-
ing in the fire plumes chapter, these fire intensity estimates could then be used
to predict temperature-time exposures for vegetation and assess the ecological
impact of a fire incident, with the possibility of really comparing predictions of
fire impact with observed fire impact. At this stage, such a complete incident
analysis has never been undertaken.

D. OTHER FIRE SPREAD ISSUES

There have been several other approaches to modeling fire spread, often moti-
vated by a need to deal with inhomogeneous distributions of fuel and varying
topography. Unlike the elliptical models, the basic ideas have come from a very
local point of view of fire spread. For example, the cellular automata approach,
which was first introduced by Green et al. (1983, 1985), considers small cells
of fuel and develops a rule approach for deciding if the fire will spread to adja-
cent cells. This presents a very efficient and simple method for implementing
into management software and provides quite reasonable fire shapes, although
there has been considerable investigation into the effect the cell shape (triangu-
lar, square, hexagonal) will have on the eventual fire shape.
There have also been several probabilistic models for fire spread, which in-
clude a probability of spread from cell to cell in a given mesh. The earliest
of these was by Kourtz and O'Regan (1971), using a square mesh and fuel of
varying moisture content in the mesh, distributed according to a Monte Carlo
method. This then required an algorithm for determining the route of the fire.
This concept was also used by Catchpole etal.(l989), who used a Markov chain
approach to allow for several different types of fuel and also by Guertin and Ball
(1990). In all of these, the concepts of transition from fuel element to fuel ele-
ment and also randomness are important elements. For this reason, it has be-
come of interest to consider the percolation approach of Albinet et al. (1986)
164 R. O.Weber

(see also Stauffer and Aharony, 1992). This arose out of developments in sta-
tistical physics (De Gennes, 1976) which were directed at finding a unified ap-
proach to spread processes with a random element. Other ecological processes
(Gardner et al, 1987; Turner et al, 1989; Reed, 1999) have also been consid-
ered as being well modeled as percolation processes. On the experimental front,
Beer (1990), Beer and Enting (1990), and Duarte (1997) have conducted labo-
ratory studies using arrays of fuel elements. They have found good agreement
between the predicted exponents from percolation theory and the clusters
found experimentally.

NOTATION
ROMAN LETTERS

A radiation intensity from fire front Wm~^

a rate of spread in the absence of wind m s~^
B^ wind effect parameter dimensionless
Q degree of curing %
C^/ wind effect parameter dimensionless
Cp specific heat at constant pressure J kg~^ K~^
E^^ wind effect parameter dimensionless
F fire danger index m s~^
/ ellipse parameter dimensionless
g ellipse parameter dimensionless
H heat of combustion of fuel J kg"^
H^ height of wind velocity measurement m
h ellipse parameter dimensionless
I index for headfire intensity W m"^
JR reaction intensity W m~^
k thermal conductivity W m~^ K~^
I heat loss coefficient J K~^ m~^
Qig heat of ignition J kg~^
q heat transfer from active combustion W m~^
Sy whole fuel bed surface area to volume ratio m~^
T temperature K
Chapter 5 Wildland Fire Spread Models 165

t time s
u average wind velocity m s~^
Uio wind velocity at a height of ten meters ms~^
V rate of spread of fire front ms~^
w mass of fuel bed per unit area kgm" 2
X fixed spatial coordinate m
X moving spatial coordinate m
y fixed spatial coordinate m

GREEK LETTERS

a radiation absorptivity m~^

(3 packing ratio dimensionless
6 angle from the horizontal radians
p density kg m"^
a fine particle surface area to volume ratio cm~^
O5 slope factor dimensionless
Ow wind factor dimensionless
X angular parameter radians
if/ angle between normal and tangent radians

SUBSCRIPTS

a ambient property
ig ignition
sm smoldering

REFERENCES
Albinet, G., Searby, G., and Staufer, D. (1986). Fire propagation in a 2-D random medium. J. Phy-
sique 47, 1-7.
Albini, F. A. (1967). A Physical model for fire spread in brush. In "Uth Symp. (Int.) on Combust.,"
pp. 553-560. The Combustion Institute, Pittsburgh.
Albini, F. A. (1976a). "Estimating Wildfire Behaviour and Effects." Gen. Tech. Rep. INT-30, USDA
Forest Service, Ogden.
166 R.O.Weber

Albini, F. A. (1976b). "Computer-Based Models of Wildland Fire Behaviour: A User's Manual."

Intermt. For. and Range Exp. Stn., Ogden.
Albini, F. A. (1981). A model for the wind-blown flame from a line fire. Combust. Flame 43, 155-
174.
Albini, F. A. (1985). A model for fire spread in wildland fuels by radiation. Combust. Sci. Technol.
42, 229-258.
Albini, F. A. (1986). Wildland fire spread by radiation—A model including fuel cooling by con-
vection. Combust. Sci. Technol. 45, 101-113.
Albini, F. A. (1996). Iterative solution of the radiation transport equations governing spread of fire
in wildland fuels. Fiziha Goreniya i Vzryva 32, 71-82.
Albini, F. A., and Stocks, B. J. (1986). Predicted and observed rates of spread of crown fires in im-
mature Jack Pine. Combust. Sci. Technol. 48, 65-76.
Alexander, M. E. (1985). Estimating the length-to-breadth ratio of elliptical forest fire patterns. In
"Proc. 8th Conf. Forest and Fire Meteorology," pp. 287-304. Soc. Am. Foresters, Detroit.
Anderson, D. H., Catchpole, E. A., De Mestre, N. J., and Parkes, I. (1982). Modeling the spread of
grass fires. J. Aust. Math. Soc. (Ser. B) 23, 451-466.
Anderson, H. E. (1968). "Sundance Fire: An Analysis of Fire Phenomena." Res. Pap. INT-56, USDA
Forest Service, Ogden.
Anderson, H. E. (1969). "Heat Transfer and Fire Spread." Res. Pap. lNT-69, USDA Forest Service,
Ogden.
Beer, T. (1990). Percolation theory and fire spread. Combust. Sci. Tech. 72, 297-304.
Beer, T. (1991a). Bushfire rate of spread forecasting: Deterministic and statistical approaches to fire
modeling. J. Forecasting 10, 301-317.
Beer, T. (1991b). The interaction of fire and wind. Boundary Layer Meteorol. 54, 287-308.
Beer, T. (1993). The speed of a fire front and its dependence on wind speed. Int. J. Wildland Fire 3,
193-202.
Beer, T. (1995). Fire propagation in vertical stick arrays: The effects of wind. Int. J. Wildland Fire 5,
43-49.
Beer, T., and Enting, 1. G. (1990). Fire spread and percolation modeling. Math. Comput. Modeling
13, 77-96.
Berlad, A. L., Rothermel, R. C, and Frandsen, W (1971). In "13th Symp. (Int.) on Combust.,"
pp. 927-933. The Combustion Institute, Pittsburgh.
Brown, J. K. (1981). Bulk densities of nonuniform surface fuels and their application to fire mod-
eling. For. Sci. 27, 667—684.
Brown, J. K. (1982). "Fuel and Fire Behaviour Prediction in Big Sagebrush." Res. Pap. INT-290,
USDA Forest Service, Ogden.
Bruce, H. D., Pong, W. Y., and Eons, W. L. (1961). The effect of density and thermal diffusivity of
wood on the rate of burning of wooden cribs. Tech. Pap. No. 63, USDA Forest Service, Ogden.
Byram, G. M. (1959). Combustion of forest fuels, "Forest Fire: Control and Use" (K. P. Davis, Ed.).
McGraw-Hill, New York.
Catchpole, E. A., and De Mestre, N. J. (1986). Physical models for a spreading line fire, Aust. For.
49, 102-111.
Catchpole, E. A., De Mestre, N. J., and Gill, A. M. (1982). Intensity of fire at its perimeter, Aust. For.
Res. 12, 47-54.
Catchpole, E. A., Hatton, T. J., and Catchpole, W. R. (1989). Fire spread through nonhomogeneous
fuel modeled as a Markov process, Fcological Modeling 48, 101-112.
Catchpole, W. R. (1987). "Heathland Fuel and Fire Modeling." PhD thesis. University of New
South Wales.
Cekirge, H. M. (1978). Propagation of fire fronts in forests. Comput. Math. Applic. 4, 325-332.
Chandler, C , Cheney, N. P., Thomas, P. H., and Trabaud, L. (1983). "Fire in Forestry, Volume I:
Forest Fire Behaviour and Effects." Wiley, New York.
Chapter 5 Wildland Fire Spread Models 167

Curry, J. R., and Pons, W. L. (1938). Rate of spread of surface fires in the Ponderosa pine type of
Cahfornia. J. Agric. Res. 57, 239-267.
da Silva, J. M. (1990). La gestion forestiere et la sylviculture de prevention des espaces forestiers
menaces par les incendies au Portugal. Rev. For. Fr. XLII, 337-345.
De Gennes, P. G. (1976). La percolation: Un concept unificateur. La Recherche 72, 919-927.
De Mestre, N. J., Catchpole, E. A., Anderson, D. H., and Rothermel, R. C. (1989). Uniform propa-
gation of a planar fire front without wind. Combust. Sci. Technol. 65, 231-244.
Dorrer, G. A. (1984). A model of the spreading of a forest fire. Heat Transfer—Sov. Res. 16,
39-52.
Drysdale, D. D. (1999). "An Introduction to Fire Dynamics." Wiley, New York.
Duarte, J. A. M. S. (1997). Fire spread in natural fuel: Computational aspects. Ann. Rev. Comput
Phys. V, 1-23.
Emmons, H. (1964). Fire in the forest. Fire Res. Ahs. Rev. 5, 163-178.
Emmons, H., and Shen, T. (1971). Fire spread in paper arrays. In "13th Symp. (Int.) on Combust.,"
pp. 917-926. The Combustion Institute, Pittsburgh.
Pons, W. L. (1946). Analysis of fire spread in light forest fuels. J. Agric. Res. 72, 93-121.
Frandsen, W. H. (1971). Fire spread through porous fuels from the conservation of energy. Com-
bust. Flame 16, 9-16.
Frandsen, W. H., and Andrews, P. L. (1979). Fire behaviour in nonuniform fuels. Res. Pap. INT-
232, USDA Forest Service, Ogden.
Fujii, N., Hasegawa, J., Pallop, L., and Sakawa, Y. A. (1980). A non-stationary model of fire spread-
ing. Appl. Math. Modeling 4, 176-180.
Gardner, R. H., Milne, B. T., Turner, M. G., and O'Neill, R. V. (1987). Neutral models for the analy-
sis of broad-scale landscape pattern. Landscape Ecology 1, 19-28.
Gebhart, B. (1971). "Heat Transfer." McGraw-HiU, New York.
Gill, A. M. (1981). Coping with fire. In "The Biology of Australian Plants" (J. S. Pate and A.J. Mc-
Comb, Eds.). Univ. W.A. Press, Perth.
Gill, A. M., Burrows, N. D., and Bradstock, R. A. (1995). Fire modeling and fire weather in an Aus-
trahan Desert. CALM Sci. Suppl. 4, 29-34.
Goldammer, J. G. (1990). "Fire in the Tropical Biota." Springer Verlag, Berlin.
Green, D. G. (1983a). Shapes of simulated fires in discrete fuels. Ecol. Model. 20, 21-32.
Green, D. G. (1983b). The ecological interpretation of fine resolution pollen records. New Phytol.
94,459-477.
Green, D. G., Gill, A. M., and Noble, I. R. (1983). Fire shapes and the adequacy of fire spread mod-
els. Ecol Model. 20, 3 3 - 4 5 .
Green, D. G., House, A. P. N., and House, S. M. (1985). Simulating spatial patterns in forest eco-
systems. Mathematics and Computers in Simulation 27, 191-198.
Green, D. G., Tridgell, A., and Gill, A. M. (1990). Interactive simulation of bushfires in heteroge-
neous fuels. Math. Comput. Modeling 13, 57-66.
Greig-Smith, R (1979). Pattern in vegetation. J. Ecol. 67, 755-779.
Grishin, A. M. (1984). Steady-state propagation of the front of a high-level forest fire. Sov. Phys.
Dokl. 29, 917-919.
Grishin, A. M., Gruzin, A. D., and Zverev, V. G. (1983). Mathematical modeling of the spreading
of high-level forest fires. Sov. Phys. Dokl. 28, 328-330.
Guertin, D. P., and Ball, G. L. (1990). Using a fire spread model to determine ignition probabihties.
Proc. IstlCFFR, Coimbra, B-13, 1-5.
Hottel, H. C , Williams, G. C , and Kwentus, G. K. (1971). Fuel pre-heating in free-burning fires.
In "13th Symp. (Int.) on Combust.," pp. 963-970. The Combustion Institute, Pittsburgh.
Hottel, H. C , Williams, G. C , and Steward, F. R. (1964). Modeling of fire-spread through a fuel
bed. In "10th Symp. (Int.) on Combust.," pp. 997-1007. The Combustion Institute, Pittsburgh.
Jones, J. C. (1993). "Combustion Science." Millennium Books, Sydney.
168 R. O.Weber

Klukas, R. W. (1973). "Control Burn Activities in Everglades National Park." Proc. Annual Tall
Timbers Fire Ecology Conf., Tall Timbers REs. Stn., Tallahassee, EL.
Konev, E. V., and Sukhinin, A. I. (1977). The analysis of flame spread through forest fuel. Combust.
Flame 28, 217-223.
Kourtz, P. H., and O'Regan, W. G. (1971). A model for a small forest fire to simulate burned and
burning areas for use in a detection model. For. Sci. 17, 163-169.
Larini, M., Giroud, R, Porterie, B. and Loraud, J. C. (1998). A multiphase formulation for fire
propagation in heterogeneous combustible media. Int. J. Heat Mass Transfer 41, 881-897.
Lee, S. L., and Hellman, J. M. (1970). Eirebrand trajectory study using an empirical velocity-
dependent burning law. Combust. Flame 15, 165-274.
Luke, R. H., and McArthur, A. G. (1978). "Bushfires in Australia." Aust. Govt. Publishing Service,
Canberra.
McAlpine, R. S., and Wotton, B. M. (1993). The use of fractal dimension to improve wildland fire
perimeter predictions. Can. J. For. Res. 23, 1073-1077.
McArthur, A. G. (1966). Weather and grassland fire behaviour. Leaflet No. 100, Aust. Eor. Timb.
Bur., Canberra.
McArthur, A. G., and Cheney, N. P. (1966). The characterization of fires in relation to ecological
studies. Aust. For. Res. 2, 36-45.
McCarter, R. J., and Broido, A. (1965). Radiative and convective energy from wood crib fires. Pyro-
dynamics 2, 65-85.
Minnich, R. A. (1983). Eire mosaics in Southern California and Northern Baja California. Science
219, 1287-1294.
Noble, I. R., Bary, G. A. V., and Gifl, A. M. (1980). McArthur's fire-danger meters expressed as equa-
tions. Aust. J. Ecol. 5, 201-203.
Noble, I. R., and Slatyer, R. O. (1980). The use of vital attributes to predict successional changes in
plant communities subject to recurrent disturbance. Vegetatio 43, 5 - 2 1 .
Ohlemiller, T.J. (1985). Smoldering combustion. Prog. Energy Combust. Sci. 11, 277-282.
Ohlemiller, T. J. (1988). Smoldering combustion. Chapter 23 (Section 1). In "SPEE Handbook
of Eire Protection Engineering," Vol. 1, pp. 352-359. National Fire Protection Association,
Gaithersburg.
Omi, P. N., and Kalabokidis, K. D. (1998). Fuels modifications to reduce large fire probabihty. In
"Proc. 3rd ICEER," Vol. II, pp. 2073-2088. Luso.
Ozisik, M. N. (1973). "Radiative Transfer." Wfley, New York.
Pagni, J., and Peterson, G. (1973). Flame spread through porous fuels. In "14th Symp. (Int.) on
Combust.," pp. 1099-1107. The Combustion Institute, Pittsburgh.
Peet, G. B. (1967). The shape of mild fires injarrah forest. Aust. For. 31, 121-127.
Reed, W.J. (2000). Forest fires and oil fields as percolation phenomena. Science, submitted.
Rothermel, R. C. (1972). A mathematical model for predicting fire spread in wildland fuels. Res.
Pap. INT-115, USDA Forest Service, Ogden.
Rothermel, R. C , and Anderson, H. E. (1966). Fire spread characteristics determined in the labo-
ratory. Res. Pap. INT-30, USDA Forest Service, Ogden.
Rothermel, R. C , and Deeming, J. E. (1980). Measuring and interpreting fire behaviour for correla-
tion with fire effects. Gen. Tech. Rep. INT-93, USDA Forest Service, Ogden.
Schmidt, W C, and Wakimoto, R. H. (1987). Cultural practices that can reduce fire hazards to
homes in the interior west. In "Proc. Symposium and Workshop on Protecting People and
Homes from Wildfire in the Interior West," pp. 131-141. USDA Forest Service, Ogden.
Stauffer, D., and Aharony, A. (1992). "Introduction to Percolation Theory." Taylor and Francis,
London.
Steward, E R. (1974). Eire spread through a fuel bed, "Heat Transfer in Fires: Thermophysics, So-
cial Aspects, Economic Impact" (P. L. Blackshear, Ed.). Wiley, New York.
Chapter 5 Wildland Fire Spread Models 169

Telisin, H. P. (1974). Some aspects of the growth and spread of fires in the open. "Heat Transfers in
Flames" (N. H. Afgan and J. M. Beer, Eds.). Wiley, New York.
Thomas, P. H. (1967). Some aspects of the growth and spread of fires in the open. Forestry 40,
139-164.
Thomas, P. H. (1971). Rates of spread of some wind-driven fires. Forestry 44, 155-175.
Turner, M. G., Gardner, R. H., Dale, V. H., and O'Neill, R. V. (1989). Predicting the spread of dis-
turbance across heterogeneous landscapes. Oikos 55, 121-129.
Van Wagner, C. E. (1967). "Calculations on Forest Fire Spread by Flame Radiation." Rep. No. 1185,
Canadian Dept. Forestry, Edmonton, Alberta.
Van Wagner, C. E. (1969). A simple fire-growth model. For. Chron. 45, 103-104.
Weber, R. O. (1989). Analytical models for fire spread due to radiation. Combust. Flame 78, 3 9 8 -
408.
Weber, R. O. (1990). A model for fire propagation in arrays. Math. Comput. Modeling 13, 95-102.
Weber, R. O. (1991). Modeling fire spread through fuel beds. Prog. Energy Combust. Sci. 17, 67-82.
Weber, R. O., and De Mestre, N. J. (1990). Fire spread on a single ponderosa pine needle: Effect of
sample orientation and external flow. Combust. Sci. Technol. 70, 17-32.
Weber, R. O., Mercer, G. N., Mahon, G. J., Catchpole, W R., and Gill, A. M. (1995). Predicting max-
imum inter-hummock distance for fire spread. In "Bushfire '95 Proceedings." Tasmania Parks
and Wildlife Service.
Weber, R. O., Mercer, G. N., Sidhu, H. S., and Gray, B. F. (1997). Combustion waves for gases and
solids. Proc. Roy. Soc. (Lond) A453, 1105-1118.
Williams, F. A. (1977). Mechanisms of fire spread. In "16th Symp. (Int.) on Combust.," pp. 1281-
1294. The Combustion Institute, Pittsburgh.
Williams, F. A. (1982). Urban and wildland fire phenomenology. Prog. Energy Combust. Sci. 8, 317-
354.
Wilhams, F. A. (1985). "Combustion Theory," 2nd ed. Benjamin Cummings, Menlo Park, CA.
This Page Intentionally Left Blank
 CHAPTER 6
Wind-Aided Fire Spread
F. E. FENDELL AND M . F. WOLFF
Space and Technology Division, TRW Space and Electronics Group,
Redondo Beach, California

1. Introduction
II. Laboratory-Scale Setup
A. Test Facility
B. Instrumentation
III. Fire Spread Model
A. Properties of the Model
B. A General Formulation
C. Approaches to Solution of the Formulation
D. Closure of the Formulation for Fire Spread in
Chaparral-Type Fuel Loading
E. Coordination of an Experimental Program
F. Contribution of Modeling to Data Correlation
IV. Preliminary Testing of the Model
A. Fuel-Bed Width
B. Fuel Arrangement
C. Inert Content
D. Moisture
E. Substratum
V. Test Results for the Effect of Wind Speed and Fuel
Loading on the Rate of Fire Spread
A. Mixed Fuel Elements
VI. Conclusions
Notation
Recommended Reading
References

I. INTRODUCTION

Fire spread in wildland encompasses very complicated phenomena, typically

entailing turbulent reacting flow and intricate fuel arrangement, with processes
Forest Fires
Copyright © 2001 by Academic Press. All rights of reproduction in any form reserved. 171
172 Fendell and Wolff

occurring over a wide range of scales. Consequently, even simple observations

and measurements are difficult, and replication is almost impossible. We present
here one approach to quantifying such challenging phenomena by undertaking
controlled, replicable laboratory-scale experiments which capture at least some
of the important phenomena of large-scale fire spread. We then develop a semi-
empirical process-based model to simulate wind-aided fire spread and test the
model by using the laboratory-scale experimental setup to generate additional
data for comparison. The model may subsequently be validated in large-scale
wildland fires; when a model which incorporates the key features of laboratory-
scale fire spread is applied to a large-scale test, attention must be concentrated
on those additional phenomena that did not play a significant role in the smaller
scale testing (e.g., see Chapter 8 in this book).
In general, a small number of extreme fire events is responsible for most of
the total area burned [e.g., 10% of the fires in southern California account for
over 90% of the area burned (Minnich, 1998)]. Thus, we focus on extreme con-
ditions of very dry fuels and high winds which may result in exceptionally high-
intensity, rapidly spreading fires involving surface fuels. The objective is to pre-
dict the spread of such wind-aided fires (i.e., given the current location of the
firefront and given requisite information about the adjacent unburned region
prior to front arrival, where will the firefront be at future times?). Traditional
approaches to this problem have constrained the firefront perimeter to an ellip-
tical shape (see Finney, 1998) or used percolation theory (Christensen et al,
1993) to speculate about probabilities of future locations of the firefront. How-
ever, such approaches often delve little into physical mechanisms and may not
readily permit the incorporation of observational data. Given the difference in
our current state of knowledge concerning low-intensity and high-intensity fire,
we do not think that it is a wise investment of effort to try to develop a tractable
fire spread model that would be applicable in all situations. Therefore, our
model focuses on exceptional conditions conducive to the initiation and per-
sistence of rapid wildland fire spread (i.e., fire spread across dry, moderately
sparse fuel under fairly sustained substantial wind of fairly constant direction).
Such conditions are exemplified by, for example, Santa Ana-wind-driven fires
in southern California chaparral (Pyne, 1982; Reid, 1994; Goldstein, 1995) and
related fires arising in other Mediterranean-climate locales (Table 1; Figure 1).
For idealized scenarios, the travel of the firefront in time is straightforward
(Richards, 1993). For a uniform fuel distribution on level ground in the ab-
sence of wind, an initially circular front advances with radial symmetry and
remains circular (Figures 2a and 2b). For a uniform fuel distribution on level
ground in the presence of a near-ground (say, at 10-m height) wind that is
steady in magnitude and constant in direction, a firefront growing from a lo-
calized ignition often becomes oval in shape (Figure 2c). The head of the fire
(where the propagation is with the wind) spreads relatively rapidly, the back or
rear of the fire (where the propagation is against the wind) spreads relatively
 Chapter 6 Wind-Aided Fire Spread 173
TABLE 1 Sites of Wind-Aided Spread in Mediterranean Fire Regimes

Site Typical vegetation Typical wind Notes

Southern Hemisphere
Chile matorral upslope no strong coastward wind
(central) (underbrush)
South Africa fynbos berg (foehn 4-24-year cycle of
(near Cape (small bush) wind blows regeneration
of Good Hope) coastward)
Australia mallee (shrub- off interior desert, degraded soil, modest
(NSW, Victoria, like eucalypt) often followed by hills, occasional summer
South Australia) cold southerly rain; SW Australian fires
burster not strongly wind-aided

California
Northern chaparral (scrub Mono 20 - 40-year cycle of regen-
Southern oak; chamise, Santa Ana eration, hilly terrain
manzanita)

Mediterranean Basin
Greece, Spain, maquis (high bora (dry cold
France shrub) winter wind off
garrigue (low Julian Alps to-
shrub) ward Adriatic);
Greece, Spain phrygana (Greece) sirocco (dry hot
tomillares (Spain) South spring
(dwarf shrubs) wind—Italy,
hilly terrain
Spain);
Spain matorral mistral, tramontane
(underbrush) (dry cold NW
wind off Alps
Italy macchia toward Gulf of
(shrub zone) Genoa—France,
Spain)

slowly, and the flanks (where the propagation is across the wind) progress only
moderately differently from the rate for the no-wind conditions. Clearly it is the
component of the wind which is locally perpendicular to the fire perimeter that
establishes whether advection carries or inhibits the travel of hot product gases
from just-burned fuel over the next-to-burn fuel to help preheat it. The long
axis of an oval fire perimeter typically continues to increase more rapidly than
the short axis. An oval perimeter frequently is narrower at the head than at the
rear (where the width is the dimension perpendicular to the wind direction),
and the width is often maximal near that windward site (along the bisector of
the burned area) that marks the point of origin. "In uniform fuel under con-
stant environmental conditions, the local rates of spread can be expected to be
174 Fendell and Wolff

FIGURE 1 Mediterranean climate locales, in which near-coastal scrublands typically have low-
fertility soils, and in which fire spread through surface vegetation is seasonally encountered, include
southern California, central Chile, much of the Mediterranean basin, the Cape region of South Af-
rica, and Australia (especially South Australia and Victoria).

M
u=o vertical cross- , vertical cross-

1
1
sectional view <
r
r .n fe sectional view

[ plan view
j plan view ]
burning area ^ - ^ . ^ 1 1.

unburned area ^ - ^ I :
burned out ^^
early later
time time
(a) (b)

wind

burning area

burning area
burned out
early
time
(c) (d)

FIGURE 2 (a) At an ignition site in a uniform expanse of unburned fuel on level terrain, under
calm, a firefront (which propagates slowly outward, against convectively induced advection) evolves
to demarcate (b) an interior region with burned-out thin fuel, from an exterior region of unburned
fuel, (c) For the corresponding event in the presence of a near-ground ambient wind sustained in
magnitude and direction, the firefront evolves to an oval configuration, (d) The rate of propagation
is appreciably enhanced along the portion of the perimeter where the wind aids spread and inhib-
ited where the wind opposes spread. The enhanced spread occurs due to hot combustion-product
gases preheating unburned downwind fuel.
Chapter 6 Wind-Aided Fire Spread 175

constant for head, flanks, and back of fire. Thus the area burned over per unit
time can be expected to grow hnearly with time, the perimeter expected to
grow at a steady rate, and the actively burning area inside the perimeter to grow
in proportion to the perimeter" (Albini, 1992, p. 44).
The oval shape of the firefront perimeter arising in simplistic scenarios has
been taken by many researchers to have the special configuration of an ellipse
(see Finney, 1998, and references cited therein). There seems no fundamental
reason for the elliptical shape to hold; in fact, the ellipse is but approximately
representative of the firefront perimeter, even for burns under uniform condi-
tions (Pyne et al, 1996, pp. 58 -59). Nevertheless, it is virtually unavoidable that
reference to ellipses permeate the ensuing discussion of the firefront-perimeter
literature. Because of the emphasis sometimes placed on elliptical configuration
for the firefront in wildland vegetation, we note that there are parameter ranges
for which the firefront is not even roughly elliptical, despite high regularity of
the fuel distribution, virtual flatness of the terrain, and constancy of the wind
magnitude and direction. In the hummock grasslands of arid Western Australia,
the patchy fuel distribution admits fire spread only for winds above a finite sig-
nificant threshold of 3 - 5 m/s or so (at 2 m height). Under a strong wind, back-
ing fire spread is minimal, flanking fire spread is modest, and head fire spread
from a localized ignition results in a downwind-widening-wedge shape for the
burned clumps of vegetation (Burrows et al, 1991; Gill et al, 1995; Cheney and
Sullivan, 1997). Because the burning of these spinifex plains, for which 30 -50%
of the ground may be covered by spinifex-type grass (with the remainder being
bare arid soil), raises significant issues, we return to spinifex later.
Of course, in realistic scenarios, neither the wind, fuel loading, nor topogra-
phy remains constant, and no useful general statement about how the configu-
ration of the fire perimeter evolves seems likely to hold. The nonuniformities
result in a typical, convoluted fire perimeter (in a plan view. Figure 3) having
concavities (indentations called "pockets") and convexities (protrusions called
"fingers") (Pyne, 1984); the possibilities are myriad. For example, buoyancy in-
duces a wind that aids upslope spread and inhibits downslope spread; yet, if the
ambient wind were blowing across the slope, then what might be considered a
flank on the basis of slope might become a head on the basis of the forced con-
vection. Such reorientation is also to be expected if the wind alters direction, as
apparently is commonly encountered in southeastern Australia bush fires when
sirocco-type winds, drawn off the hot desert interior prior to cold-front passage,
are succeeded by cooler moister southwesterly winds (southern burster) off the
ocean after cold-front passage (Pyne, 1991). Corresponding reorientation is less
common in southern California because weather fronts pass to the north, es-
pecially in dry years. However, at the end of Santa Ana-wind-driven fire spreads
in southern California, during which compressionally heated dry winds de-
scending off the Colorado plateau accelerate through narrow passes in the
southern Sierra Mountains to cause rapid propagation from the interior toward
176 Fendell and Wolff

flank

head

rear

spot fire

wind flank

FIGURE 3 Some commonly used terminology to describe phenomena associated with wind-
assisted fire spread through wildland fuel and portions of the firefront perimeter.

the coastline, the eventual return of gentler onshore marine winds may lead to
a daytime reversal of the direction of spread. If so, patches of fuel (islands or
"refugia") left unburned may then be consumed (Pyne, 1995). Since (as with
other relevant parameters) the slope of the terrain is perennial but exception-
ally rapid fire spread is dominantly strong-wind-associated, we infer that wind
incrementation of spread rate can appreciably exceed slope incrementation. Ac-
cordingly, the emphasis here is on fire spread under windy conditions.

II. LABORATORY-SCALE SETUP

We regard the conducting of well-instrumented experiments to be an essential
part of wildland fire modeling. In accord with Emmons (1971), we place em-
phasis on the control of experimental conditions to yield reproducible results.
The requirements for studying fire spread in a wildland situation are appre-
ciable: sensor distribution, maintenance, and readout; the area of surveillance
and the need for periodic updating of fuel properties and quarter-hourly (if not
subminute) updating of weather data; the requisite information collection, stor-
age, processing (including interpolation and extrapolation of the input data
from specific monitored sites to encompass the entire field of interest), display
and distribution. Furthermore, the obvious dangers of working in the types of
uncontrolled extreme wildfires of interest preclude the intentional execution
Chapter 6 Wind-Aided Fire Spread 177
of such events. Also, as mentioned previously, without adequate controls of the
various variables such as fuel loading, moisture content, and wind speeds, it is
difficult to test the roles of these factors. Finally, wildland fire situations are dif-
ficult to replicate. For these reasons, we have taken the approach of conducting
laboratory-scale experiments to develop and test a fire spread model. We use
fuel beds of regularly arranged discrete fuel elements that are well-defined, with
properties that can be systematically varied.

A. TEST FACILITY

Although a test facility with a large cross-sectional area is preferable, the larger
the cross section, the bigger is the blower needed to generate the wind that per-
mits examination of flow-assisted fire spread. The cost to build and operate the
facility increases with size; this can defeat the objective of extensive use of the
facility for data collection for the multiparametric phenomena of interest. A list
of international laboratory facilities for testing wind-aided fire spread across
discrete-fuel beds is given by Pitts (1991).
We prefer a facility that blows the airstream from upwind, through the test
section and over the test bed of fuel. Sucking air from the downwind side of
the test section is a laboratory artifice that inhibits the entrainment of air from
downwind into the hot, buoyant firefront gases (Cheney and Sullivan, 1997,
pp. 80-81).
We also prefer a facility with a movable ceiling that can translate in the di-
rection of the wind (Figure 4). Buoyant firefront gases may rise with a minimum

exhaust hood

' access windows blower

FIGURE 4 Sketch of a wind-tunnel facility with a blower that pushes ambient air through a flow-
conditioning section, to produce an approximately uniform stream of low-turbulence air. The ap-
proximately 1-m-wide, 5-m-long test section has a movable ceiling that can translate in the direc-
tion of the wind.
178 Fendell and Wolff

of obstruction if the downwind edge of the ceiling is dynamically positioned

just upwind of the propagating firefront. A movable ceiling maintains a con-
stant cross-sectional area for the portion of the test section that is upwind of
the propagating firefront so that the speed of the wind arriving at the firefront
remains invariant with firefront advance. Mounting of a video camera at the
downwind end of the ceiling furnishes feedback for positioning the ceiling dur-
ing a test and also provides an excellent overview. The height for a fixed ceiling
to be nonintrusive (so hot product gas is not artificially forced over the down-
wind, still-unburned portion of the fuel bed) is uncertain. Since the height of
the visible flames during testing in our facility (to be described later) is typi-
cally a meter or two, a factor-of-ten "clearance" would require a blower capable
of supplying winds in excess of, say, 5 m/s for a test section 1 0 - 2 0 m in height!
Alternatively, if the wind from the blower is allowed to expand into a test sec-
tion with a cross-sectional area much larger than that of the flow-preparation
duct so that the test section is effectively without a ceiling, then the wind speed
decreases along the length of the test bed. Only with a movable ceiling is a quasi
steady rate of propagation in a uniform wind speed achievable, if a quasi steady
rate exists for the given conditions. Data from our facility show that the rate of
fire spread is significantly increased when the ceiling height is held constant at
1 m relative to results obtained for the same conditions with the movable ceil-
ing; also, the rate of fire spread is slowed for results obtained for testing with-
out deployment of the ceiling, so the blower-generated wind expands into a
very large test section (Wolff et ah, 1991, Figure 9). To our knowledge, our fa-
cility is the only one with a movable ceiling, and Table 2 indicates that fire
spread rates obtained in it are less than corresponding rates measured in other
facilities with fixed ceilings. We omit citation of qualitative sketches from the
compilation included in Table 2.

B. INSTRUMENTATION

We concentrate exclusively on seeking the fire-propagation rate as a function

of the prefire fuel-bed observables, taken as uniform in space, and of the wind,
taken as constant in time. The observables include (1) the initial fuel loading,
(2) the fuel height, (3) the thickness of fuel elements, (4) the fuel moisture,
(5) the fuel-bed width, and (6) the wind speed.
To measure the rate of fire spread, we use type-k, chromel-alumel thermo-
couples inserted at regular intervals (along the centerline of the fuel bed, in the
direction of the wind) from underneath so that they protrude up through the
bed. The response time of the commercial thermocouples (OMEGA, Stamford,
CT) is about 1 s—adequate since the typical propagation rate is observed to be
below 10 cm/s.
Chapter 6 Wind-Aided Fire Spread 179
TABLE 2 Fire Spread Rate Vj Dependence on Wind Speed U (v^ ~ U")

n exponent,
Researcher jjn Fuel type Scale

Thomas (1971) 0.42-0.65'^ heather and gorse field

Wolff eta/. (1991) 0.49^ white pine toothpicks lab
Pons(1946) 0.67'^ ponderosa pine twigs lab
Steward (1974) 0.84« poplar wood shavings lab
Steward and Tennankore 0.9-1.9« birch dowels lab
(1979)
Catchpole et al. (1998) P regular and coarse excelsior, lab
ponderosa pine
needles/sticks
Gould (1991) P kerosene/eriachne grass field
Marsden-Smedley and 1.31^ buttongrass field
Catchpole (1995)
Steward (1974) 1.38« poplar match splints lab
Nelson and Adkins (1988) 1.5P slash-pine needles lab
Rothermel (1972) 1.9^ a = 97.7 cm"' model (semi-
empirical)
Burrows (1999a) 2.22^ jarrah litter lab
Burrows (1999b) 2.67^ jarrah litter field

''Data interpretation performed by Wolff et al. (1991).

''Data interpretation performed by the researchers themselves.
'Data interpretation performed by Gould (1991).
Note: The symbol a denotes surface-area-to-volume ratio.

The time histories of the centerhne thermocouples are used to ascertain when
and where a quasi steady rate of fire spread is achieved, and what that rate is.
Quasi steady propagation is achieved if the time history of a centerhne thermo-
couple is effectively identical to those of its neighbors, with temporal offsets
based on spatial separation. A typical thermocouple time history involves

1. A slow rise in temperature (interpreted as preheating);

2. A subsequent, more rapid rise (interpreted as onset of flaming); and
finally
3. A gradual fall (interpreted as forced-convective cooling of the thermo-
couple owing to the blowing).

Thermocouples are relatively inexpensive, are easily connected to a data-

acquisition system, and furnish data readily processed via automated (com-
puter) analysis, in contrast with video or infrared-camera recordings.
Thermogravimetric analysis (TGA) of white-pine toothpicks (the basic fuel
type used in our laboratory testing) indicates that 70% of the mass evolves as
gas (via pyrolysis) during gradual heating from 575 to 675 K. The speed and
180 Fendell and Wolff

turbulence level of the oncoming wind are determined by hot-wire anemome-

try. The moisture content within the fuel is determined by weighing the fuel be-
fore and after drying.

III. FIRE SPREAD MODEL

Using the previously described test facilities provides measurement of spread
rate as a function of the various test parameters; the results also guide our
thermo-gas-dynamic modeling of wind-aided fire spread.

A. PROPERTIES OF THE MODEL

Directly applying the conservation of mass, momentum, and energy to the fuel
bed and atmosphere for tracking the progression of the firefront is very difficult.
The highly disparate spatial scales, intricate geometry, and unsteady motion re-
main formidable obstacles. Typically, all cumulatively significant phenomena
occurring on scales smaller than the computational grid scale are parameter-
ized (phenomenologically formulated) in terms of grid-scale variables. In gen-
eral, it is unknown whether a valid representation of such kind is even possible,
and, even if possible, such representation is usually not known. Validation of
adopted conjectures requires detailed comparison of predictions with observa-
tions in a multitude of cases, and this demanding exercise is rarely carried out.
Instead, an approximate semiempirical model of firefront-propagation phe-
nomenology is sought (Beer, 1990,1991). For such a model to be useful in prac-
tice, the inputs should require no more detail concerning the local topography,
meteorology, and combustible matter than would normally be available. Our
objective is to predict the position of the firefront (idealized as a mathemati-
cally thin interface) at a later time from knowledge (1) of its position at an ear-
lier time and (2) of a minimal amount of (plausibly available) information about
essential properties of the environment in the vicinity of the firefront. In treat-
ing the firefront as an interface, we are tentatively regarding estimation of fire-
front thickness and time for thin-fuel-element burnout as lower priority issues.
We concentrate here on a single firefront from a single ignition; nevertheless,
the firefront, owing to the presence of firebreaks, spotting, or other circum-
stances, might become multiply-connected. Thus, burned-over areas should be
tallied, lest repeated burning of the same fuel owing to shifting wind (or merger
of two initially distinct firefronts) be erroneously permitted. Furthermore, the
firefront need not be a closed curve at any time, but we deal with the closed-
curve case here.
Chapter 6 Wind-Aided Fire Spread 101

For typical southern-California-hillside scenarios, the (immobile) combus-

tible matter consists predominantly of shrubs, such as chamise and manzanita,
and small trees. There is no noteworthy organic matter in the soil and no note-
worthy overstory of crowns. Thus, we focus exclusively on predicting firefront
propagation through surface fuels and give a lower priority to predicting flame
height. Because the time for flame to progress from one discrete fuel element to
another is much less than the temporal resolution of practical interest, we may
treat the distribution as an effectively continuous, equivalent mass of combus-
tible matter per unit planform area. Accordingly, we infer the progress of the
firefront to occur smoothly, not in discontinuous jumps.
Undertaking a simple approach first, we take the dependence of the firefront-
propagation speed to be on the values of local parameters, not on the spatial or
temporal gradients of the parameters. Nevertheless, the value of any parameter
is permitted to change in space and/or time. The procedure entails a quasi steady
approximation in that the flame speed holding for a given set of parameter val-
ues is instantaneously equilibrated to that set of values, whatever the anteced-
ent set of values. That is, if we regard the instantaneous firefront as a perime-
ter of ignition sites, the rate of fire spread is taken instantaneously to reach the
equilibrium rate of spread (into unhurned fuel) for the local values of the pa-
rameters related to propagation.
We do not incorporate spotting, the discontinuous mode of spread associ-
ated with the lofting of firebrands. If the fuel distribution varies but moderately
ahead of the current firefront position, then neglect of short-range spotting typi-
cally results in a modestly inaccurate rate of spread, since the firefront probably
soon arrives at the brand-ignited site by the continuous mode of spread. Longer
range spotting may permit the fire to traverse effectively fuel-free expanses that
would constitute firebreaks for the continuous mode of spread. We do not ad-
dress the interaction of the fire with the environment, so the nominal environ-
mental conditions are the conditions adopted for use in the model. For example,
any alteration of the wind by transit of already-burned-over expanses of fuel
is taken to be a higher order effect in describing the wind arriving at the cur-
rently burning fuel. This simplification is anticipated to require modification in
some circumstances. However, ignoring fire modification of the wind typically
may furnish an overprediction of spread rate, so the consequences are known.
We also do not explicitly incorporate the consequences of fire fighting on the
firefront-propagation model (e.g., owing to air-tanker or helicopter drops of fire
retardant or water on the fuel bed), except as those consequences may be formu-
lated in terms of parameters (such as fuel-moisture or soil-moisture content),
plausibly incorporated in the model for intervention-free spread. Of course,
tracking the fate of an attempted backfire may be regarded as little different from
tracking the firefront progress evolving from any other discrete (independent)
firestart.
182 Fendell and Wolff

B. A GENERAL FORMULATION

The firefront (interface) that encloses the area with burned thin combustible
matter is denoted by the "simple" smooth two-dimensional, plane-projection
curve r(s, t), where s denotes distance along the curve at time t (Figure 5). We
choose the parametric representation such that the burned area lies to the left
during a transit of the perimeter in the direction of increasing s. The direction of
fire spread through the fuel bed (the movement of the interface), in the imme-
diate vicinity of any point Xi{s, t) on the interface, is perpendicular to the inter-
face. The interface-translation speed F at each point on the interface is taken to
depend on the values assigned to the state variables u^^^ at that point on the in-
terface at the time of interest. The number of state variables is the minimal num-
ber of properties of the fuel bed, the meteorological environment, and the topog-
raphy necessary for a description (of the interface movement) that is of sufficient
accuracy to meet our requirements. Thus, for t > 0, with subscript i = 1 or 2,
dXi{s, t)
= F{u^^\xi{s, t ) ] , Vj(s, t)}, Xj(s, 0) given (1)
dt

local firefront
travel, P(t2-ti)
^' ^'^""^

A
outward normal, Vj (s,t)

local firefront local firefront

travel, e(t2-ti)
travel, y^{^2~^^)

C, rear A, head

local firefront
travel, FCtg-ti)
wind U

burned area
burned area during interval At
unburned area
FIGURE 5 A schematic of the increase in burned area (shaded region) due to the firefront travel
(stippled region) during the time interval {tj ~ h) under a wind of magnitude U in the direction
6 = 0. The distance s along the firefront at time t^ is measured from ^ = 0; a corresponding dis-
tance may be computed for the firefront at time tj. The rate of spread F depends (inter alia) on the
local unburned-thin-fuel loading and the component of wind normal to the front, so F varies with
6. The general rate F takes on the particular values Vj, (3, and s at the head, flank, and rear of the
fire, respectively.
Chapter 6 Wind-Aided Fire Spread 183

where Vi{s, t) denotes the outward normal. We let Xi(s, t) denote the x position
at time t of the point s along the perimeter, where x is a Cartesian coordinate
along the direction of the time-averaged wind at reference height above the
ground; ^2(5, t) denotes the corresponding y position, where the coordinate y
is perpendicular to x, in accord with the right-hand convention. We regard the
dependence of the functional F on u^^^ to be known empirically. Only the com-
ponent of the wind locally normal to the firefront is taken to aid the advance of
the local fire perimeter. In terms of the functions of computer software devel-
oped by the Fire Sciences Laboratory (Missoula, MX) of the Forest Service, U.S.
Department of Agriculture, for a two-dimensional simulation of fire growth
through surface fuels: (1) integration of Eq. (1) corresponds to the tasks under-
taken by the code FARSITE (Pyne et al, 1996; Finney, 1998); and (2) the key
functional F is furnished by the code BEHAVE (Rothermel, 1972; Andrews,
1991), which serves as a subroutine for FARSITE. Below we shall address issues
concerning mathematical accuracy and real-time implementation for FARSITE,
physical-formulation issues for BEHAVE, and limitations of a quasi steady, semi-
empirical approach. An Australian analogue of FARSITE is SiroFire (Coleman
and Sullivan, 1996). Were there sufficient record of the firefront position Xi{s, t)
during large fires, we might address the inverse approach of trying to infer the
functional F from knowledge of Xi{s, t), but such an undertaking is not cur-
rently practical.
This "parametric" model of firefront travel is probably not capable of accu-
rately resolving any fine-scale detail. Thus, in the current context, we ascribe
no credibility to details of behavior predicted by the model if the curvature of the
boundary is significant, as would be locally the case if the length of the bound-
ary between two points on it were comparable to the local radius of curvature.
Thus, prediction of large-curvature behavior is suspect as a mathematical arti-
fact of an approximate incorporation of the physics of the situation in the for-
mulation. How to resolve such artifacts is perhaps best ascertained as one gains
experience with the use of the model. To reiterate, the incidence of singularities
in curvature [usually a consequence of negative curvature ("pockets" in Fig-
ure 3) arising on the fire perimeter] may be the "most fascinating propagation
characteristics" (Sethian, 1996, p. 28) mathematically. However, their occur-
rence may reflect simply an incomplete incorporation of the physical laws.

C. APPROACHES TO SOLUTION
OF THE F O R M U L A T I O N

In practice, of course, a finite-difference approximation to the differential equa-

tion [Eq. (1)] is solved. A number of marker points distributed around the
interface are advanced by stepping forward in time, with a smaller distance be-
184 Fendell and Wolff

tween marker points As and a smaller time step At (within limits) seemingly
likely to incur smaller error because such choices permit a better approximation
to the derivative. However, in this approximate formulation, a coarser resolu-
tion could conceivably furnish a prediction closer to observation. Simplistic
procedures for treating the evolution equation include a first-order Euler time-
stepping procedure, or a second-order, predictor-corrector procedure; the order
of the numerical method also factors into the numerical error. Roberts (1989)
emphasizes a distinction between (1) the advancing of the distributed marker
points adopted to approximate the firefront curve at time t and (2) the advanc-
ing of line segments adopted to approximate the curve. The distinction seems
somewhat blurred in that a line segment must be associated with each marker
point in order to define the local normal to the curve, a necessary step to defin-
ing the local direction for advancement.
The commonly utilized computational procedure is described as "adopting
Huygens' principle" (see, e.g., Anderson et al, 1982; Knight and Coleman,
1993; Richards, 1993, and the references cited therein). The procedure usually
amounts to applying an elliptic stencil to describe how the fire grows at marker
points and drawing a smooth curve through the most advanced points on the
ellipses to obtain the updated location of the interface. The procedure seems ar-
bitrary and not particularly convenient for convoluted perimeters and, in gen-
eral, possibly highly inaccurate for the intricate nonhomogeneous situations
encountered in practice. Huygens' principle is usually encountered in the con-
text of ascertaining the evolution of an acoustic or electromagnetic wavefront.
The corresponding conjecture to permit tracking the propagation of a firefront
across a fuel bed is taking each point on the perimeter of a large fire to be an ig-
nition site for a new fire, which grows with the shape and orientation of the orig-
inal perimeter, the result being a larger fire of the same shape (Albini, 1984).
In practice, with few exceptions, the result is a larger fire of the same shape
(Albini, 1984). In practise, with few exceptions, the adopted firefront shape
evolving from an ignition site is always taken to be an ellipse, with the ignition
site taken as coincident with rear focus of the ellipse. The commitment to an el-
liptic firefront evolving from an ignition site on the firefront perimeter implies
that some already-burned fuel is treated as if it may r e b u r n — a n d then may
burn once more, because the ellipses from neighboring ignition sites typically
overlap. An ignition at the perimeter of a burned-over area is not equivalent
to an ignition within an unburned area; the experience that an ignition site
evolves to an oval firefront under a wind is derived from observation of spread
through an entirely unburned expanse of fuel. An upshot is that, under the just-
described Huygens' principle methodology for fire growth, the firefront is not
always propagated normal to itself into unburned fuel without ad hoc adjust-
ment. Finney (1998, pp. 28, 32) acknowledges that subjective spread-rate
adjustment factors are introduced to bring predictions of the Huygens' principle-
Chapter 6 Wind-Aided Fire Spread 185

based FARSITE model into agreement with observations of fire growth. Adjust-
ment is necessary because spread-rate predictions exceed the observed fire
growth for all fuel models treated. It is currently unresolved and perhaps unre-
solvable whether the discrepancy between prediction and observation is owing
to (1) shortcomings in the BEHAVE model of the spread rate for a given set
of state variables (see discussion that follows), (2) adoption of the elliptical-
stencil methodology to treat the firefront-evolution equation, (3) the spatial/
temporal resolution of the data put into the calculation, or (4) the step size used
in space and time to advance the solution—or a combination thereof. Finney
(1998) suggests that finer resolution in space and time in the modeling will
result in closer agreement between prediction and observation. In contrast,
Cheney and Sullivan (1997) suggest that, for a heading fire, wind measured
at 10 m above the ground be averaged for at least 10 minutes, preferably 1 5 -
20 minutes, for use in any currently available predictive approach, and that com-
parison of predictions of spread rate be made with measurements also taken
over a period of 15-20 minutes because accurately accounting for shorter term
variation in spread caused by gusts and lulls and spatial variations in the fuel
does not seem currently achievable.
As already noted, subdividing the firefront perimeter into small line seg-
ments (each of which is then propagated in the direction of the normal to the
arc into unburned fuel, much as a straightline firefront would advance in ac-
cord with the quasi steady fire spread rate for the local fire environment) may
result in overlap of the propagated arcs for regions of negative curvature. A
physical basis pertinent to fire spread through discrete fuel elements is needed
to guide "smoothing." Sethian (1996) categorizes the tracking of a set of marker
particles to approximate the tracking of a moving front as a discretized Lagrang-
ian approach. He suggests that a volume-of-fluid technique (Chorin, 1980) is
an Eulerian approach (describing where the front is in terms of a fixed spatial
grid), and probably superior. Only implementation can resolve questions of ac-
curacy and feasibility for tracking a complicated firefront geometry by use of
alternate discretized forms of Eq. (1). We question arbitrarily adopting changes
to the propagation-speed function F to serve mathematical convenience. Osher
and Sethian (1988) and Sethian (1996) suggest (by vague analogy, as if fire
propagation were basically a hyperbolic phenomenon) that the fire-propagation
function F be modified to the form (F — SK) , where e is a small positive con-
stant with the units of a diffusion coefficient and K is the local curvature of the
firefront; this modified form is amenable to so-called level-set methods, which
still give rise to physically implausible cusps in the firefront evolution but do
avoid mathematical multivaluedness. However, the only evidence (that fire
propagation is related to firefront curvature) cited by these authors is a specu-
lation by Markstein for flame propagation in homogeneous mixtures of reactive
gases. Markstein (1964, pp. 21-23) addresses instances in which the radius
186 Fendell and Wolff

of flame curvature is comparable to flame-structure thickness because, under

other circumstances, the role of curvature typically would be expected to be
negligible (Carrier et al., 1991). Markstein's speculation holds for phenomena
in which classical diffusional processes in the direction of spread are essential
to the mechanism of spread and in which the classical diffusional transport must
be modified from planar representation to infer the correct rate of spread. Fire
spread through chaparral-type surface fuels under wind aiding (diffusion-flame
burning in a boundary layer of reactive vapor pyrolyzed from a condensed fuel)
is not among such phenomena (see Section III.D). With respect to observations
justifying the adoption of a curvature-dependent spread rate for fire propagation
through surface-type fuels, Cheney and Sullivan (1997) do observe that if the
head of a wind-aided grassfire remains narrow, the fire spread is slower than for
a larger fire with a broad head. However, for a wind of 7 k m / h , the effect of cur-
vature on spread rate is negligible when the (say, maximum) width of the oval-
configured perimeter of the wind-aided fire is 25 m; for a wind of 14 k m / h ,
when the width is 100 m; for 21 k m / h , 150 m; for very strong winds, about
200 m. There is a tendency for regions of the wind-aided-fire perimeter with
large curvature to evolve to smaller curvature. Fires developing under strong
hot winds increase their head-fire width quickly, and the time needed to attain
the maximum rate of spread for the given fire environment (slope, loading,
wind, etc.) is about 12 minutes (Cheney and Sullivan, 1997). The lesson is that
significant effect of curvature on the quasi steady, one-dimensional spread rate
is confined to a brief interval after ignition, and to a limited locale near the head
under wind-aiding. The effect of curvature in general is insignificant for the
practically interesting temporal and spatial scales of wind-aided fire spread per-
tinent to fire-fighting efforts. Furthermore, letting the fire-propagation function
F - > F [ 1 — j(/<:)], where the function j(/<')(< 1) monotonically decreases as K
decreases, may better express the limited field observations. If the wind ceased,
aside from local effects at a large-curvature head, no inherent propensity toward
a more circular perimeter from an established oval configuration is reported, to
our knowledge; yet this is an implication of adopting the form (F — SK) for
vanishing F. In summary, the important geometric property derived from the
flamefront configuration is the direction of the local normal vector pointing into
the unburned medium. For any anomalous situation (in fire spread through a
bed of fixed surface-fuel elements) in which the firefront curvature were to per-
sist as a significant factor in firefront evolution, the simplistic formulation of
Eq. (1) is probably inadequate to treat the phenomenon. If any such anomaly
arises in an otherwise smooth evolution of the firefront configuration, physical
judgment may well be a better basis for smoothing than any mathematical for-
malism, especially a formalism that creates implausible cusps and corners. With
experience, an automated procedure for firefront propagation may be attained.
Chapter 6 Wind-Aided Fire Spread 187

D. CLOSURE OF THE FORMULATION FOR FIRE

SPREAD IN CHAPARRAL-TYPE FUEL LOADING

The major challenge for implementing the semiempirical approach of Eq. (1)
is to identify the dependence of the firefront-propagation-speed functional F
on the environment-describing variables (for fire-spread-conducive, large-area-
fire-generating conditions). We rewrite Eq. (1), which defines the firefront-
perimeter velocity to be the local normal-to the-front advance rate, as (i = 1,2)

dx
—^ ^ {VX = F{U, ^; m, . . .)Vi, x^e, 0) given (2)
ot

where U denotes the (time-averaged) magnitude of the wind velocity; 0 denotes

the angle, measured counterclockwise, from the (time-averaged) direction of
the ambient wind to the local outward normal to the firefront perimeter, VJ; m
is the mass loading of thin fuel burned with firefront passage; and other param-
eters independent of U and 6 may enter the argument of F. For simplicity, we
deal here with a homogeneous fuel distribution with respect to type, thickness,
height, and the like with a single simple ignition site, and with flat terrain, so
that a value of 6 uniquely defines a point on the firefront perimeter (Figure 6).
Hence, we can dispense with the introduction of the parameter s. An interesting
question is whether, under Eq. (2) and for a finite wind U, there exist translat-
ing, dilating, but geometrically self-similar firefront configurations. To answer
this question, we should not write any approximation to Eq. (2) which involves
F depending on anything other than a scalar function of U and 0. Discussion is
limited to a homogeneous fuel bed of fixed properties and to no slope.
We now composite a very limited number of results for special values of 6,
to constitute a smooth functional F applicable for all values of ^. We seek to for-
mulate F so that, for a given fuel bed, the results from Eq. (2) agree locally with
one-dimensional firefront-propagation rates known to hold for that fuel bed in
each of three directions: the downwind direction (0 = 0), the upwind direction
{0 = TT), and the cross-wind direction [{6 = (TT/I) and (37r/2)]. In fact, these
"one-dimensional" rates are accessible from experiment (if executed) or model-
ing (if credible) but are currently at least partly unavailable. We now undertake
to make plausible statements about spread in each of these three directions.
The peak rate of spread is in the downwind direction (6 = 0) (i.e., near site
A in Figure 5). Accurately quantifying that rate is typically regarded to have
highest priority in characterizing the functional F. In the downwind direction,
conductive-convective heat transfer is the dominant mechanism that leads to
fire spread; turbulent eddy transfer probably plays no important role in that
part of the firefront perimeter. According to laboratory testing for head-on wind-
188 Fendell and Wolff

flank (s = r.n/2)
Ucose

head (s = 0)
rear (s = r^Ti)

• X

wind U
initial firefront
burned area firefront at time At
burned area during interval At
flank (s = r^ 371/2)
unburned area
FIGURE 6 A particular example of the phenomena of Fig. 5, in which the initial perimeter of the
burned area (at a time taken to be t = 0) is a circle of radius r^. A wind of magnitude U arises at
time t = 0+ and results in a head (where spread is with the wind), flanks (where spread is across
the wind), and rear (where spread is against the wind). The fire spread during a time interval At
(stippled area) results in the burned-area perimeter sketched at time At. The x axis, the primary ray
for the angle 6, is taken to lie along the direction of the wind. If the wind changes direction to angle
6i with respect to the x axis, the revised Cartesian coordinates (xf, xj) of the point on the fire pe-
rimeter heretofore denoted (xi, X2) are identified by relations describing a simple rotational trans-
formation of coordinates through angle di\ the x* axis now is ahgned with the wind, serves as the
prime ray of the angle 6 * used in place of 6, and serves as the datum for the parametric distance s*
used in place of s.

aided fire spread across a dry uniform bed on flat terrain, F{U, 0) ~ U^^^/m^^^,
where U is the speed of the head-on wind and m is (recalled to be) the dry thin-
fuel loading (per unit planform area) burned during firefront passage (Wolff
et aL, 1991); this result is shown to be consistent with the just-discussed mecha-
nism for spread-sustaining heat transfer. Henceforth, we designate the func-
tional F(l/, 0) by the symbol Vj. The expression ultimately must be modified be-
cause clearly there is a minimum loading for which fire spread does not occur,
and clearly there may be a much reduced but perhaps still finite rate of spread
as U ^ 0. Nevertheless, the result is pertinent near site A. In Figure 5, we have
taken the wind speed U to be constant in magnitude and direction over the spa-
tial domain sketched, during the time interval ti < t < tj, in accord with our
simplified approach; we defer comment on the time averaging that would be re-
quired for so treating the wind encountered in a fire scenario.
At the flanks of the fire [0 = (rr/I) or (37r/2), e.g., near site B in Figure 5],
the local normal component of the wind is zero, but the speed of propagation
Chapter 6 Wind-Aided Fire Spread 189

is significantly larger than the speed for propagation under quiescent condi-
tions. The firefront propagation is supported by (a) turbulent eddy transfer (lat-
eral gusts) and possibly (b) the same mechanism that permits upwind transfer
{6 = 7T) (i.e., buoyancy-induced influx of fresh air down in the fuel bed) and
somewhat localized radiative transfer among fuel elements in that bed. Much
of the combustion may be of smoldering type. Cheney and Sullivan (1997) sug-
gest that burning at the flanks alternates between heading fire and backing fire.
Henceforth, we designate F(U,|7r/2|) by the symbol /3(U).
At site C in Figure 5 (i.e., at ^ = TT), there is fire spread against the wind.
This upwind propagation is attributed to mechanism (b) discussed in the last
paragraph. The upwind propagation is here taken to be dependent on 17 since
the airflow through the fuel bed competes with mechanism (b). For a fuel bed
composed of discrete combustible fuel elements. Steward (1974, pp. 342, 372)
suggests that the rate of fire spread directly against a wind U involves a balance
of (1) radiative preheating and (2) convective cooling and finite-reaction-rate
effects. Henceforth, we designate F(L/, TT) by the symbol s{U). As a composite
for F motivated by the discussion of phenomena at the three particular direc-
tions, we tentatively suggest (/x ~ 2 - 4 , as a first try):

F(l/, e) = Vj{U cos d) + P{U sin^ 6) in |^| < (TT/I) (3a)

F(U, e) = I3{U sin^ 6) + s{U cos^ d) in |^| > (TT/I) (3b)

We expect, except possibly at very low values of 17, that

s{U) < j8(L/) < Vj{U) (4)

In words, we expect the value of the spread rate F holding at the head (^ — 0)
to exceed the value of the spread rate holding at the flank [6 — (TT/I) or
(37r/2) ], which, in turn, exceeds the value of the spread rate holding at the back
{e = TT).
At any given time, the locus of the firefront is given by

^ = -cot e] = -imO (5)

" ^ 2 / t const. \ 2

This equation is derived from trigonometric relations holding for the tangent to
the firefront.
The firefront-tracking problem is obtained from trigonometric relations
holding for the normal to the firefront, specifically,

— = F cos ^ (6)
dt

—^ = F sin ^ (7)
dt
190 Fendell and Wolff

with F given by Eq. (3), together with the constraint given by Eq. (5) and ini-
tial conditions giving the firefront locus at time t = 0: Xi{6, 0), XjiO, 0). Nu-
merically, Eqs. (6) and (7) may be used to advance the firefront perimeter over
a brief interval of time At. From use of Xi[^,(0), (At)] and XjlOiO), (At)] in
Eq. (5), we obtain ^{At; Xi[^(0), (At)], XjlOiO), (At)]}. The cycle is then re-
peated: we obtain Xi[^( At), 2(At)],X2[^(At), 2(At)] from Eqs. (6) and (7), and
then^{2(At);xi[^(At),2(At)],X2[^(At),2(At)]}fromEq. (5). Of course, more
sophisticated, iterative schemes for advancing the solution in time may also be
used. A smooth interpolation for 6 is needed, especially since the number of
points on the firefront perimeter used to track the perimeter needs to increase
as the perimeter becomes longer.
For explicitness, a possible trial form, consistent with Eqs. (3a) and (3b), is
given by (for JJL = 1)

F(U, 6) = {eo cos^ 0 + c,(lJ cos 6)"^}

+ {EQ sin^ e + al/sin^ 6 exp[-bL;sin^ 6]}, \e\ < {irll) ^^

F(l/, 0) = {SQ sin^ e -^ aU sin^ 6 e x p [ - b U sin^ 6]}

+ {8oCos^^exp[-£iUcos^^]}, 1^1 > (77/2) ^^^
This form for F(U,d) implies that we have adopted the trial forms for the wind-
speed behavior at the head, flanks, and back to be

Vj{U) = eo + ClL;^^^ f3{U) = CQ + a U e x p ( - b U ) ,

e(L/) = EQexip{ — SiU)

respectively. The parameters c^, EQ, e^, a, and h are independent of U and 6, but
in general they depend on m, the mass loading of fuel, and other parameters
characterizing the fuel bed. The parameters are anticipated to be so chosen so
that, except for I/—>0, the ordering is Vj((7) > I3{U) > e(U), in accord with
Eq. (4). Marsden-Smedley and Catchpole (1995) characterize the spread rate at
the flanks and back to be 40 and 10%, respectively, of the spread rate observed
at the head of fires in Tasmanian buttongrass moorlands. Incidentally, Eqs. (8)
and (9) give F = 8o at all ^ for U —> 0: a circular perimeter remains circular as
it expands through a uniform bed on flat terrain in the absence of wind. The
form for Vj stems from the previously mentioned experiments by Wolff et al.
(1991), although 8o = 0 in those experiments; the form for j8, pertinent to the
flanks, takes the variation of spread rate with wind speed to be nonmonotonic;
the form for 8, pertinent to the back of the fire, takes the spread rate to be
monotonically decreasing with increasing wind speed.
In Eq. (10), suppose we have a fuel bed for which: ( l ) 8 o = a = 0 [no spread
in the absence of wind, or at the flanks even with wind (Burrows et ah, 1991)],
and (2) Vj{U) = CiU^^^ H{U - U^nt), where H(x) is the Heaviside unit step func-
Chapter 6 Wind-Aided Fire Spread 191

tion and U^^^f- denotes a finite minimal wind for the onset of spread under
wind aiding (addressed in Section III.F). Then F{U,0) = 0 ior U < 11^^^', for
U > U,,,,,

F(U, 6) = Ci(l/ cos ef'^ if 1^1 < cos'\U,jU)

and F{U,0) = 0 otherwise

According to the model, for a locahzed, pointhke ignition, we obtain a burned

region confined within a wedge, with vertex at the ignition site, and with each
arm of the wedge inchned at an angle cos~^{Ucnt/U) from the downwind direc-
tion. The burned region is only very roughly in the configuration of a sector of
a circle because the firefront advance along the bisector of the wedge (i.e., in
the direction of the wind) exceeds the firefront advance near the arms of the
wedge. Experimental testing is needed to check this conjecture concerning the
relation between (1) the finite minimal wind speed for flame spread onset after
line ignition perpendicular to the wind direction, and (2) the configuration of
the burn pattern emanating from the ignition of a single element in an equiva-
lent fuel bed.
If we now generalize to a case in which the wind direction may vary in space
and time, then we return to the use of the parametric length s of the firefront.
At any time, the length s is measured from the initial ray ^ = 0, which is taken
to be aligned with the wind direction at the moment of time under discussion.
For a circular perimeter of radius r^ at time t = 0 (Figure 6), with the x coor-
dinate fixing the origin of the firefront length parameter s, the initial firefront
is identified by

Xi(s, 0) = Yi COS 0{s, 0) = Vi cos(s/ri), Xjis, 0) = r^ sin(s/ri). (12)

Thenceforth, from Eq. (1),

dx(s, t) ay(s, t)
- ~ - ^ = [cos ^(s, 0 ] F , - ^ ^ = [sin ^(s, t)]F (13)
at at
[dx(s, t)/dsl
tan 0(s, t) = - ) [, (14)
^ ^ [dyis, t)/dsl
where the subscript t signifies that time t is held constant in taking the deriva-
tive with respect to the length parameter s. This particular simple scenario
would arise, for example, if there were an onset, at what we designate as time
t = 0, of a wind U in what previously had been a point-ignited spread through
uniform fuel on flat terrain under nil wind.
192 Fendell and Wolff

E. COORDINATION OF AN
EXPERIMENTAL PROGRAM

The calibration/validation of the model might be initiated against historic data,

but we suspect that any such data set is too incomplete for highly informative
comparison of prediction with observation. Probably, comparison with histori-
cal fires ought to be complemented with comparisons with field tests involving
prior inventorying and instrumentation. These field tests need not be carried
out exclusively on the very large spatial scale that may be of ultimate interest;
the algorithm may be usefully tested against relatively modest-scale fires under
spread-aiding conditions. Indeed, once the model gains a modicum of credi-
bility because its predictions compare favorably with observations, the model
might be advantageously used in test planning. However, it seems doubtful that
field testing would ever be approved for the very high fire-danger conditions
for which a validated model seems most needed. Fire can propagate through
chaparral at speeds up to about 3 m/s, with intensity of 10 MW/m of perimeter
(Albini, 1984), whereas the typical upper limit for prescribed burning is roughly
0.5 MW/m of perimeter (Luke and McArthur, 1978).
Because the model is semiempirical in that information on the functional de-
pendence of the local firefront-propagation speed on the key local state vari-
ables is taken to be input, extensive experimentation is required for model de-
velopment, even prior to model calibration/validation. We have emphasized the
importance of experimental results pertinent to the rate of fire spread at the head
of a wind-aided fire, Vj, in formulating the more general expression pertinent to
the rate of advance of the entire, often-closed firefront perimeter, F. To obtain
an expression for Vj, we emphasize the suitability of quasi one-dimensional fire-
spread experiments in which a quasi steady rate of spread is achieved (and mea-
sured) under uniform (and measured) conditions of slope, spread-aiding wind,
and fuel-bed properties. In this paragraph, we address some considerations in
carrying out such experiments. First, in the firefront frame of reference, a quasi
steady rate of spread implies that a flaming zone of finite depth is propagating
at constant speed, through a bed of sufficient length in the direction of propa-
gation; a burned-over portion of fuel bed lies upwind of the currently burning
zone, and a yet-to-be-burned portion of fuel bed lies downwind. At the very
least, in the direction of firefront propagation, the upwind and downwind por-
tions of the bed should be half as thick as the burning zone. The length of run
to attain quasi steady spread increases with wind speed. Results in which the
streamwise expanse of the flaming zone (i.e., the flame depth) increases with
the length of travel (Anderson and Rothermel, 1965) may not have attained a
quasi steady rate of spread. Also, departure from one-dimensionality is inevi-
tably encountered at the lateral edges of the fuel bed, since downwind elements
Chapter 6 Wind-Aided Fire Spread 193

are preheated from two sides in the core of the bed, but from only one side near
the lateral edges of the layer. Thus, ideally, fuel beds of several different widths
should be tested to establish that the spread rate taken to hold near the stream-
wise centerline of the fuel bed is a plausible approximation to that holding for
an indefinitely wide bed (see Section IV.A). Such bed experiments are most con-
veniently initiated by simultaneously igniting the entire upwind edge of the fuel
bed and tracking the continued progress (if any) of the firefront after the start-
ing transient dies out and a constant rate of propagation is achieved. Further,
it is preferable that experiments to obtain the rate of fire spread be conducted
on the scale of physical interest. However, we repeat for emphasis: control to
maintain constancy of the properties of the wind, fuel bed, and topography be-
comes more problematic in the field, and requirements on the amount and ro-
bustness of instrumentation increase. Because of limited control of conditions
in large-scale experiments, reproduction of results is difficult to attempt, and
even harder to achieve. Yet reproducibility seems essential for confidence that
the key experimental parameters are clearly identified, and the observations of
results are sufficiently accurate. Inevitably, the cost, difficulty, and turnaround
time in experimentation increases as the scale increases, especially for spread
under severe conditions; as a consequence, the desired fire spread data are most
accessible on smaller scale (Carrier et al., 1991; Wolff et al, 1991; Weise, 1993;
Catchpole et al, 1998). The more limited opportunity for testing on larger scale
is seemingly most advantageously used to confirm results already inferred from
the smaller scale test data, or to modify the guidance drawn from the smaller
scale test data owing to the role of phenomena (such as radiative transfer of heat,
or even stratification of the atmosphere) that tend not to contribute prominently
in smaller scale testing.
Though the rate of spread is the primary output for the intended application
to flow-assisted fire spread, other quantities, such as suitably defined "flame
height" and "flame tilt," may also be tabulated as functions of parameters char-
acterizing key prefire properties of the environment (previously categorized as
fuel-bed properties, meteorology, and topography). However, we do not em-
phasize relations among the spread rate, flame height, and flame tilt (e.g., Mon-
gia et al, 1998); such relations describe one unknown quantity in terms of an-
other unknown and, by themselves, make a limited contribution. We do ascribe
a significant role to analysis in assisting the selection of parameter values to be
tested, and in the interpretation of data, especially if the melding of data taken
on different scales to account for different contributing phenomena is entailed.
For example, by modeling, we may be able to suggest tentatively, for validation
or rejection by "one-dimensional fire spread testing," how to modify previously
obtained, semiempirical results for the rate of spread on flat terrain, to incor-
porate the effect of a uniform positive (fire spread-assisting) slope in the direc-
tion of propagation. At the firefront, if we plot the vertical profile of the en-
194 Fendell and Wolff

thalpy (in excess of the ambient enthalpy) for the gas phase above the fuel bed,
we may identify a characteristic height H for the profile. A positive slope of the
terrain, 8, may imply a smaller value for H and, therefore, an enhanced charac-
teristic heat-transfer rate from the hot gas to the portion of the fuel bed under-
going preheating to the pyrolysis-onset temperature (Fons, 1946). In fact, the
enhanced heat-transfer rate owing to slope may be comparable to that for a
higher spread-aiding wind speed U for a flat terrain with the same fuel bed. A
simple theoretical analysis might give a strong indication in what functional
form (whether an additive term or a multiplicative factor), involving which
given parameters (in addition to 6, if any), to seek to summarize the data. Such
a procedure ought to be more efficient than collecting data systematically for a
large number of environment-describing parameters and then seeking formal
correlations. On the other hand, we regard rapid fire spread modeling without
close ties to testing to have little credibility.
Attempts to bypass this demanding agenda and to predict fire-growth rate
across finite-element "parcels" by adopting conjectures about spread-rate
"probabilities" without the aid of an extensive data base [as seen in the appli-
cation of percolation theory to fire spread (Christensen et at, 1993)] seem
equivalent to guessing intuitively. One might just as well dispense with the
probabilistic formalism and directly guess the firefront positions at future times.
Without a basis for assigning probability densities, the probabilistic approach
seems an arid formalism, and one might as well simply guess the answer imme-
diately. Validation seems especially crucial for such an approach. Furthermore,
"tuning" such an approach with data seems likely to be a highly inefficient uti-
lization of the data, relative to the utilization of the data for upgrading more
physically based approaches.

F. CONTRIBUTION OF MODELING
TO DATA CORRELATION

A plethora of functional forms have been proposed for how rapidly the rate
of fire spread Vj increases with the ambient wind speed U (Table 2); the litera-
ture contains even more perplexing results concerning whether the spread rate
increases or decreases with increased thin-fuel loading (e.g., Cheney, 1981;
Weber, 1991; Weber and de Mestre, 1991) (Table 3). For guidance in the in-
terpretation of our own laboratory data for wind-aided fire spread across well-
defined arrays of thin wooden fuel elements (Figure 7), we undertake a simple
model and add complication only as a simpler theory proves inadequate to
describe the observations. As a first trial, evolved from a suggestion by Taylor
(1961), we take the rate of wind-aided spread to be that consistent with the
known rate of entrainment of gas into the weakly buoyant, two-dimensional
plume above a line source of heat (without associated release of mass or mo-
Chapter 6 Wind-Aided Fire Spread 195
TABLE 3 Fire Spread Rate Vj Dependence on Initial Fuel Loading m (vj ~ m^)

p exponent,
Researcher mP Fuel type Scale

Wolff eta/. (1991) -0.49« white pine toothpicks lab

Cheney and Sullivan (1979) Qa grass field
Rothermel (1972) 0^ semi-empirical lab; model
Burrows (1999a, 1999b) Qa jarrah litter lab; field
Marsden-Smedley and
Catchpole (1995) Qa buttongrass field
Luke and McArthur (1991) la grass field

"Data interpretation performed by the researchers themselves.

''Data interpretation performed by Gould (1991).

mentum). If we denote the strength of the hne source (or fire intensity) by the
symbol Q, with units of heat per length (of hne source) per time, then the pro-
posal is that the speed of the cross-wind U and the characteristic speed of the
buoyant updraft in the plume, W, are related by
U = 2aW= 2a[gQ/{ap^c^Tjf^'
(see, e.g., Fleeter et al, 1984). Here a denotes the dimensionless, empirical
entrainment constant introduced by Taylor [a = 0(0.1) for weakly buoyant

tilt angle of flames

A = (7t/2)-Y

weak
buoyant-gas-induced
reverse flow
uniform
airflow

substrate
-smoldering fuel
-rapidly burning fuel
•unburned fuel undergoing preheating
FIGURE 7 A schematic of the phenomenology of wind-aided fire spread across a bed of identical,
upright, regularly arranged, thin, wooden fuel elements, supported in holes in an inert substrate.
Artificial constraints are placed on the influx of air from the downwind side of the flamefront if,
in the laboratory, wind is generated by drawing from downstream, rather than by blowing from
upstream.
196 Fendell and Wolff

plumes, but a might be taken somewhat larger in order to apply the model to
the highly entraining base of the plume]; g denotes gravity; c^, the character-
istic specific heat capacity at constant pressure of the gas (mainly air) in the
plume; p^, the ambient density near ground level; and T^, the ambient tem-
perature near ground level. If la is almost of order unity, we expect that, for
W > (7, a fairly vertical plume arises. However, for 17 > W, the conditions for
our model to hold are not present, and we expect the plume to be more blown
over. The flames then are inclined from vertical and typically extend downwind
over the still-unburned portion of the fuel bed. The significance of the ratio
UV(50Q) (where U is in m/s and Q is in MW/m, and we have assigned the value
a ~ 0.4, as well as plausible values to the other factors) has been discussed for
over a half century in the literature on local discharges in a cross-flow If we
take the speed of propagation of a line fire to be that associated with a steadily
translating line source of heat, we write Q = Qmv^, where Q denotes the effec-
tive chemical heat per mass of fuel burned, m denotes the mass (per unit plan-
form area) of thin fuel burned during firefront passage, and Vj denotes the
speed of fire spread. (For a given intensity Q of a propagating line fire, the quan-
tities Vj and mQ are inversely related. However, the heat release rate per length
of fireline, Q, itself depends on fuel-preheating considerations, and thus pos-
sibly on mQ.) By substitution for Q in the preceding expression for U, after
rearrangement, Vj ~ V^lm—a result well at odds with our laboratory results
(Wolff etal., 1991). We suspect that this first-try model is inadequate because
a model based on weakly buoyant representation of processes in an above-fire
plume omits the diffusive processes occurring in the flaming gas-phase bound-
ary layer and in the fuel bed. However, for those previously discussed fuel beds
for which fire spread is sustained only for wind speed above some finite thresh-
old value (7^^!^ (see the end of Section III.D), we may be able to use the plume-
based model to estimate L^^^if It may be that

^crit ^ ^0i[gQ,J{ap^CpT^)] 1/3

In turn, we identify Q^^j^ by requiring that the flame length of a blown-over
plume be of sufficient length to span the vegetation-bare ground between a
burning hummock and its downwind neighbor, and thus to achieve contact ig-
nition of the downwind neighbor. Typically, if the center-to-center distance be-
tween neighboring hummocks is denoted s, where s^^ d (d being the diame-
ter of a hummock), then, empirically, the flame length L = pQ^, where, for L in
meters and Q in kW/m, p ~ 0.15 and q ~ 0.4 are typical values (e.g., Marsden-
Smedley and Catchpole, 1995). Upon setting L = s to identify Q^^i^, 11^^^ ~ s^'^.
Although the accuracy of the exponent is in doubt, Ucru increases as the dis-
tance between hummocks increases (Burrows et ah, 1991) because the updraft
over a fire of greater intensity must be tilted to effect flame-contact ignition of
the downwind neighbor. For small separation s, a fire of weak intensity suffices
Chapter 6 Wind-Aided Fire Spread 197
to ignite the neighbor, and U^ru is small and may be equal to even zero: molecu-
lar heat-transfer mechanisms then may permit the flames temporarily supported
by a burning element to ignite the neighbor, but such circumstances lie outside
the focus of this discussion.
To relate the spread rate more accurately to wind and loading, in our next
try we ignore gravitational effects. We develop a model in which the oncoming
wind, heated by the gas-phase-diffusion-flame burning just above the already-
pyrolyzing portion of the fuel bed, continues downwind to heat fresh unburned
fuel from ambient temperature to the pyrolysis-onset temperature T^^^ to sus-
tain the firefront propagation (Carrier etal, 1991).
In the frame of reference of a steadily propagating firefront, we take the ori-
gin of coordinates to lie at the furthest-downwind site at which the fuel-bed
surface is at the (known) pyrolysis temperature Tpy^ (Figure 8). In such a frame,
the fuel bed is translating upwind (in what we take to be the negative x direc-
tion) at a constant speed Vj. The ambient wind is flowing in the positive x direc-
tion at the constant known speed (U — Vj) ~ U, for cases of interest. We antici-
pate tentatively that the thin-fuel loading can be characterized adequately for
present purposes by the overall quantity m, recalled to be the mass of thin com-
bustible matter (per unit planform area of the bed) consumed with firefront
passage. The fuel is sufficiently dry that moisture content is not a key consid-

gas phase gas phase

T-^T
U-v.

effective A
cool '
flame position X 3
air ,
Y.'"""T^Tpyr
''' m

• pyrolysis zone preheat zone

FIGURE 8 A simplistic idealization of wind-aided fire spread across a bed of discrete fuel ele-
ments, for the purpose of modeling a quasi steady spread at the rate Vj, typically (for the conditions
of interest here) much slower than the speed (U - Vj) of the ambient temperature oncoming air
stream. The depiction is in the frame of reference of the firefront.
198 Fendell and Wolff

eration, and we tentatively ignore heat loss to inert content of the bed. Insofar
as the portion of the thin fuel loading that burns with firefront passage is un-
known prior to passage, we violate our dictum to relate Vj to parameters known
prior to the fire passage. However, by using the total presence for the consumed
loading m, we obtain a lower bound on the rate Vj (see discussion that follows).
Furthermore, fires in grass and small shrub usually remove virtually all of the
vegetation mass in the flaming combustion zone at the fire perimeter, and there
is little if any smoldering residue after the flaming zone has passed, according
to Albini (1992). More specifically, Minnich (1998, pp. 139, 147) characterizes
chaparral fires as stand-replacement burns that denude most above-ground can-
opy owing to the high continuity and thinness of the fine fuels. There is scarce
carryover fine fuels through burn cycles. On the other hand, Cheney and Sul-
livan (1997) suggest that head fires in Australian grasslands are somewhat less
efficient and leave much of the lower fuel bed unburned. This is in marked con-
trast with prairie fires in the American Midwest, which typically leave very little
unless the fuel is on the ground, and often not even then (P. Zedler, private
communication).
We take all the heat derivable by combustion of the pyrolyzate (fuel vapor
evolved from the polymeric loading m) to be released over the pyrolyzing sur-
face of the fuel bed. "The flames of a wind-driven fire are more erect and often
much taller at the front (i.e., head) than elsewhere on the perimeter" (Albini,
1992, p. 43). We ignore the fact that some of the combustible fuel vapor is not
burned, or is burned (with ambient oxygen) downwind of the pyrolyzate-front
position X = 0. Then, the conservation of energy per time per unit depth (per-
pendicular to the plane of Figure 9) is given by

VjQm = PoCpTjUY (15)

where VjQm is recalled to be the intensity; Q denotes the net heat released by
combustion per mass of vegetation; pQ denotes the density of gas near the flame;
Tj, the adiabatic flame temperature of the pyrolyzate/air diffusion flame; Cp, the
specific heat capacity (at constant pressure) of the gas near the flame; and Y, the
stand-off distance (at the pyrolysis-front position x = 0) from the surface of the
fuel bed (y = 0) of the peak gas-phase temperature Tj. The value Y is an aver-
age, since Y varies on the integral scale of the turbulence. In Eq. (15) we are
equating (1) the heat content per depth per time entering the gas phase from
the pyrolyzing portion of the fuel bed and (2) the heat content (above ambient)
per depth per time of the gas stream crossing the pyrolysis-front plane x = 0.
The datum throughout this analysis is the ambient temperature, taken to be the
same for the air and the bed, for convenience. We are ignoring any heating of
the oncoming air stream (by a warmed substratum) upwind of the fuel-bed-
burnout site, just as we are ignoring any gas-phase velocity-boundary-layer for-
mation upwind of that burnout site.
The downward heat flux (in energy per area per time) from the gas phase
Chapter 6 Wind-Aided Fire Spread 199

u
temperature profile
edge of
- boundary layer ,^^

y'^ pyrolysis
%ylv
V^ 1Y j ^
-'' |4> 1 T I'T • . 1- - ' T
iliiiii;ii=iill!ilii?Jiili^l
^ v ^
-^^f
ground ground

-<— pyrolysis zone -^- ^ preheat zone >-

FIGURE 9 A schematic (supplementary to Fig. 8) in which it is emphasized that the diffusion-
flame burning of pyrolyzate with ambient air occurs within the forced-convection boundary layer
of the adopted model. The characteristic height of the maximum temperature Tj at the onset-of-
pyrolysis front x == 0 is noted, along with a rough conjecture of the entire gas phase temperature
profile at that streamwise position; the temperature approaches its ambient value at the edge of the
boundary layer, and the onset-of-pyrolysis temperature at the fuel bed.

to the fuel bed over the preheating zone oo > x > 0 (Figures 7 and 8) is ex-
pressed by
leT,
-/I (16)

wherefedenotes the thermal conductivity of the gas and the dimensionless func-
tion /(x/Y) ^ 0 as its argument increases. This form implies that a convective-
diffusive mechanism effects the preheating of fresh fuel from ambient tempera-
ture to pyrolysis temperature T^^^. For later reference, were radiation (de Ris,
1979) the mechanism for preheating of fresh downwind fuel, if e denotes the
absorption coefficient of the hot gas and o^, the Stefan-Boltzmann constant, then,
with Y again denoting the thickness of the hot layer,

q- ^^^T]g{^ (17)

where the dimensionless function g{xlY^ —> 0 as its argument increases. This
behavior of g(x/Y) is probably best envisioned for Yfixedand x becoming large.
For the heat balance within the fuel bed for the preheating zone, it is con-
venient (for this paragraph only) to reverse the sense of the coordinate axes,
such that X is positive upwind and y is positive downward into the fuel bed;
200 Fendell and Wolff

X = 0 corresponds to that x at which the value of/(x/Y) is effectively nil. Then,

if subscripts y and | ( = x/Y) denote partial derivatives, and KJ, denotes the bulk
thermal diffusivity of the fuel bed, the temperature (above ambient) in the pre-
heating zone in the fuel bed is described by

K.Tyy - ^ T | = 0 (18)

Any flow-associated transport of heat within the fuel bed is ignored. The
Laplace-transform pair is recalled to be
TOO

h(s) = [exp(-s|)]Kl) dl
(19)
1 rioo+P
h(|)=— [exp(s|)]h(s)ds

where the real number (3 is chosen to be sufficiently large that all singularities
lie to the left of Re(s) = j8 in the complex s plane. [Re(s) denotes the real part
of the complex variable s.] Formally applying the transform to the temperature-
field equation and then solving under the condition of boundedness (for a fuel
bed approximated, for this manipulation only, to be of semiinfinite depth, for
simplicity of expression) yields, if A(s) denotes a function of integration to be
identified,
111
T(y, s) = A(s)exp Y^ (20)
\K,Y/ ^
When q is expressed as a function of x, it is denoted q; when/is expressed as
a function of x/Y, it is denoted / . Application of the boundary condition for
convective-diffusive preheating gives, after use of the Laplace transform,

q(^) = ^ / ( ^ ) = -hT~y (0, s) = h[-^) A{s) (21)

Therefore,

T(0,s)=A(s)=-T,(-l - ^ ill)

At the pyrolysis front, ^ = ^^^^ at the interface y = 0, upon inversion of the

transform,
j^ / ^ \ 1/2 Y rioo+fi J,.
Tvvr = V-Tf( - ^ N, where N ^ ^ - ^ expfs? ) ds (23)
Chapter 6 Wind-Aided Fire Spread 201

that is, N is just a dimensionless positive real number whose value depends on
details (of the heat-transfer profile) that we do not specify (because we believe
that such matters will remain accessible only by experiment for a long time into
the future). If Eq. (16) is solved for Y and substituted in Eq. (23), we obtain, if

u V hpi,c^ J V T^yr J V QmlJ J

The dependence on the square root of the ratio of the so-called conductance for
the gas phase to that for the "solid phase" is conventional in such phenomena;
we regard the product /e^^pj^c^ as a composite property of the bed and do not re-
gard it as appropriate to attempt an approximate evaluation of any one factor
in terms of other quantities that have been introduced. The appearance of the
product Qm indicates that it is the net obtainable exothermicity (per unit plan-
form area of the fuel bed) that is significant for the spread rate.
The presentation of Eq. (24) is nondimensional, but in practice no factor
other than m and U is easily varied appreciably (Albini, 1992). The relation
Vj ~ {U/my^ can hold over only a limited range of parametric values, since we
expect that (1) other processes permit finite (if slow) rates of spread in the ab-
sence of wind, or against the wind; (2) sufficiently high wind could result in
forced-convective extinction of thin fuel elements; and (3) nonpropagation may
occur for either sufficiently sparse loading (because an upwind element may
burn out without preheating its downwind neighbor to its onset-of-pyrolysis
temperature) or sufficiently dense loading (because of oxygen deprivation
within the bed, or radiative heat loss from thick fuel elements). In short, many
finite-rate processes not accounted for in the foregoing continuum-like treat-
ment of the fuel bed may enter, so only experiment can confirm for what (if any)
thin-element loading and ambient wind speed this model holds. Of course, we
would not have included this convective/diffusive-preheating development
without experimental verification (Wolff et al, 1991) that Vj -- {U/mY'^ holds
for a toothpick-type fuel bed with 0.0 < U < 4.6 m/s, 0.11 kg/m^ < m <
0.88 kg/m^ so 0 < (U/m) < 40 m^/(kg s).
Indeed, we just briefly note that the corresponding radiative-preheating de-
velopment, in which we adopt Eq. (17) in place of Eq. (16) and obtain [M is a
dimensionless positive real number with value related to the here-unspecified
details of the profile g(x/Y)]

U \ mQ J \ Tj J MsaT]
The brevity of this discussion stems from our not having observed this re-
sult in our testing. We believe (Carrier et al, 1991) that this result [i.e..
202 Fendell and Wolff

Vj ~ {U/my^] might be observed for an exceptionally heavy fuel loading, say,

m = 0 ( 2 g/cm^).
Williams (1977) notes that the rate of quasi steady, one-dimensional spread
of fire through a combustible medium is often usefully discussed by adopting
a firefront structure in which there is a nonflaming ("preheat") zone, within
which, continually, fresh fuel is warmed from ambient temperature to ignition
temperature:

Pj^Vjrhj = J, where hj = c^T^ (26)

and it is recalled that the datum for temperature is the ambient temperature; J,
with units of heat per area (perpendicular to the direction of propagation) per
time, denotes the flux furnished to the fresh fuel by convection, conduction,
and/or radiation from the flaming region (not encompassed within this exami-
nation, which is limited to a suhdomain of the complete firefront structure). The
ignition temperature Tj may be roughly identified with the onset-of-vigorous-
pyrolysis temperature T^y^ for the vegetation of interest (Albini, 1992). Williams
(1977) points out that this one-dimensional treatment is not readily applied to
scenarios in which a fuel bed yields pyrolyzate (burned in the gas phase above
the bed), as the fuel-bed surface regresses in a direction perpendicular to the
direction of propagation. Indeed, Frandsen (1971) notes the inherent two-
dimensionality of fire spread over a fuel bed (especially wind-aided fire spread)
by emphasizing the need to supplement the heat flux (to fresh fuel) in the di-
rection of propagation with a contribution involving the heat flux (to fresh fuel)
perpendicular to the direction of propagation.
Pyne et a\. (1996, p. 37) write: "Rothermel's fire spread model (1972) is the
basis for most computer-based fire management applications in the United
States, with significant use in other countries. . . . Rothermel's model was de-
veloped from a strong theoretical base in order to make its application as wide
as possible. This base was provided by Frandsen (1971) who applied the con-
servation of energy principle to a unit volume of fuel ahead of an advancing fire
in a homogeneous fuel bed. In his analysis, the fuel-reaction zone is viewed as
fixed and the unit volume moves as a constant depth toward the interface. The
unit volume ignites at the interface. Rate of spread is then a ratio between the
heat received from the source and the heat required for ignition by the poten-
tial fuel. Frandsen's equation . .. contained heat flux terms for which the mecha-
nisms of heat transfer were unknown. To solve the equation, it was necessary
to use experimental [data]." Since Rothermel's model is widely used (Andrews,
1991), we consider the model in detail. We have our reservations about seeking
universality in semiempirical models. Also, we question the citation of Frand-
sen's work as a basis for Rothermel's model, which is in the category of one-
dimensional models typified by Eq. (26); the just-described model [Eqs. ( 1 5 ) -
(25)] explicitly encompasses the role of heat transfer transverse to the direction
chapter 6 Wind-Aided Fire Spread 203

of propagation, and perhaps better includes the processes thought essential by

Frandsen. [Incidentally, the multidimensional treatment of firefront advance
traces to work by Thomas and Simms in 1963, as Frandsen (1971) himself ac-
knowledges.] One-dimensional models for fire spread through wildland vege-
tation were state-of-the-art when discussed by Emmons (1963,1965), but mul-
tidimensional treatments of spread have now been undertaken. Further, in his
one-dimensional modeling, Emmons treated the flaming zone and bed burn-
out, as well as the preheat zone, and adopted an explicit, radiative mechanism
for the preheating of fresh fuel. We are able to discern a treatment of preheat-
ing, but no clear treatment of the rest of the firefront structure, in the Rother-
mel model. Indeed, Albini (1984, p. 594), in discussing the genesis of Rother-
mel's model, writes, "the pivotal step in the development of the model was [an]
inspired conjecture." Specifically, Rothermel takes (Pyne et al, 1996), in terms
of the notation adopted earlier, and with ^ denoting the fraction of the product
mQT that serves to preheat fresh fuel to ignition (so propagation may continue),
J = mQTai + ^s + ^w) (27)
where the factor T is termed a velocity by Rothermel (though it has the units
of a frequency), and (p^ and (p^ are dimensionless empirical expressions giving
the increment to the fire spread rate owing to finite slope and finite wind, re-
spectively. Although a great variety of functional forms can be fitted to a lim-
ited amount of data, we know of no analysis that motivates adopting the form
of Eq. (27) for the quantity I. Whether the model is based around convection
and/or diffusion and/or radiation, as the primary mechanism of preheating of
fresh fuel, is not explicit. We are left uninformed about the basis of taking the
effects of slope and wind on spread rate to be additive and effectively indepen-
dent (Mongia et al, 1998). By inspection, Eq. (27) requires that fire propagate
in a situation without wind or slope, if wind or slope are to augment the spread
rate—a requirement for which counterexamples are readily demonstrated (e.g..
Burrows etal, 1991). In fact, taking (j)^ to be an additive contribution, univer-
sally proportional to a power of the wind speed U, precludes encompassing
cases in which there is an abrupt increase of spread rate above a threshold wind,
then a modest increment with further increase of wind speed (Gill et al., 1995).
Further, in the Rothermel model, the rate F is taken to be primarily an empiri-
cal function of (besides m) the (true) density of the vegetative matter, fuel-bed
depth, the ratio of surface area to volume of fuel particles, fuel mineral content,
and fuel moisture content; fuel availability is related empirically to moisture
content only. But could not elevation of the fuel bed (ambient pressure), diffu-
sional and advective processes within the vegetative matter and in the gaseous
interstices, and chemical kinetics play a comparably important role in the rate
F? Because the forms adopted for F, (f)^, and cf)^ are curve fits which give Vj cor-
rectly for the available data, but for which even the simplistic, primitive moti-
204 Fendell and Wolff

vation that went into Eqs. ( 1 5 ) - ( 2 5 ) is lacking, it would seem crucial that the
model be validated against a wider data base than was used in its formulation.
The model was distributed and in wide use before this validation was performed;
shortcomings of the model and the need for revision have since become appar-
ent (Albini, 1984). Perhaps the spread rate Vj, for use in forming the functional
F in Eq. (1), might be best ascertained by proceeding with both testing and
modeling from scratch, and regarding the Rothermel model as a historic mile-
stone. There are likely to be several other attempts to be abandoned [perhaps
including Eqs. ( 1 5 ) - ( 2 5 ) ] before an adequate statement of the dependence of
the spread rate Vj on prefire parameter values is in hand.

IV. PRELIMINARY TESTING OF THE MODEL

About 200 experiments on wind-aided fire spread across a well-defined, hori-
zontally oriented array of very small diameter ( 1 . 3 - 4 . 4 m m ) , discrete fuel ele-
ments were carried out in the previously described, specially designed and dedi-
cated firetunnel. This experimental setup allowed investigation of factors such
as fuel-bed width, fuel arrangement, inert content, moisture, and substratum.
Of course, it is the thin fuels that are burned exothermically with firefront pas-
sage. These thin fuels are quickly responsive to changes in moisture/heat and
most readily combustible. Post-firefront combustion of some of the thicker fuel
has relatively little bearing on the speed of the firefront advance.

A. FUEL-BED W I D T H

A curved firefront evolves from ignition along a straight line perpendicular to

the direction of the oncoming wind, owing to the fact that, at a given distance
downwind, those fuel elements situated near the lateral edges of a finite-width
bed are less effectively preheated by the approaching, upwind firefront than the
fuel elements situated in the interior of the bed (i.e., further from the edges).
The retardation in the rate of fire spread is manifested further and further from
the lateral edges with increasing downwind distance, so eventually a curvature
characterizes the entire firefront of any finite-width fuel bed (Figure 10). We
have noted that only the component of the oncoming wind that is locally nor-
mal to the firefront aids spread; once the flanks lag, they continue to lag. Insert-
ing structures at the lateral edges of the bed cannot replicate the perfectly non-
catalytic, perfectly slippery, perfectly adiabatic conditions holding for a perfectly
one-dimensional fire spread; ascertaining the consequences of adding physi-
cally realizable constraints at the lateral edges (Catchpole et al, 1998, Figure 3)
is problematic. Investigating the consequences on fire spread rate of increasing
Chapter 6 Wind-Aided Fire Spread 205

Wind

' 1m '
FIGURE 10 A photograph of the firefront curvature for wind-aided fire spread across a bed of
vertical white-pine toothpicks. Here, the fuel loading m = 0.11 kg/m^, the wind speed U = 1.6 m/s,
and the fuel-bed width W = 1.0 m. The site is roughly 2 m downwind from the leading edge of the
fuel bed, where the firefront was initiated along a straight line.

the bed width in the hope of extrapolating the spread rate to the value holding
for an indefinitely wide bed is probably not feasible for most facilities. Further-
more, it may be important to consider that, as tests are executed for succes-
sively wider fuel beds within a test section of fixed width, some oncoming air,
diverted around a "heat island" for a fire spread test with a narrower bed cen-
tered in the test-section, may be constrained to more two-dimensional stream-
ing for a wider bed which approaches the test-section width of the facility. The
upshot is that spread rate may be dependent on the width of the facility as well
as on the width of the fuel bed. The information that seems accessible is the
comparison of how spread rate varies with fuel loading, wind speed, and the
like for a fixed bed width within a given facility.

B. FUEL ARRANGEMENT

To avoid meticulous and tedious inventorying of fuel, one hopes that the fire
spread rate can be characterized by a single parameter, the fuel loading m, as
long as the fuel elements are wooden, "thin" [less than —0.3-0.6 cm in thick-
ness (Cheney, 1981; Albini, 1992), so the elements remain isothermal under
gradual heating], and upright. Then, recording the fuel mass before and after
flaming-firefront passage is simplified, since we have noted that virtually all the
206 Fendell and Wolff

fuel is consumed. Testing with well-defined fuel beds facilitates reproduction

of test conditions to establish error bounds. For toothpicklike fuel elements [of
density p^, (exposed) height H, and characteristic cross-sectional area cf] in-
serted upright into regularly distributed holes (drilled into a substrate), the fuel
loading m is given by [(1 — (/)) is defined as the packing ratio and p^{l — (/)),
the bulk density]

m = pfl{l - 0) = pM^d^) = PsH(Nd^/s^) (28)

The porosity (voidage) (f) is varied to change the fuel loading m; more specifi-
cally, the number of fuel elements per unit area of the bed, n, is varied—where
n = N/s^, with N denoting the number of fuel elements (however distributed)
in an area s^, where the distance s characterizes the square-grid spacing. For
heavier loading, we may insert many toothpicks in a given drilled hole; for
lighter loading, we may leave many holes unoccupied. If we further fix n
for a single-height constant-cross-sectional-area fuel-element type, then only
small-scale variability is possible within a fuel bed of a given loading. The ba-
sic "building block" is the smallest square delineated by four drilled holes; the
square (conveniently taken with one side parallel to the leading edge of the fuel
bed) is meticulously repeated to comprise the entire array for the tests. For ex-
ample, one can obtain an average of one toothpick in each hole by placing two
toothpicks in every other hole. The presence of more than one toothpick per
hole augments the possibility of shading one fuel element from some radiation
owing to the presence of another. For more complicated arrays, with mixed ele-
ments, a larger basic building block is used. While [for fixed wind speed, fixed
fuel-loading parameters (p^, H, n, d^), and fixed bed width] some variability of
the rate of fire spread, especially for lighter loadings, is observed, still, our pre-
liminary testing shows that the rate of spread may he described usefully in terms of
the macroscopic description m. For example. Figure 11 shows the sensitivity of
the rate of fire spread to microscale details, for the case in which there is the
equivalent of two white-pine toothpicks per hole. What little sensitivity there
is, is owing to a slightly faster propagation for those arrangements in which a
downwind fuel element is closer to an upwind element, with both oblique and
in-line considerations of importance. Incidentally, we regard the value of the
fuel density p^ as effectively invariant for cases of interest.
However, because so simplistic a treatment of the fuel bed may be adopted
inappropriately, in the rest of this subsection we dwell at length on circum-
stances that render the one-parameter characterization unsuitable. For example,
we have yet to discuss the consequences of altering H and/or d (within the
constraints of a "thin" fuel); these matters are examined in Section V. Cheney
and Sullivan (1997, p. 25) write, "Fuel load by itself does not influence rate of
spread of grassfires. While previous publications have reported that fuel load
directly influences a fire's rate of spread, we have found that this occurs in grass-
Chapter 6 Wind-Aided Fire Spread 207

lUU

(a)
(b)
— (c)
loo-
(d)

se -
j ^ '

n -
0.0 0.2 0.4 0.6 0.8 1.0 1.2
Xf(m)

gp • ^ — y a—ep-
YYTYT
d) (b cp Q) (b
•wYvf y
(t a di d) (b
cp gp—y gp—}&-

(a) (b)

^ -^ - *
Mt M. M. - ^ JT-
i p Q) qT -* ^
%- :% M. -^
^ % Jf- "^y d) f j) 1p
(c) (d)
FIGURE 11 Effect of the fuel-loading pattern on the rate of firefront propagation Vj for fuel load-
ing m = 0.88 kg/m^, test-bed width W = 0.55 m, and wind speed U = 0.7 m/s. The quantity Xj is
the streamwise centerline position of the firefront (downwind from the leading edge of the fuel
bed), and t is the time since ignition. The hole-center-to-hole-center spacing is 1.25 cm.

lands only if changes in fuel load also reflect changes in fuel condition, in
particular changes in fuel continuity." Of course, all fuel beds of interest in a
wildland-fire context consist of discrete combustible elements; those fuel arrays
described by Cheney and Sullivan (1997) as "continuous" have relatively small
separation distance s: if oxygen is accessible, a finite (but, for our purposes,
typically very low) rate of fire spread may be sustained even in the absence of
wind. Cheney and Sullivan (1997, pp. 24-25) report that, in grasslands typified
by upright stalks (and presumably comparable values of p^), the rate of spread
is largely independent of the values of m (the loading), H (the bed height), and
Nd^ (the fineness of the stalks) but is quite sensitive to the value of s (the "con-
tinuity"); i/all the thin fuel burned with firefront passage, so that m were iden-
tical with the initial fuel loading, then, in view of Eq. (28), these statements are
inconsistent. (See also their Fig. 4.15 on p. 32.) Cheney and Sullivan seem to
suggest that the graininess (patchiness) of a discrete-fuel distribution may be
208 Fendell and Wolff

SO pronounced that the convenient single continuum descriptor m is insuffi-

cient for fire spread rate characterization, and one must revert to description of
the fuel in terms of multiple constituent factors such as H and s, perhaps even
taking account of the variation of properties with height within the bed. Cir-
cumstances in which such detail is essential to fuel-bed description for fire
spread prediction undoubtedly arise. However, the fact that Mediterranean gar-
rigue or Australian mallee may be so discontinuously distributed that sustained
fire spread is likely only under strong surface winds does not in itself ensure
that the single descriptor m is not useful for fire spread rate estimation. For an
example discussed by Cheney and Sullivan, some grassland in the arid center
of Australia consists of combustible dense clumps (spinifex hummocks) and
bare interstices, both of roughly 1-m lateral scale, so the interstices serve as fire-
breaks, except under high wind that abets otherwise precluded fire spread. Per-
haps the closest American analogue is sagebrush, an upright aromatic shrub
with narrow grey leaves. Sagebrush grows to a height of about 0.5-3.5 m on
semidesert plateau between the Rocky Mountains and Pacific Ocean. The patchy
vegetation supports fire spread only under high wind. Another American ana-
logue is the grass tussocks on the northern-Alaska tundra. Characterizing spini-
fex hummock grassland in terms of the single parameter m, even if in general
inadequate for characterizing fire spread rate, may suffice for the special subset
of conditions pertinent to relatively rapid fire spread. In fact. Gill et al. (1995,
pp. 31-32, including Figure 6) state, "When the wind is above the threshold for
spread, the fire may spread at the same rate as in continuous fuel"—so charac-
terization of the surface-fuel bed in terms of m for fire spread rate estimation
may be no more or no less appropriate at any finite rate of spread, independently
of the value of s. Incidentally, Gill et al. (1995, pp. 32-33) discuss the frustra-
tion in using formulae for spread rate inferred from data obtained in one series
of field experiments on hummock grassland to predict the fire spread rate in an-
other series; such experience suggests that a useful contribution may be more
accessible through reproducible testing under well-defined, well-controlled
conditions in the laboratory (Section II). We reiterate that our primary focus is
on fire spread in high-continuity, thermally thin fuels such as the dense, carpet-
like, mature stands of chaparral in southern California for which graininess
(coarseness of the fuel distribution) is often not as prominent a consideration.
Cheney and Sullivan (1997) find that the rate of fire spread is independent
of the initial total thin-fuel loading (prior to firefront passage), but is highly sen-
sitive to (1) the curing state of a grassland, expressed as the fraction of dead ma-
terial in the sward, and (2) the drying out of the dead grasses. Specifically, the
rate of spread is nil for a curing state below 50% and increases monotonically
and significantly with curing state above 50%; the rate of spread is often nil for
fuel-moisture content above 20% of ovendry weight and increases monotoni-
cally and significantly for lower moisture content. If we recall that the parame-
Chapter 6 Wind-Aided Fire Spread 209

ter m refers to the loading of thin fuel burned during firefront passage (Sec-
tion III.D), sometimes termed the "available" thin-fuel loading, then perhaps
we narrow the apparent discrepancy between field and laboratory observations
with respect to the dependency of the rate of spread on the amount of fuel. We
recall (Section lll.F) that Albini (1992) comments that virtually all thin fuel
typically is burned during firefront passage in grass and small shrub; Minnich
(1998) notes that such completeness of burns holds for chaparral in southern
California. Burrows et al. (1991, pp. 196, 198) state that fire spread in spinifex
entails complete burning of all above-ground parts of the plant in the flaming
zone, though they confuse the issue by also discussing flaming hummocks be-
hind the main head fire. However, Cheney and Sullivan (1997) found that fast-
spreading grass fires left appreciable thin fuel unburned in Australian field tests.
Marsden-Smedley and Catchpole (1995) make the same observation concern-
ing fire spread in Tasmanian buttongrass moorlands. Accurate prediction, over
the range of fire spread conditions, of the available (as distinct from the total)
thin-fuel loading is sometimes challenging. Unfortunately, the rate of fire spread
may be related to a quantity which, while a convenient gross descriptor, is not
always known prior to firefront passage.
Laboratory experiments with poorly reproducible fuel beds (Burrows, 1999a,
1999b) result not only in large scatter in the spread-rate data but also in no
clarification of the spread-rate dependence on available fuel.

C. INERT CONTENT

The presence of noncombustible material in the fuel bed (effectively including

higher moisture-content, thicker fuel that does not burn with firefront passage)
may introduce heat-sink and/or wind-retardation effects. When the inert mate-
rial is well-mixed with combustible polymers, an oxygen-deprivation effect may
also arise, so that vigorous burning is inhibited and copious smoke may be pro-
duced. This oxygen-deprivation effect is not readily investigated with the up-
right-fuel-element/upright-inert-element arrangement adopted here for exami-
nation. We consider fuel beds in which finishing nails are regularly distributed
(12.8 kg/m^) (Figure 12), in the midspan only, for three tests (Figure 13) in
which the discrete fuel elements are white-pine toothpicks (0.22 kg/m^). Both
upwind and downwind of the midspan, the white-pine fuel loading is identical
to that of the midspan, but no nails are present. Figures 14 and 15 present the
results of other tests with nails present; the inserts symbolize the loading of the
midspan, with a left-leaning mark denoting a nail-filled hole in the substrate, a
right-leaning mark denoting a toothpick-filled hole, and a small dot with no
mark denoting an empty hole. We define R to be the ratio of the number of nails
to the number of toothpicks.
FIGURE 12 A side view of the midspan (of a pretest fuel bed) consisting of common nails inter-
spersed regularly amid white-pine toothpicks.

Toothpicks Only Nails & Toothpicks, Toothpicks Only

240- (R=0) (R=1) (R=0)

200- Vx = 0.7 cm/s

A J\ Is
J\ J,s J\
J\ A J\
J\ J\ J\
160 -

Vf = 0.8 errl/S
120-
-
80 - u—-
v^ = 1.5 cm/s
40-
1 www

0- r*-^ • l i ^i " 1—''' 1 1 ' \ 1 '

u—-
' \
0.0 1.0 2.0 3.0 4.0
X (m)

FIGURE 13 The firefront position Xj vs. time t for tests conducted with white-pine toothpicks at
a fuel loading m = 0.22 kg/m^, a wind speed U = 2.3 m/s, and a fuel-bed width W = 0.55 m. The
ratio R (the number of nails divided by the number of toothpicks) is unity for the midspan of the
fuel bed and zero elsewhere. The inset depicts the nail-and-toothpick arrangement in the midspan,
where left-leaning lines represent nails and right-leaning lines represent toothpicks.
 Toothpicks Only Nails & Toothpicks Toothpicks Only
240- (R=0) (R=0)

v^ = 2.3 cm/s V| = 1.5 cm/s

(R=0.5) (R=1)
200- JJJJJ JJ
— YTTTTTTT

11 »
160 u ^
^f = 3.5 cm/s (R=0.33) \ \

120- I'j'rn'n
Tj'jj'j'n
y^ • •
\ V/^ . . - •
- 4.9 cnn/s
80- "f
jjjjjjj
(R=0) VA ^ n° °
o
'j-rj'j'j-j'i
'J'J'J'J-J'J'J r:; o o ^
40- 11 ^
u ^
0- ^ i - i U . . • ' 1—'' 1 1 ' — 1 1 — '' 1
—1
0.0 1.0 2.0 3.0 4.0
X (m)

FIGURE 14 Same as Fig. 13, except the ratio R takes on the successive values, 0, 0.33, 0.50, and
1.0 (for the midspan only) for the four cases tested. In the inset, a dot without a line denotes an
empty hole in the substratum.

Toothpicks Only Nails & Toothpicks Toothpicks Only

240 (R=0) (R=1) (R=0)

200 i 1 WWW

U = 3.4 m/s
160

1 UJUl
120 1 www
U = 2.5 m/s

80 H

40

0.0 4.0
X (m)

FIGURE 15 Same as Fig. 13, but with the ratio R = 1 for two tests with different wind speeds,
U = 2.5 m/s and U = 3.4 m/s.
212 Fendell and Wolff

Figure 13 illustrates that the pattern of loading of inert material alters the
fire spread rate, even for fixed total amount of combustible and inert material.
The ratio R in the midspan is unity in all three tests shown, but the rates of fire
spread differ by a factor of two in the midspan. The preheating capacity of a
burning fuel element is strongest on a neighbor located immediately downwind;
if that neighbor is an inert nail, the fire spread is slowed. Therefore, for these
tests, the arrangement of the fuel has an effect on the resulting fire spread rate,
and the fuel bed cannot be characterized by a single loading value, as indicated
earlier for an arrangement of upright combustible elements only.
Figure 14 presents results forR = 0, 0.33, 0.5, and 1, with m and U held con-
stant; Vj decreases monotonically as R increases. Increasing the amount of inert
material in the bed significantly reduces the fire spread rate. The relative im-
portance of (1) the heat-sink effect and (2) the wind-retardation drag (so the
effective wind speed U is reduced), owing to the presence of the nails in the
midspan, is uncertain.
In Figure 15, the spread rate Vj decreases in the midspan (where nails are
present) when the wind speed U is increased from 2.5 to 3.4 m/s, whereas an
increase in wind speed increases the spread rate in the nail-free regions. The
observed behavior in the midsection may be evidence of a so-called finite-
Damkohler-number effect in fire spread across discrete fuel elements, where
the Damkohler number D^ is defined to be the ratio of a characteristic reaction
rate to a characteristic flow rate. For D^ sufficiently small, the chemical reaction
is extinguished ("chemically frozen" flow); if D^ is sufficiently large, the chemi-
cal reaction proceeds at chemical-equilibrium rates; for intermediate values of
Di, transport rates and reaction rates are competitive, and a faster flow implies
a slower rate of chemical reaction. Upwind and downwind, presumably Dj is
sufficiently large for the flow to be in chemical equilibrium; the faster flow is
responsible for a faster rate of spread under rate-of-preheating-controlled con-
siderations. In the midsection, the temperature may be reduced owing to the
presence of a heat sink (the nails), and the chemical-reaction rate typically de-
creases exponentially as the temperature is reduced. Rather than being under
preheating-mechanism control, the spread rate is under reaction-rate control,
and an enhanced wind speed implies a reduced spread rate.
The presence of effectively noncombustible material seems typically not a
major issue in the chaparral context of primary interest.

D. MOISTURE

The ambient moisture content of the white-pine fuel elements used in the test-
ing is roughly 8%. The maximum amount of water which can be absorbed by
the (commercial) white-pine toothpicks is roughly 30% of the dry toothpick
Chapter 6 Wind-Aided Fire Spread 213

240 Moisture above ambient

Test
Fuel Substrate
- « - 82 0% 0%^ v^ = 2.0 cm/s
200 - B - 96 20-30% ~1%* - (Test 96)
+ 113 4-13% ~1%*
o 140 9-15% ~1%*
160 H • 150 ~ 0% 10%
% at surface is much higher
120

80

40 H V| = 2.8 cm/s

4.0
X (m)

FIGURE 16 The streamwise centerhne firefront position Xj as a function of the time t for tests
with 0.55-m-wide beds of 4.6-cm-high white-pine toothpicks with pretest fuel loading m =
0.44 kg/m^, at a wind speed U = 2.5 m/s. The moisture content is expressed as a mass percent
above the initial mass (where the initial mass includes ambient moisture, typically 6 to 8% of the
oven dry mass). The moisture enhancement is achieved by pretest confinement in a saturated
environment.

mass, whereas chamise chaparral may have a ratio of water mass to dry mass of
40 to 50% in midsummer.
The rate of fire spread is substantially reduced when additional moisture
(greater than ambient conditions) is added to the toothpicks or the substrate
(Wolff et al, 1991, Section 3.5; Figures 16 and 17). However, as noted earlier,
our primary interest is the extreme fire spread conditions associated with Santa
Ana wind episodes (after a long warm dry summer) during which the relative
humidity may drop to —10% or lower for days.

E. SUBSTRATUM

The rate of fire spread observed in the laboratory testing appears to be indepen-
dent of the substrate composition, at least for the clay and ceramic materials
tested. The ceramic has density of 425 kg/m^, heat capacity of 1130 J/(kg K),
and thermal conductivity of 0.080 W/(m K) at 447 K and 0.223 W/(m K) at
1225 K. The respective properties of the clay are 1750 kg/m^, 1000 J/(kg K),
214 Fendell and Wolff

IVIoisture above ambient wet substrate wet substrate

Test
240- Fuel Substrate ^ + & wet fuel
- o 0% 0%
200- + 5-7% ~1%* -f-^
• -2% 10% • +
* % at surface is much higher • +
160- • +
+
" + '*' dry
120-
+
-
m ^ a
80- + • °
+ rn °
• + • °
40-
. -h
+ ^ •
D °
•+ n D
- • +3 • ° "
0- ^ ^ 1 1 , , ,
0.0 1.0 2.0 3.0 4.0
x^(m)

FIGURE 17 The same as Fig. 16, but for tests at a fuel loading m = 0.22 kg/m^ and for a wind
speed U = 1.0 m/s. The retardation of the rate of fire spread \j owing to substrate moisture is dis-
tinguished roughly from the retardation owing to the fuel moisture content.

and 0.585 W/(m K) at room temperature. Although the square root of the con-
ductance of the clay is about three times that of the ceramic at 1255 K, in fact
the two inert materials yield about the same fire spread results. However, as just
noted, retention of even residual liquid water by either material can lead to fire
spread rate results distinctly altered from those obtained for the dry substrate.

V. TEST RESULTS FOR THE EFFECT

OF WIND SPEED AND FUEL LOADING
ON THE RATE OF FIRE SPREAD

As noted in Section IIl.D, and in accord with the results obtained from a diffu-
sive-convective-preheating model in Section III.F, our laboratory testing with
arrays of regularly arranged, upright, effectively identical white-pine toothpick-
type discrete fuel elements gives good agreement with the formula (Figure 18)

V. = C{U/m) 111 (29)

The premultiplier C is a function of H, d, inert content, water content, and bed

width, but was not found to be a function of wood species within the limits of
Chapter 6 Wind-Aided Fire Spread 215
10.0-

0.50
8.0 Vf=1.1 (U/m)

E
o

0.0 1.0 2.0 3.0

U/m (m3/s-kg)
FIGURE 18 Quasi steady firefront propagation speed Vj through upright white-pine toothpicks
as a function of the ratio of wind speed U to fuel loading m, from tests with 0 < U ^ 4.6 m/s,
0.11 kg/m^ < m < 1.76 kg/m^ and fuel-bed width W - 0.55 m.

our testing. The obviously finite range of the parameters U and m for which this
formula has been shown to hold was stated in Section III.F.
In Section III.F, we showed that Eq. (29) is consistent with control of fire
spread rate by convective-diffusive heat transfer from hot gas to unburned fuel
elements. The dominance of the heat-transfer mechanism is anticipated to hold
over some range (U/m)^ < (U/m) < {U/m)^, where (l//m)^—>0, and {U/m)^,
was not encountered, for the range of parameters tested (Figure 18). At the up-
per limit, the forced-convective strain out of the burning of thin elements (Fen-
dell, 1965), and/or the inefficiency of preheating downwind elements for too
sparse loading probably intrudes. At the lower limit, the vigor of buoyant ascent,
and/or the inaccessibility of sufficient oxygen for completion of fuel-vapor oxi-
dation (Fendell and Kung, 1993), and/or the inefficiency of radiative preheat-
ing of downwind elements, may intrude, so the spread rate is small or nil.

A. MIXED FUEL ELEMENTS

If, for U = 2.5 m/s and m ~ 0.42 kg/m^, one uses 4.4-mm-diameter elements
to obtain the same fuel loading achieved with 1.3-mm elements, the spacing be-
tween elements is large for the thicker elements, and fire spread is not readily
216 Fendell and Wolff

1.3 & 4.4 mm fuel

240 O O O B O B O c9 A (B
OOO O OO OOO O
O ID O B O e) O B O B

O <S O B d o O B O B
200 H
Vx = 1.6 cm/s
0.41 kg/m2

160

120 H : 2.8 cm/s

80

40 H

0.44 kg/m2

0.0 1.0 2.0 3.0 4.0

x^(m)

FIGURE 19 The streamwise centerline firefront position as a function of time for tests with a
0.55-m-wide bed, composed of 4.6-cm-high elements, at a wind speed U = 2.5 m/s. In the inset,
empty circles designate an unfilled hole in the ceramic substrate, a vertical line denotes a hole occu-
pied by a 4.4-mm-diameter birch dowel, and an inclined line denotes a hole occupied by a 1.3-mm-
diameter white-pine toothpick.

sustained. This suggests that perhaps the thickness for elements to behave as
thin fuel is below 4.4 mm. (The range between 3 and 6 mm is marginal.) Add-
ing roughly 25% by mass of the thinner elements between the thicker elements
facilitates sustained fire spread.
For example, Figure 19 juxtaposes a test involving 1.3-mm (27% by mass)
and 4.4-mm (73% by mass) fuels, with a test involving entirely 1.3-mm-
diameter fuel. The rate of fire spread in the mixed fuel bed is 1.6 cm/s, and in
the thin-element-only bed, 2.8 cm/s. If we adopt the test with only thin fuel as
a baseline and adopt a mass-weighted diameter of 3.6 mm for the mixed-fuel
test, then (Wolff et al, 1991) from the generalized relation (H is recalled to de-
note fuel-element height in the prefire bed)

p-- 2/3 (30)

^r
we obtain Vj = 1.5 cm/s—approximately what is observed. Since the multipli-
cative constant C is found to change significantly as specific fuel-element prop-
erties are altered (with m held constant), we must forsake some of the simplic-
Chapter 6 Wind-Aided Fire Spread 217

ity of Eq. (29). If H and/or d vary from the nominal values used to assign C in
Eq. (29), we must adopt Eq. (30).
Whereas both thick and thin fuel elements burned simultaneously in the
just-discussed mixed-fuel test (Figure 19), in another test, only the thin fuel ele-
ments burned completely and a majority of the thicker fuel mass remained after
firefront passage. The fuel loading was the same in both mixed-bed tests, but
in the incomplete-burn test, all the thin toothpicks were moved to the columns
between the columns with the thick elements. (By definition, a column con-
sists of regularly spaced holes forming a straight line which is parallel to the
oncoming-wind direction.) The wind speed was 2.5 m/s in the complete-burn
test and 4.6 m/s in the incomplete-burn test. This example, in which a fire, in
a bed with multidiameter fuel elements, consumed only the thinner elements
and left the thicker elements virtually intact, mimics the burning of only those
elements less than about 6 mm in thickness during firefront passage in wild-
land vegetation.
A bed of regularly arranged toothpicks is hardly a close surrogate for a chap-
arral stand, especially one that has experienced patch burning. Nevertheless,
with respect to the pertinence of the just-described testing, in chaparral-type
brush, we reiterate that most of the vegetation is consumed with firefront pas-
sage (Albini, 1992), especially for high-intensity fire (Pyne, 1984, p. 117). Fur-
thermore, the range of height in southern California chaparral is limited to
about 1.5-3 m, with 2 m being a typical characterization of mature chaparral.
Also, because of the typical absence of much grazing, harvesting, "thinning," or
patch burning in southern California, over large expanses the loading m in-
creases with the passage of years, without high nonuniformity in the loading or
size at a given time at a given site. Although the appellation (evergreen-shrub-
type) chaparral admits the presence of multiple species, nevertheless large
stands are often fairly uniform.

VI. CONCLUSIONS
On the chaparral-covered hillsides of southern California, autumnal episodes
of persistent strong low-humidity warm winds, after the vegetation is desiccated
during the long dry summer, can result in wind-aided fire spread over large
areas. An annual critical fire season is encountered in other Mediterranean-
climate locales on several continents.
To predict the evolution of the firefront that typically separates burned and
unburned areas of vegetation, one could advance arc segments of the firefront
perimeter. Each segment is to be advanced along its normal, according to the
equilibrium rate of spread holding for the local, instantaneous conditions. This
is to be carried out at the back and flanks of the perimeter, but it seems espe-
218 Fendell and Wolff

cially important to quantify accurately the advance of the head of the wind-
aided fire.
In a plausible, semiempirical approach, the equilibrium rate of spread may be
ascertained by reproducible, fast-turnaround laboratory testing of well-defined
fuel beds in dedicated well-instrumented facilities, along with an integrated
program of field burning. Simultaneously pursuing tractable, analytic, funda-
mentally sound models of the test phenomena may afford insight into the de-
sign of experiments and the interpretation of data.
Although the equilibrium rate of spread may not be fully achieved in the
buildup phase of a firefront evolving from an ignition, for the high-spread-rate
conditions of primary interest here, the fire in minutes typically achieves a size
in which the equilibrium rate is an excellent approximation. Results from well-
defined smaller scale experiments can be used to develop a capability to antici-
pate the evolution of the firefront configuration of a larger scale fire. It is this
larger scale fire not suppressed in its early stage that challenges fire-management
capacity at the urban-wildland interface.
However, by themselves, a limited number of minimally instrumented field
events conducted in heterogeneous, possibly discontinuous, incompletely
characterized fuel beds, which are burned under uncontrollable conditions,
constitute isolated, perhaps unrepeatable anecdotes, not a data base to assist
quantification.

NOTATION

ROMAN SYMBOLS

Cjy specific heat capacity of a fuel bed J kg~^ K~^

Cp specific heat capacity at constant pressure of gas J kg"^ K~^
near the flame
Di (first) Damkohler number
d thickness of a combustible element in a fuel bed m
F firefront propagation speed ms~^
g gravity m s~^
H height of a fuel bed m
hj ignition enthalpy J kg"^
J heat per area per time to preheat fuel (1-D model) J m~^ s~^
k thermal conductivity of gas J m"^ s~^ K"^
ki, thermal conductivity of a fuel bed J m~^ s~^ K~^
Chapter 6 Wind-Aided Fire Spread 219

L flame length m
m thin-fuel loading burned during firefront passage kg m~^
n number of fuel elements per unit area in a regu- m~^
lar arrangement of upright fuel elements
Q effective exothermicity per mass of vegetative fuel J kg~^
burned
Q line-fire intensity J m~^ s"^
Q^rif^ minimal line-fire intensity for fire spread J m~^ s~^
q heat flux from gas phase downward to the pre- J m~^ s~^
heating portion of a fuel bed
R ratio characterizing presence of inert mass to
combustible mass in a fuel bed
s distance between fuel elements in a regular m
arrangement
s distance along a firefront perimeter m
Tj adiabatic flame temperature of pyrolyzate/air K
diffusion flame
Tj ignition temperature K
'^pyr pyrolysis-onset temperature K
T^ ambient temperature K
t time s
U ambient wind speed m s~
U^^jt minimal ambient wind speed for fire spread m s~
(Vj)j rate of change of Xj with time m s"
Vj rate of spread at the head of a wind-aided fire m s"
W updraft in a buoyant plume m s"
Xj coordinates of a point on the firefront perimeter m
a-^ 1,2)
Xi X coordinate of a point on the firefront perimeter m
Xj y coordinate of a point on the firefront perimeter m
X Cartesian coordinate parallel to the ground m
Y stand-off distance (at the pyrolysis front), from m
the surface of the fuel bed, of the peak gas-phase
temperature
y Cartesian coordinate normal to the ground m
220 Fendell and Wolff

GREEK SYMBOLS

a Taylor entrainment constant

j8 rate of spread at the flank of a wind-aided fire m s"^
r a reaction rate in Rothermel's theory s~^
r ( s , t) two-dimensional curve delineating firefront
position
e rate of spread at the back of a wind-aided fire m s~^
8 small parameter describing diffusion m^s~^
e radiative-absorption coefficient m~^
9 angle subtended by the local outward normal to
the firefront perimeter and the wind direction
K local curvature of the firefront perimeter m~^
Kj, effective thermal diffusivity of a fuel bed m^ s"^
^'j components of local outward normal vector to
the firefront perimeter (i = 1,2)
^ fraction of exothermicity that serves to sustain
propagation, in Rothermel's theory
Ps (true) density of vegetation kg m~^
pb bulk density of vegetation kg m~^
Po density of gas near the flame kg m~^
Poo ambient gas density kg m~^
a Stefan-Boltzmann constant Jm~^s"^K"'^
4> porosity (voidage)
(f)^ term to account for the effect of slope on the rate of
head-fire spread, in Rothermel's model
4>^ term to account for the effect of wind on the rate of
head-fire spread, in Rothermel's model

ACKNOWLEDGMENTS
We are grateful to G. F. Carrier of Harvard University for his suggestions, guidance, and en-
couragement throughout our work on fire dynamics. We also thank F. A. Albini of Montana State
University and S.J. Pyne of Arizona State University for many helpful comments on the manuscript.
Chapter 6 Wind-Aided Fire Spread 221

RECOMMENDED READING
Albini, F. A. (1984). Wildland fires. Amer. Scientist 72, 590-597.
Albini, F. A. (1992). Dynamics and modeling of vegetation fires: Observations. In "Fire in the
Environment—The Ecological, Atmosphere, and Climatic Importance of Vegetation Fires"
(P. J. Crutzen and J. G. Goldammer, Eds.), pp. 39-52. John Wiley, New York.
Carrier, G. F., Fendell, F. E., and Wolff, M. F. (1991). Wind-aided firespread across arrays of dis-
crete fuel elements. I. Theory. Combust. Sci. Tech. 75, 31-51.
Glassman, I. (1996). "Combustion," 3rd ed. Academic Press, New York.
Lyons, J. W (1985). "Fire." Scientific American, New York.
Pyne, S.J. (1984). "Introduction to Wildland Fire—Fire Management in the United States." Wiley-
Interscience, New York.
Wolff, M. F., Carrier, G. F., and Fendell, F. E. (1991). Wind-aided firespread across arrays of dis-
crete fuel elements. II. Experiment. Combust. Sci. Tech. 77, 261-289.

REFERENCES

Albini, F. A. (1984). Wildland fires. Amer. Scientist 72, 590-597.

Albini, F. A. (1992). Dynamics and modeling of vegetation fires: observations. In "Fire in the
Environment—The Ecological, Atmosphere, and Climatic Importance of Vegetation Fires"
(P. J. Crutzen and J. G. Goldammer, Eds.), pp. 39-52. John Wiley, New York.
Anderson, D. H., Catchpole, E. A., de Mestre, N. J., and Parkes, T. (1982). Modeling the spread of
grass fires. J. Austral. Math. Soc. (Ser. B) 23, 451-466.
Anderson, H. E., and Rothermel, R. C. (1965). Influence of moisture and wind upon the charac-
teristics of free-burning fires. In "Tenth Symposium (International) on Combustion," pp. 1009-
1019. Combustion Institute, Pittsburgh.
Andrews, P. L. (1991). Use of the Rothermel fire spread model for fire danger rating and fire be-
havior prediction in the United States. In "Conference on Bushfire Modelling and Fire Danger
Rating Systems—Proceedings" (N. P. Cheney and A. M. Gill, Eds.), pp. 1-8. CSIRO Division
of Forestry, Yarralumla.
Burrows, N. D. (1999a). Fire behavior in jarrah forest fuels: 1. Laboratory experiments. CALM-
Science 3(1), 31-56.
Burrows, N. D. (1999b). Fire behavior in jarrah forest fuels: 2. Field experiments. CALMScience
3(1), 57-84.
Burrows, N., Ward, B., and Robinson, A. (1991). Fire behavior in spinifex fuels on the Gibson Des-
ert Nature Reserve, Western Australia. J. Arid Environments 20, 189-204.
Beer, T. (1990). The Australian national bushfire model project. Mathematical and Computer Mod-
eling 13(12), 49-56.
Beer, T. (1991). Bushfire-control decision support systems. Environment International 17,101-110.
Carrier, G., FendeU, F., Chen, K., and Cook, S. (1991). Evaluating a simple model for laminar-
flame-propagation rates. II. Spherical geometry. Combust. Sci. Tech. 79, 229-245.
Carrier, G. F., Fendell, F. E., and Wolff, M. F. (1991). Wind-aided firespread across arrays of dis-
crete fuel elements. I. Theory. Combust. Sci. Tech. 75, 31-51.
Catchpole, W R., Catchpole, E. A., Butler, B. W, Rothermel, R. C , Morris, G. A., and Latham, D.J.
(1998). Rate of spread of free-burning fires in woody fuels in a wind tunnel. Combust. Sci. Tech.
131, 1-37.
Cheney, N. P. (1981). Fire behavior. In "Fire and the Austrahan Biota" (A. M. Gill, R. H. Groves,
and I. R. Noble, Eds.), pp. 151-175. Australian Academy of Science, Canberra.
222 Fendell and Wolff

Cheney, P., and Sullivan, A. (1997). "Grassfires—Fuel, weather and fire behavior." CSIRO,
Collingwood.
Christensen, K., Flyvbjerg, H., and Olami, Z. (1993). Self-organized critical forest-fire model: Mean-
field theory and simulation in 1 to 6 dimensions. Phys. Rev. Let. 71, 2737-2740.
Chorin, A.J. (1980). Flame advection and propagation algorithms. J. Comp. Phys. 35, 1-11.
Coleman, J. R., and Sullivan, A. L. (1996). A real-time computer application for the prediction of
fire spread across the Australian landscape. Simulation 67, 230-240.
de Ris, J. (1979). Fire radiation—a review. In "Seventeenth Symposium (International) on Com-
bustion," pp. 1003-1016. Combustion Institute, Pittsburgh.
Emmons, H. W. (1963). Fire in the forest. Fire Ahstr. Revs. 5, 163-178.
Emmons, H. W. (1965). Fundamental problems of the free burning fire. In "Tenth Symposium (In-
ternational) on Combustion," pp. 951-964. Combustion Institute, Pittsburgh.
Emmons, H. W. (1971). Fluid mechanics and combustion. In "Thirteenth Symposium (Interna-
tional) on Combustion," pp. 1-18. Combustion Institute, Pittsburgh.
Fendell, F. E. (1965). Ignition and extinction in combustion of initially unmixed reactants. J. Fluid
Mech. 21, 281-303.
Fendell, F. E., and Kung, E. Y. (1993). The pyrolyzation of vegetation by brief intense radiation.
J. Thermophys. Heat Transf. 7, 510-516.
Finney, M. A. (1998). FARSITE: fire area simulator—model development and evaluation. Research
Paper RMRS-RP-4. USDA Forest Service, Rocky Mountain Research Station, Fort Collins.
Fleeter, R. D., Fendell, F. E., Cohen, L. M., Gat, N., and Witte, A. B. (1984). Laboratory facility for
wind-aided firespread along a fuel matrix. Combust. Flame 57, 289-311.
Eons, W. L. (1946). Analysis of fire spread in light forest fuels. J. Agric. Res. 72, 93-121.
Frandsen, W. H. (1971). Firespread through porous fuels from the conservation of energy. Com-
bust. Flame 16, 99-16.
Gill, A. M., Burrows, N. D., and Bradstock, R. A. (1995). Fire modeling and fire weather in an Aus-
tralian desert. CALMScience (W. Australian J. Conservation Land Management) 24 (Supple-
ment 4), 29-34.
Goldstein, P. (1995). The wise man of the mountains. Los Angeles Times Mag. (Feb. 12), 20-23,
33-34,36.
Gould, J. S. (1991). Validation of the Rothermel firespread model and related fuel parameters
in grassland fires. In "Conference on Bushfire Modelling and Fire Danger Rating Systems—
Proceedings" (N. P. Cheney and A. M. Gill, Eds.), pp. 51-64. CSIRO Division of Forestry,
Yarralumla.
Knight, I., and Coleman, J. (1993). A fire perimeter expansion algorithm based on Huygens' wave-
let propagation. Int. J. Wildland Fire 3, 73-84.
Luke, R. H., and McArthur, A. G. (1978). "Bushfires in Australia." Australian Government Pub-
lishing Service, Canberra.
Markstein, G. H. (1964). Perturbation analysis of stability and response of plane flame fronts. In
"Nonsteady Flame Propagation" (G. H. Markstein, Ed.), pp. 15-74. Pergamon, New York.
Marsden-Smedley, J. B., and Catchpole, W. R. (1995). Fire behavior modelling in Tasmanian but-
tongrass moorlands. II. Fire behavior. Int. J. Wildland Fire 5, 215-228.
Minnich, R. A. (1998). Landscapes, land-use and fire policy: Where do large fires come from. In
"Large Forest Fires" (J. M. Moreno, Ed.), pp. 133-158. Bookhuys, Leiden.
Mongia, L. M., Pagni, P. J., and Weise, D. R. (1998). "Model Comparisons with Simulated Wildfire
Flame Spread Data." Western States Section Spring Meeting, Paper WSS/CI 985-68. Combus-
tion Institute, Pittsburgh.
Nelson, R. M., and Adkins, C. W. (1988). A dimensionless correlation for the spread of wind-driven
forest fires. Can.]. For. Res. 18, 391-397.
Osher, S., and Sethian, J. A. (1988). Fronts propagating with curvature-dependent speeds: Algo-
rithms based on Hamilton-Jacobi formulations. J. Comput. Phys. 79, 12-49.
Chapter 6 Wind-Aided Fire Spread 223

Pitts, W. M. (1991). Wind effects on fire. Prog. Energy Combust. Set 17, 83-134.
Pyne, S. J. (1982). "Fire in America—A Cultural History of Wildland and Rural Fire." Princeton
University Press, Princeton.
Pyne, S.J. (1984). "Introduction to Wildland Fire—Fire Management in the United States." Wiley-
Interscience, New York.
Pyne, S. J. (1991). "Burning Bush—A Fire History of Australia." Henry Holt, New York.
Pyne, S. J. (1995). "World Fire—The Culture of Fire on Earth." Henry Holt, New York.
Pyne, S. J., Andrews, P. L., and Laven, R. D. (1996). "Introduction to Wildland Fire," 2nd ed. John
Wiley, New York.
Reid, R. (1994). After the fire. Amer. Scientist 82 (1; Jan.-Feb.), 20-21.
Richards, G. D. (1993). The properties of elliptical wildfire growth for time-dependent fuel and
meteorological conditions. Combust. Sci. Tech. 92, 145-171.
Roberts, S. (1989). "A Line Element Algorithm for Curve Flow Problems in the Plane." Center for
Mathematical Analyses Report CMA-R58-89. Australian National University, Canberra.
Rothermel, R. C. (1972). "A Mathematical Model for Predicting Firespread in Wildland Fuels."
Research Paper lNT-115. USDA Forest Service, Intermountain Forest and Range Experiment
Station, Ogden.
Sethian, J. A. (1996). "Level Set Methods—Evolving Interfaces in Geometry, Fluid Mechanics,
Computer Vision, and Materials Science." Cambridge University, Cambridge.
Steward, F. R. (1974). Firespread through a fuel bed. In "Heat Transfer in Fires: Thermophysics/
Social Aspects/Economic Impact" (P. L. Blackshear, Ed.), pp. 315-378. Scripta, Washington.
Steward, F. R., and Tennankore, K. N. (1979). "Fire Spread and Individual Particle Burning Rates For
Uniform Fuel Matrices." Report, Fire Science Center. University of New Brunswick, Fredericton.
Taylor, G. I. (1961). Fire under influence of natural convection. In "International Symposium on
the Use of Models in Fire Research" (W G. Berl, Ed.), Publication 786, pp. 10-28. National
Academy of Sciences—National Research Council, Washington, DC.
Thomas, P. H. (1971). Rates of spread of some wind-driven fires. Forestry 44, 155-175.
Weber, R. O. (1991). Modeling firespread through fuel beds. Prog. Energy Combust. Sci. 17, 67-82.
Weber, R. O., and de Mestre, N. J. (1991). Buoyant convection: a physical process as the basis
for fire modeling. In "Conference on Bushfire Modelling and Fire Danger Rating Systems—
Proceedings" (N. R Cheney and A. M. Gill, Eds.), pp. 43-50. CSIRO Division of Forestry,
Yarralumla.
Weise, D. R. (1993). "Modeling Wind and Slope-Induced Wfldland Fire Behavior." PhD disserta-
tion. University of California, Berkeley.
Williams, F. A. (1977). Mechanisms of firespread. In "Sixteenth Symposium (International) on
Combustion," pp. 1281-1294. Combustion Institute, Pittsburgh.
Wolff, M. E, Carrier, G. F., and Fendefl, F. E. (1991). Wind-aided firespread across arrays of dis-
crete fuel elements. II. Experiment. Combust. Sci. Tech. 77, 261-289.
This Page Intentionally Left Blank
 CHAPTER 7
Fire Plumes
G. N. MERCER AND R . O . W E B E R
School of Mathematics and Statistics, Uniyersity College UNSW,
ADFA Canberra, Australia

I. Introduction
A. Alternative Modeling Approaches
B. Application to Wildland Fires
C. Fire Plume Temperatures
II. Modeling Fire Temperature Maxima
A. Introduction
B. Plume Studies
C. Temperature Measurements
D. A Three-Region Model
E. Fitting Experimental Data
E Time Dependence—Temperature Rise and Fall
G. Total Time-above-Temperature
H. Using Time above AT to Estimate Death
III. Plumes above Eires in a Cross Wind
A. Introduction
B. Mathematical Formulation
C. Wind Profile
D. Temperature-Time Profiles
E. Discussion
References

I. INTRODUCTION
Ecological studies of the impact of v^ildland fires need to be concerned with the
temperatures to which plants will be subjected. Even though many measure-
ments of the temperatures obtained within and above wildland fires have been
made, such as those by Trabaud (1979), to date no one has been able to account
for the full variation of temperature with height. Indeed, sampling heights have
usually been restricted to less than a few meters and only the plume region
above the fire has been modeled. This may suffice for studies of crown death.
Forest Fires
Copyright © 2001 by Academic Press. All rights of reproduction in any form reserved. 225
226 Mercer and Weber

such as those conducted by Van Wagner (1973), but it is insufficient for a full
understanding of fire effects upon vegetation. Knowing the temperature profile
above a fire is important in studying the impact of fire on vegetation. Issues
such as leaf scorch, seed death, and stem death all rely on a complete knowl-
edge of the heat exposure of the vegetation.
What is a plume? Anyone who has observed rising smoke or condensed wa-
ter has seen a plume in action, and anyone considering a fire will immediately
realize that the hot gases from that fire will rise and have an impact upon vege-
tation above the fire (as well as obviously consuming the vegetation in the fire).
In broad terms, a plume is characterized by rising buoyant fluids, the buoyancy
forces arising from intense localized sources of heat (and/or mass). Compre-
hensive textbooks on fluid mechanics, such as Yih (1969), or on heat transfer,
such as Gebhardt (1971), will contain some discussion of plumes, normally
in connection with natural convection above point, line, and plane sources of
heat. The discussion will usually focus on the conservation of mass, momentum,
and energy and the expression of these conserved quantities as mathematical
equations, the Navier-Stokes equations. There are many ways of analyzing the
Navier-Stokes equations, including numerical methods (similar to those em-
ployed for mesoscale meteorological modeling as in Chapter 8 in this book).
Alternatively, it is sometimes possible to derive approximate solutions, which
can be very convenient for application to fire situations, particularly when an
estimate of the temperatures will suffice. One such approximate solution is a
"similarity solution" of the Navier-Stokes equations, where the partial differen-
tial equations are simplified by the use of suitable variables, and then the equa-
tions can be simplified. The validity of using an approximate solution can be
justified by comparison with careful laboratory experiments, which allow one
to determine the conditions for which the theory and the phenomena are well
understood. Although a valuable technique, similarity solutions will not be used
in this chapter, as the mathematical level is more advanced than required for an
understanding of the ecological effects of fire. Another, related approximation
technique is "dimensional analysis," where the processes involved are consid-
ered in terms of the fundamental variables and the dimensions of these vari-
ables. Balancing these dimensions can then lead to valuable relationships for
quantities of interest, particularly observables such as temperature rise. The
most fundamental equation for modeling the ecological effects of fire, Eq. (1),
can be derived from dimensional analysis. (In fact, it also arises as a similarity
solution of the Navier-Stokes equations with appropriate boundary conditions.)
Readers are encouraged to consult the references, such as Yih (1969), for the
details of these mathematical techniques.
Early theoretical and experimental work on plumes was conducted during
World War II and was inspired by the need to disperse fog from airfields, both
in a still atmosphere and in the presence of appreciable wind (see Yih, 1951,
Chapter 7 Fire Plumes 227

1953; Rouse et al, 1952; Thomas, 1963). Out of this early theoretical and experi-
mental work there arose an understanding that plumes arising from sources of
heat only (no mass source at all) could be well described by similarity laws
familiar to engineers and scientists (especially in the area of fluid mechanics).
The success for both laminar and turbulent flows inspired researchers such as
Thomas (1963, 1965) to attempt to use these results to model fire plumes, both
in the laboratory and in the field. Van Wagner (1973, 1975) used this model of
a fire plume to try to predict scorch heights in forests but only had limited suc-
cess as the parameters and functional relationships did not turn out as neatly
as anticipated. There could be many reasons for this, including the inappro-
priateness of a plume model with no mass source for modeling a fire or the
extended three-dimensional nature of a fire as compared to the idealized line
source for the plume model. Whatever the reasons, the landscape fire commu-
nity was sufficiently "buoyed" by the results to continue trying to gain insight
into temperature rise above fires by using relatively simple plume models (e.g.,
Mercer et al, 1994; Weber et al, 1995a,b).
The most significant relationship for turbulent plumes above line sources of
heat (Yih, 1953) can be written as

AT = ^ ^ (1)
z
Here AT is the temperature increase above ambient at a height z above the line
source of intensity I (measured in power output per unit length). The con-
stant le is a proportionality factor, which Van Wagner (1973) determined from
13 outdoor fires in various forests. If units are used such that AT is measured
in degrees Kelvin, z is measured in meters, and I is measured in kilowatts per
meter, then the constant of proportionality, in SI units, is (M. E. Alexander,
1998):
k = 4.47 Km^'^kW-^^^ (2)
Naturally this value is not to be used as a "universal constant." Rather it was de-
termined by fitting 13 data points and should only be used as a guide. Further-
more, the two-thirds power of the intensity will not be strictly accurate for the
plume above a fire. It also is best used as a guide because it originally arose in
a slightly different context and has been adopted for fire plume research in the
spirit of elementary mathematical modeling.

A. ALTERNATIVE MODELING APPROACHES

Mathematical models of fire plumes can be constructed and presented in a va-

riety of ways. Possibly the simplest is to adopt a relationship such as Eq. (1) and
228 Mercer and Weber

fit data sets to determine how well the temperature rise is described by such a
simple relationship. This empirical approach can be statistically intensive, but
it offers the advantage of a well-defined result, hopefully with confidence in-
tervals. For the particular case of Eq. (1), Yih (1953) and Van Wagner (1973),
among others, have shown it to provide a reasonable description of turbulent
plumes, at least in light wind situations.
An alternative approach is to follow Morton et al. (1956); see also Morton
(1965) and Williams (1982). This involves a consideration of the main forces
affecting the dynamics of a plume and allows one to approximate the equations
for conservation of mass, momentum, and energy. The approximate equations
can then be solved, usually numerically, and the results expressed graphically.
This method can be used to good advantage in more complex situations, such
as wind-blown plumes, and will be discussed later in this chapter. The disad-
vantage is that the final model needs to be implemented in computer software,
albeit reasonably simple software, making it a little more cumbersome to use.
The most sophisticated approach is to consider a complete (as far as this is
possible) description of the fluid mechanics of a source of heat (and mass) dis-
charging into the atmosphere. This generally involves solving a set of nonlinear,
coupled, partial differential equations, the Navier-Stokes equations, and pos-
sibly a turbulence model for the fluid motion associated with a fully developed
fire plume. Modern computational facilities are quite able to be used efficiently
and reasonably quickly to solve problems of this nature; however, it needs to be
noted that a great deal of "submodeling" is required and that the end result can
really only be used by experts. By submodeling, we mean that the equations as
solved have important parameters which need to be pre-determined, the bound-
aries of the computational domain need to be treated carefully and appropriate
conditions must be imposed on fluid velocities and temperature at these bound-
aries, the source of heat (and mass) must be prescribed in a realistic manner,
and the turbulence model must be suitable for a large range of spatial scales in-
cluding any vegetation on the surface and the full size of the plume. An ex-
ample of the implementation of such a model is described by Morvan et al
(1998), and further examples of this approach can be found in Chapter 8 in this
book on coupling atmospheric and fire models. Finally, it should be mentioned
that there has been considerable research into plumes in the context of build-
ing or structure fires. Some of this can be accessed from the survey paper by
Baum (1999) and recent papers such as Trelles et al (1999).

B. APPLICATION TO WILDLAND FIRES

Fires in the field occur over a large range of environmental conditions, with
variations in topography, vegetation, and weather. A proper detailed model
would need to be able to account for all of these, at least in principle. Certainly
Chapter 7 Fire Plumes 229

the comprehensive description of the transport of scalar quantities (such as

heat) and vector quantities (such as momentum) will require a full understand-
ing of the fluid flow as it interacts with vegetation. Fundamental studies of this
have been completed; see for example the review by Raupach and Thom (1981)
or the book by Kaimal and Finnigan (1994). However, the subject of airflow
in vegetation has been found to be quite complex, even in the absence of any
strong sources of heat and mass such as a fire. For this reason, we must con-
sider the subject of detailed fluid mechanical computations of fire plumes and
the interaction with vegetation to be in its infancy.
Consequently, for the remainder of this chapter, we will refrain from very de-
tailed modeling of fire plumes. Rather we shall consider the observables, princi-
pally temperature and velocity, and how they are affected by the source strength
(fire intensity) and environmental conditions (principally wind). Additionally,
we shall consider the impact a particular fire event (rather than a fire regime)
can have upon the different sorts of vegetation often encountered in a landscape
mosaic.

C. FIRE PLUME TEMPERATURES

Thermocouple measurements are the primary means for determining temper-

ature rise in and around a fire. Considerable effort has been expended by many
researchers (Stocks and Walker, 1968; Trabaud, 1979; Gifl and Knight, 1991;
Xanthopoulos, 1991) in devising a measurement system which is both robust
enough to withstand repeated field work and sensitive enough to give accurate
temperature readings. The resolution has relied on a compromise for thermo-
couple size, with the need for wires to be reasonably strong and also fairly quick
to register temperature changes. With an accepted instrument and experimen-
tal method, one can then set about the arduous task of gathering temperature
data in the laboratory (e.g., Xanthopoulos, 1991), or the field (e.g., Gill and
Knight, 1991).

II. MODELING FIRE TEMPERATURE MAXIMA

A. INTRODUCTION

In this section we describe a model for the maximum temperature reached as a

function of height as a wildland fire passes. The fuel bed, the flaming region
above the fuel bed, and the fire plume are all included in the model, but sepa-
rate temperature-height functions are used in each of the three regions. Con-
tinuity of the temperature and the heat flux (through the temperature gradient)
across the borders of the regions can then be used to eliminate unnecessary
230 Mercer and Weber

constants. In this way, it is possible to match a constant-temperature region in

the fuel bed with the region of flaming (described by an exponential function
characteristic of a reaction-diffusion process) and the classical turbulent plume
above the fire.

B. PLUME STUDIES

Yih (1953) calculated the temperatures reached in a turbulent plume resulting

from an idealized line source of heat and derived Eq. (1). We note that Eq. (1)
is only applicable directly above the stationary source, x = 0. Elsewhere, the
generalization of this equation, also found by Yih (1953), is

AT = exp(-x^/i8V) (3)

where x is the horizontal distance from the source and (3 is an entrainment con-
stant (/3 = 0.16 according to Lee and Emmons, 1961).
Although, strictly speaking, Yih's (1953) results apply only for a stationary
line source in a quiescent atmosphere, Thomas (1963) and subsequently Van
Wagner (1973) have used Eq. (1) in modeling the temperature rise above wild-
land fires. As one reaches a height far above a real wildland fire, it can be rea-
sonably approximated as a line source, and it appears to move only very slowly.
Thus, Thomas (1963) and Van Wagner (1973) have had some success—Van
Wagner (1973) in particular with providing a first model for crown death.
A laboratory study of stationary pool fires by Kung and Stavrianidis (1982)
provides one of the best experimental tests of the applicability of the dimen-
sional analysis of Yih (1953) to fires. The experimental results show impressive
agreement in the plume region. However, there is a significant region of mea-
sured temperature increases, in and around the flames, where the theoretical
results are inadequate.
Our main motivation is to provide a model for the entire temperature profile
for wildland fires. The two main benefits of this are
i. The ability to predict maximum external temperature rise to which
vegetation would be subjected at any height,
ii. A clarification of the height above a fire at which plume theory can be
reliably applied.
Item (i) depends upon the combustion characteristics of fuel types as well as
the fluid mechanics of the fire, and we can only provide a partial realization of
this here. Item (ii) depends solely upon the fluid mechanics and we show that
our model admits an understanding of the significance of the height at which
the classical plume theory becomes applicable.
Chapter 7 Fire Plumes 231
Other factors which one might hke to include are the movement of the fire,
fire depth, wind profiles, and terrain.

C. TEMPERATURE MEASUREMENTS

The model presented earlier [see Eq. (3)] was fitted to temperature measure-
ments made in a series of experiments in Ku-Ring-Gai National Park near Syd-
ney in New South Wales, Australia. The measurements involved the use of a
single vertical array of sheathed Type K (chromel-alumel) thermocouples to
measure temperature rises above ambient in experimental fires in heathy fuels.
Fires were lit only on days of very light winds. The depth of the fuel varied
greatly, from 0.5 to 2 m, and the flames ranged in height from 1 to 10 m. A dis-
cussion of the utility of such measurements can be found in Gill and Knight
(1991). Typical results are as shown schematically in Figure 1. A detailed re-
port of the experiments is not included here; readers are referred to the detailed
report on the experiments, as opposed to the temperature measurements, that
can be found in Bradstock and Auld (1995).
It was found that many of the experiments gave a similarity of form, despite
quite different fuel depths, which is very encouraging for the development of a

I II III

(D
§
(0

Q.
E

<

Height above ground (z)

FIGURE 1 Typical form of the curve of the temperature rise above ambient versus the the height
above the ground.
232 Mercer and Weber

universal model. This was the original aim of Thomas (1963) and Van Wagner
(1973), founded on the expectation that dimensional reasoning provides rela-
tionships which are scale-invariant. The key problem that arose was the inabil-
ity to determine a universal value for the plume constant [k in Eq. (1)], and
indeed there is no reason to expect that in such a simple model k could be a
universal constant. In the present analysis, we wish to shift the emphasis away
from trying to find universal constants. Rather, it is our expectation that the full
temperature versus height profile can be understood with a three-region model,
as described in the next two sections with the Ku-Ring-Gai fires as a case study.
Furthermore, it will allow a novel comparison of fires and their potential im-
pact upon vegetation.

D. A THREE-REGION MODEL

A typical curve of temperature rise above ambient, AT, versus height, z, can be
divided into three regions, as shown schematically in Figure 1.

I: AT, = K, 0 ^ Z^Zd (4)

II: AT„ = Ke-"("-^'\ Zd^ Z^ Zp (5)
III: AT,,, = C/z, z^ z„. (6)

where K, C, a, Zd, and Zp are constants which need to be determined. In Region I,

it is anticipated that the presence of combusting sohd will create a constant high-
temperature region which extends through a height Zd, perhaps comparable to
the fuel bed depth. In Region II the flames mix with entrained air and an expo-
nential decrease in temperature rise, following a Gaussian distribution, is as-
sumed. Region III is the plume region and extends above a height Zp', hence the
plume equation (1) is used (but with the intensity I absorbed into the constant C).
To reduce the number of constants which need to be determined, and to pro-
vide a smooth AT versus z curve, the temperature rise and the gradient will be
matched across the boundaries between regions:

AT,(2 = Zd) = AT„(2 = Zd) (7)

(AT,) = ^(AT„) (8)

dz
^Tuiz = Zp) = AT„,(z = Zp) (9)

(AT„) = ^(ATn,) (10)

dz
Chapter 7 Fire Plumes 233

There is a little algebra which needs to be done (see Weber et at, 1995b), but then
these conditions determine two of the constants in terms of the other three:

C = KZpe-''^^^-^'^' (11)

IZpiZp - Zd)

Therefore, the model for AT versus z consists of Eqs. (4)-(6) subject to

Eqs. (7)-(10). The following three remaining parameters need to be found:

i. K, the maximum temperature reached anywhere. It may be possible

to estimate this from a combustion calculation, assuming a certain
proportion of total heat generated is lost to the atmosphere,
ii. Zd, perhaps related to fuel bed depth or zone of persistent flame,
iii. Zp, perhaps related to the height of the flames in the zone of flame
flickering.

It is valuable to have these guiding roles for K, Zd, and Zp when one comes to fit
wildland fire data. A detailed survey of AT versus z data from wildland fires
is required to fully justify these guiding roles for K, Zd, and Zp. In this context,
one should note the extreme paucity of published data which uses thermo-
couple array or other temperature measurement means. The most detailed stud-
ies known to the authors are Tunstall et a\. (1976), Van Wagner (1975), and
Williamson and Black (1981), none of which, in their current form, can be com-
pared with our three-zone model.

E. FITTING EXPERIMENTAL DATA

To determine the parameters in the three-region model, seven of the experi-

mental fires are now examined in detail. The three model parameters were first
determined approximately by simply viewing the AT versus z plots of the ex-
perimental data. This instantly provided a smooth curve which gave a good fit
to the data. Further refinement using a simple least-squares routine to minimize
the error was then done. This usually provided a small improvement to the origi-
nal fit. In Figures 2 and 3, we present the results of curve fitting from two of the
experimental fires. For the fire presented in Figure 2, the fuel bed was quite deep
and the flames quite high; hence, we chose K = 700°C, Zd= 1.5 m, and Zp =
2.15 m. For the fire presented in Figure 3, the fuel bed was much shallower,
although the flames were of a similar size; hence, we chose K = 790°C, Zd =
0.25 m, and Zp = 2.00 m.
234 Mercer and Weber

O 800-
r 0
0 0
600-
500-

400-
o\
300-
Q.
Ta = 30°C
E
.0^ 200- zd = 1.5m
zp = 2.15m

0.1 0.2 0.30.4 0.60.8 2.0 3.04.0 6.08.0

Height above ground (m)
FIGURE 2 Log-log plot of the temperature rise for a deep fuel bed.

The ability of the curves to fit the experimental data with a minimum of fuss
is most impressive. It should be stressed that the model is not yet predictive.
Indeed detailed measurements in a given fuel type would be required to cali-
brate the model prior to using it in a predictive sense. Environmental factors
such as wind, humidity, and fuel moisture would also need to be taken into ac-
count. Despite this, even prior to any calibration, the results provide an insight
into where the fire plume region begins. Namely, for the heath communities in
Ku-Ring-Gai, NSW, it would seem that 2 m is the minimum height at which
plume theory can be successfully applied.

o 800- u n n

c
600- OS.
500-

o 40O oV
03
0)
300-
Q. o\
E Ta = 3 0 ° C
.05
200- zd = 0.25m \ 0

zp = 2.00m
d

0.1 0.2 0.30.4 0.60.8 2.0 3.04.0 6.08.0

Height above ground (m)
FIGURE 3 Log-log plot of the temperature rise for a shallow fuel bed.
Chapter 7 Fire Plumes 235

The three regions identified here could match those identified by McCaffrey
(1979) for fires burning natural gas above ceramic plates in the laboratory (viz.,
a continuous-flame region, an intermittent flame region, and a plume region).
In wildland fires, the fuel is solid—unlike the fuel in McCaffrey's fires—so it is
possible that our regions, especially Region I, may be related to fuel-bed char-
acteristics such as depth. Region I is effectively missing in shallow fuels, such
as litter (Stott 1986, author's unpublished data), but readily identified in deeper
fuels, such as shrublands (present work and also Trabaud, 1979) and tall grasses
(Tunstall et al, 1976). Therefore, the situation described here is a more gen-
eral one.

F. TIME D E P E N D E N C E —
TEMPERATURE RISE AND FALL

Not only is the temperature profile in the plume above a fire of interest, but the
temperature-time history is also of critical importance (particularly in regard
to vegetation survival). Assuming that a wildland fire is moving at a rate of
spread U and that it approximates a line source at position x = Ut, then we may
recast Eq. (3) as

AT = - e x p ( - t ^ / i 8 V ) (13)

where ^ = ^lU, A = kf'^, and t = 0 at the time of maximum temperature.

Thus, t is negative prior to the arrival of the fire and positive after the fire has
passed. This equation gives the temperature rise as the fire approaches but is
not valid for the temperature fall.
To determine the temperature fall, note that, from simple Newtonian theory,
the cooling rate of a hot object is related to its temperature elevation above am-
bient by
dT/dt = -y{T - TJ (14)
where y is a constant for a given fuel. Integration of (14) yields the following
expression for the AT as a function of time:
AT = Bexp(-7t) (15)
with the obvious condition that B is equal to the ^T^^ and thus B = AJz nec-
essarily, where A/z was defined for the temperature rise. This should apply ex-
actly to fires where the combustion residence time is reasonably short. From a
fluid-mechanical analysis of the cooling phase, assuming Newtonian cooling, a
value for y of order 0.1 is obtained. This is much greater than values fitted from
236 Mercer and Weber

experimental data (see discussion that follows) and is thus equivalent to much
faster cooling than occurs in practice. The answer to this lies in the fact that
real wildland fires leave a trail of partially combusted fuel, from smoldering ash
through coals to actively flaming logs; hence, the combustion residence time is
not necessarily short. Therefore, y obtained from a least-squares fit to data from
such fires will be an "effective" cooling.
In addition to the experimental fires in Ku-Ring-Gai Chase National Park,
near Sydney, data were also fitted using experiments at the CSIRO Kapalga Ex-
perimental Station in Kakadu National Park, Northern Territory. The Ku-Ring-
Gai fuels were shrublands, varying in depth from 0.5 to about 2 m (Weber et ah,
1995b), while the Kapalga vegetation consisted of mainly long dry grass and in-
termittent shrubs and eucalypts to about 15 m maximum height (Moore et ah,
1995). Data from seven Ku-Ring-Gai fires and 20 Kakadu fires, obtained over a
3-year period, were analyzed. Here we choose to only use results at the 3.5- and
6.0-m level above ground since the 1/^ plume theory is more likely to apply and
the moving fire is more like a line source when viewed from that height. Fig-
ures 4 and 5 show the results of fitting Eqs. (13) and (15) to one of the Ku-Ring-
Gai fires at 3.5 and 6.0 m above ground level, respectively. Both curves were
constrained to pass through A T ^ ^ . Note the rapid rise into maximum (within
60 s) after the fire's onset, followed by a slow fall in temperature over more than
300 s following the maximum. However, at large positive values of time, tem-
peratures tend to an elevated level above those predicted by the Newtonian
cooling model. This is reflected in the fitted values for the cooling parameter y

300
o

0)
!Q
E 200
CO

o
CO

100
CO
CD
Q.

0 100 200 300

Time from maximum temperature (s)
FIGURE 4 Temperature-time curve and data for the Ku-Ring-Gai fire at 3.5 m.
Chapter 7 Fire Plumes 237

300
o

200
o
CO

100

-100 0 100 200 300

Time from maximum temperature (s)
FIGURE 5 Temperature-time curve and data for the Ku-Ring-Gai fire at 6.0 m.

which were 0.013 and 0.009 at the 3.5- and 6.0-m levels, respectively. These are
an order of magnitude smaller than the theoretical Newtonian value of about
0.1. The fluctuations in AT which occur during cool-down probably correspond
to small-scale flaring of fuel elements and also to slow fluid-mechanical phe-
nomena connected with air entrainment. Figures 6 and 7 depict temperatures
against time for a Kakadu fire, for 3.5 m and 6.0 m above ground level. The

100

o
80
0 r +\
E 60

o
03 40
0

20
1 -fct /
0

-100 0 100 200 300

Time from maximum temperature (s)

FIGURE 6 Temperature-time curve and data for the Kakadu fire at 3.5 m,
238 Mercer and Weber

100
O

E
CO

o
CO

Q.
E
.0)

0 100 200
Time from maximum temperature (s)
FIGURE 7 Temperature-time curve and data for the Kakadu fire at 6.0 m.

scaling is quite different from that of the Ku-Ring-Gai fire, yet the same quah-
tative behavior is evident. Again, the model underpredicts the decline because it
is unable to accommodate prolonged smoldering or persistent combustion after
the firefront has passed. The fitted value of y was 0.016 at 3.5 m and 0.013 at
6.0 m. In both fires, the fit around the model is very good.

G. TOTAL TIME-ABOVE-TEMPERATURE

See also Chapter 14 in this book by Dickinson and Johnson. The simple ap-
proach to predicting thermal death of plant materials in fires would be to take
the length of exposure above certain temperatures and compare the results with
those from constant temperature exposure in a furnace or water bath. Given ex-
perimental temperature data or fitted curves such as those in Figures 4 - 7 , the
total time for which a given value of AT is exceeded may be read directly off the
graph. However, with a little algebra, we may express time above AT as a func-
tion of AT as follows: From Eq. (13) the time above AT during which the tem-
perature is rising is
t, = Pz(ln{A/zAT)y (16)
From Eq. (15), the time above AT during which the temperature is falling is
(1/y) ln{A/zAT) (17)
Chapter 7 Fire Plumes 239

"1 1 \ r
50 100 150 200 250 300
Temperature above ambient (°C)
FIGURE 8 Duration of the fire above a given temperature for the Ku-Ring-Gai fire at 3.5 m.

Therefore, the total time above AT is

t,,, = PziHA/z^T))"^ + (l/y)ln(A/2AT) (18)
Fitted values of A, /3, and y were inserted into Eq. (18), and the resulting curves
are shown in Figures 8-11.
It is also possible to obtain an estimate of time above AT directly from the
thermocouple data. This is fraught with problems as it requires an interpola-
tion between data points. However, in order to evaluate our method of fitting
temperature-time curves, it seemed appropriate to compare the two methods

50 100 150 200 250 300

Temperature above ambient (°C)
FIGURE 9 Duration of the fire above a given temperature for the Ku-Ring-Gai fire at 6.0 m.
240 Mercer and Weber

Temperature above ambient (°C)

FIGURE 10 Duration of the fire above a given temperature for the Kakadu fire at 3.5 m.

of finding time above AT. Hence the "data" points on Figures 8-11 exhibit the
same trend as the curve from Eq. (18).

H. USING TIME ABOVE AT TO ESTIMATE DEATH

Laboratory measurements of temperature-time exposures which cause leaf

death usually come from bathing the sample in a constant temperature and
measuring the time till death (see Chapter 14 in this book for more on this
topic). This is not a true representation of the temperature exposure in a fire,
due to many factors including the variability in thermal environment associ-

20 40 60 80 100
Temperature above ambient (°C)

FIGURE 11 Duration of the fire above a given temperature for the Kakadu fire at 6.0 m.
Chapter 7 Fire Plumes 241
ated with wildland fires and tlie different thermal properties of water and air
but, for obvious reasons, is the convenient experiment to perform. It provides
curves hke those in Martin et al. (1969), characterized by the equation
In t^ = a — bT (19)
where t^ is the time to death at an exposure temperature T.
A way in which these laboratory curves might be used together with our
temperature-time relationship [given by Eq. (18)] to predict leaf death and
other fire effects follows. It is the heat flux and the ability of the vegetation to
dissipate heat that governs the temperature rise of a sample. However, the flux
is difficult (if not impossible) to estimate for a given fire, at a given height and
time, even with temperature information. Hence, in the absence of detailed un-
derstanding of the fluid mechanics, we are forced into considering only the
temperature information available to us.
We first notice that the lethal time at a constant temperature, T^gy^^^, will be
more than the lethal time at a varying temperature, T^^^O' where the minimum
is always equal to or greater than T^^^^^. It is then clear that we can perform a di-
rect comparison of the time above a given temperature curve from a fire with
the death curves found in the laboratory. This provides a bound on the effects
of the fire on vegetation. Namely, if the time-temperature curve is ever at a
higher temperature than the death curve, then the vegetation being considered
will perish. However, if the death curve is always above the time-temperature
curve, we cannot be certain of the fate of the vegetation. These possibilities are
shown together in Figure 12.

dead
\ \
\ \^dead

'^'^^<:^::^ death curve

^\unknown

Temperature
FIGURE 12 A typical death curve and some of the possibiUties for the time-temperature curves
and the effect on vegetation.
242 Mercer and Weber

A better method to determine the fate of the vegetation, particularly in this

uncertain zone, consists of the following. Divide the time-temperature curve
into discrete temperature ranges and determine the time spent within a particu-
lar range. The ratio of this calculated time to the laboratory-measured time to
death at a representative time in the same range is then determined. This allows
for both the heating and cooling phases that the vegetation is subjected to. If
we assume additivity of these exposures, then the sum of these ratios will give
an indication of the likelihood of death. Mathematically, this can be expressed
as a "death number," D,
^ t(Ti_i <T <TA

where t{Ti^i < T < T^) is the time spent in the temperature range (Ti_i, Tj)
and t^{Tf) is the laboratory measured time to death of a representative temper-
ature (T*) in the range (Tj_i, Tj). If the sum of these ratios, the death number
(D), is greater than 1, then death is likely, and if the sum is less than 1, then the
vegetation is likely to survive. Of course, for values near 1, the outcome is still
uncertain. The larger the number, the closer together the discrete temperature
ranges considered; hence, the narrower the temperature range, the more accu-
rate this method will be. In the limit as the width of the temperature range tends
to zero, Eq. (20) can be rewritten

which, using the model for the external temperature [Eq. (18)] and the model of
Martin et al. (1969) for the time to death [Eq. (19)], gives the death number as
I3z 1 \ _. I - e'
- dr (22)
2 y -T7
What is needed are the laboratory-measured time to death of the vegetation
(leaves, fruit, stem) for as many different temperatures as possible to determine
t^(T) and the external temperature profile either from experiments or the model
detailed previously. The fires considered here both have a maximum tempera-
ture well above 60°C; hence, leaf scorch will occur at the heights in question.
The effects of fire exposure on the fruits and stems could be determined using
the preceding method. Unfortunately, at present, the data needed to use Eq. (20)
or to fit to the model of Martin et al. (1969) [Eq. (19)] to find t^{T) are not avail-
able in the literature. Mercer et al (1994) have used the model outlined here as
the external temperature input to their model for the temperature exposure of
seeds in woody fruits. Judd and Ashton (1991) have conducted experimental
heating of seeds to consider survival, and Bradstock et al. (1994) have examined
banksia seed survival during wildland fires. All this work is really the prelimi-
Chapter 7 Fire Plumes 243

nary stage of these investigations, and there is a need for mathematical model-
ing and experimental studies to be conducted together to provide reliable meth-
ods for predicting the full impact of wildland fires on vegetation.

III. PLUMES ABOVE FIRES IN A CROSS WIND

A. INTRODUCTION

Here a more detailed two-dimensional model of the plume above a line fire is
presented by considering the balancing of mass, momentum, and energy asso-
ciated with the fire and a cross wind. The case where the fire influences the
wind profile, which is possible for intense fires, is not considered here. For the
sake of this model, the wind profile is an input parameter. This has advantages
because different wind profiles such as over grasslands or in tree canopies are
easily incorporated. Of interest here is how a cross wind affects the plume
above a line fire in both tilting the plume and with increased entrainment into
the plume. For a comprehensive approach where the wind profile and fire are
coupled, see Chapter 8 in this book.
The previous model was based on the assumption that the wind was light
enough that it had little impact on the plume above the fire. Van Wagner (1973)
derived correlations for the scorch height of vegetation above a fire in a light
wind. This work was based on the assumption that the plume is simply tilted
without being distorted and did not take into account the increased entrain-
ment into the plume due to the cross wind. Nevertheless, it was used to try to
predict the scorch height for vegetation above wind-blown wildland fires and
is an interesting example of a relatively simple correlation model. A model of
the wind-blown flame from a line fire was presented by Albini (1981a) who re-
placed entrainment to the plume velocity (this is the no-wind vertical plume
assumption for entrainment) by accretion proportional to the cross-flow veloc-
ity. This work concentrated only on the flame and was not extended to the
plume above the fire. Fleeter et al. (1984) considered Gaussian distributions for
the velocity and temperature above the fire and analyzed the case when the
cross-wind velocity is comparable to the entrainment velocity.
There have been numerous detailed (but similar) models developed to de-
scribe buoyant plumes from point heat sources (often referred to as axisymmet-
ric plumes) in a cross flow (Schatzmann, 1979; Davidson, 1986a,b; Krishna-
murthy and Hall, 1987, to name but a few). In contrast, buoyant plumes above
a line heat source in a cross flow have had considerably less attention. The model
used here is developed in a similar manner to these point heat source (axisym-
metric) plume models which are often used to model the discharge from stacks
but can also be used to model the plume above a pool fire. The models consist
244 Mercer and Weber

of a system of coupled ordinary differential equations derived from the conser-

vation of mass, momentum, and energy of the plume. There are generally two
types of distribution of velocity and temperature (or density) used in these mod-
els: top-hat distributions where the velocity and temperature are considered to
be constant across the plume and have the ambient level outside the plume and
Gaussian where the velocity and temperature are taken to have Gaussian dis-
tributions across the plume. It has been shown by Davidson (1986a) that the
differences between the top-hat and Gaussian approaches for the point heat
source (axisymmetric) plume are small over the range of parameters of practi-
cal interest. The top-hat approach, which requires the solution of a simpler set
of equations, will be used here. If more detail than the mean plume behavior
(obtained from the top-hat model) is required, then it can be determined by
fitting Gaussian distributions from the top-hat results (Davidson, 1986a).

B. MATHEMATICAL FORMULATION

The situation modeled here is an idealization of a blown line fire plume as

shown in Figure 13. The plume is considered to be initially rising vertically with
velocity Wj, half width bj, temperature Tj, and density pj. These initial condi-
tions are taken to be far enough above the flames so that the detailed structure
of the flame is not required. Values for these initial conditions must be obtained
from measurements or from flame models. The ambient atmosphere is at T^ and
has density p^ and a height-dependent wind velocity Ua{z). Mass entrainment

Wi, b| T, Pi

FIGURE 13 A schematic of the plume and cross wind showing the nomenclature used.
Chapter 7 Fire Plumes 245

of the ambient air into the plume is represented by the hnear sum of the hori-
zontal and vertical components. That is, for the plume that is at an angled to
the flow, the entrainment speed, v^, is given by
V, = a(w - Ua cos 8) + pUa sin d (23)
where w is the velocity of the plume. For the case of the axisymmetric plume,
values for a and /3 are documented with a ~ 0.1 and /3 ~ 0.5 (Krishnamurthy
and Hall, 1987). For the line plume, there are less data available in the litera-
ture for the value of these parameters, and further measurements are needed to
generate reliable estimates for them. Throughout this paper, the value of a will
be taken as 0.16 as in Lee and Emmons (1961), and 13 will have the value 0.5.
In a similar manner to that described in Davidson (1986a) for the axisym-
metric plume, the governing equations are derived from balancing thefluxesof
momentum and energy. Conservation of mass gives
d{pbw)
dT^ = PaVe (24)

where s is the arc length along the plume centerline and p^ and b are the den-
sity and half width, respectively. Due to the top-hat assumption, the density
(temperature) and velocity are assumed constant across the width. Conserva-
tion of s-momentum gives
d(Ppbw^)
^ = Pa^e^a COS 6 - b{Pp - Pa)g siu 6 (25)

Conservation of r-momentum (r is the direction normal to s) gives an equation

for the trajectory angle
de _ Pa^e^a Sm 0 + b{pp ' Pa)g COS 6

ds Ppbv\^'^
The conservation of thermal energy gives

1 = Pa^e^a (27)
ds
where T^ is the plume temperature. Rearranging (27) by using (24) gives
= -ppbM/ s m 6—- (28)
ds ^ dz
where AT = Tp — T^ is the difference in the temperature of the plume above
the ambient air and dTjdz is the measure of the temperature stratification in
the ambient atmosphere which, for the purposes of this model, can be taken as
246 Mercer and Weber

zero; that is, here we are considering a neutral atmosphere. This is easily altered
if a stratified or unstable atmosphere is to be considered. The ideal gas law gives

Pp = Pay (29)

and the plume trajectory is defined by

^ = cos ^ (30)
as
dz
- ^ = sin^ (31)
as
The model then consists of solving the seven equations ( 2 4 ) - ( 2 6 ) and ( 2 8 ) -
(31) for the seven unknowns b, w, T^, p^, ^, x, and z as functions of distance
along the plume trajectory subject to the initial conditions
b = bj, w = Wj, T = Tj, p - Pi,
(32)
e = 77/2, X = 0, z = 0
Introducing the change of variables

yi = pp^^. ji = Pp^^^. y-i = 0.

y^ - PphwAT, y5 = ^. y6 = Z
and using Eq. (29), the system of equations can be written

dyi
^ = '^'^ (34)

dyi ,, , yij4 ^ s i n y 3
(35)
ds yi la
dy3 _ pqVe^a siny3 ^ y^y^ P c o s y 3
(36)
ds yi y\ Ta
dy^
(37)

— = 0033.3 (38)

dy6
— = smy3 (39)

In this form, it is a straightforward matter to solve this system of six coupled

first-order ordinary differential equations subject to the initial conditions given
Chapter 7 Fire Plumes 247

by Eq. (32) in the appropriate new variables given by Eq. (33). Here the so-
lutions were determined using the NAG (1991) routine D02BBF which inte-
grates a system of first-order differential equations using a Runge-Kutta-Merson
method.
To obtain the simplified version for the no-wind case, it is a simple matter of
setting Ua = 0 and y^ = 6 = TT/I in the preceding equations and noting that
Eqs. (36) and (38) are no longer relevant because there is no plume angle or
horizontal distance to consider. The arc length variable s also then equates to
the height above the source that is ^ = s, and Eq. (39) is also not needed.

C. W I N D PROFILE

For the equations of the model to be solved, the wind profile Ua{z) must be speci-
fied. One of the strengths of this model is that any horizontal wind profile can be
used and hence a variety of conditions can be modeled. For example, the plumes
above grassland fires or fires above and below canopies, all of which have differ-
ent velocity profiles, can each be modeled. In this chapter, no attempt is made
to model the impact of the fire upon the prevailing wind field. Near the fire, this
is unrealistic because there can be substantial effects upon the wind field due to
the fire. However, a much more detailed model that takes into account all of the
fluid mechanics and possibly also the combustion properties of the fire would
be needed to accurately predict the effect of the fire on the velocity profile (the
fluid mechanics is included in the discussion in Chapter 8 in this book).
As a first (somewhat unrealistic) approximation, we consider the wind to be
constant with height. This is the situation assumed by Van Wagner (1973) and
Fleeter et al. (1984). By using a similarity analysis, it has been shown for the
uniform wind case that the plume is self-similar (Raupach, 1990); hence, away
from the source, the midline trajectory is a straight line as a function of down-
wind distance, that is z oc x. This agrees with the experimental work of Rama-
prian and Haniu (1989) for two-dimensional plumes in a cross-flow Shown in
Figures 14a and 14b are plots of the plume midline trajectory and width for the
initial conditions h^ = 0.2 m, w^ = 1.0 m/s, T^ = 900 K with T^ - 300 K for
U^ = 1, 3 m/s, respectively. Clearly, away from the source, the trajectory is a
straight line as expected from theory and experimental results. The effect of in-
creasing the wind strength is also evident. With increasing wind strength, the
plume is blown over more and hence temperatures at a given height above the
heat (fire) are lower as the velocity increases. Therefore, for a fire with a given
fixed heat release rate that is independent of the wind strength, vegetation above
the fire is exposed to higher temperatures than in a higher wind case. Of course,
this does not take into account the effect the wind can have on the heat release
rate of the fire.
248 Mercer and Weber

(a) U3 = 1 m/s (b) U3 = 3 m/s

30-

20-

10-

0- ' 1 1 1 1 1 '
0 10 20 30 40 0 10 20 30 40
x/b i x/b i

FIGURE 14 Plume trajectory and width for a constant wind.

Two different and more realistic wind profiles are more relevant to the pres-
ent model. The first is a logarithmic profile as used by Albini (1983), namely,
z - 0.64V\
In
o.i3y /
Uaiz) = 1/H (40)
H^a64V\
0.13V /
where V is the vegetation height and l/^ is the velocity measured at a height H.
This velocity profile is applicable above a vegetation cover of trees or shrubs.
More applicable to grassland or bare-ground is a power law profile

Z±_ZoY (41)
Uaiz) = UH
H
where n is usually taken as 1/7 (Albini, 1981b, 1983). Here ZQ is the base of the
plume and z is measured from this point. Figures 15a and 15b are plots of the
plume midline and width for the initial conditions as for Figure 14 but with a

(a) UH = 1 m/s (b) U„ = 3 m/s

30-

20-

10-

0- n 1 1 1 1 '
0 10 20 30 40
x/b,
FIGURE 15 Plume trajectory and width for a power law wind profile with l/^ given at H = 10 m
and the plume starting aiZo — 1 m.
C h a p t e r ! Fire Plumes 249

power law velocity profile with 11^= 1,3 m/s at H = 10 m. The differences
between this more realistic profile and the uniform velocity profile considered
earlier are evident comparing Figures 14 and 15. The plume for the power law
velocity profile is more upright close to the ground as is expected from the
lower wind velocities in this region when compared to the uniform velocity
case. Far enough above the height where the initial conditions are given, the
midline trajectory is approximately linear with downwind distance since the
wind velocity is only slowly varying far enough above the source. The major
difference between the results for the two different wind profiles is the height
to which the plume extends. With a more realistic power law velocity profile,
the plume is substantially higher than with a uniform velocity profile. This re-
sults in a significant difference when modeling the impact of the fire on vegeta-
tion above it and should therefore be taken into account in any such models.

D. TEMPERATURE-TIME PROFILES

Before the effect of the fire plume on vegetation can be determined, the tem-
perature-time profiles at various heights above the fire must be calculated.
These temperature-time profiles can then be used as inputs into models for the
effect of temperature exposures on vegetation (Mercer et al, 1994). In the pres-
ent model, the fire is assumed stationary; hence, temperature versus downwind
distance profiles can be determined. To convert from this to a temperature-time
profile, an estimate must made of the rate of spread of the fire. The distance
from the stationary fire can be related to time exposure due to a moving fire. In
many circumstances, the rate of spread of fire is small compared to the wind
speed; hence, the fact that the fire is moving has little effect on the results pre-
sented. In the case when the rate of spread is not small compared to the wind
speed, the velocity profile should be altered so that it is relative to the moving
fire; that is, the frame of reference should be with the moving fire. In the no-
wind case, this is similar to having a wind in the opposite direction to that of
the moving fire.
As described in Davidson (1986a), Gaussian distributions can be fitted to
the top-hat results of this model to give the distribution within the plume as

To = T, + J ( T , - Tje-(^'^^^^) (42)

where r is the distance normal to the centerline trajectory, N is the plume

edge criterion which is usually taken to be 2, and A^ is the spreading ratio of
mass and heat to momentum which here is taken to be 1 (Davidson, 1986a).
Figures 16a and 16b show the temperature versus downwind distance for the
250 Mercer and Weber

(a) UH = 1 m/s ( b ) UH = 3 m/s

2.5
9 200 O 200
d. Q.
E Is.o E
•" 100 100
\\\io.o
0 10 20 30 40 0 10 20 30 40
x/b, x/b;
FIGURE 16 Temperature vs. nondimensionalized downwind distance (x/b^) for the plumes given
in Figure 15 at nondimensional heights of zlhi = 2.5, 5.0, and 10.0.

plumes given in Figures 15a and 15b at three nondimensional heights z/h^ = 2.5,
5.0, 10.0. With increasing wind strength, the maximum temperature reached
at any height decreases, and the profiles become more skewed as the plumes are
bent over farther. For the lightest wind considered, U^ = 1 m/s, the plume is
not far from symmetrical (the no-wind case), whereas for l/^ ~ ^ i^/s, there is
considerable skewing of the plume downwind.

E. DISCUSSION

This model can be used to predict the temperature-time curves above line fires
and hence impact on vegetation above the fire. Certain inputs are needed be-
fore the model can successfully be used. The ambient wind velocity profile is
needed. This can either be from velocity profiles outlined earlier (or similar an-
alytic profiles) or from measured or predicted velocity profiles depending on
the terrain and vegetation involved. Estimates for the initial plume conditions
bj, Wj, and Tj are also required. These can be obtained from either measure-
ments of experimental fires or from models such as the one by Albini (1981a).
These will depend on the fire and hence on many parameters such as the fuel
type, the loading and moisture content, the wind velocity, the ambient temper-
ature, and the relative humidity to name but a few. Indeed the initial conditions
can be related to the heat of the fire

CpV^iPiC^i - Ta) dx (43)

which simplifies to

Q = 2CpW,p,AT,b, (44)
Chapter 7 Fire Plumes 251

where ATJ == Tj — T^ is the difference in temperature of the plume from the air.
So if an estimate for the heat release rate of the fire is known, then only two of
Wj, Tj, or hi are needed; the third can be estimated from Eq. (44). The most
difficult of these initial conditions to determine is Wj, the initial velocity in the
plume. The initial plume and temperature are easier quantities to measure ex-
perimentally or theoretically from fire models.
The present relatively simple model can be used as an aid in predicting fire
impact upon vegetation in many situations. A common reason for using pre-
scribed fire in some parts of the world are fuel reduction burns used to mini-
mize the intensity of wildfires. These have an impact on the surviving vegeta-
tion that needs to be estimated before the burns are conducted. Typically fuel
reduction burns are conducted in light winds as an aid to their control. They
therefore have a larger impact on the vegetation above the fire than a fire of simi-
lar intensity would in a stronger wind where the plume is bent farther and does
not impinge on the vegetation to such a degree. Of course, the effect of the wind
on the fire should also be taken into account. This model can then be used to
determine temperature impacts such as the height of scorch of the leaves and
potential death of canopy stored seed [in conjunction with models like the one
in Mercer et al. (1994)].
Since this is only a plume model, no attempt has been made to model the
flame of the fire. Models such as the one presented in Albini (1981a) should be
used to obtain details of the flaming zone. The present model takes no account
of small-scale phenomena such as flare ups; hence, its use is confined to broad
areas rather than small areas such as an individual fuel element. Phenomena
such as smoldering of fuel after the main fire has passed and the cooling of
vegetation and the ground, all of which can possibly increase temperature ex-
posure of the vegetation above the fire, are not considered.
On a larger scale than that considered here, this model can also be incorpo-
rated into a model for spotting ahead of a line fire. In this case, the assump-
tion that the atmosphere is neutral is not justifiable; hence, the dTjdz term in
Eq. (28) should not be taken as zero. An estimate for this temperature strati-
fication in the atmosphere would be needed which would depend upon the
atmospheric conditions prevailing at the time. The velocity within the plume
has been calculated, and this can be used to determine the height that burning
brands will reach before they fall back and possibly ignite a fire downwind of
the main fire front. The model presented here differs from previous models
of firebrand trajectories in two ways. First, the present model includes the en-
trainment into the plume due to the cross wind. This is in contrast to existing
models for firebrand trajectories (Lee and Hellman, 1969, 1970; Albini, 1979,
1981b, 1983) which use the no-wind assumption that the entrainment velocity
is proportional to the vertical velocity. This will overpredict the height to which
the firebrand will rise since the plume rises higher in the no-wind case than in
252 Mercer and Weber

the case with wind present. It is more reahstic to include the entrainment as
presented in this model in determining the firebrand's actual trajectory. Second,
in the present work, a line fire is considered as opposed to previous work by
Lee and Hellman (1969) and Albini (1979, 1981b) who considered a spot fire
(axisymmetric source) and Albini (1983) who considered a buoyant line ther-
mal as a right elliptical cylinder. The line source is appropriate where the spot-
ting distance is relatively short so that, on the scale of the distance the firebrand
has traveled, the fire appears as a line source. For spotting a large distance from
the fire, a point source is accurate enough because at a large distance from the
fire structure (spot fire or line fire) is less relevant.

NOTATION

ROMAN LETTERS

b plume half width m

c constant mK
Cp specific heat J kg-^ K-i
9 gravitational acceleration m s~^
H height of wind velocity measurement m
I line fire intensity kWm"^
k constant
K constant K
N plume edge criterion dimensionless
n power law parameter dimensionless
Q heat release W
r coordinate normal to s m
s distance along plume m
AT temperature rise above ambient K
T temperature K
Ua(z) ambient wind velocity m s~^
V vegetation height m
Ve entrainment speed m s~^
W plume velocity m s~^
X horizontal distance m
z vertical distance from plume base m
Chapter 7 Fire Plumes 253

Zo height of the plume base m

Zd height related to the fuel bed depth m
Zp height related to the start of the plume m

GREEK LETTERS

a entrainment coefficient m~^

j8 entrainment coefficient dimensionless
A change from ambient
A^ spreading ratio of mass and heat versus momentum
6 plume trajectory angle from the horizontal
-3
p density kg m

SUBSCRIPTS

a ambient property
i initial property
p plume property
G Gaussian

REFERENCES
Albini, F. A. (1979). "Spot Fire Distance from Burning Trees—A Predictive Model." United States
Department of Agriculture, Forest Service, General Technical Report INT 56. Intermountain
Forest and Range Experimental Station, Ogden.
Albini, F. A. (1981a). A model for the wind-blown flame from a line fire. Combustion and Flame 43,
155-174.
Albini, F. A. (1981b). "Spot Fire Distance from Isolated Sources—Extensions of a Predictive
Model." USDA Forest Service Research Note INT 309. Intermountain Forest and Range Experi-
mental Station, Ogden.
Albini, F. A. (1983). Transport of fire-brands by line thermals. Combust. Sci. Tech. 32, 277-288.
Alexander, M. E. (1998). "Crown Fire Thresholds in Exotic Pine Plantations of Australia." PhD
thesis. Australian National University.
Baum, H. R. (1999). "Large Eddy Simulations of Fires—From Concepts to Computations." Soci-
ety of Fire Protection Engineers, 1999 Arthur B. Guise Medal Lecture.
Bradstock, R. A., and Auld, T. D. (1995). Soil temperatures during experimental bushfires in re-
lation to fire intensity: Consequences for legume germination and fire management in south-
eastern Australia. J. Appl. Ecol. 32, 76-84.
Bradstock, R. A., Gill, A. M., Hastings, S. M., and Moore, P. H. R. (1994). Survival of serotinous
254 Mercer and Weber

seedbanks during bushfires: Comparative studies of Hakea species from south-eastern Aus-
tralia. Aust.J. Ecol 19, 276-282.
Davidson, G. A. (1986a). Gaussian versus top-hat profile assumptions in integral plume models.
Atmospheric Environment 20, 471-478.
Davidson, G. A. (1986b). A discussion of Schatzmann's integral plume model from a control vol-
ume viewpoint. J. Climate Appl. Meteorol. 25, 858-867.
Fleeter, R. D., Fendell, F. E., Cohen, L. M., Gat, N., and White, A. B. (1984). Laboratory facihty for
wind-aided firespread along a fuel matrix. Combustion and Flame 57, 289-311.
Gebhardt, B. (1971). "Heat Transfer." McGraw-Hill, New York.
Gill, A. M., and Knight, I. K. (1991). Fire Measurement. In Conference on Bushfire Modeling and
Fire Danger Rating Systems (N. R Cheney and A. M. Gill, Eds.), pp. 137-146. CSIRO Division
of Forestry, Canberra.
Judd T. S., and Ashton D. H. (1991). Fruit clustering in the Myrtaceae: Seed survival in capsules
subjected to experimental heating. Aust.J. Bot. 39, 241-245.
Kaimal, J. C , and Finnigan, J. J. (1994). "Atmospheric Boundary Layer Flows: Their Structure and
Measurement." Oxford University Press, Oxford, UK.
Krishnamurthy, R., and Hall, J. G. (1987). Numerical and approximate solutions for plume rise.
Atmospheric Environment 21, 2083-2089.
Kung, H-C, and Stavrianidis, P. (1982). Buoyant plumes of large-scale pool fires. In "19th Sympo-
sium on combustion," pp. 905-912. The Combustion Institute (Pittsburgh).
Lee, S-L., and Emmons, H. W (1961). A Study of natural convection above a line fire. J. Fluid Me-
chanics 11,353-369.
Lee, S-L., and Hellman, J. M. (1969). Study of fire-brand trajectories in a turbulent swirling natu-
ral convection plume. Combustion and Flame 13, 645-655.
Lee, S-L., and Hellman, J. M. (1970). Fire-brand trajectory study using an empirical velocity-
dependent burning law. Combustion and Flame 15, 265-274.
Martin, R. E., Cushwa, C. T., and Miller, R. L. (1969). Fire as a physical factor in wildland man-
agement. In "Proc. 9th Ann. Tall Timb. Fire Ecol. Conf.," pp. 271-288.
McCaffrey, B. J. (1979). "Purely Buoyant Diffusion Flames: Some Experimental Results." U.S. Na-
tional Bur. Standards Rep. No. NBSIR 79-1910.
Mercer, G. N., Gill, A. M., and Weber, R. O. (1994). A time dependent model of the fire impact of
seeds in woody fruits. Aust.J. Bot. 42, 71-81.
Moore, P. H. R., Gill, A. M., and Kohnert, R. (1995). Quantifying bushfires for ecology using two
electronic devices and biological indicators. CALM Science Supplement 4, 83-88.
Morton, B. R. (1965) Modeling fire plumes. In "Tenth Symposium (International) on Combustion,"
pp. 973-982. The Combustion Institute (Pittsburgh).
Morton, B. R., Taylor, G. 1., and Turner, J. S. (1956). Turbulent gravitational convection from main-
tained and instantaneous sources. Proc. Roy. Soc. (Lond.) A234, 1-15.
Morvan, D., Porterie, B., Larini, M., and Loraud, J. C. (1998). Numerical simulation of turbulent
diffusion flame in cross flow. Combust. Sci. Tech. 140, 93-122.
NAG (1991). "Numerical Algorithms Group Fortran Library," Mark 15, Vol. 2. NAG, Oxford, UK.
Ramaprian, B. R., and Haniu, H. (1989). Measurements in two-dimensional plumes in cross flow.
J.FluidsEng. I l l , 130-138.
Raupach, M. R. (1990). Similarity analysis of the interactions of bushfire plumes with ambient
winds. Math. Comput. Modeling 13, 113-121.
Raupach, M. R., and Thom, A. (1981). Turbulence in and above plant canopies. Ann. Rev. Fluid
Mech. 13, 97-129.
Rouse, H., Yih, C. S., and Humphreys, H. W (1952). Gravitational convection from a boundary
source. Tellus 4, 201-210
Schatzmann, M. (1979). An integral model of plume rise. Atmospheric Environment 13, 721-731.
Chapter 7 Fire Plumes 255

Stocks, B., and Walker, J. (1968). Thermocouple errors in forest fire research. Fire Technol. 4,
59-62.
Stott, P. (1986). The spatial pattern of dry season fires in the savanna forests of Thailand. J. Biogeog.
13,345-358.
Thomas, P. H. (1963). The size of flames from natural fires. In "9th Symp. (Int.) on Combustion,"
pp. 844-859. The Combustion Institute (Pittsburgh).
Thomas, P. H. (1965). Buoyant diffusion flames: Some measurements of air entrainment, heat trans-
fer and flame merging. In 10th Symp. (Int.) on Combustion," pp. 983-996. The Combustion
Institute (Pittsburgh).
Trabaud, L. (1979). Etude du comportement du feu dans la garrigue de chene kermes a partir des
temperatures et des vitesses de propagation. Ann. Sci. Forest. 36, 13-38.
Trefles, J., McGrattan, K. B., and Baum, H. R. (1999). Smoke transport by sheared winds. Combus-
tion Theory and Modeling 3, 323-341.
Tunstall, B. R., Walker, J., and GiU, A. M. (1976). Temperature distribution around synthetic trees
during grass fires. For. Sci. 22, 269-276.
Van Wagner, C. E. (1973). Height of crown scorch in forest fires. Can. J. Forest Res. 3, 373-378.
Van Wagner, C. E. (1975). Convection temperatures above low intensity forest fires. Can. For. Ser-
vice Bi-monthly Res. Notes 31, 21 and 26.
Weber, R. O., GiU, A. M., Lyons, P. R. A., and Mercer, G. N. (1995a). Time dependence of tempera-
ture above wildland fires. CALM Sci. Suppl 4, 17-22.
Weber, R. O., Gifl, A. M., Lyons, P R. A., Moore, P H. A., Bradstock, R. A., and Mercer, G. N.
(1995b). Modeling wildland fire temperatures. CALM Sci. Suppl. 4, 23-26.
Wilhams, F. A. (1982). Urban and wildland fire phenomenology. Prog. Energy Combust. Sci. 8, 317-
354.
Williamson, G. B., and Black, E. M. (1981). High temperature of forest fires under pines as a se-
lective advantage over oaks. Nature 293, 643-644.
Xanthopoulos, G. (1991). "A Model for the Prediction of Crowning in Forest Fires." PhD thesis.
University of Montana.
Yih, C. S. (1951). Free convection due to a point source of heat. In "Proc. 1st U.S. Nat. Congre.
Appl. Mech.," pp. 941-947.
Yih, C. S. (1953). Free convection due to boundary sources. In "Fluid Models in Geophysics," Proc.
First Symposium on the use of Models in Geophysics, pp. 117-133. U.S. Gov't Printing Office,
Washington.
Yih, C. S. (1969). "Fluid Mechanics." McGraw-Hill, New York.
This Page Intentionally Left Blank
 CHAPTER 8
Coupling Atmospheric
and Fire Models
MARY A N N JENKINS
Department of Earth and Atmospheric Science, York University, Toronto, Ontario, Canada

TERRY CLARK AND JANICE C O E N

National Center for Atmospheric Research, Mesoscale and Microscale Meteorology,
Boulder, Colorado

I. Introduction
II. Vorticity Dynamics in a Fire
III. Coupling between Atmosphere and Fire
IV. The Elements of Fire Modeling
V. Modeling the Atmosphere
VI. The Coupled Fire-Atmosphere
Modeling Approach
A. The Importance of Ignition Tracing on
Fire-Atmosphere Coupling
VII. Idealized Studies of Wildfire Behavior
A. Fire-Scale Convection Drawing the Fire Line
B. Fire-Scale Vortices Disrupt the Fire Line
C. Microscale Vortices Affecting Fire Spread
D. The Effect of Atmospheric Stability and W i n d
Changing with Height on a Fire Propagating
over a Small Hill
E. Sensitivity of Fire Behavior to Fuel
Characteristics and Configuration
VIII. Infrared Observations of Fires
IX. Conclusions and Future Work
Appendix I. Circulation and Vorticity
Appendix II. Development of Vertical Rotation in
a Frictionless Fluid
Appendix III. Generation of Vertical Motion in
Rotating Convective Cells
Notation
References

Forest Fires
Copyright © 2001 by Academic Press. All rights of reproduction in any form reserved. 237
258 Jenkins et a\.

I. I N T R O D U C T I O N

The interactions of forest fires and air flow are highly nonlinear. The heat and
moisture supplied through the burning of ground and canopy fuel during a for-
est fire create extreme levels of buoyancy forcing. The horizontal gradients of
buoyancy produce vortices or fire whirls of tornado strength, which in turn af-
fect the nature of the fire spread through advection of hot gases and burning
material, while fire vortices enhance mixing of air with the flame which leads
to higher flame temperatures, increased combustion efficiency, and greater in-
tensity. Winds at the fire scale can be either strongly modified or even solely
produced by the fire, depending on the level of atmosphere-fire coupling. This
coupling or feedback occurs over spatial scales from tens of meters at the flame
front to kilometers on the scale of the total burn area.
In this chapter, we attempt to describe a fairly recent and major advance in
the modeling of wildfires: the coupling of a cloud-resolving numerical predic-
tion model with a simple fire-spread and wildfire behavior model, so that the
atmosphere-fire is treated as a single, dynamical system. With this modeling
approach, it is possible to simulate the small-scale atmosphere-fire interactions
and feedbacks that are important to wildfire behavior, especially severe wildfire
behavior, and the possible impacts of evolving, larger scale atmospheric forcing
on the fire and vice versa.
A "small" fire is more or less a surface fire, in steady state, that grows in size
but not in intensity. Small fires spread through more or less constant heat trans-
fer by radiation with additional heat transfer by convection, or winds, at the
head of the fire. Small fires are governed by the properties of the fuel bed in
which the fires burn and the air movements within and a few meters above the
fuel bed. A small fire becomes a "large" fire when the fire's intensity becomes
sufficiently great to produce a convection column stronger than the ambient
wind field. A fire that multiplies its rate of energy output many times in a short
period and burns with an intensity far out of proportion to apparent burning
conditions is called a "blowup" fire.
Large fires do not conform to fire behavior expected of the far more frequent,
low-intensity surface fire. Large fires are said to "make their own weather" be-
cause they can noticeably alter the temperature, humidity, and wind fields in
their vicinity. A large fire exhibiting extreme behavior can put up a convection
column to a height of 7000 m or more. Large fires and blowup fires have higher
fire-line intensities (defined as a heat flux or the rate of fire line spread times
the heat per unit area generated by the available fuel), where high fire-line
intensities are of two forms: a high rate of spread when the fire is wind driven
or high rate of heat release when the fire is convection dominated. Large fires
have different modes of propagation and distribution patterns than the ordi-
nary surface fire. Their fire fronts can have significantly higher spread rates (up
Chapter 8 Coupling Atmospheric and Fire Models 259

to 61 m/min as opposed to 6 m/min for surface fires with steady state lieadfire
spread rates; McRae et al, 1989) and can accelerate under the influences of con-
vection. Convective processes in the column can produce extreme fire behav-
ior such as crown fires or large-scale spotting caused by firebrands lofted ahead
of the fire by wind. In the aftermath, the burn pattern may show complex bound-
aries and inexplicably unburned islands. It is the task of fire researchers to ex-
plain such behavior and patterns, and ultimately to predict them.
In some cases, a large fire becomes a conflagration, a fire that takes on storm
characteristics with associated strong, mesocyclonic rotation. Massive fire
whirls up to 400 m wide and several thousand meters in height can develop on
the outer perimeter. Whirls on the edge of the burn can be powerful enough
to rip out standing trees and send them up into the convection column. They
can travel for considerable distances (4 km), removing slash pieces and organic
material in their path, exposing the bare mineral soil underneath (McRae and
Stocks, 1987; McRae and Flannigan, 1990). Large pieces of slash (4 m long,
20 cm in diameter) can be driven into the ground (45 cm into a sandy soil), much
like spears. There is postfire evidence of multiple whirls within these burns.
Although relatively rare, blowup fires, large fires, and conflagrations are re-
sponsible for the major loss of life in forest fires (Chandler et al, 1983), and in
the western United States, 1% of the largest fires account for 80 -96% of the area
burned (Strauss et al, 1989). Large-scale forest fires result in billions of dollars
in financial losses annually in the United States alone. They can locally disrupt
complete ecosystems and can, in cases of extremely large fires, eject significant
amounts of smoke high into the atmosphere with possible long-range air qual-
ity and long-term global meteorological effects.
Each blowup fire or conflagration raises the question of how to recognize the
conditions causing extreme fire behavior and how to predict them. The transi-
tion from a small fire to a large fire is typically sudden, 15 minutes or so, and at
present, the causes for the sudden transition from an ordinary, steady state to
large or blowup fire are not known. Wildfire behavior models developed by the
forestry community have traditionally been used to predict fire spread rate and
heat release for a prescribed set of fuel, slope, and wind conditions and are de-
signed primarily to assess the behavior and risk of surface fires. These models
are not capable of assessing the complex behavior and risk of fires that have
progressed past the small stage.
In the following sections, the basics of coupled fire-atmosphere numerical
prediction, an approach that is capable of simulating the complex behavior of
large and small fires, are presented. The vorticity dynamics that play an impor-
tant role in determining fire behavior are described. To define what is meant
by fire-atmosphere coupling, a nondimensional Froude number, the ratio be-
tween the kinetic energy of the flow in a fire and the potential energy provided
by the fire, is presented and its importance to fire dynamics and extreme fire
260 Jenkins et ah

behavior is discussed. The elements of a fire model and of a three-dimensional

numerical atmospheric prediction model, and how these models are coupled to
produce a physically based dynamical fire-atmosphere model appropriate for
wildfire modeling, are described. The sources of error in a numerical approach
to coupled fire-atmosphere prediction, the problems of resolution, and the
technical and physical difficulties to overcome are discussed. Recent numerical
modeling studies of wildfire behavior using a coupled atmosphere-fire model
are presented, and how infrared video technology is used to observe and analyze
important physical features and properties of fire-scale convection from a safe
distance is described. Finally, we discuss where the coupled fire-atmosphere
modeling approach is, where it is likely to go in the near future, and what it
should become.

II. VORTICITY DYNAMICS IN A FIRE

When the fire's intensity becomes great enough so that the flow responds
strongly to the heating supplied by the fire, the fire can force its own fire-scale
circulations. Convective eddies are seen as horizontal rolls or vertical vortices.
In a convectively dominated fire, intense vertical rotors can develop, with wind
speeds up to 100 m s~^ (Corlett, 1974; King, 1964). To understand the convec-
tive processes that come into play, we must understand the dynamics of the
fluid and how these rotating rolls or vortices develop.
Vertical shear in the wind (wind speed changing with height) provides a
source of rotation about a horizontal axis. To understand why, look at the sit-
uation depicted in Figure 1. Imagine a paddle wheel placed in the wind field,
with the wind coming from the left. Since the wind hitting the bottom of the
paddle wheel is stronger than the wind hitting the top, the wheel spins in a
counterclockwise direction (Figure la). Similarly, parcels of air in the shear
wind field rotate because the bottoms of the parcels are moving faster than the
tops. This spin of individual air parcels or particles is called vorticity. When the
winds begin to interact with a strong updraft, as shown in Figure lb, the rota-
tion about the horizontal axis is tilted, becoming rotation about a vertical axis.
Similarly, the rotation of individual air parcels in the fluid is also tilted into the
vertical producing vertical vorticity.
Strong horizontal gradients in buoyancy near the surface along the leading
edge of the fire front produce especially strong horizontal rotation or vorticity.
When the horizontal vorticity is tilted into the vertical by the fire updraft, very
intense vertical vorticity can develop where the large buoyancy gradients and
horizontal gradients of the vertical wind coexist.
Vertical vorticity is also produced by the solenoidal effect. In the absence of
other forces, fluid parcels or particles will not spin when density and pressure
fields are constant or when they change but in the same direction. But when
Chapter 8 Coupling Atmospheric and Fire Models 261

a)

updraft

FIGURE 1 Spin about a horizontal axis, illustrated by the vortex tube, is present when wind
speeds change (in this case decrease) with height (a). When the vertically sheared winds are
deflected by a strong updraft (b), the axis of spin is tilted into the vertical. For the conditions de-
picted in (a) and (b), the resulting upflow has a counterclockwise spin to the north of the updraft
core and a clockwise spin to the south.

horizontal gradients of pressure and density occur and are not parallel, fluid
parcels experience opposing forces over their lengths, that combine to exert a
torque, causing them to rotate, producing vertical vorticity.
To illustrate this, consider what happens to a small rectangular parcel or bar
of fluid in the situation depicted in Figure 2. First imagine a pressure gradient
in the fluid, perpendicular to the parcel and uniform along its length. In this
case, the parcel simply moves through the fluid from high to low pressure with-
out rotating. Now imagine that the fluid density is not constant but is larger at
one end of the parcel than the other. The forces Fy^and Fy^ acting on either end
of the parcel are approximated by
1 Ap
'^ Pi Ay' '^ Pi Ay
262 Jenkins et al.

p, > p .
P2 p. p, >p,

_\ s^ p f^y-i < i^y^i

y \ \

u. "V
\
\

FIGURE 2 Production of vertical vorticity by the solenoidal effect. See text for explanation.

Since
1 1

then
|F.,I < \Py.\
The torque apphed by the forces, Fy^ and F^^, causes the fluid bar to rotate coun-
terclockwise. Here p is fluid density, and Ap is the difference in fluid pressure
over distance Ay, where y is the horizontal north-south direction.
Once tilted into the vertical or generated by the solenoidal effect, vertical
vorticity can be intensified by the processes of horizontal convergence and ver-
tical stretching. Imagine an air parcel enclosed by a surface. When additional
air flows horizontally through this surface into the parcel, we say that mass con-
verges into the parcel or that the parcel experiences horizontal mass conver-
gence. More air enters the air parcel (u + Auorv + AV, where u and V are wind
components in the horizontal east-west or x and north-south or y directions,
respectively, and Au, Av, are changes in u, v) than leaves it (u or v). If no ad-
justment in its physical parameters follows, the parcel experiences compression
or, what is the same thing, an internal pressure increase. The parcel expands to
relieve this outward-directed pressure force. But the horizontal mass influx pre-
vents horizontal expansion to a great degree, so the parcel expands or stretches
vertically.
Vertical stretching and horizontal convergence also occur in the following
way. Imagine a vertically spinning air parcel as a short, wide vortex tube. The
vortex tube will stretch vertically if the top rises faster than the base. The dif-
ference in rise is Aw/A^, where A^ is distance in the vertical and Aw is the
change in vertical velocity w over A^. When Aw/Az is positive, there is more
chapter 8 Coupling Atmospheric and Fire Models 263

air leaving the top of the air parcel (w + Aw) than entering the bottom (w).
This is called divergence, and the tube is being stretched vertically. But the par-
cel's mass must be conserved. Conservation of mass therefore requires hori-
zontal convergence, which shrinks the tube's radius, narrowing the diameter of
the spinning air parcel.
A fundamental law of physics requires that an air parcel's angular momen-
tum about its vertical axis, which is proportional to its momentum (mass times
velocity) and to its distance from the axis, be conserved. As the air parcel's dis-
tance from the central axis of rotation decreases, its velocity must increase.
This is the principle of conservation of angular momentum and can be thought
of as the "ice skater effect." The air parcel begins to spin faster, just as a figure
skater spins faster when she pulls her arms into her body.
Horizontal convergence and vertical stretching narrow the diameter of the
spinning air parcel, further increasing its rotation as the angular momentum of
the air parcel, now rotating over a smaller distance, is conserved (i.e., the ice
skater effect). Vertical vorticity therefore is intensified.
Vorticity can also increase or decrease at any given locality just by air with
different vorticity blowing into its region. Local vorticity increases when re-
placed by air of greater vorticity and decreases when replaced by air of lesser
vorticity. This process is called advection, and it works in both horizontal di-
rections and in the vertical direction. The term advection is normally reserved
for horizontal transport, whereas vertical advection is known as convection.
Finally, turbulent or frictional drag in a fluid can affect vorticity. Turbulent
or frictional drag can sometimes increase rotation and, very importantly, some-
times reduce or even eliminate rotation in a fluid. Friction slows wind near any
solid boundary such as the ground, and in this case friction always acts in the
opposite direction from the wind. Frictional drag against the ground reduces
rotation or causes spin-down regardless of wind direction.
A vortex is air moving around a central axis about as fast as it moves toward
and along the axis. The development and intensification of a vertical vortex is
a combination of multiple processes working together: the generation of verti-
cal vorticity by the solenoidal effect and tilting, the dynamic-pipe effect, advec-
tion of vorticity in the vertical, and convergence and vertical stretching (Ap-
pendices II and III).
Figure 1 illustrates the generation of vertical vorticity by tilting. Once the
horizontal vorticity in Figure la is tilted into the vertical (Figure lb), the local
vorticity owing to this tilting will be positive to the north of the updraft core
and negative to the south. As a result, a counterrotating vortex pair is estab-
lished with counterclockwise rotation to the north and clockwise rotation to
the south of the initial updraft.
Along the rotating column of each vortex, the pressure field is in balance
with the strongly curved wind field. The inwardly directed force acting on air
264 Jenkins et a\.

parcels as a result of the reduced pressure at the center of the column is coun-
tered by the outwardly directed centrifugal force resulting from the parcels'
rotation about the center. In this condition, called cyclostrophic balance, air
moves easily around and along the axis of the vortex, but radical motions to-
ward or away from the axis of rotation are strongly suppressed. The rotating
column acts as a "dynamic pipe"; it is like a hose of a vacuum cleaner, except
that instead of being channeled by the wall of the hose, the airflow in the vor-
tex is constrained by its own swirling motion.
The low pressure in the center of the vortex provides a vertically directed
pressure gradient force (the centrifugal pump effect described in Appendix III).
Since air is accelerated from high to low pressure, an updraft develops from be-
low, or a downdraft develops from above, in the rotating core. When the ver-
tical pressure gradient force directs air parcels upward, the air flowing along
the vortex's axis is sucked in through its lower end, and as air parcels converge
into the base of the pipe, they turn and accelerate upward. In doing so, they are
stretched vertically (as in Figure 3). Stretching narrows the diameter of the vor-
tex, further increasing the angular speed of its winds and the vorticity of its air

FIGURE 3 Convergence and stretching intensify the vertical rotation initiated by tilting and ex-
tend it upward. Once rotation is vertical, a balance between the inwardly directed pressure gradi-
ent force and the outwardly directed centrifugal force develops. The centrifugal force suppresses
flow toward or away from the axis of rotation (the "dynamic pipe" effect), while the low pressure
at the center of rotation provides a vertically directed force. When the force is directed upward, the
winds converge at the lower end of the rotating updraft. The enhanced convergence compresses
the column and stretches it upward. Because the column is narrower, it must spin faster to con-
serve its angular momentum.
Chapter 8 Coupling Atmospheric and Fire Models 265

parcels, as the angular momentum of the air, now rotating over a smaller dis-
tance, is conserved (i.e., the ice skater effect).
Thus, for an upwardly directed pressure gradient force, the dynamic pipe in-
tensifies the rotation at its lower end, which in turn extends the pipe downward
as the centrifugal force gets strong enough to establish cyclostrophic balance.
Owing to inertia, inflow that gets sucked in through the pipe's lower end can ac-
tually overshoot its equilibrium radius, conserving its angular momentum and
picking up speed as it approaches the center of the core before turning sharply
to spiral upward. As a result, the vortex is narrower at the bottom, and the
highest wind speeds are found in a small ring at the base of the vortex.
Frictional drag against the ground upsets the cyclostrophic balance of the
vortex by slowing the wind and reducing the centrifugal force to zero. But the
inward radial pressure-gradient force does not vanish there. As a result, strong
inflow close to the ground, driven by the unbalanced inward pressure gradient
force, transports fluid parcels closer to the axis than would have otherwise been
possible without friction, whereupon the flow turns vertical. The resulting wind
speeds can be significantly greater than those of the circulation in the vortex
above the surface friction layer.
Note that while vertical progression due to the dynamic pipe effect can take
place, the vortex can also build upward or downward through vertical advec-
tion of vorticity and intensify by convergence and stretching. When vorticity
and radial inflow are constant with height, high angular momentum air ap-
proaches the axis from below and aloft simultaneously. The vortex in this case
forms nearly independently of height. In a fire, however, buoyancy forcing sup-
plied through the burning of surface fuel and the generation of near-surface
vertical vorticity by the solenoidal effect and tilting likely provide that the ver-
tical vorticity and radial inflow are maximized near the ground. Apparently
due to vertical advection of vorticity and horizontal convergence and vertical
stretching, the fire vortex forms from the ground upward. It is also observed
that fire vortices can be short-lived. One possible explanation for this is that the
pressure gradient forcing can boost updraft strength at one level, quickly de-
stroy it at another. A detailed examination of where and what processes domi-
nate to build and destroy fire vortices has yet to be done.
It is also possible that, as a result of powerful updrafts in the cores of vor-
tices, compensating downdrafts form between the updraft cells, which, if strong
enough, can separate and move the centers of convection to the right and left
of the original updraft depicted in Figure lb.
The vortex dynamics just presented are purposely idealized or simplified to
convey the essence of the problem. A seemingly simple complication arising
in reality is that rotating vortices are observed to sometimes break apart, as
discussed in the next paragraph. Exactly how vortices restructure themselves
in such cases is unknown and something that fire modeling may eventually
resolve. Neither is the observed flow field necessarily a one-cell vortex charac-
266 Jenkins et al

terized by an updraft at the core of the vortex. Powerful vortices can also de-
velop an internal multiple vortex structure that has a downdraft in the central
core surrounded by two or more subsidiary vortices with updrafts along their
vertical axes. To learn more about the complications of the internal structure of
laboratory vortices, we refer the interested reader to Church and Snow (1979)
and Gall (1982).
Vertical rotation within fires can result from at least two similar but differ-
ent mechanisms. Adopting the terminology from Clark et al. (1999), the first
mechanism is enhanced vortex tilting. In this case, the tilting of horizontal vor-
ticity is believed to result from a local region of buoyancy that acts strongly
enough to break or disconnect the tilted vortex, leaving one or two rotating,
standing hot fire whirls or plumes, which transport heat almost vertically into
the atmosphere.
The second mechanism is the "turbulent burst"- or "horseshoe"- or "hair-
pin"-shaped vortex. The turbulent burst is a concept developed in the study of
boundary layer turbulence (e.g., Moin and Kim, 1982; Kim and Moin, 1986).
The boundary layer is defined as the portion of the atmosphere in which the flow
field is strongly influenced by interaction with the surface of the Earth. Again
vortex tilting occurs, but in a manner that results in an unbroken vortex that
moves both vertically and horizontally; as the vortices rise and come closer to-
gether, their combined motion results in the vortex tilting forward at a rela-
tively sharp angle giving a hairpin shape. When the horizontal motion is strong
enough, the rotors shoot or burst forward over unburned fuel, sometimes for
considerable distances. It is believed that this type of vortex tilting led to the
hairpin vortex observed by Radke et al. (2000). Observations captured a flame
shooting horizontally as far as 100 m in only 2 s (Figure 4), and the fire fighters
wisely stayed clear of this particular fire behavior at the time of the observations.
For a more in-depth description, a mathematical treatment of vorticity is
given in Appendix I, the development of vertically rotating convective cells is
described in Appendix II, and the development of an updraft in a rotating con-
vective cell is described in Appendix III.

III. COUPLING BETWEEN

ATMOSPHERE AND FIRE
As discussed in previous sections, winds at the fire scale can be either strongly
modified or even solely produced by the fire, depending on the level of atmo-
sphere-fire coupling. This coupling or feedback occurs over spatial scales from
tens of meters at the flame front to kilometers on the scale of the total burn area.
Therefore, the knowledge of the interaction of the fire with the atmosphere is
Chapter 8 Coupling Atmospheric and Fire Models 267
WIFE FI1T3 Data WIFE FI1T3 Data
Frame=920 Tfme=30.667 (S) Thrs=40.0 Frame=935 Time=31167 (S) Thrs=40.0
860,

752

645,

537

430,

322

215

107

0
WIFE F11T3 Data WIFE FI1T3 Data
Frame=950Time=31.667 (S) Thrs=40.0 Frame=965Time=32.167 (S)Thrs=40.0

0.0 215.0 430.0 645.0 860.0 0.0 215.0 430.0 645.0 860.0
107.5 322,5 537.5 752.5 107.6 322.5 537.5 752.5
X(M)
to J- .
0.0 L. .
200.0 240.0
••iiiiii
280.0 320.0 360.0

FIGURE 4 Airborne infrared imagery for fire spread analysis shows "hairpin" vortex (approxi-
mate location 753 m iny direction and 540 m in x direction) bursting forward at 70 m in 2 s. (Af-
ter Radke et al, 2000.)

needed to understand and predict fire behavior correctly. It is first necessary to

define this feedback or coupling between the atmosphere and the fire.
In grass fires, it has been observed that the importance of wind speed dimin-
ishes after a critical speed is reached. Above certain speeds, further increases in
wind speed do not stimulate additional increases in rate of spread or fire line
intensity. To interpret this transition with increasing wind speed, Byram (1973)
suggested a convective Froude number, F^. The Froude number is defined as
the ratio of the kinetic energy of the air and potential energy provided by the
268 Jenkins et ah

fire. The measure of kinetic energy is the relative speed of the air passing over
the fire, UQ = U — Vj, where U is the ambient wind speed and Vj is the fire spread
(i.e., rate of spread at the head of a fire). The measure of the potential energy
used for the fire is the fire-line width in the mean wind direction times the buoy-
ancy of the heated air. One possible formulation for F^ (Clark et ah, 1996a) is

F -
ul
<AT>

Here T is temperature and g is acceleration due to gravity. The < A T > is the
average or mean value of the temperature anomaly AT over the region of in-
tense heating and through Wj, the width of this region. Buoyancy is defined
as g{<AT>/T). The Froude number is a nondimensional number. The val-
ues of UQ, WJ, T, and < A T > can therefore differ without the combination
Ul/[g{<AT>/T)Wj] being different. Providing other governing parameters are
also similar, we can expect the same behavior for flow with the same Froude
number (i.e., all such flows are then dynamically similar).
An increasing F,. means that for increasing ambient wind speeds the fire de-
posits less heat into parcels of air as they move over the fire, until at some am-
bient wind speed the effect can be neglected. The process can be compared to
a freight train of partly filled grain cars moving under a loading chute (Chand-
ler et al, 1983). If the train moves too slowly, the cars fill up and are unable to
carry off as much grain as is being fed to them. Once the train reaches a speed
at which the cars do not become filled, it is able to carry away all the grain, and
an increase in speed will not change its ability to do so. To illustrate large F^,
simply run your hand over a burning match; when performed at comfortably
high values of F^., the interaction between hand and fire is negligible.
F^ can be used to gauge the level and type of coupling between the fire and
atmosphere. The smaller the F^, the larger the effect of buoyancy forcing on the
acceleration of the flow. When FJ ~ 1, the flow enters a regime where both fire
and dynamics are important. When F^ < 1, the flow responds strongly to the
heating supplied by the fire, and the fire regime is plume-driven or convection
dominated. If fire convection is to control the larger-scale dynamics, values of
F^ should be small. In high-wind-speed cases, as in wind-driven fires, F^ > 1,
and the flow is not dominated by the heating supplied by the fire. For typical
values of Uo = U - Vj ^ 5 m s'\ T -- 300 K, AT ^ 50 K, and Wj ^ 50 m, the
F^ ~ 0.3. This implies that even for small, stable fires in weak or moderate
winds, the effect of buoyancy forcing on the acceleration of the flow is signifi-
cant, and the wind and fire are essential parts of the same system and cannot be
separated. Also implied is that small Fj is a necessary but not sufficient condi-
tion for severe coupled fire-atmosphere behavior.
Chapter 8 Coupling Atmospheric and Fire Models 269

IV. THE ELEMENTS OF FIRE MODELING

Regardless of the level of sophistication, current forest fire models require and
provide, in one form or another, certain basic information. This includes ini-
tial total mass and dry mass of fuels (mass per unit area); initial moist/dry mass
ratio of fuels; the rate at which fuel, once dry, burns (mass per unit area per unit
time); the amount of energy provided per mass of dry fuel burnt; and the ratio
of latent (moist) to sensible heat of this energy. The fire model outputs are sen-
sible and moisture (latent) heat from the burning fuels (energy per second). Cer-
tain additions may be necessary: a percentage of the sensible heat provided by
the fire dries the fuel; canopy ignition can occur when sensible heat flux from
ground fuels exceeds a threshold value (energy per mass); to emulate smolder-
ing, a random component can be added to the burn rates; and surface fuel is ig-
nited by contact, where ignition contact follows the predicted fire spread rate
and direction.
The ability to predict the rate of fire spread is recognized as the most impor-
tant single requirement for successful fire suppression. Current models of for-
est fire spread assume that radiative heat transfer and convection, or wind, are
the mechanisms that transmit heat from the flaming zone to adjacent fuels. Ra-
diative heat transfer is increased by fuel changes that convert more of the reac-
tion intensity into effective heat flux. Convection transfer is strengthened by in-
creasing winds that can also increase both convective and radiant heat transfer.
The important effects of terrain on fire propagation speed are also included.
There are many problems with current wildfire and fire spread rate models,
but the major drawback is that, to apply them successfully, information on fuel,
topography, and winds at the local, fire-scale level is required, information that
is presently not available to fire managers and fighters. And because convection
is a key mechanism for fire propagation after the incipient stage, knowledge of
the small-scale atmospheric circulations in and around a wildfire is primary. In
most applications, however, single values of wind speed and direction are ap-
plied over a large area and long time period. Therefore, microscale processes
that affect the fire spread rate and involve the variable winds within the fire and
adjacent fuel are not treated, whereas the formulae currently used to link the
fire spread rate to the ambient winds ignore any feedback between the small-
scale air motions and the fire. The solution to these deficiencies is to couple fire
behavior and fire-spread-rate models to a dynamical atmospheric model.

V. MODELING THE ATMOSPHERE

To model the atmospheric circulation in and around a wildfire properly, a

prognostic three-dimensional numerical model based on fundamental fluid
270 Jenkins et ah

dynamics and capable of resolving convective-scale motions is required. A

model of fundamental fluid dynamics consists of the Navier-Stokes equations
of conservation of momentum, conservation of mass, and conservation of
energy. Three dimensions are essential as fire dynamics are inherently three-
dimensional due to advection, the solenoidal effect, vortex tilting, horizontal
convergence, and turbulence or frictional drag.
The Navier-Stokes equations are smooth analytical functions, or differential
equations, which in the case of conservation of momentum describes the in-
stantaneous relation between position, velocity, and acceleration throughout
the motion. A numerical model discretizes these differential equations to work
at a finite number of regularly spaced locations called grid points. Each grid
point holds the average value for a volume of surrounding air. This volume is
called a grid cell or grid volume. To determine the functions at a future time,
the equations are rearranged into forecast equations, and the functions are fore-
cast for one-time step, utilizing present (and sometimes past) conditions to de-
termine future conditions. Each operation in the analytical equations is ex-
pressed as a finite difference between grid variables. Computational time is
spent for each grid point in the domain, and the computation is repeated for
a succession of time steps in order to reach a forecast of possibly several hours
duration. In a Cartesian coordinate system, the sizes of the grid cells in three
directions are Ax (east-west). Ay ( n o r t h - s o u t h ) , and A:^ (vertical).
There are several sources of error in a numerical forecast. One is round-off
error. The computer is able to store only a limited amount of information; as
a result, real numbers are only approximated. Truncation is another source of
error. An analytical variable is represented at a grid point as a function of its
values at other grid points, the result being an infinite sum of terms, each of
greater power of Ax or At (i.e., a Taylor's series expansion). Eor practical rea-
sons, the computer retains the terms of the lowest power, the series is trun-
cated, and higher power terms (which slightly improve accuracy) are neglected.
Another error source is numerical instability, when the numerical solution rap-
idly diverges from the correct solution. Truncation is one cause, but numerical
instability can also occur if the wind speeds are large, the grid size small, and
the time step At not small enough. A necessary requirement for stability in the
X direction with ambient wind speed U is

At < ^—^
U
with similar requirements in the y and z directions.
A numerical model of the atmosphere-fire system must use, from an atmo-
spheric scientist's perspective, a grid mesh with small Ax, Ay, and A:^ values;
currently typical sizes of grid cells in the domain of the fire can be tens of me-
ters. Therefore, to preserve numerical stability, the time step must be also be
Chapter 8 Coupling Atmospheric and Fire Models 27 i

reduced. Typical time-step durations At are on the order of fractions of a second

in the fire domain. A small time step has consequences for the total run time of
the calculation and, since error is accumulated by the many more time steps re-
quired to reach a particular time, for the accuracy of the numerical solution.
The errors in initial conditions also affect the long-range predictability. The
Navier-Stokes equations are nonlinear; as a result, the system is very sensitive
to initial conditions. Such sensitivity means that a substantially different fore-
cast can result from only slightly different initial conditions and that the fore-
cast always becomes inaccurate with time. To minimize this error, initial con-
ditions are specified as accurately as possible, and only short-term forecasts are
considered trustworthy.
And lastly and most importantly, the physics of the fire-atmosphere system
must be faithfully represented by the model formulation. The final source of er-
ror is then how accurately subgrid scale atmospheric and combustion processes
are currently determined in terms of grid-scale (and therefore model forecasted)
parameters. This is a central problem in numerical forecasting, and it is dis-
cussed further in the next section. Ideally the model framework should also be
amenable to adding chemistry for pollution or smoke dispersal studies.
Because the computational demands of coupled atmosphere-fire modeling
are great, additional features that diminish computational load while providing
an accurate forecast are necessary.
One such feature is a vertically stretched, terrain-following coordinate sys-
tem. In this system, the lower grid levels follow the contoured surface, while the
vertically stretched coordinate allows a variable vertical resolution to the solu-
tions of the equations near the surface and in regions of rough terrain.
Another such feature is two-way interactive grid nesting, also called grid re-
finement or multilevel grid nesting. Diverse scales of motion influence a wild-
fire's behavior, from the scale of the regional weather pattern (1000 km) down
to the local and fire-generated flow (tens of meters and less), and multiple lev-
els of interactive-grid refinement allow the numerical model to run over such a
large range of scales. A large outer domain with large-sized grid cells and time
steps, and initialized with coarse resolution meteorological data from, say, Gen-
eral Circulation Forecast Models, determine satisfactorily the largest scale atmo-
spheric circulation. Inside or nested in this domain are successively smaller and
smaller domains, each containing topography at the grid resolution of that do-
main, while fields are matched between the boundaries of the coarser grid and
finer grid domains. Eventually the meteorology is forecast in the smallest do-
main, where fine-scale grid cells are used to resolve the fire's convective-scale
motions.
A further refinement is an adaptive grid mesh in which the finite difference
grid is refined in selected regions that may be changed in time to follow evolv-
ing features such as the moving fire front.
272 Jenkins et at

As in Section II, the description of a three-dimensional numerical atmo-

spheric prediction model presented is purposely simplified to convey the es-
sence of the approach. Although there are many more aspects involved, they
are beyond the scope of this text and are therefore not included.

VI. THE COUPLED FIRE-

ATMOSPHERE MODELING APPROACH
We have seen that even for small, stable fires coupling between the fire and the
atmosphere can be significant. Temperature, humidity, and wind fields are no-
ticeably altered by the fire, which in turn alter the fire. Once temperature and
humidity fields are altered, pressure gradient and buoyancy forces come into
play, and the velocity field changes. Velocity field changes affect fire behavior
and fire spread rates, which can advance the fire front, igniting more fuel, re-
leasing sensible and latent heat, and repeating the process.
The coupling between the fire and the atmosphere occurs mainly through
the fluxes of sensible and latent heat from the ground fuel and canopy burns.
Radiation is also important, both in preheating the fuels ahead of the fire and
shortening their ignition time and to the atmospheric circulation in and above
a wildfire. To couple current fire and numerical atmospheric prediction mod-
els, radiation and sensible heat and moisture fluxes associated with the fire are
included as subgrid-scale sources of heat in the atmospheric model's conser-
vation of energy equation. Wind is also modified by the forest, especially the
vertical wind profile, and the effect of surface and forest canopy drag on the
velocity field is a subgrid-scale momentum source in the atmospheric model's
conservation of momentum equation.
The heat fluxes and radiation from the fire are absorbed by the air, and a
method for incorporating this effect in the atmospheric model is required. A
very simple approximation is that the amount of radiation absorbed into the
atmosphere decays exponentially with height, where the depth of extinction
may range from 50 to 100 m in the vertical. In reality, the interaction between
radiation and the atmosphere is much more complex than this and is not well
understood. For example, an accurate treatment of flame front radiation onto
unburned fuel and resulting fire spread strongly depends on a good knowledge
of the vortex dynamics at the fire line. To understand the impact of radiation
on fire spread, rotors produced by the mechanism of enhanced vortex tilting
might be approximated as a radiating vertical wall of flame, whereas the forward
bursts of flames in hairpin vortices cannot be reasonably approximated in this
way. To date, this type of interaction has not been taken into account.
Yet, however they are handled, the effects of convection, turbulent mixing
and frictional drag, and black body radiation occurring on scales not resolved
by the numerical atmospheric model must be represented accurately.
Chapter 8 Coupling Atmospheric and Fire Models 273

Because current wildfire formulae are mainly algebraic and based on statis-
tical or empirical ideas and do not describe the microscale processes that affect
the fire spread rate and involve the variable winds within the fire and adjacent
fuel realistically, coupling the fire model with the atmospheric model cannot
completely accomplish the task of including the feedback between the small-
scale air motions and the fire. The solution is to develop a physically based, dy-
namic wildfire model. Yet, even if a fully physically based wildfire model were
available to couple with a fully physically based atmospheric model, the tech-
nical and conceptual difficulties to be overcome are tremendous.
Within fires, there are small vortical structures on the millimeter scale that
tightly bend the flame fronts, up to vortex structures on length scales of many
meters. The time scales range from milliseconds in submillimeter eddies to the
puffing frequencies on the order of seconds. In addition to these fire-convective
length and time scales are diffusive length and time scales, from the submilli-
meter and submillisecond range required for resolution of chemical species dif-
fusion gradients in flame zones, to the tens of meters and thousands of seconds
required to resolve the length and time scales of fire/object interaction. More
than five orders of magnitude in temporal and spatial resolution are required to
simulate directly all the relevant scales within the fire. Coupling a numerical
fire simulation with a numerical atmospheric model adds a further five or more
orders of magnitude in resolution when attempting to incorporate atmospheric
forcing by meteorological conditions on the regional weather scale.
Given the present level of computational resources, no numerical simulation
tool is capable of simultaneously resolving all required length and time scales
associated with forest fire phenomena. The problem is easily illustrated by con-
sidering the following: A numerical model basically divides three-dimensional
space into grid cells with volume ~(Ax)^, and the computations will locate a
moving fluid particle within one of these cubes. If greater resolution and accu-
racy are desired, a finer mesh with smaller grid cells is used. To improve the
accuracy tenfold for instance, the size of the grid cell must be Ax/10 instead of
Ax. There are 10^ = 1000 such small cubes in a single large one, all of which
must be allocated some place in the memory of the computer. This means that
the computations which can trace a parcel's location within 100 m in 1 s will run
for almost 20 minutes to pinpoint the same trajectory within 10 m. By this time,
of course, the model forecast can be meaningless, the parcel having in reality
long since left the computed position. Alternatively, to find the answer within
1 s, some way must be found to perform the same computations 1000 times
faster, which cannot be achieved by technological means only. Today, speed in-
creases tenfold, not a thousandfold, from one generation of computers to the
next. (The current observation is that computer speed and memory capacity
double about every 18 months.) This problem is not specific to coupled fire-
atmosphere modeling; in every area of science, there are a great many compu-
tations that cannot be performed now or in any foreseeable future.
274 Jenkins et ah

Basically there are two approaches to dealing with the numerical model-
ing of phenomena operating on a large range of scales. One is starting with the
smallest length scales and resolving all the physical processes u p to the largest
length scales that can be computed. The other is starting with the largest length
scales and resolving downscale with as many grid cells that can be computed.
Although the first approach is preferable from a scientific perspective, the larg-
est length scales within a fire model that currently can be numerically modeled
are on the order of centimeters. Even with the tremendous growth in both speed
and memory for computing hardware, this approach will likely be limited to
length scales significantly smaller than the resolution of the smallest grid used
by coupled atmosphere-fire models. The second approach uses a technique
known in the atmospheric science community as subgrid-scale parameteriza-
tion. Physical mechanisms with length and time scales below the minimum that
can be resolved by numerical models are expressed either empirically or mathe-
matically in terms of observed variables or model forecasted parameters in an
ad hoc attempt to represent the physical phenomena as realistically as possible.
Subgrid-scale parameterization is not specific to atmospheric numerical mod-
eling; this is also the technique used by combustion scientists to model turbu-
lence in the flow, combustion processes, soot formation, and thermal radiation
of combustion products in a fire, along with the turbulent eddy viscosity or fric-
tion near solid surfaces. Attempts are made to develop transport models that
connect fire propagation rates to the full conservation equations of energy, mo-
mentum, species concentrations, mass, and turbulence and that represent the
essence of a combination of many microscale processes without resolving each
process in complete detail (Linn, 1997). In both atmospheric and fire combus-
tion numerical modeling research, subgrid-scale parameterizations have proven
to represent successfully many unresolved physical processes in the system to
the degree that is needed, and the development of and improvements to param-
eterizations are ongoing efforts.

A. T H E IMPORTANCE OF IGNITION TRACING

ON FIRE-ATMOSPHERE COUPLING

One major technical difficulty with a coupled fire-atmosphere model is igni-

tion tracing. In a numerical model, the fuel is necessarily divided into grids.
The simplest way to advance the fire through the fuel grid mesh is to ignite en-
tire fuel cells on contact. For example, four tracers are assigned to a grid cell
once the ground fuel for that particular cell is ignited. Based on the prescribed
spread rate, the tracers move parallel with and against, and normal right and
left of, the tracer-advecting wind. Once a ground tracer reaches an unburned
grid box, that fuel cell is ignited, and four new tracers are initiated at the point
Chapter 8 Coupling Atmospheric and Fire Models z75

of contact. Tracers should not move into burned out or burning grid boxes.
One of the immediate problems with this approach is that ignition tracers do
not necessarily maintain or move to increase the fire perimeter within a fuel
box, and negative distances can be covered. This happens when the local fire
convection changes the wind direction, and an ignition tracer that was moving
with the wind suddenly finds itself in an opposing wind condition. With this
simple approach to ignition parameterization, it is difficult to prevent the model
ignition tracer moving back into burned or burning regions within a fuel grid
cell. Obviously this simple ignition parameterization fails to maintain or move
the ignition pattern to increase the fire perimeter within a fuel box in a realis-
tic fashion.
Another result of this simple approach to ignition is that the problems of
grid mesh size are exacerbated. Model resolution is always an issue, but it is es-
pecially so for the fire-atmosphere system where model resolution has an ex-
tremely important influence on heat transport away from the fire. W h e n the
entire fuel cell is allowed to ignite on contact and the model resolution is too
coarse, the model fluid becomes inefficient at transporting heat, artificially
changing the heat balance.
W h e n the fire begins, there is initially no convection to remove heat effi-
ciently, causing very high air temperatures to build near the surface. Once the
convection is well established, the temperature equilibrates to significantly
lower values as the convective heat transport balances the heat flux from the
fire. The time the fire takes to establish this balance in a numerical simulation
using a 20-m grid resolution is approximately 1-2 minutes. After convective
adjustment, the atmosphere transports heat away from the fire such that the
convection is in balance with its heat source. The sensible heat flux into the
atmosphere is calculated as

where c^ is the specific heat capacity at constant pressure p, p is air density, and
w' and T' are the respective vertical velocity and temperature perturbations or
fluctuations from the mean. The < > denotes the average or mean value over
the region of intense heating. W h e n using a 20-m grid resolution, realistic nu-
merical model values for the maximum vertical velocity anomaly Aw are ap-
proximately 2 0 - 3 0 m s " \ and for the maximum temperature anomaly AT are
approximately 40 - 80 K over the region of intense heating. A rough estimate of
Fj supplied by these model perturbations is, therefore,
F, ^ Cp < p A w A T > « 0.8 to 2.4 MW m'^ (1)
which easily accounts for the —0.8 MW m~^ of sensible heat flux typically
emitted by the fire. AT is limited to values as low as 40 - 8 0 K in a forest fire due
to the efficient heat transport by convection, even though the flames may have
radiant temperatures of —800-1600 K.
276 Jenkins et a\.

To establish a horizontal fire resolution when the entire fuel cell ignites on
contact, take Axji^^j < VJT, where r is the time it takes the fire to burn the small
fuel, and Vj is the fire spread rate. A fire line typically moves 1 0 - 2 5 m in time
T. For Vj ~ 0.5 m s~^ and r ~ 50 s, Axji^^j < 25 m should provide a continuous
fire behavior. If Axj^^^j is too large, the heat flux from the fire is discontinuous,
which forces air to always readjust to the fire as it moves through the fuel grid
mesh. Because the convective adjustment time is comparable to the time re-
quired for the small fuel to burn, namely about 100 s, the effects of the heat flux
from a model with coarse fuel resolution will not average out with time. A
Axji^gj = 10 m horizontal resolution for the fire is generally sufficient to avoid
discontinuities in the numerical model solutions when the entire fuel cell ig-
nites on contact. This relatively small grid cell necessarily increases the com-
putational load.
If horizontal and vertical resolutions in the atmospheric model are substan-
tially reduced, then the heat flux is input into the model atmosphere in an un-
realistic fashion, and the model fluid becomes inefficient at transporting heat.
The numerical results are much higher temperature fluctuations and a much
smaller convective Froude number. An almost immediate model blowup fire
with uncontrollable fire spread rates in excess of 50 m s"^ can occur. The igni-
tion parameterization and resolution must not dictate the values of F^ and F^,
and thereby the overall fire behavior.
An effective tracer ignition parameterization is instead where the subgrid-
scale ignition pattern moves through each fuel cell at the fire spread rate di-
rected by the fire-scale winds. What is needed and what has been recently
developed is an improved tracer ignition parameterization in which the coor-
dinates of ignition tracers define a particular burn area within the cell. These
coordinates are time dependent and allow fires of arbitrary orientation and
shape to move through a mesh of fuel grids at fire spread rates, with minimal
adverse effects on the model that result when the entire fuel cell is allowed to
ignite on contact. This type of ignition parameterization provides spatially
smooth predictions of heat and moisture fluxes from the fire and can perform
at higher grid resolutions. Larger grid mesh values can be used with this igni-
tion approach, although it is still necessary to consider the influence of model
resolution on heat transport away from the fire, and to resolving the fine-scale
convective processes that maintain or move the ignition pattern to increase the
fire perimeter within a fuel box in a realistic fashion.
Figure 5 shows an example of this type of tracer ignition for a spot fire ex-
panding outward under zero wind conditions. The plot shows a region of 21-
by-21 fuel cells with the solid circular contour outlining the fire line. The in-
sert shows the regularly shaped fire line constructed from the shapes provided
by four ignition tracers within each fuel cell nearest the fire line. Most of the
fuel cells inside the fire line are fully ignited except for those near the fire line
which are only partially ignited.
Chapter 8 Coupling Atmospheric and Fire Models 277
21

21

^ L
\ —L

FIGURE 5 Circular-shaped fire line simulated for a spot fire in zero wind. The circular solid con-
tour represents the fire line, and the interior rectangles represent the individual burnt or burning
fuel cells. See text for explanation.

VII. IDEALIZED STUDIES

OF WILDFIRE BEHAVIOR

Despite the difficulties to overcome, efforts at numerical coupled atmosphere-

fire modeling are currently helping to improve our understanding of wildfire
behavior and will in the future predict it. In the following we describe the nu-
merical results from model simulations of Clark et a\. (1996a,b) and Coen et al.
(1998). Their simulations produced a number of realistic features of natural
fires. These include the bowing of fire lines into conical shapes, the formation
of fire whirls that touch down within the fire line and break it apart, microscale
vortices in the fire line that strongly accelerate the local winds and fire spread
rate, how negative vertical wind shear can contribute to extreme fire behavior,
and the deflection along the ridge of a line fire propagating up a relatively sharp
slope. We also describe the numerical results from model simulations by Bos-
sert et al. (2000) and Linn et al. (2000). Their simulations demonstrate the sen-
sitivity of fire behavior to different fuel characteristics and configuration. The
dependent fuel parameters studied were fuel load, moisture, and type, and a
two-storey fuel canopy. We discuss the use of the infrared video camera, a re-
mote sensing device that provides high spatial and temporal observational data
of wildfire winds and heat fluxes.
278 Jenkins et a\.

A. FIRE-SCALE CONVECTION
DRAWING THE FIRE LINE

The perimeter of a moving fire line normally does not spread evenly, but with
a bias. The center spreads slightly more rapidly than the ends; a line of fire nar-
rows and forms a head. In moderate ambient winds, prescribed burns ignited
in a line almost always develop a parabolic shape to the fire front. This feature
occurred in Clark et a\. (1996a,b), in a series of idealized numerical experi-
ments designed to test the effect of different constant background winds on the
evolution of a short ( ~ 400-m-long) fire line.
In weak ambient winds ( < 3 m s~^), the model results show a chaotic and
broken fire front. Under these conditions, air feeding the hot convection column
comes directly from below. The vorticity dynamics described in Section VII.B
(and Appendix 11) produce vertical vortices embedded in the fire line, and the
erratic fire-scale, fire-generated winds break up the fire line.
Under moderate ambient wind conditions (~ 3 to 5 m s"^), numerical simu-
lations produce a stable fire line, where moderate winds move the mean surface
effects of convection ahead of the fire line. The convection cell aloft remains
just downstream of the fire, drawing the fire with it, and its near-surface conver-
gence pattern induces a smoothly, continually evolving, parabolic shape to the
fire line. Figure 6a shows the evolution of a fire line in a constant background
or ambient wind of 3 m s~^ Figure 6b shows that a slightly faster constant am-
bient wind speed of 5 m s~^ accentuates this effect. This convective feedback is
a linear fire-scale phenomenon, in that smaller scale interactions between de-
veloped flow do occur, but the overall effect remains.
Once a fire line is long enough, it cannot sustain a single convective updraft
column. The essentially two-dimensional updraft along a linear fire line expe-
riences along-line roll instabilities that can cause the line to break into multiple
convective updrafts or towers. Providing conditions similar to those required
to form the local parabolic shape exist, multiple protrusions or fire heads form.
An example of this (Figure 7) occurred in the 1985 Onion fire in Owens Val-
ley of California (C. George, personal communication). The fire line developed
protrusions or fingers that were spaced about 1 k m apart and that appeared only
with an ambient wind. When the mean wind died down, the fuel between the
fingers burned out, forming a more linear and stalled smoldering fire line. With
the return of the wind, fire spread resumed with fingers reforming. Numerical
simulations with a coupled atmosphere-fire model reproduce this phenome-
non. Modeling results show one convective finger developing along an initially
400-m fire length and two convective fingers along an 800-m fire line.
These simulations demonstrate a useful aspect of a numerically coupled
atmosphere-fire model: the ability to examine the effect of a single variable on
the fire behavior. In these experiments only the background wind was changed.
Chapter 8 Coupling Atmospheric and Fire Models 279
a)U„ = 3 m s ' b ) U , = 5mE
1 • ' • 1 ' ' •

: ,
S '
0.6 y
' 1 1
!
0.4 4 '.

0.0 1 . i—i—i . . .

0.0 0.2 0.4 0.6 0.2 0.4

X(km) X(km)

100 200 300 400 500

Heat flux (kWm")

FIGURE 6 Numerical simulations of a fire line (heat fluxes greater than 0.1 MW m~^) in con-
stant background winds of (a) UQ = 3 m s"^ and (b) UQ = 5m s"\ (After Clark et al, 1996a.) The
bottom left to right axis is the x axis, the vertical axis on the left is the y axis, and units of x, y are
km. The straight line parallel to the y axis is the initial fire line.

FIGURE 7 The 1985 Onion sage brush fire in Owens Valley, California. See text for description.
(Courtesy of C. George, USDA Forest Service, Intermountain Research Station, Missoula, Montana.)
280 Jenkins et al.

Other ambient meteorological conditions were set to neutral. No topography,

Coriolis force (deflection of moving air by rotation of the Earth), or diurnal ef-
fects were included. The fuel was homogeneous so that any scale selection or
breakup in fields and fire line were due to the dynamic and thermodynamic in-
teractions between the fire model and the meteorology. Different fuel properties
had been suggested as being responsible for the fingering fire behavior shown
in Figure 7. The simulations show that the ambient wind strength, not the
properties of the fuel complex, was responsible for the differences in fire line
behavior.

B. FIRE-SCALE VORTICES DISRUPT THE FIRE LINE

In prevailing winds, the leeward side of the convection column is a preferential

spot for the development of tornadic fire whirls and, under the right wind con-
ditions, these fire-scale vortices can disrupt the fire line. Clark et al. (1996b) set
up numerical experiments to test the effect on a short (400-m-long) fire line
of the change in ambient wind speed with height. Two ambient wind profiles
were used (Figure 8), one a constant westerly wind (3 m s~^), the other a hyper-

1200

u (m/s)
FIGURE 8 Two idealized wind profiles used to demonstrate fire behavior caused by coupling be-
tween convective motions and ambient vertical winds. See text for explanation.
Chapter 8 Coupling Atmospheric and Fire Models 281

FIR7CR FIR7CS1

-0.10 -0.05 0.00 0.05 0.10

FIGURE 9 Fire lines (solid outlines of heat fluxes greater than 0.1 MW m~^) and vertical vortic-
ity (shading) are shown for (a) the constant wind and (b) the hyperbolic tangent wind profiles in
Figure 8. Vectors denote local winds at 15 m above ground level. (After Clark et al, 1996b.)

bolic tangent profile westerly (3 m s"^) at the surface, decreasing with height
to 500 m above the ground, and then reversing direction to an easterly wind
( - 3 ms~^) aloft.
Figure 9 shows the near-surface vorticity patterns and fire line ignitions for
the two experiments. The constant ambient wind profile case develops a para-
bolic fire line shape as described in Section VILA, whereas the hyperbolic tan-
gent profile case shows a fire line separated at its center.
Both cases produced a counterclockwise rotating vortex ahead of the fire
line and south of the updraft core and a clockwise rotating vortex ahead of the
fire line and north of the updraft core. The rotating vortices are the result of tilt-
ing the horizontal shear in the convection into the vertical by the fire's central
updraft core, the process that is depicted in Figure 1.
In the constant ambient wind profile, the vortices, once developed, touch
down, but well in front of the fire line. In the hyperbolic wind profile, the
vortices develop in front of the fire line, but eventually the winds above 500 m
advect the vortices back into the fire line, where they touch down in the fire.
The counterrotating winds are sufficient to overcome the ambient winds, caus-
ing reversed flow and breaking of the fire line. The mathematical explanation
for this development is given in Appendix II. Although the effect of vorticity
dynamics on the fire line propagation is obvious, there is no evidence from the
282 Jenkins et ah

numerical results that the events in the second wind profile case weakened the
fire more than the first.

C. MicROSCALE VORTICES
AFFECTING FIRE SPREAD

The smallest scale vorticity at the fire front is important to the dynamics of fire
spread. Tilting into the vertical of the small-scale negative shear (wind speeds
decreasing with height) close to the surface provides increased winds in the di-
rection of the fire spread. This is the situation depicted in Figure 1 and can re-
sult in a marked increase in the rate of fire spread (as illustrated by the right-
pointing arrow in Figure lb). It is simple for a numerical model to determine
when and where in the fire domain maximum fire spread rates are. The strongest
fire spread rates occurred for the second wind profile simulation (i.e., Figure 8
and Figure 9b), and the sequence of events was tracked by the numerical model.
The vorticity advected back into the fire and behind the fire line when the
easterly wind aloft interacted with the fire; as a result, chaotic and amplified fire
behavior set in at about 16 minutes into the simulation. Figure 10 shows a time
sequence of vertical velocity after the amplified behavior set in. A fire-induced
downdraft in the rear flow produced near-surface vertical shear that moved
through the fire front, adjacent to an existing vertical rotor. Vertical tilting of
this shear by local updrafts intensified the rotation rate of vertical rotor lead-
ing to anomalously large fire spread rate Vj values. The vertical wind increased
to well over 5 m s~^ at 15 m above ground level. Model wind speeds are neces-
sarily averaged over a large grid volume (20 m on a side in this case). Unlike
model winds, tornadic whirls at 10-50 m in diameter in real fires obtain up-
drafts and horizontal velocities of 50-100 m s~\

D. T H E EFFECT OF ATMOSPHERIC STABILITY

AND W I N D CHANGING WITH HEIGHT ON
A FIRE PROPAGATING OVER A SMALL HILL

Observations, laboratory experiments (Weiss and Biging, 1996), and computer

models (Linn, 1997; Coen et al, 1998) show that when fire spreads on sloped
terrain, it spreads faster uphill on steeper slopes. This is due to a combination
of dynamic effects, the relative importance of each still being debated (Baines,
1990). One effect is that the fire's convection increases momentum and wind
velocities near the ground. Another involves the fluid dynamics of flow over
Chapter 8 Coupling Atmospheric and Fire Models 283

FR7CD1
W(m/s) Z=15.0(m)
lOm/s

Time =23:55 min Time =24:05 min

~^~^--^^-^^a.>i.N».\ \ \ ]

11

^ 1.0
> ^Y^'-^^^ f^^^t^ j^^^^ ^ 1
^ y^ />/ / ^ ATt'^^ ^ ^ j
0.9 > / // / / (/ \ f v ^ V ]

^•'••••••-- ^ ^ T ; - 1

0.50 0.60 0.70 0.50 0.60 0.70

Time =24:00 min Time =24:10 min

1' [ ' '^^' -1
'
• ^ - ^ - ^ - ^ • • • • 1
" ^ " ' • • ' • • • ^ • '

-> -^ —^ ^^ ^^^ >». ^i \ \ -^ -^ --^ ^ > i . >^ ^ Ni \ V

-^ -» -^ ^^>^ "^ N. \ \ \J 1^^ ^> -^-»^ Q > ^ \ \ \ j

1.1 1.1
1 ^ ^ ^^^.^ ^ *^ ^ M
E^
^ 1.0 1.0
> \^^7)/
^ ^ / / / /^ sT-^ W j
\^^ ^j ^
0.9 > / / / > ' / * ] / \ l\ ^/V \ 0.9f.\^ /(/ Jr x^ fsSK^ >1
r^(C"^j
/ / / / /
/11 - ff^ ^ j
/ / \ y^/^ " j
1 / i 1 i \ i 1 11 11 1 1 1 M 1 1 M
0.50 0.60 0.70 0.50 0.60 0.70
X(l<m) X(km)

-4
FIGURE 10 The vertical velocity (units m s~^) from a numerical simulation of an initially 400-
km-long fire line in a hyperbolic tangent wind profile where L/Q ^ ± 3 m s~\ See text for explana-
tion. Four times are shown and frames are 5 s apart. Solid outlines are for heat fluxes greater than
0.1 MW m^^. Vectors denote local winds at 15 m above ground level. (After Clark et a/., 1996b.)
284 Jenkins et ah

hills, which leads to accelerations and decelerations in wind velocities even

without the presence of a fire. A third effect is attributed to radiation and con-
vection preheating and drying the fuel ahead of the fire (Rothermel, 1972) and
to flames being brought closer to the fuel and propagating the fire by contact.
Figure 11 presents the results of an idealized simulation by Coen et al. (1998)
of a fire line propagating over a small Gaussian-shaped ridge with a relatively
sharp slope (height 200 m and half-width 300 m) and extending north-south
over the entire model domain. The hyperbolic tangent wind profile of Figure 8
was again used, but with stable atmospheric conditions below 500 m and un-

a)Time=25:0 b)Time=35:0
10 m/s

0.5 1.0 1.5 0.5 1.0 1.5

C)Time=45:0 d)Time=55:0

X (km) AX = AY = 20 m X (km)
FIGURE 11 A near-surface horizontal cross section of the vertical vorticity (shaded) associated
with a fire line (solid oudine of heat fluxes greater than 0.1 MW m"^) as it flows over a smafl ridge
at 25, 35, 45, and 55 minutes into the simulation. Ambient atmospheric conditions are a 3 m s~^
low-level, westerly (from the left), stable flow. The vectors indicate the winds at the 15 m height.
The west-east profile of the hifl is shown above the main figure.
Chapter 8 Coupling Atmospheric and Fire Models 285

Stable above. ^ These simulations demonstrated the unusual interactions that oc-
cur between a fire and its atmospheric environment when spreading over moun-
tainous terrain for the chosen atmospheric conditions. Figure 11 shows that,
as the fire moved up the slope, the right flank weakened and died, and then,
instead of continuing to propagate down the other side, the left flank veered
sharply along the ridge line. Since the flow driving the fire forward was sym-
metrical and cannot be the source of asymmetry, this fire behavior suggests that
an instability amplified some small asymmetric disturbance.
After bowing forward into the expected bow-shaped fire line (Figure 11,
time = 25 min), other interesting features appeared in this simulation. These
included a large fire-scale rotation just in the lee of the ridge (Figure 11, time =
45 min), into which flowed a string of vertical vortices with alternating sign
that formed initially at the rear edge of the active (left) flank of the fire.
The atmospheric situation for the simulation shown in Figure 11 is depict-
ing a type of air flow known as nocturnal drainage flow. At night, air cools. Cold
air is heavier than warm and can flow downslope with speeds of 3 to 20 m s~\
These flows are characterized by a shallow, stable layer of cool air spreading
quickly near the surface under a layer of less stably stratified air that is either
still or traveling in the opposite direction. Figure 11 demonstrates how com-
plex the interactions between the fire and the atmosphere can become even for
common atmospheric flows like low-humidity, nocturnal drainage flows. The
results of this simulation also raise a crucial question: the possible impossibil-
ity of predicting fire behavior when a sudden amplification of small, transient
instabilities in the flow leads to such deviant behavior.

E. SENSITIVITY OF FIRE BEHAVIOR TO FUEL

CHARACTERISTICS AND CONFIGURATION

A method for creating a more general operational atmosphere-fire model is to

use a physics, as opposed to empirically, based fire transport and behavior model
to represent the processes of forest fire combustion and spread rate. One such
model is the FIRFTEC (Linn, 1997; Linn and Harlow, 1998). The FIRETEC is
based on conservation of mass, momentum, species, and energy. It includes a

^whether an atmosphere at a location is stable or not depends on the resulting motion of a par-
cel of air that is displaced vertically from its initial position. If the parcel moves further away from
its initial position, then the atmosphere is unstable and convection occurs; if the parcel returns
toward its initial position, then the atmosphere is stable; and if the parcel does not move once dis-
placed, then the atmosphere is neutral. The magnitude of the parcel's vertical acceleration is deter-
mined primarily by the temperature or density difference between the air parcel and its environ-
ment at a given location, time, and pressure level.
286 Jenkins et al.

representation of fuel pyrolysis, turbulent transport of combustion products,

and radiative preheating of fuel due to the approaching flame front.
The FIRETEC has been coupled with the HIGRAD, an atmospheric dynam-
ics model developed to handle the sharp temperature and flow gradients encoun-
tered in the vicinity of a wildfire (Reisner et al, 1998, 2000). Using FIRETEC/
HIGRAD, Bossert et al. (2000) simulated a small portion of a 16,000-acre fire
in steep terrain (Calabasas fire, October 21 to 22, 1996, Corral Canyon, CA).
FIRETEC allows a fully three-dimensional characterization of fuel, and in this
study the dependent fuel parameters were fuel load, moisture, and type. The
results show how the fire responded to four different fuel types. Even though
the fuel load and moisture contents were variable among the four fuel types, the
numerical results were not dramatically different in three of the four cases. In
contrast, one fuel type, which had the lightest fuel load and highest moisture
content, produced radically different fire behavior. The fire never achieved suf-
ficiently high enough temperatures to sustain itself up the sloped terrain and
died after 8 minutes of simulation. This result shows that the FIRETEC model
is sensitive to the specification of fuel load and moisture. Bossert et al (2000)
also produced one simulation showing steeper topography overwhelming any
detailed specification of fuels, resulting in an equivalent of a blowup fire. Con-
vective heating of a steeper portion of the slope by hot gases from the fire's plume
caused the entire portion of the slope to be engulfed by fire after only one addi-
tional minute of simulated time.
FIRETEC was also used to consider multistorey fuel canopies involving
ground and crown fuels. Linn et al (2000) show FIRETEC/HIGRAD simula-
tions of fire sensitivity to live moisture in the canopy fuel; as expected and in
agreement with observations, high moisture in the canopy fuel and the fire was
confined to surface; low moisture in the canopy fuel and the fire crowned. These
results demonstrate the potential of using a physics-based fire behavior model
coupled with an atmospheric prediction model for exploring such phenomena
as crowning. Linn et al (2000) also simulated a grass fire in which FIRETEC/
HIGRAD caused irregular fuel consumption that produced a nonhomogeneous
fire front and burn pattern. The simulated burn pattern (Figure 12) is similar
to patterns sometimes observed in real fires: streaks of burned and unburned
fuel, perpendicular to the fire front and parallel to the ambient wind.

VIII. INFRARED OBSERVATIONS OF FIRES

There is no substitute for direct observations of fire dynamics and behavior.
Unfortunately severe fires are almost inaccessible to planned observations, and
forestry researchers are required to build their models on sometimes shaky ob-
Chapter 8 Coupling Atmospheric and Fire Models 287

FIGURE 12 Side view image of a three-dimensional simulation of a tall-grass fire in 4 ni s

bient winds blowing perpendicular to the fire front. (After Linn et a\., 2000.)

servational foundations. Theoretical descriptions of convection patterns and

behavior in fires still rely heavily on laboratory and computer simulations. This
is changing. Advances in understanding come from the application of remote
sensing instrumentation. It is now possible to probe fires from a safe distance
with Doppler radar and lidar, and most recently, infrared or IR video.
Although data describing the fire spread rate and some qualitative aspects of
wildfire behavior exist, none has revealed the very small convective time and
spatial scales in the fire. The intense vortices of various strengths and sizes in
fires have been observed both experimentally (Church et al., 1980) and in model
simulations (Heilman, 1992; Heilman and Fast, 1992; Clark et al, 1996a,b). As
discussed in the introduction, observations in one case study (McRae and Flan-
nigan, 1990) show that fire vortices can rip out and loft standing trees. A pool
fire simulation produced counterrotating vortices to characterize the structure
of rising smoke (McGrattan et al., 1996). Similar vortices were observed in fires
using lidar and radar where the scale of the observed vortices was on the order
of 100 m (Banta et al, 1992).
The fire convection is rarely more than a few meters, lasting rarely more
than a split second. It is necessary to observe fire-scale features on time scales
of 0 . 0 5 - 0 . 1 s and spatial scales of 1-3 m. The long dwell time, the low scan-
ning rate, and the spatial resolution of the Doppler radar (—10 /xm) and lidar
( ~ 3 mm) do not provide wind speed measurements on such small scales (Banta
etal, 1992).
However, current IR camera capability provides high-frequency (1/30 s) and
high resolution ( 0 . 0 5 - 0 . 1 6 m) radiant temperature images that are being used
288 Jenkins et al.

to derive wind fields to provide clues to motions on scales of the order of meters
and time scales of fractions of a second. Infrared video cameras have a temper-
ature measurement accuracy of better than ±2°C at temperatures up to 2000°C
and can resolve temperature differences of less than O.FC. The IR camera views
a shallow depth into the flame front and variabilities in the distribution of hot
soot particles provide the information necessary to derive two-dimensional im-
ages of flow Analysis of these images provides estimates of velocities in the ver-
tical and in one horizontal direction. Measurements in crown fires in the Cana-
dian Northwest Territories suggest updrafts of 10 to 30 m s~\ downdrafts of
- 1 0 to - 2 0 m s~\ and horizontal motions of 5 to 15 m s~^ throughout the fire
(Clark etfll., 1999).
IR temperatures and vertical winds are used to estimate sensible heat fluxes
from the fire. Following Eq. (1), the sensible heat flux is

¥,^c^<pv,{T,^-T,^)> (2)

where w is the vertical velocity at the location of IR temperature measurement

TjR, and Tj^ is the time and area average of Tj^. Heat fluxes calculated according
to Eq. (2) then can be compared with expected values based on fuel loading and
observed burnout times to corroborate the image flow analysis.
It is possible to infer from an IR image flow analysis a number of physical
mechanisms contributing to the fire spread rate. IR camera imagery appears to
represent such fine-scale fire structures as hairpin vortices or turbulent bursts
and the tilting of horizontal vortices leading to counter rotating convective
towers where the estimated vertical vorticity is 4-10 s~^ (Clark et al., 1999).
These fire structures determined by IR video measurements are being compared
to and used to corroborate computer simulations such as those described in the
previous section.

IX. CONCLUSIONS AND FUTURE WORK

In this chapter, we describe the coupling of an atmospheric prediction model
with an empirical fire spread model so that the fire-atmosphere is treated as a
single, dynamic system. This is a fairly recent and major advance in the model-
ing of wildfires. Unlike the wildfire models that have traditionally been used to
assess the risk, behavior, and spread of fires, this model is capable of simulating
the convective-scale fire-atmosphere circulations, including the small-scale
vortex dynamics, that play an important role in influencing a wildfire's overall
behavior, as well as fire spread rates. And unlike traditional wildfire models, the
coupled fire-atmosphere modeling approach can include the possible impacts
of evolving large-scale atmospheric forcing on the fire and vice versa.
Chapter 8 Coupling Atmospheric and Fire Models 289

There are two major problems to overcome before this approach can be used
to predict forest fire behavior. First, we must learn to faithfully represent the
physics of the fire-atmosphere system to the degree that is needed. Second, we
must develop observational systems and techniques to provide the detailed ob-
servations of fire structures that are required to improve and validate the nu-
merical results of the fire-atmosphere model. Real-time simulations of wildfire
events may be possible in the near future using newer computer technology, pro-
vided also that topography, initial weather data, and fuel data can be gathered
quickly for use in the model, and new visualization techniques be developed to
display results in real time.
We have discussed a few numerical modeling results from publications (Clark
et al, 1996a,b; Coen et al, 1998; Bossert et al, 2000; Linn et al, 2000) directed
at improving our understanding of wildfire behavior. To gather the observa-
tions to validate numerical model results from simulations like these, we have
briefly described the measurement technique of infrared video imagery. The
current analysis of IR video imagery is capable of estimating two-dimensional
wind fields and heat fluxes from the fire.
One weakness of the coupled fire-atmosphere modeling approach is the
currently necessary employment of empirical formulae for forest fire models.
The accuracy of ignition spread is dependent on the fire spread rate model,
while the heat and moisture fluxes absorbed by the air are dependent on the ac-
curacy of the fire model. Eventually forest fire models and fire spread-rate for-
mulae must be replaced with more physics-based treatments of microscale pro-
cesses within the fire and adjacent fuel. This is an extremely difficult exercise.
Forest fires have complicated chemistry, radiation, and combustion properties.
The interactions between forest fires and air flow are highly nonlinear. Diverse
scales of motion influence a wildfire, and at almost every scale the physics re-
quire fully dimensional temporal and spatial dynamics.
The ultimate goal is to develop a fully comprehensive coupled atmosphere-
fire model of forecast potential in the operational mode. This is not possible at
the present time or in the near future. The reasons are that many physical pro-
cesses in the fire-atmosphere system remain poorly understood and that com-
putational demands of such a modeling approach far exceed the capability of
present-day computer technology. The most important use today of physically
based fire behavior and transport models is to identify conditions where the
current operational empirically based forest fire models are not appropriate. Al-
though unsuitable for faster-than-real-time applications, coupling a fully physi-
cally based treatment of the fire combustion and transport with an atmospheric
prediction model has the advantage of being able to examine complex wildfire
behavior under any set of conditions. Existing fire behavior and fire spread
models are faster and easier to use than physically based models and are quite
adequate for predicting fire behavior, but only under certain circumstances.
290 Jenkins et al

Physically based models will be used to extend existing operational models

and to develop new operational models that are suitable for a wider range of
conditions.
In the meantime, present capabilities are sufficient to model components of
the fire-atmosphere system with a considerable degree of sophistication, as the
combustion and atmospheric science research communities show Given the
recent remarkable advances in computer technology, significant development
will be made as many model refinements are added.
Finally, we emphasize that convection is the dominant mechanism of fire
propagation after the incipient stage of a fire and, despite these objections to
the current state of numerical fire modeling, knowledge of the small-scale at-
mospheric circulations in and around a wildfire is the single most important
and elemental improvement that can be made to current wildfire modeling.

APPENDIX I. CIRCULATION AND VORTICITY

Circulation is defined as

C ^ (bV'dt

where the line integral of fluid velocity V is performed over a closed contour,
dl is an element length of the contour, and A is the area enclosed by the con-
tour. The line integral is performed in a counterclockwise or positive direction.
Circulation is a scalar quantity and is the macroscopic measure of rotation for
a finite area of fluid.
The vorticity vector il is defined as

fl = V X V

the curl of the velocity vector field [&*obar*{V}|ucarrhd|.63:] Here V = ui + yj

+ wfe is the three-dimensional velocity field and

^ d d d
V= ^ + ^ + ^
dx dy dz
is the three-dimensional del operator, where i, j , and k are the unit vectors di-
rected along the x, y, and z Cartesian axes. We are interested in rotation about
a horizontal axis or the vertical component of vorticity. Therefore we take

- -^ - „ -^ dv du
k'fl = k'V XV =
dx dy
The quantity ^ = k' V X V = dv/dx — du/dy is called relative vorticity (i.e.,
relative to the ground).
Chapter 8 Coupling Atmospheric and Fire Models 291
While circulation is a macroscopic measure of rotation of the fluid, vortic-
ity is a microscopic measure of rotation or "spin" of individual particles in a
fluid. Relative vorticity is defined as

du §V'dl
f lim
dx
or the circulation for an infinitesimally small horizontal area A (A —> 0).
To get a physical feeling for vorticity, we use natural coordinates and con-
sider the flow profile in Figure LI. The natural coordinate system is defined by
the orthogonal set of unit vectors t,h,h, where t is parallel to the horizontal ve-
locity at each point, h is normal to the horizontal velocity and directed so that
it is positive to the left of the flow direction, and k is directed vertically upward.
s(x, y, t) is the curve followed by a parcel moving in the horizontal plane, and
the horizontal velocity is V = Vt, where the horizontal speed V is a nonnega-
tive scalar defined by V = ds/dt.
The circulation about the circuit in Figure I.l is

C = V[8s + d{8s)] V + —8n 8s

dn

where the negative sign ( —) specifies flow direction. Also 8n sin(5j8) ~ 8n 8(3
d{8s) for small angle 8/3. So

av
C = V8s + Vd(8s) - V8s 8n8s

dV
= V8n8l3 8n8s
dn

\ 8s dnj

FIGURE I.l Circulation for an infinitesimal loop in the natural coordinate system.
292 Jenkins et al.

(a) (b)

FIGURE 1.2 Two types of two-dimensional flow: (a) the rate of change of wind speed normal to
the direction of the flow, and (b) the turning of wind following the motion of the air parcel. In (a), a
pinwheel in the flow spins cyclonically. In (b), a pinwheel in the flow spins anticyclonically.

But R8s = 6j8 where R = radius of curvature following the parcel motion
so that

C = 8n 8s
dn
Since the relative vorticity is defined as

^ = hm -

we have
C V dV_
lim
8n 8s->0 8n 8s R dn
where V/R is curvature vorticity and dV/dn is shear vorticity. Curvature vortic-
ity is the turning of the wind along the parcel motion. Shear vorticity is the rate
of change of wind speed normal to the direction of the flow. Figure 1.2 illus-
trates both shear (a) and curvature (b) vorticity. When a small paddle wheel is
placed in the flow in Figure 1.2a, it turns in the counterclockwise or cyclonic
direction. When a small paddle wheel is placed in the flow in Figure 1.2b, it
turns in the clockwise or anticyclonic direction. Vorticity ^ is positive for cy-
clonic motion and negative for anticyclonic motion.

APPENDIX II. DEVELOPMENT OF VERTICAL

ROTATION IN A FRICTIONLESS FLUID
The material and concepts discussed in Section II of this chapter are based on
the following mathematical representation of vorticity dynamics. In a nonrotat-
Chapter 8 Coupling Atmospheric and Fire Models 293

ing reference frame (i.e., the Coriolis force due to the Earth's rotation is ne-
glected) , conservation of momentum is expressed as

dV dV ,^ „ , ^ 1
^ — + (V.V)V=--Vp-.»e

in a frictionless (or inviscid) fluid. Here V = ui + yj + wHs the three-dimen-

sional velocity field and

„ a a a
V= — + — + —
dx dy dz
is the three-dimensional del operator, where i, j , and k are the unit vectors di-
rected along the Cartesian x, 3/, and z axes, t is time, p is density, p is pressure,
and g is the acceleration due to gravity. The force due to friction is neglected.
The vector identity

^ ^ /V'V\ -^ -^
( y . V)V - V( 1- V X V XV

is used to rewrite the conservation of momentum equation as

av /v-v\ -> -^ Vp
_ . _ v ( — ) +vxa-J-<,k (I1.A,
where ft = V X V. Taking the curl (V X) of Eq. (11.A), and recalling that the
curl of the gradient vanishes, gives

aft ^ ^ Vp
^Vx(VXft)-VX-^ (Il.B)

Let ^ ==feX ft = fe • V X V, the vertical component of the vorticity (Appen-

dix 1). Taking k • of Eq. (Il.B) gives

— = fe.VX (VX ft) - k'V(-) X Vp (ll.C)

dt \p J
the vertical vorticity equation. After some algebraic manipulation, it is possible
to rewrite Eq. (ll.C) as

at dz

\ dz J p \dx dy dy dx
294 Jenkins et a\.

where V^ = ui + v/e is the two-dimensional horizontal wind and Vj^ is the

horizontal gradient in the x-y plane. Each term in Eq. (II.D) has a physical
interpretation.
Term d^ldt is the local time change of vertical vorticity. Rather than attach
our coordinate system to moving fluid parcels, we determine the rate of change
of t, for the fluid at a fixed point or location on the Earth's surface. Each term
on the right-hand side of Eq. (11.D) contributes to the local rate of change of ^.
Term — V^ • V,^^ is the advection of vorticity by the horizontal wind. For ex-
ample, wind blowing from a region of positive vorticity toward a region of
negative vorticity contributes positively to the local vorticity change. Term
— \^{d^ldz) is the advection of vorticity by the vertical wind, and its effect is
similar to horizontal advection of vorticity except that it redistributes vorticity
f in the vertical direction.
Term — f V^ • Vj^ represents changing vorticity due to horizontal divergence.
It is the fluid analog of the change in angular velocity resulting from a change
in the moment of inertia of a solid body when angular momentum is conserved.
If there is positive horizontal divergence, the area enclosed by a chain of fluid
parcels increases with time, and if the circulation is to be conserved, the aver-
age vorticity of the enclosed fluid must decrease. This term is called the diver-
gence or stretching term. If it is assumed that the three-dimensional flow is non-
divergent, V • V = 0, then
d\M du dv
dz dx dz
Substituting this into the divergence term gives
„ ^ dw
dz
Vertical motion is responsible for the "stretching" effect. When vertical motion
narrows the diameter of the area enclosed by the chain of fluid parcels (hori-
zontal convergence), vorticity increases.
Term k • {dVjJdz X V,^w) represents the generation of vertical vorticity by
the tilting of horizontally oriented components of vorticity into the vertical by
a nonuniform vertical motion field. This term is called the twisting or tilting
term and is illustrated in Figure 1 and described in Section II.
T e r m - f e - V ( l / p ) X Vp - {l/p^)[{dp/dx){dp/dy) - (ap/ay)(ap/ax)] iscalled
the solenoidal term. It is illustrated in Figure 2 and described in Section II. To
illustrate possible production of vertical vorticity by the solenoidal term, con-
sider what happens to a small rectangular parcel or bar of fluid in the situation
depicted in Figure 2. Imagine a pressure gradient in the fluid, perpendicular to
the parcel and uniform along its length; in this case, the parcel moves through
Chapter 8 Coupling Atmospheric and Fire Models 295

the fluid from high to low pressure. Now imagine that the fluid density is not
constant, but is larger at one end of the parcel than the other. The forces Fy^ and
Fy^ can be written

Since

^'>''^' };^i
then

and the torque applied by the pressure gradient forces, Fy^ and Fy^, causes the
fluid bar to rotate counterclockwise. Thus when the gradient of pressure Vp
and the gradient of density Vp are not parallel, the parcel experiences a torque
that causes it to rotate, producing vorticity.
The advection terms distribute existing ^, while the stretching or divergence
term "spins up" or "spins down" existing vorticity. Only the tilting and sole-
noidal terms generate vertical vorticity ^ in Eq. (II.D) in an inviscid fluid. And
although friction in the fluid is not included in this discussion, turbulent or fric-
tional drag can be both a source and sink of local rotation.

A. VERTICALLY ROTATING CONVECTIVE

CELLS IN A MEAN W I N D FIELD

Imagine a flow field as in Figure II. 1 consisting of a single convective updraft

embedded in a background flow U which is a function of height z only and west-
erly in the positive i direction. Let

11 = —-j + (t'(x,y, z, 0 , V = U{z)i + V'(x,y, z, t),

dz

where it is assumed for simplicity that density does not vary in the horizontal
and is a function of height only. The overbar (~) denotes the background or
base state, and the prime C) denotes fluctuations due to convection or pertur-
bations from the background or base state. Substituting fl, V, p, and p into the
right-hand side of Eq. (II.D) gives
296 Jenkins et al.

FIGURE II. 1 Development of rotation in a single convective updraft (shown by arrows) along
a fire front embedded in a background with mean wind u decreasing with height (shown by wind
arrows in the upper left corner). Heavy solid lines show vortex lines with sense of rotation shown
by circulation arrows. Plus and minus signs indicate cyclonic and anticyclonic rotation caused by
vortex tube tilting.

dC
VL
dt - -n-v,r -u dx dz
(II.E)
dw' dU ^
+ + k- an
X VhW'
dy dz dz
For this flow field, there is no generation of vertical vorticity by the sole-
noidal effect, and in Eq. (II.E) vertical vorticity is produced only by tilting the
horizontal vorticity into the vertical:

dW dU dVh „ ,\ dw' dU dw' du' dw' dV

+ k' — - X Vhw' = +
dy dz dz J dy dz dy dz dx dz
The terms on the right-hand side of this expression are, respectively, the tilting
by the perturbed differential vertical motion field {dw'/dy, dw'/dx) of linear hori-
zontal vorticity {dU/dz) and perturbed horizontal vorticity {du'/dz, dv'Idz).
For forced convection, such as in a wildfire, the perturbation portion of the
dependent variables can be as large or larger than the mean or basic state por-
tion. W h e n perturbation variables and products of perturbation variables domi-
nate, the effect by the tilting of horizontal vorticity of the base state, dU/dz, is
Chapter 8 Coupling Atmospheric and Fire Models 297

negligible compared to the tilting of horizontal shear in the convective flow,

du'ldz and dv'/dz-
Say fire convection produces vertical wind shears where du'/dz > 0. This
happens for east winds (u' < 0) stronger at the surface, decreasing with height,
or west winds (u' > 0) weaker at the surface, increasing with height. This posi-
tive vertical wind shear of u' coupled with dWIdy > 0 south of the updraft core
and dw'/dy < 0 north of the updraft core produces—once tilting of horizontal
vorticity has occurred—two centers of vertical vorticity. The one south of the
updraft core has counterclockwise or positive rotation; the one north has clock-
wise or negative rotation. This is similar, but opposite to, the situation depicted
in Figure 1(c) (Section II). If the perturbation product (dw'/dy){du'/dz) is the
dominate tilting term in Eq. (II.E), then d^'/dt is increased south and decreased
north of the updraft core.
Once ^' exists, both advection of the perturbed vorticity by the basic state
flow, —U{d^'/dx), and by the perturbed horizontal wind, — VJ^ • V^j^', can con-
tribute to the local time change of perturbation vorticity owing to the air mo-
tion. If a background wind U blows from a region of positive perturbation vor-
ticity (f' > 0) toward a region negative or no perturbation vorticity (^' ^ 0),
the advection of perturbed vorticity by the basic state flow contributes posi-
tively to the local perturbation change, and d^'/dt in Eq. (II.E) is increased. For
the situation depicted in Figure II. 1, the mean wind u will advect fire-generated
vorticity ^' east of the fire at low levels, and west of the fire at upper levels.
And once ^' exists, the stretching term —^'^h'^h can either increase or
decrease ^' locally for horizontal winds converging or diverging, respectively,
while the vertical advection term —w'{d^'/dz) can redistribute vorticity in the
vertical. For example, upper-level vorticity can be advected to the surface by
convective downdrafts.
Density does vary in all three directions, and the solenoidal effect in
Eqs. (II.D) and (II.E) generates perturbed vertical vorticity that also is advected
by the background wind and convectively perturbed flow and is increased or
decreased by horizontal convergence and stretching. And finally, the fluid is
not frictionless. Although friction and turbulence can increase local rotation,
they are also crucial for the destruction of local rotation.

APPENDIX III. GENERATION OF VERTICAL

MOTION IN ROTATING CONVECTIVE CELLS
Even rotating convective cells with relatively low buoyancy can produce low-
level updrafts that are as strong or even stronger than their high-buoyancy
counterparts. The reason for this is that the strength of vertical motion in ro-
tating convective cells is not only dependent on buoyancy forcing. Acceleration
298 Jenkins et a\.

in the core of a rotating convective cell is principally governed by buoyancy

and the vertical gradient of perturbation pressure.
The total buoyancy of an air parcel is due to thermal (sensible, latent, and
radiational) heating and the effect of condensate loading, while vertically di-
rected pressure gradient forces are attributed to the so-called "centrifugal pump"
effect. Section II and Appendix II describe how cyclonic and anticyclonic ver-
tically rotating vortices can develop in a fire. As soon as a vertical vortex is es-
tablished, vertical motion in the center of the vortex develops. To understand
the generation of the vertical motion, take the V • of Eq. (II.A) to yield

72/ P ,(v.v)
- -V' + V-(VX 11) (III.A)

where V-V = 0 is assumed. Terms -V^[(V-V)/2] and V(y X H) represent

dynamical forcing, and we are interested in the effect of dynamical forcing on
the pressure field in a vertical vortex. We use cylindrical coordinates (r. A, z)
centered on the axis of rotation (Figure III.l) to investigate the pressure varia-
tion in a vortex. Here A is azimuthal angle, J;^ is the unit vector in the azimuthal
direction (with positive counterclockwise), z is vertical height, r is the radial
distance from the axis of rotation, and i ^ is the unit vector in the r direction.
For pure rotation about a vertical z axis, the horizontal velocity field and
vertical component of vorticity in cylindrical coordinates are, respectively,

V VAJ and k'VL = ^ = (rvj

r dr

FIGURE III.l The cylindrical coordinate system.

Chapter 8 Coupling Atmospheric and Fire Models 299
Therefore,

VXn=^^(r,v,)i,

Assuming that the vertical scale is much larger than the radial scale, the Lapla-
cian V^ in cylindrical coordinates is approximately

r dr dr
Substituting into Eq. (III.A), the dynamical component of the pressure pertur-
bation in the vortex is expressed as
1 a / r dpdyn^ r a(vi/2)i 1 a
)1----
r + — _v.-(rv,)_
r dr\po dr ar r ar
(IIl.B)

r dr
where p = Po and is considered constant in the radial direction. Integrating
Eq. (III.B) with respect to r gives

1 ^Pdyn vl
Po dr
an equation for cyclostrophic balance, where the pressure gradient force
{l/po){dp^yJdr) is always directed inw/ard, toward the center of rotation, while
the centrifugal force v\/r is always directed outward. The low pressure at the
center of (either clockwise or counterclockwise) rotation provides a pressure
gradient force to balance the outward centrifugal force.
This dynamically generated low pressure at in the center of the vortex pro-
duces a vertically directed pressure gradient force. Since air is accelerated from
high to low pressure, the result is the development of an updraft from below,
or a downdraft from above, in the rotating core. This is the centrifugal p u m p
effect.

NOTATION

ROMAN LETTERS

Cp specific heat capacity of dry air at constant p J kg-' K-'

g magnitude of acceleration due to gravity ms"^
300 Jenkins et al

p pressure Pa
s distance along a parcel trajectory m
t time s
u, V, w wind components in x (eastward), 3/ (north- m s~^
ward), and z (vertical) directions
Vj rate of spread at the head of a wind-aided fire m s~^
x,y, z distances in the east-west, n o r t h - s o u t h , and m
vertical directions; Cartesian coordinates paral-
lel to, and perpendicular to, the ground
C fluid circulation m^s~^
F a force kg m s~^
Fg sensible heat flux W m"^
Fy convective Froude number
T absolute temperature K
Ti^ infrared T measurement K
U ambient air wind speed ms~^
UQ relative speed of air passing over the fire m s~^
V speed in natural coordinates m s~^
V fluid velocity ms~^
Wj width of region of intense heating m
T burn time of small fuel s
At time step s
Aw w anomaly over region of intense heating; m s~^
change in w over A:^ ms~^
Axji^gj fuel grid cell size in horizontal direction m
Ax, Ay, A:^ distances or grid cell sizes in x, y, and z m
directions
AT T anomaly over region of intense heating K
^ vertical component of relative vorticity s~^
p air density kg m~^
fl vorticity vector s~^
V three dimensional gradient or del operator m~^
r, j , k unit vectors along x, y, and z axes
Chapter 8 Coupling Atmospheric and Fire Models 30 i

CYLINDRICAL COORDINATES

t, n unit vectors normal and parallel to fluid parcel

trajectory
jS angular direction of the wind or fluid flow
R radius of curvature following the parcel motion
A, r asimuthal angle, radial distance from axis
Jx, ix unit vectors in the asimuthal and r directions
Ux velocity component in the asimuthal direction

SPECIAL NOTATION

a background or base state; time and area average

<> horizontal average or mean value
' fluctuations or perturbations from a background
or base state; fluctuations or perturbations from
the horizontal average
^ unit vector

REFERENCES
Baines, P. (1990). Physical mechanisms for the propagation of surface fires. Math. Comput. Model-
ing, 13,83-94.
Banta, R. M., Olivier, L. D., Holloway, E. T., Kropfli, R. A., Bartram, B. W., Cupp, R. E., and Post,
M.J. (1992). Smoke column observations from two forest fires using Doppler lidar and Doppler
radar. J. Appl Meteorol 31, 1328-1349.
Bossert, J. E., Linn, R. R., Reisner, J. M., Winterkamp, J. L., Dennison, P., and Roberts, D. (2000).
Coupled atmosphere-fire behavior model sensitivity to spatial fuels characterization. Preprints
of "3rd Symp. Fire and Forest Meteorology," Amer. Meteor. Soc. 80th Annual Meeting, Los An-
geles, CA, pp. 21-26.
Byram, G. M. (1973). Combustion of forest fuels. "Forest Fire, Control and Use" (A. A. Brown and
K. P Davis, Eds. 2nd ed., pp. 155-182. McGraw-Hill, New York.
Chandler, C , Cheney, P., Thomas, P., Trabaud, L., and Williams, D. (1983). "Fire in Forestry For-
est Fire Behavior and Effects." John Wiley and Sons, New York.
Church, C. R., and Snow, J. T. (1979). The dynamics of natural tornadoes as inferred from labora-
tory simulations. J. Rech. Atmos. 12, 111-133.
Church, C. R., Snow, J. T., and Dessens, J. (1980). Intense atmospheric vortices associated with a
100 MW fire. Bull. Amer. Meteor. Soc. 61, 682-694.
Clark, T. L., Radke, L., Coen, J. L., and Middleton, D. (1999). Analysis of small-scale convective dy-
namics in a crown fire using infrared video camera imagery. J. Appl. Meteorol. 38, 1401-1420.
Clark, T. L., Jenkins, M. A., Coen, J., and Packham, D. R. (1996a). A coupled atmospheric-fire
model: Role of the convective Froude number and dynamic fingering at the fire line. Int. J. Wild-
land Fire 6(4), 177-190.
302 Jenkins et al.

Clark, T. L., Jenkins, M. A., Coen, J., and Packham, D. R. (1996b). A coupled atmospheric-fire
model: Convective feedback on fire line dynamics. J. Appl. Meteorol. 35, 875-901.
Coen, J. L., Clark, T. L., and Hall, W. D. (1998). Simulations of the effect of terrain on fire behav-
ior: Experiments using a coupled atmosphere-fire model. Preprints of "2nd Symp. Fire and
Forest Meteorology," Amer. Meteor. Soc, Phoenix, AZ, pp. 87-90.
Corlett, R. C. (1974). Fire violence and modeling (Chapter 6). In "Heat Transfer in Fires: Thermo-
physics. Social Impacts, Economic Impact." Scripta Book Company, Hemisphere Publishing
Corporation, Washington, DC.
Gall, R. L. (1982). Internal dynamics of tornado-like vortices. J. Atmos. Sci. 39, 2721-2736.
Heilman, W. E. (1992). Atmospheric simulations of extreme surface heating episodes on simple
hills. Int. J. Wildland Fire 2, 99-114.
Heilman, W. E., and Fast, J. D. (1992). Simulations of horizontal roll vortex development above
lines of extreme surface heating. Int. J. Wildland Fire 2, 55-68.
Kim, J., and Moin, P. (1986). The structure of the vorticity field in turbulent channel flow. Part 2.
Study of ensemble-averaged fields. J. Fluid Mech. 162, 339-363.
King, A. R. (1964). Characteristics of a fire-induced tornado. Australian Meteorological Magazine
No. 44, pp. 1-9.
Linn, R. R. (1997). "A Transport Model for Prediction of Wildfire Behavior." Los Alamos National
Laboratory thesis LA-13334-T, New Mexico State University.
Linn, R. R., and Harlow, F. H. (1998). FIRETEC: A transport description of wildfire behavior. Pre-
prints of "2nd Symp. Fire and Forest Meteorology," Amer. Meteor. Soc. 78th Annual Meeting,
Phoenix, AZ, pp. 14-19.
Linn, R. R., Bossert, J. E., Harlow, R, Reisner, J. M., and Smith, S. (2000). Studying complex wildfire
behavior using FIRETEC. Preprints of "3rd Symp. Fire and Forest Meteorology," Amer. Meteor.
Soc. 80th Annual Meeting, Los Angeles, CA, pp. 15-20.
McGrattan, K., Baum, H. R., and Rehm, R. G. (1996). Numerical simulation of smoke plumes from
large oil fires. Atmos. Environ. 30(24), 4125-4136.
McRae, D. J., and Stocks, B.J. (1987). Large-scale convection burning in Ontario. "Proc. 9th ConL
on Fire and Forest Meteorology," San Diego, pp. 23-29.
McRae, D. J., Stocks, B. J., and Ogilvie, C. J. (1989). Fire acceleration on large-scale convection
burns. "Proc. 10th Conf. on Fire and Forest Meteorology," Ottawa, pp. 101-107.
McRae, D. J., and Flannigan, M. D. (1990). Development of large vortices on prescribed fires. Can.
J. For. Res. 20, 1878-1887.
Moin, P., and Kim, J. (1982). Numerical investigation of turbulent channelflow.J. Fluid Mech. 118,
341-377.
Radke, L. F., Clark, T. L., Coen, J. L., Walther, C , Lockwood, R., Riggan, P. J., Brass, J., and Hig-
gins, R. (2000). The WildFire Experiment: Observations with Airborne Remote Sensors. Can.
J. Remote Sensing 26(5), 406-417.
Reisner, J. M., Bossert, J. E., and Winterkamp, J. L. (1998). Numerical simulations of two wildfire
events using a combined modeling system (HI-GRAD/BEHAVE). Preprints of "2nd Symp. For-
est and Fire Meteorology," Amer. Meteor. Soc. 78th Annual Meeting, Phoenix, AZ, pp. 6-13.
Reisner, J. M., Swynne, S., Margolin, L., and Linn, R. R. (2000). Coupled-atmosphere fire model-
ing using the method of averaging. Mon. Wea. Review, 128, 3683-3691.
Rothermel, R. C. (1972). "A Mathematical Model for Predicting Fire Spread in Wildland Fuels."
Research Paper INT-115. USDA Forest Service, Intermountain Forest and Range Experiment
Station, Ogden, UT.
Strauss, D., Bednar, L., and Mees, R. (1989). Do one percent of forest fires cause ninety-nine per-
cent of the damage? For. Sci. 35(2), 319-328.
Weiss, D. R., and Biging, G. S. (1996). Effects of wind velocity and slope on flame properties. Can.
J. For. Res. 26, 1849-1858.
 CHAPTER 9
Surface Energy Budget
and Fuel Moisture
KENNETH E . KUNKEL
Illinois State Water Survey, Champaign, Illinois

I. Introduction
II. Evapotranspiration Processes and the
Meteorological Controlling Factors
A. Radiation
B. Atmospheric Water Vapor Content
C. Temperature
D. Vertical Mixing (Wind and Stability)
III. Estimation of Potential Evapotranspiration Rates
A. Penman-Monteith Method
B. Radiation Models
C. Pan Evaporation
D. Temperature Models
IV. Functional Dependence of PET and AFT
V. Characteristics of PET
A. Seasonal Cycle
B. Elevation Dependence
C. Topographic Effects
D. Understory ET
VI. Near-Surface Environment
VII. Models of Land-Surface Interactions
VIII. Remote Sensing of the Surface Energy Budget
IX. Fire Weather Rating Systems
A. United States
B. Canada
Notation
Suggested Reading List
References

Forest Fires
Copyright © 2001 by Academic Press. All rights of reproduction in any form reserved. 303
304 Kenneth E. Kunkel

I. INTRODUCTION
The effect of weather on fuel moisture has traditionally been estimated using sta-
tistically based empirical relationships that are a function of temperature and
relative humidity (Simard, 1968; Van Wagner, 1987). These relationships have
provided a useful approach to estimate the danger of forest fires. However, be-
cause they are not based on fundamental physical principles, it may not always
be appropriate to use them as a tool to study fuel-drying processes. There is no
assurance that these relationships will hold in a general sense. In fact, the differ-
ences in systems of Canada, the United States, and other countries suggests that
any particular set of these relationships may not be generally applicable to the
other regions. In Chapter 4 in this book, a detailed description of the principles
governing moisture within fuels is provided. This chapter builds on that dis-
cussion by describing the broadly applicable physical principles of evaporation
as it relates to fuel drying. It is hoped that this information can be used by the
ecological researcher to more accurately describe the fuel conditions that are
encountered in field studies and to identify the important processes and mecha-
nisms responsible for those fuel conditions.
The moisture content of forest fuels is dependent on a number of factors. This
chapter will examine the meteorological factors that influence variations in fuel
moisture content. Specifically, the meteorological factors that affect evaporation
rates are discussed in detail. A number of methods for estimating evapora-
tion rates are described, with a primary emphasis on the widely used Penman-
Monteith formulation. This formulation is used to describe the functional de-
pendence of evaporation rates on important meteorological variables. Also, the
state of the atmosphere in close proximity to the Earth's surface and its impor-
tance to the environmental conditions experienced by fuels are discussed. Fi-
nally, the chapter discusses state-of-the-art models that are used in atmospheric
circulation models to describe the complex interactions between the land sur-
face and the atmosphere. Such models may be of use to the ecological researcher
in describing evaporative drying of fuels.
In living fuels, foliage and twigs play an important role in forest fire be-
havior and can exhibit significant fluctuations in moisture content (Chandler
et al, 1983). The moisture content is determined by the relative balance be-
tween evaporative losses and supply of water from the root system. Water is lost
through transpiration, which is controlled by meteorological factors and the de-
gree of stomatal opening. Water uptake by the roots is controlled by soil char-
acteristics, soil moisture content, and the vertical distribution of roots. When
soil moisture is high, water loss is approximately balanced on a daily time scale
by water uptake. When soil moisture is deficient, root uptake of water is re-
stricted and may not balance transpiration losses, leading to gradual decreases
Chapter 9 Surface Energy Budget and Fuel Moisture 305

in fuel moisture content. The diurnal cycle of transpiration is of large ampli-

tude, peaking in early afternoon, and this leads to changes in fuel moisture con-
tent during the day (Chandler et al, 1983).
The moisture content of dead fuels is controlled by different processes (Pyne,
1984), although a similar set of meteorological factors provide the atmospheric
control on these processes. Moisture cannot be replenished by root uptake.
Moisture replenishment can occur by absorption through direct contact with
liquid water. Moisture exchange in the gaseous phase depends on the fuel's hy-
groscopic properties and is driven by the difference in water vapor pressure be-
tween the fuel particle interior and the atmosphere in immediate contact with
the fuel. Fine fuels, such as grass, leaves, needles, and ground litter, can rapidly
change moisture content over time scales of hours. Medium (sticks, branches)
and coarse (logs) fuels change moisture more slowly, on time scales of weeks
to months for medium fuels and months to years for coarse fuels (Brown and
Davis, 1973). For these fuels, the persistence of meteorological conditions is an
important factor in determining fire risk.

II. EVAPOTRANSPIRATION PROCESSES

AND THE METEOROLOGICAL
CONTROLLING EACTORS
The evaporation of water from the soil or from the surface of living or dead fu-
els or the transpiration of water through leaves will be referred to herein by the
common term "evapotranspiration" (ET). Fundamentally, ET can occur when
the water vapor pressure of the air layer immediately adjacent to the surface
containing liquid water is lower than the water vapor pressure of the liquid wa-
ter. If the adjacent air layer and liquid water surface are in thermal equilibrium
(often the case), then ET can occur when the water vapor pressure of the adja-
cent air layer is below its saturated value. The saturation water vapor pressure
is a function of temperature.
The latent heat of vaporization of water is high, with a value of 2454 J g~^
at 20°C, varying slightly with temperature. The rate of ET is often limited by
availability of energy. In a closed system, an unsaturated near-surface air layer
will be moistened and cooled by ET and quickly reach saturation, stopping the
process of ET. In the atmosphere, ET is sustained principally by two processes.
The first process is the absorption of electromagnetic radiation at the surface,
which raises the temperature of the liquid water and the adjacent air. Thus, the
value of saturation of the adjacent air is increased, in essence increasing the ca-
pacity of the air for further evaporation. The second process is ventilation of the
surface-air interface, which replaces the moistened air near the interface with
306 Kenneth E. Kunkel

drier and/or warmer air. Several meteorological variables are of particular im-
portance in determining the magnitude of these two processes. These are short-
wave and longwave radiation, temperature, atmospheric water vapor content,
and vertical mixing.

A. RADIATION

The primary source of energy for ET is electromagnetic radiation from the sun
which is concentrated principally in the visible and near-infrared (shortwave)
portion of the spectrum, at wavelengths less than 3 X 10~^ m. The amount
of radiation that is available for ET is determined by several processes, as de-
scribed later.
The amount of incoming solar radiation at the top of the atmosphere varies
with latitude and season based on the E a r t h - s u n geometry. Solar radiation is
attenuated as it passes through the atmosphere by direct absorption and scat-
tering by gases, particulates, and, most importantly, clouds. Globally averaged,
about 54% of the solar energy at the top of the atmosphere reaches the Earth's
surface (Salby, 1996). At the surface, some of the solar radiation is reflected back
to space. The reflectivity of the surface varies considerably with surface type
and is represented by a parameter called the albedo, which is the fractional part
of the radiation that is reflected. Except for snow-covered surfaces, the albedo of
most surfaces is less than 0.3. For example, Betts and Ball (1997), in the BOREAS
experiment (Sellers et al, 1997), measured summertime albedos of 0.20 over
grass, 0.15 over an aspen forest, and 0.083 over a coniferous forest.
The rate of emission of radiation by matter is governed by the Stefan-
Boltzmann law,
I = Eo-T^ (1)
where J = energy flux ( W m ~ ^ ) , s = emissivity, a = Stefan-Boltzmann con-
stant (5.67 X 10"^ W m~^ K""^), and T = temperature of the matter (K). At the
range of temperatures of the atmosphere and the earth's surface, this emission
is in the infrared (longwave) portion of the spectrum, mostly at wavelengths
> 3 X 10"^ m. Most land surface types have high emissivities, with forests in
the range of 0.97—0.99 (Oke, 1978). Although the cloud-free atmosphere is
nearly transparent at visible wavelengths, it is semiopaque at infrared wave-
lengths and absorbs and reemits a significant amount of infrared radiation to-
ward the surface. The magnitude of infrared radiation emitted by the cloud-free
atmosphere is largely determined by the temperature and water vapor content
in the lowest portion of the atmosphere. However, clouds have a much higher
emissivity than the cloud-free atmosphere. Thus, the amount and type of cloud
cover is an important factor in determining the magnitude of atmospheric in-
Chapter 9 Surface Energy Budget and Fuel Moisture 307

frared radiation. The upward longwave radiation, J^, above the Earth's surface
is the sum of the emission by the Earth's surface, governed by Eq. (1), and the
reflection of downward longwave radiation from the atmosphere, J^. Usually,
the absorptivity (1 — reflectivity) of the Earth's surface is assumed to be equal
to the emissivity. Thus,

I, = s,aT: + (1 - eX (2)
where s^ = emissivity of the surface and T^ = temperature of the surface (K).
A portion of the electromagnetic radiation that reaches the Earth's surface is
transformed into nonradiative forms of energy, including latent heat through
the ET process, heat storage in the soil and biomass, and direct heating of the
atmosphere through conduction. The amount of radiation available for trans-
formation into these other forms of energy is called the net radiation (R„) and
is defined as

R, = S,(l -a) +1,-1, (3)

where S^ = downward solar radiation at the Earth's surface (W m"^) and a =
albedo. Eigure 1 shows the diurnal evolution of R„ on a typical sunny sum-
mer day in the central United States, along with the individual components in
Eq. (3), as measured by a SURERAD station (see discussion that follows).
There is a strong diurnal dependence to R„, with the magnitude peaking near
midday and dropping to negative values at night, primarily in response to the

Bondville, IL - June 2,1998

110a

900J

^ 700H

500

300

100H

-100
^ ^ ^ ^ ^
< < a: Q:
<
CJ CD CVl CD OJ

Time (C.S.T)
FIGURE 1 Daily variation in the components of the surface radiation budget at the SURFRAD
site at Bondville, Illinois, on June 2, 1998. Components are R„ (medium solid), S^ (thick solid), aS^
(thin solid), I, (thin dashed), and I^ (thick dashed).
308 Kenneth E. Kunkel

variation of incoming solar radiation. Infrared radiation (J^, J^) also varies
through the day, although with a lesser amplitude than solar radiation.
The daily accumulated radiative energy available to be transformed into other
forms of energy, found by integrating R„, is positive in Figure 1. The magnitude
of the daily integrated R^ varies with latitude, the seasonal cycle, and meteoro-
logical conditions (particularly cloud cover). However, the climatology of R„ is
not well specified because the measurement of R„ is technically difficult; data
have typically been obtained only for short duration field experiments. Even
in carefully controlled experiments, commercially available sensors have been
found to exhibit biases (Hodges and Smith, 1997). To address the need for cli-
matic observations of radiation, the Surface Radiation Budget Network (SURF-
RAD) was established in the United States in 1993 by the U.S. Department of
Commerce's National Oceanic and Atmospheric Administration. The SURF-
RAD is providing accurate, long-term, continuous measurements of the surface
energy budget at six sites in climatologically diverse regions of the United States.
Figure 2 shows the annual cycle in 1997 of the daily average values of the radi-
ation budget components at Goodwin Creek, Mississippi. All components ex-
hibit a pronounced seasonal cycle, with values peaking in the summer. Daily R„
peaks at about 13 MJ m~^ in July and falls to a minimum of less than 2 MJ m~^
in December. Daily integrated values of J^ and J^ are much higher than S^ be-
cause infrared radiation occurs at night as well as during day. However, these
largely cancel, and the net infrared radiation (J^ — J j is smaller than the net
shortwave radiation throughout the year.

Goodwin Creek, MS-1997

Month
FIGURE 2 Annual cycle of the daily average values of the components of the surface energy bud-
get at Goodwin Creek, Mississippi, during 1997. Components are R„ (medium solid), S^ (thick
sohd), aSd (thin solid), J, (thin dashed), and I^ (thick dashed).
Chapter 9 Surface Energy Budget and Fuel Moisture 309

Fort Peck, MT-1997

50n

's
-^40-
E / —
' _ ^^
•D 30- ^- \^
-D ^ ^ 'a ^

^ ^^"^
X 20-
3
^ ^ ^ '- — -^ ^ ^^^^"^^
>^ ^^"""^'^^"^ Rn
P10-
CD
c
LU
asT*"^
.. • • — 1 1 1 1 1 1 1 " l ., 1 1
U

-10-
CO ^ ZJ OJ Q.
LJ. ^
CD
^ ^ -> 5 (^
CO O
Month
FIGURE 3 Annual cycle of the daily mean values of the surface energy budget components dur-
ing 1997 at Fort Peck, Montana. Components are R„ (medium solid), S^ (thick solid), aS^ (thin
solid). Is (thin dashed), and I^ (thick dashed).

Figure 3 shows the radiation components for a more northern location:

Ft. Peck, Montana. The seasonal cycle is qualitatively similar, but there are some
significant differences. Incoming solar radiation peaks in summer at about
24 MJ m~^, slightly higher than in Mississippi. This is a consequence of longer
days and lower cloud cover at Ft. Peck, which offsets the higher solar elevation
angles at Goodwin Creek. The amplitude of the annual cycle is higher at the
more northern Montana location. The rather high values of reflected solar radia-
tion (aS^ in January, February, and March, reaching a peak of about 10 MJ m~^
in March, are a result of snow cover and the high albedo associated with snow
cover. The peak value of R„ in summer is lower in Montana than in Mississippi.
A comparison of Figures 2 and 3 indicates that this is primarily a result of lower
values of incoming longwave radiation at the Montana site, probably due to
lower water vapor content and lower air temperatures. During the cold season,
net radiation is negative at Ft. Peck due to high values of reflected shortwave
radiation.
There are no comparable long-term SURFRAD sites over a forest canopy.
However, short-term experiments indicate that the radiation budget over a for-
est canopy is qualitatively similar to what is shown in Figures 2 and 3. For
example, McCaughey et aZ. (1997) measured the longwave components of the
surface energy budget over a jack pine {Pinus banksiana) forest in Saskatchewan,
Canada, during the warm season. Net longwave radiation was very similar to
what is seen in Figure 3 for Fort Peck, Montana.
310 Kenneth E. Kunkel

At ground level within a forest canopy, net radiation is reduced sharply be-
low that at the top of the canopy because of shielding by the canopy. The mag-
nitude of the reduction in R„ depends upon tree density, tree architecture, and
the season. J. M. Chen et a\. (1997) measured net radiation in a relatively sparse
aspen (Foipulus tremuloides) forest. Net radiation at ground level was about half
of the above-canopy value before leaf emergence and about a quarter of the
above-canopy value after leaf emergence. Baldocchi and Vogel (1996) measured
net radiation in a deciduous forest and a boreal jack pine (Pinus banksiand) for-
est. Net radiation at the forest floor in the deciduous forest was less than 10%
of the value above the canopy. Net radiation at the floor of the jack pine forest
was 10-15% of the above-canopy level. Black and Kelliher (1989) measured net
radiation in a Douglas fir (Pseudotsuga menziesii) stand and found that below-
canopy net radiation was 13-16% of the above-canopy levels. These selected
examples suggest that the available energy for evapotranspiration at the forest
floor is usually a small fraction of what is available at the top of the canopy.

B. ATMOSPHERIC WATER VAPOR CONTENT

The saturation value of atmospheric water vapor content is a highly nonlinear

function of temperature. This dependence can be expressed as (Buck, 1981)

es = Aexp[BT/(C + T)] (4)

where es = saturation value of water vapor pressure (hPa), T = temperature

(°C), and A, B, and C are constants. Over water, the values of these constants
are A = 6.1121, B = 17.502, and C = 240.97. Over ice, the values are A =
6.1151, B = 22.452, and C = 272.55. A graphical representation of Eq. (4)
is shown in Figure 4, over water and over ice. The value of es approximately
doubles for every 10°C increase in temperature. At temperatures below 0°C,
there are small differences in es over water and over ice.
Evaporation rates in the air layer adjacent to a flat water surface are propor-
tional to the difference (es — e), where e = actual water vapor pressure, when
water is freely available and the water surface is in thermal equilibrium with the
adjacent air layer. This difference is referred to as the water vapor pressure defi-
cit. Thus, evaporation rates are highly dependent on temperature because of the
relationship of es to T.
Observations of atmospheric water vapor content are usually expressed in
terms of dew point temperature (T^, °C) or relative humidity (RH, %). The ac-
tual water vapor pressure (e) is related to T^ by

e = A exp[BTy(C + T^)] (5)

Chapter 9 Surface Energy Budget and Fuel Moisture 311

ou-

70-
A /\
CO
CL
60- y
^ /
O X
3
(0
CO
50- y
Q)
Q.
40- y^
O
Q. ^y
CO
30- v ^
>
C
o ^^^
^^^
2 20- ^^y^
3
CO
CO 10-
^ ^ ' ^
0- 1 \ \ \ \ \ \ \ \ 1 \
-20 -15 -10 -5 5 10 15 20 25 30 35 40
Temperature (°C)

0)
Q.

O
Q.
CO
>

CO
CO

-10
Temperature (°C)
FIGURE 4 Dependence of the saturation water vapor pressure (hPa) on temperature (°C) (A) over
water for T > -20°C and (B) over water and ice for T < 0°C.

Relative humidity is defined in terms of tlie mixing ratio (v), which is the ratio
of the mass of water vapor to the mass of dry air in a volume. The definition is
RH(T) = 100% yhs(l) (6)
where vs (T) is the saturation mixing ratio at the air temperature T. The rela-
tionship of vapor pressure to mixing ratio is (List, 1949)
0.62197 e/(P - e) (7)
312 Kenneth E. Kunkel

where P = air pressure (hPa). Relative humidity can then be expressed in terms
of vapor pressure as
e(P es)
RH = 100% (8)
es{P - e)
Water vapor content can also be defined as the ratio of the mass of water
vapor to the mass of moist air in a volume, known as the specific humidity (q).
This can be expressed in terms of the mixing ratio as
q = v/(l + v) (9)
Relative humidity is strongly dependent on temperature through its depen-
dence on es. In most situations, there is little diurnal variation in e but large
variations in RH. RH usually reaches a maximum in the early morning around
sunrise when temperatures are at a minimum and falls to a minimum in mid af-
ternoon around the time of the daily maximum temperature. This behavior is
illustrated by the daily average profile for International Falls, Minnesota, in July
(Figure 5). There is relatively little variation in water vapor pressure, ranging
from 14.6 hPa in the early morning to 16.4 hPa in late morning, a total percent-
age change of 11%. By contrast, relative humidity varies from near 90% in the
early morning to near 50% in late afternoon.
Some measures of FT use relative humidity as one indicator of drying rates.
By itself, RH may be a deceptive indicator because of its dependence on tem-
perature. Figure 6 shows the dependence of the water vapor pressure deficit on

International Falls, MN - July

100
18
HgO Vapor pressure
16 Cfl
14 a.
12 0
=3
(0
10 CO
Relative humidity 8 0)

6 O
Q.
4 (T1
2 >

<
CD CD

Time of day
FIGURE 5 The diurnal cycle of relative humidity (%, solid line) and atmospheric water vapor
pressure (hPa, dashed line) during July at International Falls, Minnesota. These are averages of
hourly observations for the period 1961-1990.
Chapter 9 Surface Energy Budget and Fuel Moisture 313

10%

30%

50%

70%

90%

T 1^
5 10 15 20 30 35 40

Temperature (°C)

FIGURE 6 The dependence of the water vapor pressure deficit (hPa) on temperature (°C) for rela-
tive humidity values of 10, 30, 50, 70, and 90%.

temperature for selected values of relative humidity. The water vapor pressure
deficit, and thus the ET rates, increases exponentially with temperature, fol-
lowing Eq. (4), for constant RH.

C. TEMPERATURE

The dependence of evaporation on temperature is primarily through the de-

pendence of es on temperature. Temperature may also have indirect effects be-
cause longwave radiation increases with temperature. However, as illustrated
in Figures 2 and 3, net longwave radiation (I^ — J j is only weakly dependent
on season (temperature).

D. VERTICAL MIXING ( W I N D AND STABILITY)

Evaporation moistens and cools the air near the surface, which has a negative
feedback on evaporation rates. Mixing of air near the surface has the net effect
of transporting moist air upward and replacing it with drier and/or warmer air
from aloft, maintaining evaporation. One factor affecting the rate of mixing is
wind speed. Evaporation rates tend to increase with increasing wind speed,
other factors being equal. A second factor is the vertical air density gradient
which can enhance or suppress mixing.
314 Kenneth E. Kunkel

1. Mixing Processes
Mixing in the lowest few meters of the atmosphere is driven primarily by the
interaction of the surface with the atmosphere. The Earth's surface is aerody-
namically rough and interacts with moving air. This interaction results in the
transfer of momentum between the surface and the atmosphere. The wind speed
near the surface is reduced by this momentum transfer. Momentum transfer oc-
curs through two processes. The first is friction; air in direct contact with sur-
face elements is slowed by frictional forces. Vertical mixing processes transfer
this slow moving air stream upward away from the surface to a region of faster
moving air streams and vice versa. This exchange slows the faster moving air
and accelerates the slower moving air. The second process is through the pres-
sure force; high-pressure builds up on the windward side of surface elements
that are exposed to the wind which then slows the air upwind of the element.
This process does not involve exchange of air and momentum is transferred
without direct contact with the surface.
Mixing occurs either by molecular diffusion in laminar flow or by turbulence.
Momentum transfer to the surface results in a vertical gradient in wind speed.
This gradient is often sufficiently large that laminar flow is unstable, and turbu-
lent motion results. Turbulent mixing is much more efficient than molecular dif-
fusion. The atmospheric transport of momentum usually occurs at much higher
rates than provided by molecular diffusion because the atmosphere near the sur-
face is generally turbulent, except immediately adjacent to the surface. In a shal-
low layer (of the order of 1 cm thickness) near surface objects, proximity to the
surface prevents turbulent motion and the wind flow is laminar; the transfer of
momentum, heat, and water vapor occurs by molecular diffusion in this shal-
low, laminar-flow layer.

2. Atmospheric Stability
The second factor affecting the rate of mixing, in addition to the interaction of
the wind with the surface, is the stability of the atmosphere. Because pressure
decreases with height, the temperature of a rising parcel of air will decrease with
height at a rate known as the "adiabatic lapse rate," as governed by the gas law,
whose value is 9.8°C km~\ If temperature decreases with height at a slower
rate than adiabatic, then a rising parcel will find itself at a lower temperature
(higher density) than the surrounding environment and will tend to sink back
to its original level. This suppresses vertical mixing and the atmosphere is re-
ferred to as "stable." If the temperature decreases with height more rapidly than
adiabatic, then a rising parcel of air will find itself at a higher temperature (lower
density) than the surrounding environment and will accelerate upward. This
enhances vertical mixing and the atmosphere is referred to as "unstable." It has
Chapter 9 Surface Energy Budget and Fuel Moisture 315

been convenient within the science of meteorology to define stabihty condi-

tions through use of a quantity known as the potential temperature (0) which
is defined as
0.287
Po
e{z) = T{z] (10)
-Piz)
where z = height, T(z) has units of K, and PQ is the pressure of a reference level,
usually taken to be 1000 hPa. In essence, the potential temperature is a value
that has been adjusted for the changes in temperature that occur due to pres-
sure changes when a parcel rises or sinks in the atmosphere. When expressed
in terms of 6, the criteria for stability become

— < 0 unstable
dz

— =0 neutral (11)
dz
^^
— > 0 stable
dz
In a stable atmosphere, the character of the wind flow, whether turbulent or
laminar, is determined by a balance between the tendency for vertically sheared
flow to break down into turbulence and the suppression of vertical motions by
stable buoyancy. This characteristic is often expressed in a parameter known as
the gradient Richardson number (Ri), defined as

where g = acceleration of gravity, and V = horizontal wind speed. When Ri is

greater than about 0.25, turbulent motion ceases, and the flow becomes lami-
nar. In this case, exchange of momentum, heat, and water vapor is markedly
reduced. In the opposite case, when the potential temperature decreases with
height, the atmosphere is buoyantly unstable, and convection transfers mo-
mentum and enhances the rate of turbulent mixing.
The lowest layer of the atmosphere, where turbulence driven by momentum
and heat exchange at the surface is present, is often referred to as the planetary
boundary layer (PBL). During the daytime, when incoming solar radiation cre-
ates an unstable temperature profile, the depth of the PBL is typically 1-2 km
(Panofsky and Button, 1984). At night, under a stable temperature profile, the
PBL depth is often less than 100 m, particularly when the wind speed is low. The
lowest part of the PBL is referred to as the surface layer. The surface layer is not
precisely defined; within this layer, vertical variations of momentum, heat, and
316 Kenneth E. Kunkel

moisture fluxes are smaU (< 10%) and, as a first approximation, can be assumed
to be constant with height. As a rule of thumb, the surface layer occupies the
lowest 10% of the PBL (Panofsky and Dutton, 1984).

3. Vertical Wind Profile

The turbulent transport of momentum between the surface and the atmosphere
creates a vertical gradient in wind speed. Within the surface layer, the wind
speed typically increases with the logarithm of height, as documented by nu-
merous field experiments. Mathematically, the wind speed (U) dependence on
height (z) can be expressed as (Panofsky and Dutton, 1984)

Uiz) = fMiz - d)/zo] - 4'm} (13)

where u^ = the friction velocity (m s"^), d = displacement height (m), ZQ =

roughness length (m), k = von Karman constant (= 0.4), and il/^ is a correc-
tion factor for stability effects. The friction velocity is a measure of the rate of
momentum transport. The displacement height is a measure of the shielding of
the Earth's surface by a vegetative canopy. For short grass and short crops, d is
small or negligible. However, a dense forest canopy is very effective at shielding
the free atmospheric flow from the surface; from the viewpoint of the atmo-
sphere, the effective surface is much higher than the actual ground level. Typi-
cally, in dense crops or forests, the displacement height is 60-80% of the aver-
age height of the vegetation (Thom, 1975).
The roughness length is a measure of the efficiency of the interactions be-
tween the free atmospheric flow and the surface. For a very smooth, flat sur-
face, such as a sandy desert, the interaction between the free atmospheric flow
and the surface is minimal, and the corresponding roughness length of that
surface is also small. By contrast, surfaces that have many undulations which
provide opportunities for airflow to penetrate between the undulations and be
slowed by interactions with these undulations are much more efficient at trans-
ferring momentum and the corresponding roughness length is much larger.
Forests typically have rather large roughness lengths because of the tall trees
that comprise the forest and the many openings that allow air to penetrate into
the forest canopy. Interestingly, very dense forests where air cannot penetrate
readily down into the canopy can have a smaller roughness length than a more
open forest where the wind flow can more easily penetrate deeper into the can-
opy. In general, the roughness length is a function of the height of the surface
roughness elements, their density, and their flexibility. The roughness length in
general will increase with increasing height of the surface elements. The rough-
ness length will also increase with the increasing density of the surface elements
up to a point above which the roughness length will decrease as the density be-
Chapter 9 Surface Energy Budget and Fuel Moisture 317

comes very high. Also the roughness length generally decreases with increas-
ing flexibility of vegetation, since the pressure force mechanism for momen-
tum transport will be more effective for rigid objects than for flexible objects.
As roughness increases, turbulence and vertical mixing rates increase. Garratt
(1977) and Oke (1978) provide typical values of Zo for a wide range of surfaces.
Forests have values usually in the range of 1-6 m. By contrast, the roughness
height of short grass is of the order of 1 cm.
The logarithmic wind profile applies to the free atmosphere above the can-
opy. Within the canopy, the vertical wind speed gradient is dependent on the
structure and density of the canopy (Fritschen, 1985). Denser canopies are
characterized by lower wind speeds than more open canopies. The gradient
within the canopy has often been parameterized with an exponential function
(Cionco, 1965):
U{z) = l/(h) exp[/3(l - z/h)] (14)
where h = height of the canopy, and j8 is a parameter called the wind velocity
attenuation coefficient. For forest canopies, j8 is typically in the range of 2 - 5
(Cionco, 1978), These values of j8 indicate that wind speeds at the floor of a for-
est are a small fraction of values at the top of a forest, thus reducing ventilation
and FT rates.
Water vapor is to a first approximation a passive element in the surface mix-
ing process. When momentum is transported through friction, in which slow-
moving air molecules near surface objects are exchanged with faster moving air,
water vapor molecules are also exchanged. However, the transport of momen-
tum by the pressure force process does not result in exchange of air; therefore,
the transport of water vapor is less efficient than momentum transport.

III. ESTIMATION OF POTENTIAL

EVAPOTRANSPIRATION RATES
A. PENMAN-MONTEITH METHOD

1. General Formulation
The energy balance at the air-surface interface in nonprecipitating conditions
is usually written as
R^ = LE + H^ G (15)
where L = latent heat of evaporation (J kg~^), E = water vapor flux
(kg m"^ s"^), H = sensible heat flux (W m~^), and G = soil heat flux (W m"^),
as illustrated in Figure 7. Measurements of the energy budget and application
318 Kenneth E. Kunkel

t Rn'

.^^

S, S,(1-a) K Is LE'^
4 I A

FIGURE 7 Schematic of the components of the surface energy budget.

of this formula have focused both on the interface between the top of the can-
opy and the free atmosphere (denoted by superscript c) and on the forest floor-
atmosphere interface (denoted by the superscript s). This expression ignores
photosynthesis, heat storage in plant tissue and the canopy volume, and meta-
bolic activity, all of which tend to be smaller than the experimental error ( < 10%
of R„). Although this approximation is usually adequate and will be used in this
chapter, it is important to point out that heat storage in plant tissue and the
canopy volume can be significant in tall forests, particularly when net radiation
is low For example, Denmead and Bradley (1985) made estimates of this stor-
age term for a mature Ponderosa pine (Pinus ponderosa) forest and found that
it usually represented > 10% of the net radiation, nonnegligible if high accuracy
is required.
Chapter 9 Surface Energy Budget and Fuel Moisture 319

The product LE is referred to as the latent heat flux. The latent heat flux is
usually directed upward from the surface to the atmosphere. However, at night,
condensation can occur, in which case the flow of moisture is toward the sur-
face; however, the values of condensation are usually much lower than evapo-
ration. Sensible heat flux (H) is usually directed upward into the atmosphere
during the day and downward toward the canopy at night. Likewise, the soil
heat flux is usually directed downward into the ground during the day and to-
ward the surface at night. The exact partitioning of radiative energy into these
components is dependent on meteorological factors, the characteristics of the
soils, and the characteristics of the fuel.
The sensible and latent heat fluxes are often parameterized in terms of the
near-surface gradients of temperature and vapor pressure as follows (Bonan,
1996):
H = pc^{e, - ^ J / r , (16)
and

LE = '4^ (17)
where 6^ — air potential temperature, d^ = potential temperature of surface,
es^ = saturation water vapor pressure at the surface, y = CpP/(0.622L), p = air
density (kg m~^), c^ = heat capacity of air at constant pressure (J kg~^ K~^),
r^ = aerodynamic resistance (sm~^), r^ = stomatal or surface resistance
(s m~^), and P = pressure (hPa). T^ and e are measured at some standard height.
For the practical problem of estimating fluxes, these formulae are difficult to
apply because T^ and es^ are not easily measured and thus not typically available
to make such estimates. Thus, many methods have been developed to estimate
fluxes from more readily available measurements. In particular, numerous for-
mulae have been developed to estimate "potential" evapotranspiration (PET)
rates (i.e., that which would occur if soil water were not limited and meteoro-
logical factors only limit the rate of ET). One of the most successful and widely
used of these is the Penman formula (Penman, 1948):
HK - G) - pc{es - e)/r^
PET = (18)
A+ 7
This equation was modified slightly by Monteith (1965, 1981) to produce the
following formula for actual evapotranspiration (AET):
A(R, - G) + pc^{es - e)/r,
AET = '—-- (19)
A + 7(1 + rjr,)
where A = d{es)/dT (hPa K"^). These formulae are quite accurate for a wide
range of meteorological conditions because they incorporate all the primary
320 Kenneth E. Kunkel

meteorological factors and are based on fundamental energy conservation prin-

ciples. The first term in the numerator, called the radiation term, accounts for
the direct absorption of radiative energy at the air-surface interface. The sec-
ond term in the numerator, the advection term, accounts for additional evapo-
ration due to ventilation of the surface (the variable r^ incorporates the wind
speed; see the discussion that follows). This ventilation rate is primarily a func-
tion of the water vapor pressure deficit of the free atmosphere and the rate of
turbulent mixing, which is embodied in the variable r^. It is important to note
that the water vapor pressure deficit used in Eqs. (18) and (19) is that at a stan-
dard measurement height, not at the air-surface interface. These often are quite
different.

2. Aerodynamic Resistance
The rate of vertical diffusion of water vapor between the surface and the free
atmosphere is parameterized by r^ and r^, using an analogy to electrical circuits
in Ohm's law (Chapter 4 in this book) where the evaporation rate (current) is
determined by the water vapor pressure deficit (voltage difference) divided by
the inverse of the diffusion rate (resistance). The aerodynamic resistance (r^) is
inversely proportional to the rate of vertical mixing of air near the surface. In
general, it is a function of the wind speed, the roughness of the surface, and the
thermal stability of the near-surface atmosphere. The vertical profiles of tem-
perature and water vapor content follow a logarithmic dependence similar to
Eq. (12); r^ can be derived from these logarithmic forms and expressed as

'a{z) = V^;^[Hiz - d)/Zo} - IpmlHiz - d)/Zo} - ikmX^] (20)

where if/mx^ is the stability correction for wind (m), temperature (h), or wa-
ter vapor (w), depending on whether r^ is used to estimate momentum, sen-
sible heat, or latent heat fluxes, respectively. This expression for r^ is based on
a fundamental understanding of the nature of the turbulent near-surface atmo-
sphere, an understanding that is supported by numerous field experiments.
Strictly speaking, the general form for r^ [Eq. (20)] is applicable to fluxes of
water vapor, heat, and momentum. However, as noted earlier, the efficiency of
water vapor and heat transport is less than that of momentum near the surface
because the pressure force does not act to transport water vapor and heat. In
essence, the resistance for water vapor and heat transport is larger than for mo-
mentum transport. This is sometimes referred to as the "excess" resistance. Ex-
perimental evidence indicates that the efficiency of water vapor transport is
similar to that of heat transport and these are usually assumed to be equal. Ex-
perimental observations (Chamberlain, 1968; Garratt and Hicks, 1973) have
shown that the magnitude of this excess resistance is dependent on the flexibil-
Chapter 9 Surface Energy Budget and Fuel Moisture 321

ity of surface elements. Surface elements that are rigid, such as fallen logs, tree
trunks, etc., are efficient at exchanging momentum by the pressure force, and,
in this case, the difference between the efficiency of momentum and water vapor/
heat transport is large. By contrast, flexible surface elements, such as the leaves
of trees and grasses, are not very efficient at transporting momentum through
the pressure force because they bend in the wind. In this case, the efficiencies
for momentum and water vapor/heat transport are more nearly equivalent. A
variety of expressions have been developed to parameterize this effect. Perhaps
the most common and simplest approach is to assign a roughness length for wa-
ter vapor and heat transport that is smaller than that for momentum. The val-
ues of :^o given previously are for momentum. For water vapor or heat, a typi-
cal value is ZQM, = ZQII, where ZQW is the roughness height for water vapor or heat
(Rowntree, 1991; Shuttleworth, 1991). However, experimental evidence sug-
gests that Zow is not a constant for a particular surface and, in fact, varies with
environmental conditions (Blyth and Dolman, 1995; Hignett, 1994; Sun and
Mahrt, 1995). Based on work by Zilitinkevich (1970), the relationship of ZQW to
ZQ can be expressed as

ln^ = 0.13f^''' (21)

where v = kinematic viscosity of air with a value of about 1.5 X 10~^ m^ s~\
The quantity (U^ZQ/^) is the roughness Reynolds number, which can be inter-
preted as the Reynolds number of the smallest turbulent eddy. In this relation-
ship, the ratio of ZQ to ZQ^ increases with increasing roughness and with increas-
ing wind speed (through U:jc).
The excess resistance can be incorporated in Eq. (20) as follows:

raiz) = ] ^ M i z - d)/zo} - 4'mIHiz - d)/zo.} - 4'k..] (22)

A number of expressions have been derived for ij/^ and (//j^^, based on field ex-
periments. These are often expressed as a function of a parameter known as the
Monin-Obukhov length (L^o), defined as
ulpcT

where L^^ has the units of length and is similar to the gradient Richardson
number (Ri) in expressing the relative magnitude of wind shear forces to buoy-
ancy forces. When the absolute magnitude of sensible heat flux (H) is large com-
pared to the momentum flux, the absolute value of L^^ is small, and vice versa.
The sign of L^^ is indicative of the stability (negative for unstable conditions,
positive for stable conditions). Commonly used expressions for the wind ((/^^)
322 Kenneth E. Kunkel

and temperature/water vapor (i/^h,w) profile correction factors are (Paulson,

1970)
1 + X^ \ / 1 + X
^m[{z - d)/L^,] = In
2 J\ 2
77
2 tan ^x + for L ^ , < 0 (24)

[z-d)
ilj^iiz - d)/L^,] = -5 for L ^ , > 0 (25)

where X = [1 - I6{z - d)/L,^X^' and

1
^hAiz - d)/L^,,] = 2 In {! + ( ! - 16(z - d)/L^J^^} for L^, < 0
(26)

^Kwliz - d)/L^,] = -5{d - z)lK, for L^, > 0

(27)

3. Stomatal or Surface Resistance

The variable r^ is the resistance of the surface to the transport of water vapor.
When r^ > 0, the surface limits ET below that of the potential rate. For dead
fuels, this can be interpreted as inversely proportional to the rate of diffusion of
water to the surface of the fuel. When the diffusion rate is low, it is this factor
that controls ET rates, rather than the availability of energy and the rate of tur-
bulent mixing. For leaves, r^ is the stomatal resistance. The stomatal resistance
reflects the physiological control on transpiration by the opening and closing
of stomates in the leaf which is part of the process of photosynthesis. A number
of meteorological factors influence the magnitude of r^ including solar radiation,
soil moisture, water vapor pressure deficit, and temperature. Because it is de-
pendent on solar radiation, r^ exhibits a pronounced and characteristic diurnal
dependence as stomates open and close in response to increases and decreases
in solar radiation. Stomates also respond to deficient soil moisture by partially
closing to conserve water. When soil moisture is not limiting, minimum val-
ues of Tj are achieved in the middle of the day and are a function of the plant
type. Crops such as corn, soybean, and wheat typically have minimum val-
ues around 50 s m~\ whereas forests have higher minimum values, typically
around 100 s m~^ or greater (Rowntree, 1991).
Some recent experiments have provided a number of observations of r^ ap-
plicable to forests. Blanken et at (1997) found that daytime mean values of
stomatal resistance over an aspen forest varied widely, from about 50 to >200
s m~\ with a typical daytime mean value of around 100 s m " \ Chang et al.
(1997) measured stomatal resistance values of 100-150 s m"^ over a poplar
Chapter 9 Surface Energy Budget and Fuel Moisture 323

(Populus trichocarpa X P. tacamahaca) forest. Baldocchi et al. (1997) measured

stomatal resistance over a boreal jack pine forest in Saskatchewan, Canada, and
found rather high values, generally in excess of 200 s m"^ and often exceeding
500 s m~\ They also obtained measurements over a well-watered deciduous
forest. Values of stomatal resistance were in the range of 150-400 s m~\ some-
what lower than the jack pine forest. They argued that the high values of stoma-
tal resistance in the jack pine forest reflect an adaptation of the plant to an en-
vironment of relatively low water availability and limits on decomposition and
nutrient cycling in the boreal region where the jack pine measurements were
taken. There is also experimental evidence that stomatal resistance increases
when the above-canopy vapor pressure deficit exceeds approximately 10 hPa;
above this threshold, stomatal resistance varies in such a way that transpiration
remains relatively constant. This has been observed over an aspen forest (Hogg
et al, 1997), in tropical rainforests (Meinzer et al, 1993; Granier et al, 1996),
in temperate forests and woodlands (Lopushinsky, 1986; Price and Black, 1989;
Goulden and Field, 1995), and in a Holm-oak savanna (Infante et al, 1997). This
may reflect a physiological response of the plant to maintain leaf water poten-
tials above the point where plant damage would occur (Tyree and Sperry, 1988;
Sperry and Pockman, 1993; Goulden and Field, 1995; Goulden et al, 1997).

4. Application of Penman-Monteith Formulae

The practical application of Eqs. (18) and (19) to estimate PET and AFT is some-
times difficult because the required meteorological variables are often not avail-
able. For example, the routine hourly observations taken by National Meteo-
rological Services at airports do not include any components of the radiation
budget. By contrast, a number of specialized networks that have been estab-
lished by universities, state agencies, and other organizations do include mea-
surements of downward solar radiation and soil heat flux. However, even many
of these networks do not measure net radiation.
For this reason, practical application of Eq. (17) often requires estimation of
radiation components. Meyers and Dale (1983) developed a method to estimate
downward solar radiation from the cloud cover observations that are taken at
airports. Petersen et al (1995) applied this method to develop an historical so-
lar radiation climatology for the midwestern United States. The following meth-
ods can be used to estimate R„ if hourly solar radiation observations are avail-
able or if hourly cloud observations are available to estimate solar radiation
using the method of Meyers and Dale (1983) and Petersen et al (1995).
A variety of investigators have found that simple regressions between R„ and
Sj provide reasonably accurate estimates of R„ with r^ values greater than 0.9
(Shaw, 1956; Monteith and Szeicz, 1961; Fritschen, 1967; Nielsen et al, 1981;
Zhong et al, 1990; Pinker and Corio, 1984; Pinker et al, 1985; Kustas et al,
324 Kenneth E. Kunkel

1994; Kaminsky and Dubayah, 1997). However, these studies have produced
widely varying regression coefficients. This suggests that the exact relationship
between R„ and 5j may be rather stable for a particular location but will vary
from location to location depending on other factors, probably including the
latitude, temperature and humidity climatology, and land-use. Even at a spe-
cific location, the relationship will likely vary with season. Use of this approach
is problematic because it may be uncertain which study's regression coefficients
should be adopted.
A second method avoids some of these uncertainties by making independent
estimates of the longwave components of the radiation budget. In this method,
the net radiation R„ is calculated following Weiss (1983) by
R„ = Sll -a)+ FXh - h) (28)
where F^ is a function of cloud cover. J^ is given by Brutsaert (1975) as
I, = I.24(e/TJ°^«(^T:) (29)
where the units are hPa (e) and K (T^). Is is estimated by
I, = 0.9ScrTl + (1 - 0.98)1^ (30)
F^ is given by
F, = 0.4 4- 0.6 Cj (31)
where Cj = cloud transmission. Soil heat flux G is estimated as
G = 0.15R„ (32)
This method is probably more accurate than the use of linear regressions be-
tween S^ and R„ but does require additional data (water vapor pressure, tem-
perature, cloud cover). Cloud cover data may not be readily available for non-
National Weather Service sites. Also, both I^ and I^ are estimated using air
temperature at a standard height even though infrared emissions are deter-
mined by the surface temperature (for I j and the air temperature above the site
(for J^). This further reduces the accuracy.
Another approach to the estimation of PET and AET using the Penman
method has been developed by Doorenbos and Pruitt (1977). This publication
provides convenient tables and procedures, particularly for estimating the ra-
diation term.

B. RADIATION MODELS

Empirical studies have indicated that the left- and right-hand terms of the nu-
merator in Eqs. (18) and (19) are correlated. There is a physical basis for this
finding since increasing radiation (left-hand side) will tend to increase the tem-
Chapter 9 Surface Energy Budget and Fuel Moisture 325

perature and thus the atmospheric water vapor pressure deficit (right-hand
side). In humid chmates, the left-hand term is typically much larger than the
right-hand term. This has led to a class of models to estimate PET from radia-
tion. Representative of this class is the Priestly-Taylor (1972) equation

A(R„ - G)
PET = a - V (33)
A+7
where a is an empirical constant to which they assigned a value of 1.26, based
on analysis of field data. Other studies have been performed to estimate a, and
most of them have found values near to that of Priestly and Taylor (1972) for
situations where stomatal resistance is low. However, values of a are often
somewhat lower for forests because of higher values of stomatal resistance. For
example, Blanken et al. (1997) measured a value of 1.22 over a hazelnut under-
story in an aspen forest. However, the average value of a above the forest can-
opy was only 0.91 with an upper limiting value of 1.11, indicating that sto-
matal control of forest transpiration substantially reduced actual evaporation
below the potential value that would be calculated using a standard value of
a = 1.26.
Although application of this model requires measurements of radiation, it
does not require measurements of humidity which is difficult to measure with
high accuracy. This is the main attraction of this class of models. It also does
not require estimates of the aerodynamic resistance and the associated uncer-
tainties in estimating surface parameters such as ZQ and d.

C. PAN EVAPORATION

Pan evaporation is measured at scattered stations around the world and pro-
vides a relatively straightforward approach to the estimation of PET. Its primary
attraction is that it is a direct measure of evaporation and integrates the effects
of all the meteorological controlling factors. The standard Class A pan is 121 cm
in diameter and 25.5 cm deep. Data are obtained by measuring the depth of wa-
ter in the pan and subtracting from the previous day's depth. Periodically, water
is added to the pan. This is a simple and straightforward measurement. How-
ever, pan evaporation data must be used with care because, in general, they pro-
vide an overestimate of water loss from vegetated surfaces due primarily to two
factors. First, the water volume and the pan itself absorb a greater fraction of
incoming solar radiation (has a lower albedo) than vegetation. This raises the
temperature of the water above the ambient air temperature and causes exces-
sive rates of evaporation. Second, the Class A pan is raised slightly above the
land surface which enhances the turbulent transport of water vapor from the
326 Kenneth E. Kunkel

water surface to the atmosphere. These factors cause pan evaporation to be typi-
cally 20-40% higher than PET.
Correction factors, called pan coefficients, have been developed for the
United States by Farnsworth and Thompson (1982) and Farnsworth et ah
(1982). When applied to the pan evaporation data, these factors provide rea-
sonably accurate estimates of PET. For the United States, these factors are typi-
cally in the range of 0.65-0.8, the lowest values applying to the sunny regions
of the southwest and the highest values to the northern Great Lakes and west
coast. Doorenbos and Pruitt (1977) also provide procedures to estimate pan co-
efficients, given the general climatic regime of an area and the location of the pan.

D. TEMPERATURE MODELS

There is a correlation between temperature and radiation and also between tem-
perature and water vapor pressure deficit. This has been the underlying founda-
tion for development of models that require only temperature. One such model
is the Blaney and Criddle (1950) equation which provides monthly values of
PET in mm:
PET = QD(0.46 T^ + 8) (34)
where C^ = an empirically determined consumptive use coefficient, D =
monthly mean percentage of annual daytime hours of the year, and T^ = mean
monthly temperature in °C. Values of C^^ have been determined for irrigated
crops and typically range from 0.6 to 0.9. This method has been applied widely
for irrigation applications in the western United States.
Thornthwaite (1948) developed a measure of PET that was used to classify
climatic regimes, defined as
10 T^
PET^ = 1.6 (35)

where

s= ^{Tj5y''' (36)
m=l

and
a = 6.75 X 10~^ s^ - 7.71 X 10"^ s^ + 1.79 X 10"^ s 4- 0.49 (37)
where PET^ is the value of PET in cm for month m and T^ is the mean tem-
perature (°C) for month m. This formula is reasonably accurate in the humid
eastern United States but underestimates PET in arid regions.
Chapter 9 Surface Energy Budget and Fuel Moisture 327

The exact relationship between PET and temperature depends on general

climatic conditions, particularly arid vs. humid climates, and any temperature
model must be tailored to these conditions to minimize uncertainties and er-
rors. Even with care, such models will likely produce estimates with significant
errors when meteorological conditions are climatically unusual for a location.
The real attraction of such models is that temperature data are widely available.
However, their use is justifiable only when data are limited to temperature and
when accuracy requirements are not high.

IV. FUNCTIONAL DEPENDENCE

OF PET AND AET
The Penman-Monteith equation is used here to illustrate the dependence of PET
and AET on selected meteorological variables. Figure 8 illustrates the depen-
dence on temperature for selected values of dewpoint temperature and selected
nominal values for R„, (7, P, and r^. AET increases rather rapidly with tempera-
ture, primarily reflecting the increase in es with temperature. However, there is
also a contribution from the partitioning of net radiative energy between sen-
sible and latent forms of energy. As temperature increases, an increasing frac-
tion of net radiation is partitioned into latent heat energy. AET decreases with
increasing T^, reflecting decreases in the saturation vapor pressure deficit.

0.14.

0.12H

Temperature (°C)
FIGURE 8 The dependence of the actual evapotranspiration on temperature for values of dew-
point temperature of 0°C, 10°C, and 20°C. Nominal values of R„ = 400 W m " ^ U = 5 m s " \ P =
1000 hPa, and r^ — 100 s m~^ were used to make these calculations with Eq. (19).
328 Kenneth E. Kunkel

Wind speed (m s^)

FIGURE 9 The dependence of the actual evapotranspiration on wind speed for values of dew-
point temperature of 0°C, 10°C, and 20°C. Nominal values of R„ = 400 W m"^ T = 30°C, P =
1000 hPa, and r^ = 100 s m"^ were selected for these calculations using Eq. (19).

Figure 9 illustrates the dependence on wind speed for selected values of T^

and nominal values for R„, T, P, and r^. There is a linear dependence on wind
speed. The sensitivity to wind speed increases with decreasing T^ (increas-
ing water vapor pressure deficit). For these nominal values, the radiation term
alone is about 0.047 mm hr~^ (y intercept). The advection term can be compa-
rable to, or even exceed, the radiation term for moderate wind speeds and low
values of T^.
The dependence of AFT and PFT on temperature for selected values of r^ is
illustrated in Figure 10. The curve for r^ = 0 indicates PFT. As r^ increases,
AFT decreases below PFT. At higher values of r^, the rate of change in AFT with
increasing temperature is rather small. This illustrates that, as the surface re-
sistance increases, AFT rates become less dependent on atmospheric forcing
and more dependent on the characteristics of the fuel that determine the sur-
face resistance.
Figures 8-10 illustrate that AFT is a sensitive function of temperature, at-
mospheric water vapor content, wind speed, and stomatal or surface resistance.
The use of the Penman-Monteith equation [Fq. (19)], which incorporates all of
these variables, provides a suitable framework for accurate estimation of AFT.
By contrast, simpler approaches, such as the radiation or temperature models,
do not incorporate all these variables and may result in large errors under cer-
tain conditions.
Chapter 9 Surface Energy Budget and Fuel Moisture 329
0.18

0.16H

Temperature (°C)
FIGURE 10 The dependence of actual evapotranspiration on temperature for values of r^ = 0,
50 s m~\ 100 s m " \ and 300 s m~\ Nominal values of R„ = 400 W m"^ T^ = 10°C, U = 5m s~\
and P = 1000 hPa were selected for these calculations using Eq. (19). The curve for r^ = 0 is the
potential evapotranspiration.

V. CHARACTERISTICS OF PET
The drying of forest fuels is related to the relative magnitudes of AET and of the
amount and frequency of precipitation events. Long-term measurements of
AET are not widely available. However, AET is approximately equal to PET
when soil moisture levels are high. Even under deficient soil moisture levels,
AET usually varies proportionally with PET. Therefore, an analysis of PET,
which can be estimated from the standard meteorological observations avail-
able at many locations, provides insight into factors that affect forest fire risk.

A. SEASONAL CYCLE

The seasonal variation of PET is a key factor in determining the risk of forest
fires. Of particular importance is the relative magnitude of PET compared to
precipitation. When PET exceeds precipitation, soil moisture can be depleted,
and both live and dead fuels can lose moisture content, increasing the risk of
fires. Figure 11 shows the seasonal cycle of PET at International Falls, Minne-
sota, based on the Penman-Monteith formula using the methodology described
earlier to estimate R„. The climate at this location can be classified as humid.
330 Kenneth E. Kunkel

International Falls, MN: 1961-1990

F 14
o J ^^^^^V
o 12
] Potential E l / \.
B 10
Q.

o 81 / / "^- \
CI)
Q_
K 6
J / /
LU J / ^ >. \
"cO 41 / ^/" Precipitation
c
Q
1 > ^ ^ ^ ^V "^ ^
?
C)
a.
1 I . I . I 1 1 1 I . I — \ — 1 1 — . 1* .—
D) Q.
^O > o
CD o CD
^ s. < < CO O z Q

Month
FIGURE 11 Annual cycle of average monthly potential evapotranspiration (solid) and total pre-
cipitation (dashed) for International Falls, Minnesota, based on 1961-1990 data.

Values of PET are quite low during the colder half of the year, rising rapidly
during the spring and reaching a peak in mid summer. Precipitation exceeds
PET during much of the colder half of the year, but during the late spring and
summer months PET exceeds precipitation, on average. As a result, there is nor-
mally a depletion of soil moisture during the summer months. Often, the high-
est fire danger occurs in mid to late summer in this area, when soil moisture
has reached its climatological minimum. The drying of dead fuels is not as sen-
sitive to total precipitation as it is to the number of days with precipitation. This
variable remains high through the summer months with about one day in three
receiving some precipitation.
Variations in climatic conditions from year to year can be significant. Fig-
ure 12 shows monthly PET and precipitation at International Falls during 1980.
During a very dry spring, PET exceeded precipitation substantially, resulting in
early depletion of soil moisture. This can enhance the danger of spring fires be-
fore vegetation has broken dormancy when it is more flammable.

B. ELEVATION DEPENDENCE

In mountainous regions, PET decreases with increasing elevation, primarily be-

cause of the dependance of temperature on elevation. During summer, tempera-
ture decreases on average by about 6°C km~^ in much of the western United
States and Canada. Even though actual water vapor pressure also decreases with
height, the saturation vapor pressure usually decreases more rapidly, leading
Chapter 9 Surface Energy Budget and Fuel Moisture 331
International Falls, MN: 1980

Month
FIGURE 12 Monthly potential evapotranspiration (solid) and total precipitation (dashed) for In-
ternational Falls, Minnesota, during 1980.

to a decrease in water vapor pressure deficit with increasing height. In addi-

tion, mountainous areas usually experience more cloudiness, reducing the net
radiation.
An example is shown in Figure 13, which compares PET (estimated from
pan evaporation data) at Flagstaff, Arizona (el. 2110 m) and Phoenix, Arizona

30

25J

S, 20-
I- Phoenix

LU

0)

o
Q- 10-
5H

04
Q.
Q. 0 CD
C/D Q
<
Month
FIGURE 13 Annual cycle of monthly average potential evapotranspiration (cm) for Flagstaff, Ari-
zona (light) and Phoenix, Arizona (dark).
332 Kenneth E. Kunkel

(el. 323 m). The differences are due to both lower temperatures and more
cloudiness at Flagstaff compared to Phoenix.

C. TOPOGRAPHIC EFFECTS

Topographical effects can cause significant local variations in the surface energy
budget, specifically incoming solar radiation. South-facing slopes will receive
substantially greater amounts than north-facing slopes. These differences are
dependent on the latitude and the time of year. Oke (1978) presents estimates
of direct-beam solar radiation for a latitude of 40° at the time of the equinoxes
and the solstices. He shows that differences between south-facing and north-
facing slopes are smallest at the time of the summer solstice and largest at the
winter solstice. Figure 14 shows an estimate of the ratio of solar radiation on a
30° south-facing slope to the radiation on a horizontal surface as a function of
time of year at a latitude of 40°N, based on Jordan and Liu (1977). The ratio
is highest at the winter solstice and is greater than 1 through most of the year.
However, the ratio is actually less than 1 around the time of the summer solstice,
indicating that a horizontal surface receives slightly more solar radiation than
a 30° south-facing slope.
Measurements by Aisenshtat (1966) in the Turkestan Mountains (41°N)
during September also illustrate the effect of slope. In this case, R„ was nearly
three times as great on a 33° south-facing slope as a 31° north-facing slope.
Evapotransporation was nearly twice as large on the south-facing slope.

CO

o
CO

T 1 I r T 1 r
Jan. Feb. Mar. Apr. May Jun. Jul. Aug. Sep. Oct. Nov. Dec.
Month
FIGURE 14 Annual cycle of the ratio of monthly average solar radiation on a 30° south-facing
surface to solar radiation on a horizontal surface at a latitude of 40°N.
Chapter 9 Surface Energy Budget and Fuel Moisture 333

It is obvious that slope can have a large effect on fuel moisture content and
fire risk. Fuels on south-facing slopes will, in general, tend to have lower mois-
ture contents than those on north-facing slopes.

D. UNDERSTORY E T

Within a forest canopy, both radiation and ventilation are reduced by canopy
shielding, as described earlier. Observations indicate that the magnitude of un-
derstory ET varies widely depending on the density of the understory, the type
of understory vegetation, and soil moisture. Black and Kelliher (1989) reviewed
field experiments and found that the ratio of understory ET to total ET varied
from <0.05 to >0.50. Denmead and Bradley (1985) obtained measurements in
a 16-m-high mature Ponderosa pine forest and found that understory ET was
<20% of total ET. Furthermore, they found that much of the understory ET oc-
curred from intermittent gusts of wind that ventilated the canopy. Baldocchi
and Meyers (1991) measured ET at the floor of a deciduous forest and reported
on the average diurnal dependence during spring and fall. Their daily average
values peaked at about 25 W m~^, or only about 0.04 mm hr~\ Black and Kel-
liher (1989) reported that the ratio of understory to total ET often increases as
soil moisture decreases, perhaps reflecting increases in stomatal resistance of
the overs tor y trees as soil moisture decreases.

VI. NEAR-SURFACE ENVIRONMENT

Chapter 4 in this book describes the relationship of the moisture content of dead
fuels to relative humidity and temperature. As illustrated in Figures 4-13, the
equilibrium moisture content is strongly dependent on the environmental con-
ditions experienced by the fuel. Routine meteorological observations are taken
at standard heights. For example, in the United States the standard height for
measurements of maximum and minimum temperature in the National Weather
Services's Cooperative Observer Network is 1.5 m. However, the actual mete-
orological conditions at the fuel surface are often quite different because large
vertical gradients are a common feature of the near-surface atmosphere, due to
the momentum, heat, and water vapor exchange that occurs at the air-surface
interface. These gradients can be particularly large for ground-based fuels. A
critical factor in determining the magnitude of the vertical gradients is the na-
ture of diffusion and mixing. Within the laminar sublayer, the mixing rates are
determined by molecular diffusion. Because this is quite slow, the vertical gra-
dients are largest in this laminar sublayer. Above the laminar sublayer, mixing
rates increase rapidly because the atmosphere is turbulent. Immediately above
334 Kenneth E. Kunkel

2 1 2

1.5 • E, 1.5

X
1-

0.5 -

0
J X
1

05

0
(3 0.5 1 1.5 2 26 28 30 32 34 36 38 40
Speed (m s') Temperature (° C)
o , 2

.5 •
i; 1.5

1 • 1

X 0.5 • X 0.5

0 • 1 1 ^^"^"^^^^^ 1 1 0
15 16 17 18 19 20 20 25 30 35 40 45

Vapor pressure (h Pa) Relative humidity (%)

FIGURE 15 Vertical profiles of wind, temperature, water vapor pressure, and relative humidity
for H - 300 W m"2 and LE = 150 W m~\

the laminar sublayer, the size of turbulent eddies is restricted by proximity to

the surface. The dominant size of eddies becomes larger as the distance from the
surface increases. Thus, turbulent mixing rates tend to increase with height. As
a consequence, the vertical gradients gradually decrease with height.
Above the laminar sublayer, the logarithmic profile that applies to the height
dependence of wind speed is also observed for temperature and water vapor
content. The exact dependence is a function of the magnitude of the fluxes of
heat, momentum, and water vapor, as well as the characteristics of the surface,
particularly the roughness height. Figure 15 shows the vertical profiles of tem-
perature, humidity, and wind for typical mid-summer conditions over a surface
that is not shielded by a forest canopy. These profiles have been calculated us-
ing empirically determined formulae that include the effects of stability (Panof-
sky and Button, 1984). As can be seen, the vertical gradients are large near the
surface, gradually decreasing as height increases. Although both temperature
and water vapor content increase as the surface is approached, the net effect is
a decrease in relative humidity because of the exponential dependence of satu-
ration water vapor pressure on temperature.
Figure 15 does not show what happens across the laminar sublayer. For tem-
perature, observational evidence shows that the gradients can be very large. This
is not surprising since exchange of heat occurs through molecular diffusion.
Figure 16 shows an example of the diurnal dependence of the temperatures of
Chapter 9 Surface Energy Budget and Fuel Moisture 335

Boardman, OR - June 6,1992

70

Ta (potato)

T3 (range)

12:00 AM 6:00 AM 12:00 PM 6:00 PM 12:00 AM

Time
FIGURE 16 Temperature (°C) of the ground surface and the atmosphere at a height of 2 m on
June 6, 1992, for a semiarid grassland (dark) and for a potato field (light) in eastern Oregon.

the surface and of the atmosphere at a height of 2 m for a semiarid grass range-
land. The temperature difference peaks at solar noon with values in excess of
30°C. This is probably typical of ground fuels. Figure 16 also shows the tem-
perature difference for a nearby irrigated potato field. By contrast with the dry
rangeland, the temperature difference over the potato field never exceeds 10°C.
This contrast reflects two factors that influence this gradient. First, the range-
land is very sparsely vegetated, and momentum transport occurs largely by in-
teractions with the solid, nonflexible ground surface rather than with the sparse
grass. In the potato field, the atmospheric interactions are primarily with the
flexible plants. Thus, the "excess" resistance to heat transport is higher for the
rangeland than for the potato field, supporting a larger temperature gradient.
The second factor is the aridity of the surface and the partitioning of net radia-
tion into latent and sensible heat fluxes. The rangeland was quite dry, and most
of the available energy was partitioned into sensible heat flux. In the irrigated
potato field, latent heat flux was higher than sensible heat flux. Thus, evapora-
tive cooling of the surface was significant in the potato field, reducing the tem-
perature gradient.
Following Eq. (16), the surface potential temperature can be expressed as a
function of the sensible heat flux, stability, wind speed, and surface roughness
length, similar to Zeng and Dickinson (1998), as

e, = diz) + (38)

where r^{z) takes the form of Eq. (22).

336 Kenneth E. Kunkel

Measurement of the surface temperature is relatively straightforward using

small in situ probes or infrared thermometry. However, measurement of the sur-
face water vapor content is much more difficult and such measurements have
not been reported. It seems likely that the trends indicated in Figure 15 would
continue across the laminar sublayer; that is, there is likely to be a substantial
decrease in relative humidity at the fuel surface corresponding to the increase
in temperature shown in Figure 16. Thus, ground-based fuels that are exposed
to the sun experience a considerably lower relative humidity during daytime
than what would be indicated by measurements at a standard measurement
height. The surface water vapor content can be expressed in a similar manner
to Eq. (36) as

LErJz)
e, = e{z) + ^^ (39)
pCp

VII. MODELS OF LAND-

SURFACE INTERACTIONS
The preceding sections illustrate some of the complexities involved in describ-
ing mathematically the interactions of the atmosphere with the land surface
and how these interactions affect drying of fuels. A great deal of effort has been
devoted within the atmospheric sciences community to the development of
mathematical models and associated computer software packages that can be
used to describe the temporal behavior of surface fluxes of heat, momentum,
and water vapor and their dependence on meteorological conditions and land
surface conditions. Some of these have been developed into community mod-
els that are available to interested researchers. A brief description of their capa-
bilities, structure, and limitations is given here. In recent years, there has been
a systematic effort to improve these models through intercomparisons among
these models and comparisons with field studies (T. H. Chen et al, 1997). This
has led to significant model improvements.
The discussion here will focus on two models, the Biosphere-Atmosphere
Transfer Scheme (BATS) (Dickinson et al, 1981,1986,1993) and the Land Sur-
face Model (LSM) (Bonan, 1996). Both of these are community models available
through the National Center for Atmospheric Research (NCAR). Most other
models share many of the features of these two models. These models are quite
complex, which precludes a comprehensive description. The modeling of one
of the simpler cases, bare ground, will be described. This case can be applied to
the problem of modeling the environmental conditions (temperature and rela-
tive humidity) experienced by a duff layer, including the diurnal dependence
of those conditions.
Chapter 9 Surface Energy Budget and Fuel Moisture 337

In the absence of snow cover, the ground surface energy budget can be writ-
ten as follows by combining Eqs. (3) and (15):

Sa(l - a) + I^- I,- H- LE - G = 0 (40)

Several of these terms are directly or indirectly a function of the surface tem-
perature, Tj. To specify the surface fluxes, a value of T^ must be found that sat-
isfies Eq. (40). The upward longwave radiation, J^, is parameterized by Eq. (2).
The sensible heat flux, H, is parameterized by Eq. (16). The expression for la-
tent heat flux [Eq. (17)] does not explicitly contain T^ but is parameterized in
the BATS model as

LE = Lph[q,{T,) - q]/r^ (41)

where qs(Tj is the saturated specific humidity at the surface temperature and h
is a soil wetness factor which ranges from 1 when the soil is wet to 0 when it is
dry. This expression for LE is a function of T^. To determine the ground heat
flux at the surface, G(0), it is necessary to solve for the heat flow in the soil.
This is described next. Once this is solved, G can be parameterized as follows:

G{0) = - ^ \ '— (42)

where T^ and A^^ are the temperature and thickness, respectively, of the first
soil layer and K^i is the thermal conductivity (W m~^ K"^) of the first soil layer.
If Sj and I^ are specified (e.g., they are assumed not to be a function of T^),
the surface temperature can be solved using the preceding parameterizations
for I5, H, LE, and G as functions of T^. Since these forms contain higher order
polynomials, exponential functions, and other complex functions, it is not pos-
sible to find an analytical solution. In the LSM (Bonan, 1996), T^ is found by it-
eratively solving the equation

/ dl, dH dLE dG \ ,
5/1 - a ) + I ^ - I , - H - L E - G - [ ^ + + + AT, = 0

(43)

where AT, = T""^^ — T" and the superscript n indicates the iteration. The time
step for solution of this equation is for the order of 1 h or less.
The soil temperature profile is determined by discretizing the soil column
into multiple layers. The soil heat flux at depth z is

G{z) = Uz)-^ (44)

338 Kenneth E. Kunkel

One-dimensional energy conservation requires that

ST, a ax.
(45)
^^^'Kz
where pg{z) and cj^z) are the soil density and soil heat capacity, respectively, at
depth z. In the LSM, six layers of geometrically increasing thicknesses (10, 20,
40, 80, 160, and 320 cm) are used. Equation (45) is solved by numerical meth-
ods for Tg with the boundary conditions that G{z) == 0 at the bottom of the soil
column.
The volumetric soil water content affects the density and heat capacity of the
soil and the latent heat flux, as expressed in the parameter h [Eq. (41)]. Thus,
accurate estimation of the surface temperature requires calculation of soil wa-
ter. The soil water budget must account for the movement of water vertically
and for ET and precipitation at the soil-atmosphere interface. In the LSM, the
change in soil water content due to the movement of water is based on the Rich-
ards equation:

dV^_ _d_
(46)
dt dz "^ dz dV,

where V^ is the volumetric soil water content (mm^ mm~^), K^ is the hydraulic
conductivity (mm s~^), and ^ ^ is the soil matrix potential (mm). This equa-
tion is combined with the accounting for evaporation and precipitation and nu-
merically implemented for the six-layer soil column to solve for volumetric soil
water.
This system of equations [Eqs. ( 4 0 ) - ( 4 6 ) ] can be used to simulate the diur-
nal dependence of surface temperature and humidity, given a knowledge of ex-
ternal meteorological forcing. It is important to specify properly key parameters
such as Pg{z), cj^z), ^g(^), and Kj^{z), since these can be quite different for or-
ganic layers compared to mineral soils. The multilayer modeling of the soil col-
u m n in LSM and many other models provides a convenient framework for mod-
eling the commonplace situation of an organic layer overlying a mineral soil.
The modeling of the surface energy budget is much more complex when a
vegetative canopy is present and will not be described in detail here. In general
terms, models such as BATS and LSM break this system into four major com-
ponents: the plant, the atmosphere above the canopy, the atmosphere within the
canopy, and the ground surface. They then explicitly account for interactions
between the vegetation and the within-canopy air volume, between the within-
canopy air volume and the ground surface, and between the wi thin-canopy air
volume and the atmosphere above the canopy. Radiation exchange can occur be-
tween the atmosphere and the plants, between the atmosphere and the ground
surface (through gaps in the canopy overstory), and between the plants and the
Chapter 9 Surface Energy Budget and Fuel Moisture 339

ground surface. Exchange of heat and water vapor occurs between the plants
and the within-canopy air volume, between the within-canopy air volume and
the ground surface, and between the wi thin-canopy air volume and the atmo-
sphere above the canopy. Each of the processes is parameterized, and a set of
state variables for each component of the system is calculated.
In addition to the modeling of radiative and sensible fluxes, another impor-
tant process that is parameterized by these models is the interception of pre-
cipitation by the canopy. This can be very important in forest canopies where
the surface area of the vegetation is very large. Rain will coat the vegetation with
a film of water before water will fall to the ground. This water can then evapo-
rate and return to the atmosphere. This process typically reduces the amount
of rain reaching the ground surface by 10 - 5 0 % . The BATS and LSM models ex-
plicitly track the storage of water on vegetative surfaces. The maximum value
of this storage is assumed to be a linear function of the stem and leaf area of the
vegetative canopy.
Pressing research problems, particularly the need to accurately model future
climate change, are motivating work to improve these models. The ecological
researcher can benefit from these advances by applying current and future land
models to simulate the environmental conditions experienced by fuels.

VIII. REMOTE SENSING OF

THE SURFACE ENERGY BUDGET
There often exists substantial spatial variability in the surface energy budget,
arising from spatial variations in a number of conditions, including precipita-
tion distribution, soil water holding characteristics, and vegetation character-
istics. Thus, point measurements of the surface energy budget may be repre-
sentative only of a relatively small area surrounding a measurement site. Also,
comprehensive measurements of the components of the surface energy budget
are taken at very few sites because of the cost and difficulty. For these reasons,
there has been considerable research investigating the use of remote sensors on
airborne or satellite platforms to make such measurements.
One widely used technique relies on the parameterization of H expressed in
Eq. (16). In this relationship, H is proportional to the difference in temperature
between the surface and the overlying atmosphere. Thermal infrared measure-
ments are routinely obtained by several satellites, providing an estimate of the
surface radiative temperature ( T j . The air temperature (T^) can be estimated
from the surface observing network. The aerodynamic resistance (r^) must be
estimated, requiring values for the roughness height and wind speed. Thus,
some knowledge of the characteristics of the surface is necessary.
340 Kenneth E. Kunkel

The latent heat flux (LE) is often estimated as a residual of the terms in
Eq. (15). This requires an estimate of the radiative terms in Eq. (3). Values of
S^ and albedo can be retrieved from geostationary satellites (Tarpley, 1979; Gau-
tier et al, 1980; Diak and Gautier, 1983) using methods in which cloudiness is
the major factor modulating solar radiation at the surface. The infrared radia-
tion terms (l^ and I j can be estimated from thermal radiation satellite measure-
ments. The soil heat flux (G) is usually small compared to H and can be esti-
mated with sufficient accuracy using a combination of the measurement of T^
and a model, such as Eq. (42).
Although the basic approach (Jackson et al, 1977; Seguin and Itier, 1983)
for estimating H has changed little, there have been a number of variations, re-
sulting in a rather large number of models (Olioso et al, 1999; Schmugge and
Becker, 1991). These models have been somewhat successful in providing large-
scale estimates of H and LE (Lagouarde, 1991).
The accuracy of remote sensing estimates of surface fluxes using thermal IR
measurements is limited by the precision of the measurement of T^, which is of
the order of 2 - 3 K. In many situations, the temperature difference (T^ — T^) is
of the same order of magnitude. In these situations, the estimates of the surface
fluxes are not highly accurate. Typically, under the best circumstances, the sen-
sible heat flux can be estimated to within ±30 W m~^ (Seguin et al, 1999). A
complication arises when the surface is partially covered with vegetation. Often,
the vegetative elements and the bare ground have very different radiative tem-
peratures. Intercomparison of some of these models suggests that, with partial
vegetative cover, separate accounting for the fluxes from the bare ground and
from the vegetative elements provides superior results (Zhan et al, 1996). Es-
timates of vegetative cover using the Normalized-Difference Vegetation Index
(NDVI) can provide more accurate estimates of the components of the surface
energy budget (Boegh et al, 1999; Inoue and Moran, 1997).
There has also been a significant amount of work investigating the use of re-
mote sensors to estimate surface soil moisture. One approach relies on passive
microwave radiometry. The surface brightness temperature in the microwave
portion of the spectrum is a highly sensitive function of volumetric soil mois-
ture. Aircraft measurements using microwave radiometers have achieved ac-
curacies of ±3% in measurement of the volumetric soil water (Jackson et al,
1995). One limitation of passive microwave approaches is that satellite sensors
can achieve only rather crude resolution (10-30 km). Another proposed ap-
proach utilizes active microwave sensors which can obtain multifrequency,
multipolarization data at rather high spatial resolution. This approach shows
promise of achieving accurate estimates of soil moisture at high resolution
(Bindlish and Barros, 2000).
Despite some of the limitations and uncertainties, remote sensing of the sur-
face energy budget holds potential as a tool to assess the variability of fuel mois-
ture on regional scales (1-100 km).
Chapter 9 Surface Energy Budget and Fuel Moisture 341

IX. FIRE WEATHER RATING SYSTEMS

Fire weather rating systems incorporate measures of drying that reflect the
physical basis for ET described in this chapter. To illustrate this, the United
States and Canadian rating systems are briefly discussed.

A. UNITED STATES

The National Fire Rating System of the United States is based in part on indices
of the fuel moisture of both dead and live fuels. The relationships among en-
vironmental conditions and fuel moisture are empirically based. The drying of
dead fuels under constant conditions is assumed to follow an exponential de-
cay curve. Dead fuels are stratified into four categories of time response: 1, 10,
100, and 1000 h. A key concept in this system is the equilibrium moisture con-
tent (EMC) which is the "moisture content dead fuels would obtain if left in a
steady state environment long enough to obtain equilibrium (no net moisture
exchange)" (Bradshaw et al, 1983). The EMC is calculated from temperature
and relative humidity using regression equations developed by Simard (1968).
These equations are
EMC =
^0.03299 + 0.281073 RH - 0.00058 RH T RH < 10
2.22749 + 0.160107 RH - 0.0148 I 10 < RH < 50
,21.06060 4- 0.005565 Rtf - 0.0035 RH T - 0.483199 RH RH > 50
(47)
Figure 17 shows this relationship as a function of RH for various values of tem-
perature, indicating that the EMC is highly sensitive to RH and slightly sensi-
tive to temperature.
An important consideration in the application of the preceding formula is
that the relative humidity and temperature values are those at the surface of
the dead fuel, not at the normal observational height of instruments. As noted
in the previous section, these values can be quite different, particularly during
sunny daytime conditions. To account for the near-surface gradient, the rating
system uses adjustment factors that are applied to the observed temperature and
relative humidity; these are a function of sky cover conditions and are shown
in Table 1 (Bradshaw et al, 1983). As this table shows, the adjustment for tem-
perature and relative humidity are large under sunny conditions but quite con-
sistent with observations, as illustrated in Figures 15 and 16. These adjusted
factors are used in the EMC regression equations. These corrections are applied
to mid-afternoon observations when relative humidity reaches its minimum
value and fuel moisture content would be at a minimum.
342 Kenneth E. Kunkel

30
. - V

o^
?sj T=10C
26 H T = 30C
C
CD
?4] T = 50C
C
o
22-^
o 20 ^
(D18-^
16H
o 14H
b 1?H
b 10^
ZJ
^ 8H
'5 6 H
LU 4 -
2 1
<c 10 20
II1111II1111| 11 II1111 ii111II111|i 11 n-prm I I I II I M 1 1 1 1 1 I I 1 1 1 I I I
80 90 100
30 40 50 60 70
Relative humidity (%)
FIGURE 17 Equilibrium moisture content (%) of dead fuels as a function of relative humidity for
values of temperature = 10, 30, and 50°C as calculated in the U.S. National Fire Rating System.

For 1- and 10-h fuels, mid-afternoon observations are used to estimate fuel
moisture content. This reflects the substantial variation of RH within the daily
cycle (Figure 5) and the need to incorporate this into the rating system for fast
response fuels. To compute fuel moisture content for 100-h fuels, a weighted
24-h average EMC value is used; this is calculated from the day's maximum and
minimum temperature and maximum and minimum relative humidities. For
1000-h fuels, EMC is averaged over 7-day periods.
Two types of classes are used for live fuels: grass forbs and woody shrubs. The
system's developers found that the 1000-h fuel moisture function responded
to meteorological conditions in a similar fashion to that expected for live fuel.
The system thus applies this function to estimate live fuel moisture. Of course,
plants are adapted to their environment. Those adapted to drier conditions tend
to use moisture more slowly than those adapted to wetter conditions. Thus, the
fire danger rating system uses parameters that are a function of climatic zone, as

TABLE 1 Temperature and Relative Humidity Adjustment Factors

for Calculation of Equilibrium Moisture Content

Fractional Temperature Relative humidity

sky cover adjustment (°F) adjustment

0.0-0.1 + 25 X 0.75
0.1-0.5 + 19 X 0.83
0.5-0.9 + 15 X 0.91
0.9-1.0 +5 X 1.0
Chapter 9 Surface Energy Budget and Fuel Moisture 343

determined by the Thornthwaite climate classification scheme. In this system,

there are four climate classes: arid and semiarid, subhumid, humid, and wet.

B. CANADA

The Canadian Forest Fire Weather Index System (Van Wagner, 1987) uses
three separate indices for three distinct dead fuel classes:
1. Fine Fuel Moisture Code (FFMC), which represents the moisture con-
tent of litter and other fine fuels, with a time lag of less than one day.
2. Duff Moisture Code (DMC), which represents the moisture content of
loosely compacted organic matter, with a time lag of 12 days.
3. Drought Code (DC), representing a deep layer of compact organic mat-
ter with a time lag of 52 days.
There are a number of basic similarities between the Canadian system and
the U.S. system, although some of the functional relationships are different. In
the Canadian system, an equilibrium moisture content is calculated which is a
function of relative humidity and temperature. This is used in the computation
of the FFMC. The functional dependence exhibits hysteresis; specifically, the
equilibrium moisture content is lower during wetting than during drying. Fuel
moisture content approaches the equilibrium value in an exponential manner,
much as in the U.S. system. The rate of change is a function of wind speed, rela-
tively humidity, temperature, and (in the case of the DMC and DC) the day
length (which is a surrogate for radiation). Thus, all the major meteorological
factors that affect FT are incorporated in an empirical manner.
The DMC uses a constant equilibrium moisture content of 20%. The rate of
drying is a function of temperature, relative humidity, and an empirical day
length factor. This function does not include any wind speed dependence. This
is a reflection of the control on FT by surface resistance ( r j . When r^ is large,
the aerodynamic resistance, which incorporates the wind speed dependence, is
relatively less important. Evaporative drying for the DC index is based on an
estimate of potential evaportransporation which is calculated from an empirical
function of noon time temperature and a day length factor.

NOTATION

ROMAN LETTERS

a albedo dimensionless
A,B,C constants used in formula relating es to T
344 Kenneth E. Kunkel

b soil wetness factor ( 0 - 1 )

Cp heat capacity of air at constant pressure
Cg heat capacity of the soil Jkg-^K-i
Cj cloud radiation transmission dimensionless
Q empirically-determined consumptive use
coefficient
d displacement height m
D monthly mean percentage of annual daytime %
hours of the year
e atmospheric water vapor pressure hPa
es saturation value of atmospheric water vapor hPa
pressure
E water vapor flux k g m -^s-^
EMC equilibrium moisture content
ET evapotranspiration kg m ^ s \
W m " ^ or
m m h~^
F^ function relating net infrared radiation to cloud dimensionless
cover
G soil heat flux Wm"^
h height of the vegetative canopy m
H sensible heat flux Wm"^
J infrared radiation Wm"^
k von Karman constant ( = 0.4) dimensionless
Kg thermal conductivity of the soil Wm"^ K"^
Kfj soil hydraulic conductivity m m s~^
L latent heat of evaporation Jkg-i
L^o Monin-Obukhov length m
P air pressure hPa
PQ reference level pressure hPa
PET potential evapotranspiration W m"^ or

PET^ accumulated value of PET for month m cm

q specific humidity kgH^O/kg
moist air
Chapter 9 Surface Energy Budget and Fuel Moisture 345

r resistance sm ^
R„ net radiation Wm"^
RH relative humidity %
Ri gradient Richardson number dimensionless
S^ downward solar radiation at the earth's surface Wm"^
T temperature °C
T^ dew point temperature °C
T^ mean monthly temperature
-1
u^ friction velocity ms
U{z) wind speed at height z ms
V atmospheric water vapor mixing ratio kg H2O/
kg dry air
vs saturation water vapor mixing ratio kg H2O/
kg dry air
Vs volumetric soil water content mm^ mm"^
z height above ground surface or depth in soil m
Zo roughness length for momentum transport m
Zow roughness length for water vapor transport m

GREEK LETTERS

empirical constant used in Priestley-Taylor dimensionless

formula for PET
wind velocity attenuation coefficient
A des/dT hPa K"^
8 emissivity dimensionless
y Cp P/(0.622L) hPa K~^
V kinematic viscosity of air m^s-^
soil matrix potential mm
vertical profile adjustment factor for tempera- dimensionless
ture, wind, and water vapor
p air density kgm"^
(J Stefan Boltzmann constantt (5.67 X W m'^ K~^
10"^Wm"^K"^)
0 potential temperature K
346 Kenneth E. Kunkel

SUBSCRIPTS

a air or aerodynamic
g soil
h temperature
m wind
s surface of soil or vegetative canopy
w water vapor

SUGGESTED READING LIST

Campbell, G. S., and Norman, J. M. (1998). "An Introduction to Environmental Physics." Springer-
Verlag, New York.
Panofsky, H. A., and Dutton, J. A. (1984). "Atmospheric Turbulence: Models and Methods for En-
gineering Applications." John Wiley and Sons, New York.
Rosenberg, N. J. (1974). "Microclimate: The Biological Environment." John Wiley and Sons,
New York.
Schmugge. T. J., and Andre, J.-C, Eds. (1991). "Land Surface Evaporation: Measurement and Pa-
rameterization." Springer-Verlag, New York.

REFERENCES
Aisenshtat, B. A. (1966). Investigations of the heat budget of central Asia. In "Sowremennye Prob-
lemy Klimatologii" (M. I. Budyko, Ed.), pp. 83-129. Meteorol. Gidrol, Leningrad.
Baldocchi, D. O., and Meyers, T. P. (1991). Trace gas exchange at the floor of a deciduous forest:
I. Evaporation and COj efflux. J. Geophys. Res. 96, 7271-7285.
Baldocchi, D. D., and Vogel, C. A. (1996). Energy and CO2 flux densities above and below temper-
ate and broad leafed forest and a boreal pine forest. Tree Physiol. 16, 5-16.
Baldocchi, D. D., Vogel, C. A., and Hall, B. (1997). Seasonal variation of energy and water vapor ex-
change rates above and below and boreal jack pine forest canopy. J. Geophys. Res. 102, 28,939-
28,951.
Betts, A. K., and Ball, J. H. (1997). Albedo over the boreal forest. J. Geophys. Res. 102,28,901-28,909.
Bindlish, R., and Barros, A. P. (2000). Multifrequency soil moisture inversion from SAR measure-
ments with the use of lEM. Remote Sens. Environ. 71, 67-88.
Black, T. A., and Kelliher, F. M. (1989). Processes controlling understory evaporation. Phil. Trans.
Roy. Soc. (Lond) B. 324, 207-231.
Blaney, H. F., and Griddle, W O. (1950). "Determining Water Requirements in Irrigated Data." U.S.
Department of Agriculture Soil Gonservation Service Technical Paper No. 96.
Blanken, P. D., Black, T. A., Yang, P G., Neumann, H. H., Nesic, Z., Staebler, R., den Hartog, G., No-
vak, M. D., and Lee, X. (1997). Energy balance and canopy conductance of a boreal aspen for-
est: Partitioning overstory and understory components. J. Geophys. Res. 102, 28,915-28,927.
Blyth, E. M., and Dolman, A. J. (1995). The roughness length for heat of sparse vegetation. J. Appl.
Meteorol 34,583-585.
Chapter 9 Surface Energy Budget and Fuel Moisture 347

Boegh, E., Soegaard, H., Hanan, N., Rabat, P., and Lesch, L. (1999). A remote sensing study of the
NDVI-Tg relationship and the transpiration from sparse vegetation in the Sahel based on high-
resolution data. Remote Sens. Environ. 69, 224-240.
Bonan, G. B. (1996). "A Land Surface Model (LSM Version 1.0) for Ecological Hydrological, and
Atmospheric Studies Technical Description and User's Guide." NCAR Technical Note NCAR/
TN-417+STR. National Center for Atmospheric Research, Boulder.
Bradshaw, L. S., Deeming, J. E., Burgan, R. E., and Cohen, J. D. (1983). "The 1978 National Fire
Danger Rating System: Technical Documentation." General Technical Report lNT-169, U.S. De-
partment of Agriculture, Forest Service, Intermountain Forest and Range Experiment Station,
Ogden.
Brown, A. A., and Davis, K. R (1973). "Forest Fire: Control and Use." McGraw-Hill, New York.
Brutsaert, W. (1975). On a derivable formula for long-wave radiation from clear skies. Water Resour.
Res. 11,742-744.
Buck, A. L. (1981). New equations for computing vapor pressure and enhancement factors. J. Appl.
Meteorol. 20, 1527-1532.
Chamberlain, A. C. (1968). Transport of gases to and from surfaces with bluff and wave-like rough-
ness elements. Quart. J. Roy. Meteorol. Soc. 94, 318-332.
Chandler, C , Cheney, P., Thomas, P., Trabaud, L., and Williams, D. (1983). "Fire in Forestry: Vol-
ume One. Forest Fire Behavior Effects." John Wiley and Sons, New York
Chang, H., Simmonds, L. P., Morison, J. I. L., and Payne, D. (1997). Estimation of transpiration by
single trees: Comparison of sap flow measurements with a combination equation. Agric. For.
Meteorol. 87, 155-169.
Chen, J. M., Rich, P. M., Gower, S. T., Norman, J. M., and Plummer, S. (1997). Leaf area index of
boreal forests: Theory, techniques, and measurements. J. Geophys. Res. 102, 29,429-29,443.
Chen, T. H., Henderson-Sellers, A. Milly, P. C. D., Pitman, A. J., Beljaars, A. C. M., Polcher, J., Abra-
mopoulus, F., Boone, A., Chang, S., Chen, F., Dai, Y., Desbouough, E. E., Dickinson, R. E., Du-
menil, L., Ek, M., Garratt, J. R., Gedneny, N., Gusew, Y. M., Kim, J., Koster, R., Kowalczyk, E. A.,
Loval, K., Lean, J., Lettenmaier, D., Liang, X., Makfouf, J.-F., Mengelkamp, H.-T., Mitchell, K.,
Nanonova, O. N., Noilhan, J., Robock, A., Rozenzweig, C , Schaake, J., Schlosser, C. A., Schulz,
J.-P, Shao, Y., Shimakin, A. B., Verseghy, D. L., Wetzel, P, Wood, E. F., Xue, Y., Yang, Z.-L., and
Zeng, Q. (1997). Cabauw experimental results from the Project for Intercomparison for Land-
Surface Parameterization Schemes. J. Climate 10, 1194-1215.
Cionco, R. M. (1965). A mathematical model for air flow in a vegetative canopy. J. Appl. Meteorol.
4,517.
Cionco, R. M. (1978). Analysis of canopy index values for various canopy densities. Boundary-
Layer Meteorol. 15, 81-93.
Denmead, O. T., and Bradley, E. F. (1985). Flux-gradient relationships in a forest canopy. In "The
Forest-Atmosphere Interaction" (B. A. Hutchison and B. B. Hicks, Eds.), pp. 543-561. D. Rie-
del, Dordrecht.
Diak, G., and Gautier, C. (1983). Improvements to a simple model for estimating insolation from
GOES data. J. Clim. Appl. Meteorol. 22, 505-508.
Dickinson, R. E., Henderson-Sellers, A., and Kennedy, P.J. (1993). "Biosphere-Atmosphere Trans-
fer Scheme (BATS) Version le as Coupled to the NCAR Community Climate Model." NCAR
Technical Note NCAR/TN-387+STR. National Center for Atmospheric Research, Boulder.
Dickinson, R. E., Henderson-Sellers, A., Kennedy, P. J., and Wilson, M. F. (1986). "Biosphere-
Atmosphere Transfer Scheme (BATS) for the NCAR Community Climate Model." NCAR Tech-
nical Note/TN-275 + STR. National Center for Atmospheric Research, Boulder.
Dickinson, R. E., Jaeger, J., Washington, W M., and Wolske, R. (1981). "Boundary Subroutine for
the NCAR Global Climate Model." NCAR Technical Note/TN-173 + lA. National Center for At-
mospheric Research, Boulder.
348 Kenneth E. Kunkel

Doorenbos, J., and Pruitt, W. O. (1977). "Crop Water Requirements." F.A.O. Irrigation and Drain-
age Paper 24. F. A. O., Rome.
Farnsworth, R. K., and Thompson, E. S. (1982). "Mean Monthly, Seasonal, and Annual Pan Evapo-
ration for the United States." NOAA Technical Report NWS 34. U.S. Department of Commerce,
Washington, DC.
Farnsworth, R. K., Thompson, E. S., and Peck, E. L. (1982). "Evaporation Atlas for the Contiguous
48 United States." NOAA Technical Report NWS 33. U.S. Department of Commerce, Washing-
ton, DC.
Fritschen, L. J. (1967). Net and solar-radiation relations over irrigated field crops. Agric. Meteorol
4,55-62.
Fritchen, L.J. (1985). Characterization of boundary condtions affecting forest environmental phe-
nomena. In "The Forest-Atmosphere Interaction" (B.A. Hutchison and B. B. Hicks, Eds.), pp. 3 -
23. D. Riedel, Dordrecht.
Garratt, J. R., and Hicks, B. B. (1973). Momentum, heat and water vapour transfer to and from nat-
ural and artificial surfaces. Quart. J. Roy. Meteorol. Soc. 99, 680-687.
Garratt, J. R. (1977). "Aerodynamic Roughness and Mean Monthly Surface Stress over Australia."
CSIRO Aust. Div Atmos. Phys. Tech. Paper No. 29. CSIRO, Melbourne.
Gautier, C , Diak, G., and Masse, S. (1980). A simple physical model to estimate incident solar ra-
diation at the surface from GOES satellite data. J. Appl. Meteorol. 19, 1005-1012.
Goulden, M. L., Daube, B. C , Fan, S.-M. Sutton, D. J., Bazzaz, A., Munger, J. W, and Wofsy, S. C.
(1997). Physiological responses of a black spruce forest to weather. J. Geophys. Res. 102,
28,987-28,996.
Goulden, M. L., and Field, C. B. (1995). Three methods for monitoring the gas exchange of indi-
vidual tree canopies: Ventilated-chamber, sap-flow and Penman-Montieth measurements on
evergreen oaks. Funct. Ecol. 8, 125-135.
Granier, A., Hue, R., and Barigah, S. T. (1996). Transpiration of natural rain forest and its depen-
dence on climatic factors. Agric. For. Meteorol. 78, 19-29.
Hignett, P. (1994). Roughness lengths for temperature and momentum over heterogeneous terrain.
Boundary Layer Meteorol. 68, 225-236.
Hodges, G. B., and Smith, E. A. (1997). Intercalibration, objective analysis, intercomparison and
synthesis of BOREAS surface net radiation measurements. J. Geophys. Res. 102, 28,885-28,900.
Hogg, E. H., Black, T. A., den Hartog, G., Neumann, H. H., Zimmermann, R., Hurdle, P.,
Blanken, P D., Nesic, Z., Yang, P C , Staebler, R. M., McDonald, K. C , and Oren, R. (1997). A
comparison of sap flow and eddy fluxes of water vapor from a boreal deciduous forest. J. Geo-
phys. Res. 102, 28,929-28,937.
Infante, J. M., Rambal, S., and Joffre, R. (1997). Modeling transpiration in Holm-oak savanna: Scal-
ing up from the leaf to the tree scale. Agric. For. Meteorol. 87, 273-289.
Inoue, Y., and Moran, M. S. (1997). A simplified method for remote sensing of daily canpy transpi-
ration—A case study with direct measurements of canopy transpiration in soybean canopies.
Int. J. Remote Sens. 18, 139-152.
Jackson, R. D., Reginato, R. J., and Idso, S. B. (1977). Wheat canopy temperature: A practical tool
for evaluating water requirements. Water Resour. Res. 13, 651-656.
Jackson, T. J., LeVine, D. M., Swift, C. T., Schmugge, T. J., and Schiebe, F. R. (1995). Large area
mapping of moisture using the ESTAR passive microwave radiometer in Washita '92. Remote
Sens. Environ. 53, 27-37.
Jordan, R. C , and Liu, B. Y. H. (Eds.). (1977). "Applications of Solar Energy for Heating and Cool-
ing of Buildings." American Society of Heating, Refrigerating, and Air-Conditioning Engineers,
Inc., New York.
Kaminsky, K. Z., and Dubayah, R. (1997). Estimation of surface net radiation in the boreal forest
and northern prairie from shortwave flux measurements. J. Geophys. Res. 102, 29,707-29,716.
Chapter 9 Surface Energy Budget and Fuel Moisture 349

Kustas, W. P., Pinker, R. T., Schmugge, T. J., and Humes, K. S. (1994). Daytime net radiation esti-
mated for a semiarid rangeland basin from remotely sensed data. Agric. For. Meteorol. 71, 337-
357.
Lagouarde, J.-P. (1991). Use of NOAA AVHRR date combined with an agrometeorological model
for evaporation mapping. Int. J. Remote Sens. 12, 1853-1864.
List, R. J. (1949). "Smithsonian Meteorological Tables," 6th ed. Smithsonian Institution Press,
Washington, DC.
Lopushinsky, W. (1986). Seasonal and diurnal trends of heat pulse velocity in Douglas-fir and pon-
derosa pine. Can. J. For. Res. 16, 814-821.
McCaughey, J. H. , Lafleur, P. M., Joiner, D. W., Bartlett, P. A., Costello, A. M., Jelinski, D. E., and
Ryan, M. G. (1997). Magnitudes and seasonal patterns of energy, water, and carbon exchanges
at a boreal young jack pine forest in the BOREAS northern study area. J. Geophys. Res. 102,
28,997-29,007.
Meinzer, F. C , Goldstein, G., Holbrook, N. M., Jackson, P., and Cavelier, J. (1993). Stomatal and
environmental control of transpiration in a lowland tropical forest tree. Plant Cell Environ. 16,
429-436.
Meyers, T. P., and Dale, R. F. (1983). Predicting daily insolation with hourly cloud height and
coverage. J. Clim. Appl. Meteorol. 22, 537-545.
Monteith, J. L. (1965). Evaporation and the environment. Symp. Soc. Exp. Biol. 19, 205-234.
Monteith,J. L. (1981). Evaporation and surface temperature. Quart.]. Roy. Meteorol. Soc. 107,1-27.
Monteith, J. L., and Szeicz, G. (1961). The radiation balance of bare soil and vegetation. Quart. J.
Roy. Meteorol. Soc. 87, 159-170.
Nielsen, L. B., Prahm, L. P., Berkowicz, R., and Conradsen, K. (1981). Net incoming radiation es-
timated from hourly global radiation and/or cloud observations. J. Climatol. 1, 255-272.
Oke, T. R. (1978). "Boundary Layer Climates." Methuen, London.
Olioso, A., Chauki, H., Courault, D., and Wigneron, J.-P. (1999). Estimation of evapotranspiration
and photosynthesis by assimilation of remote sensing data into SVAT models. Remote Sens. En-
viron. 68, 341-356.
Panofsky, H. A., and Dutton, J. A. (1984). "Atmospheric Turbulence: Models and Methods for En-
gineering Applications." John Wiley and Sons, New York.
Paulson, C.A. (1970). The mathematical representation of wind speed and temperature profiles in
the unstable atmospheric surface layer. J. Appl. Meteorol. 9, 857-861.
Penman, H. L. (1948). Natural evaporation from open water, bare soil, and grass. Proc. Roy. Soc.
(Lond)A. 193, 120-145.
Petersen, M. S., Lamb, P. J., and Kunkel, K. E. (1995). Implementation of a semi-physical model for
examining solar radiation in the Midwest. J. Appl. Meteorol. 34, 1905-1915.
Pinker, R. T., and Corio, L. A. (1984). Surface radiation budget from satellites. Mon. Wea. Rev. 112,
209-215.
Pinker, R. T., Corio, L. A., and Tarpley, J. D. (1985). The relationship between planetary and sur-
face net radiation. J. Clim. Appl. Meteorol. 24, 1272-1268.
Price, D. T., and Black, T. A. (1989). Estimation of forest transpiration and COj uptake using the
Penman-Montieth equation and a physiological photosynthesis model. In "Proceedings of Work-
shop on Estimation of Areal Evapotranspiration," pp. 213-228. lAHS Pub. 177, Vancouver.
Priestley, C. H. B., and Taylor, R.J. (1972). On the assessment of surface heat flux and evaporation
using large scale parameters. Mon. Wea. Rev. 100, 81-92.
Pyne, S. J. (1984). "Introduction to Wildland Fire: Fire Management in the United States." John
Wiley and Sons, New York.
Rowntree, P. R. (1991). Atmospheric parameterization schemes for evaporation over land: Basic
concepts and climate modeling aspects. In "Land Surface Evaporation: Measurement and Pa-
rameterization" (T.J. Schmugge andJ.-C. Andre, Eds.), pp. 5-29. Springer-Verlag, New York.
350 Kenneth E. Kunkel

Salby, M. L. (1996). "Fundamentals of Atmospheric Physics." Academic Press, San Diego.

Schmugge, T. J., and Becker, F. (1991). Remote sensing observations for the monitoring of land-
surface fluxes and water budgets. In "Land Surface Evaporation: Measurement and Parameter-
ization" (T.J. Schumgge andJ.-C. Andre, Eds.), pp. 337-347. Springer-Verlag, New York.
Seguin, B., and Itier, B. (1983). Using midday surface temperature to estimate daily evaporation
from satellite thermal IR data. Int. J. Remote Sens. 4, 371-383.
Seguin, B., Becker, F., Phulpin, T., Gu, X. R, Guyot, G., Kerr, Y., King, C., Lagouarde, J. P., Ottle, C.,
Stofl, M. P., Tabbagh, A., and Vidal, A. (1999). IRSUTE: A minisateUite project for land surface
heat flux estimation from field to regional scale. Remote Sens. Environ. 68, 357-369.
Sellers, R J., Hafl, R G., Kelly, R. D., Black, A., Baldocchi, D., Berry, J., Ryan, M., Ranson, K. J.,
Grill, P. M., Lettenmaier, D. P., Margolis, H., Cihlar,J., Newcomer,J., Fitzjarrald, D.,Jarvis, P. G.,
Gower, S. T., Halliwell, D., Williams, D., Goodison, B., Wickland, D. E., and Guertin, R E.
(1997). BOREAS in 1997: Experiment overview, scientific results, and future directions. J.
Geophys. Res. 102, 28,731-28,769.
Shaw, R. H. (1956). A comparison of solar radiation and net radiation. Bull. Amer. Meteorol. Soc. 37,
205-206.
Shuttleworth, W.J. (1991). Evaporation models in hydrology. In "Land Surface Evaporation: Mea-
surement and Parameterization" (T.J. Schmugge andJ.-C. Andre, Eds.), pp. 93-120. Springer-
Verlag, New York.
Simard, A.J. (1968). "The Moisture Content of Forest Fuels—1." Inf. Rep. FF-X-14. Forest Fire Re-
search Institute, Ottawa.
Sperry, J. S., and Pockman, W T. (1993). Limitation of transpiration by hydraulic conductance and
xylem cavitation in Betula occidentalis. Plant Cell Environ. 16, 279-287.
Sun, J., and Mahrt, L. (1995). Determination of surface fluxes from the surface radiative tempera-
ture. J. Atmos. Sci. 33, 1110-1117.
Tarpley, J. D. (1979). Estimating incident solar radiation at the surface from geostationary satellite
data. J. Appl. Meteorol. 18, 1172-1181.
Thom, A. S. (1975). Momentum, mass and heat exchange of plant communities. In "Vegetation
and the Atmosphere, Vol. 1. Principles" (J. L. Monteith, Ed.), pp. 57-109. Academic Press,
New York.
Thornthwaite, C. W. (1948). An approach toward a rational classification of climate. Geogr. Rev.
38,55-94.
Tyree, M. T., and Sperry, J. S. (1988). Do woody plants operate near the point of catastrophic xylem
dysfunction caused by dynamic water stress? Plant Physiol. 88, 574-580.
Van Wagner, C. E. (1987). "Development and Structure of the Canadian Forest Fire Weather In-
dex System." Forestry Technical Report 35. Canadian Forestry Service, Ottawa.
Weiss, A. (1983). A quantitative approach to the Pruitt and Doorenbos version of the Penman equa-
tion. Irrig. Sci. 4, 267-275.
Zeng, X., and Dickinson, R. E. (1998). Effect of surface sublayer on surface skin temperature and
fluxes. J. Climate 11, 537-550.
Zhan, X., Kustas, W P., and Humes, K. S. (1996). An intercomparison study on models of sensible
heat flux over partial canopy surfaces with remotely sensed surface temperature. Remote Sens.
Environ. 58, 242-256.
Zhong, M., Weill, A., and Taconet, O. (1990). Estimation of net radiation and surface heat fluxes
using NOAA-7 satellite infrared data during fair-weather cloudy conditions of MESOGERS-84
experiment. Boundary Layer Meteorol. 33, 353-370.
Zilitinkevich, S. S. (1970). "Dynamics of the Atmospheric Boundary Layer." Leningrad Gidro-
meteor, Leningrad.
 CHAPTER lO
Climate, Weather,
and Area Burned
M. D . FLANNIGAN
Canadian Forest Seryice, Edmonton, Alberta, Canada

B. M. WOTTON
Canadian Forest Service, Sault Ste. Marie, Ontario, Canada

I. Introduction
II. Weather and Area Burned—Synoptic Surface Features
III. Weather and Area Burned—Upper Air Features
A. Upper Air—Circulation
B. Vertical Structure of the Atmosphere
IV. Teleconnections
V Future Warming and Area Burned
VI. Summary
References

I. INTRODUCTION
Forest fires are strongly linked to weather and climate (Flannigan and Harring-
ton, 1988; Johnson, 1992; Swetnam, 1993). Fire has been an integral ecological
process since the arrival of vegetation on the landscape. For the purposes of
this chapter, we will define weather as short-term processes that result in varia-
tions in the atmospheric conditions ranging from minutes to a fire season. Pro-
cesses that influence the atmosphere over time periods longer than a fire sea-
son will be defined as climate. There are several factors that control the climate
and weather at any one location. These factors include variations in solar radia-
tion due to latitude, distribution of continents and oceans, atmospheric pres-
sure and wind systems, ocean currents, major terrain features, proximity to wa-
ter bodies, and local features including topography (see Trewartha and Horn,
1980, for more details). As climate varies, the corresponding weather variables
can vary in magnitude and direction.

Forest Fires
Copyright © 2001 by Academic Press. All rights of reproduction in any form reserved. 351
352 Flannigan and Wotton

The objective of this chapter is to highhght the connection between chmate/

weather and the area burned by forest fires. We have used examples primar-
ily from Canada or North America to illustrate our points. This chapter is di-
vided into sections that describe the relationships between surface weather and
area burned, upper air features and area burned, and teleconnections and area
burned. We also try to identify key knowledge gaps in the fire and weather/
climate relationship. This chapter closes with a short discussion on how global
change might influence forest fire activity and area burned in the 21st century.

II. WEATHER AND AREA BURNED-

SYNOPTIC SURFACE FEATURES
The day-to-day weather can dramatically influence fire behavior and area
burned. In fact, much of the area burned in a region during any given year oc-
curs on just a few days of severe fire weather (Nimchuk, 1983; Harvey et al,
1986). This has led to many studies over various spatial and temporal scales
that try to relate the weather to fire. These studies can be broken into two broad
categories, namely, case studies and what we will call area burned studies for
the purpose of this paper. Case studies examine the conditions associated with
an individual fire or an outbreak of fires and usually cover a short period of
time and a relatively small area (approximately the fire area). Area-burned
studies relate the weather to the area burned from numerous fires over many
years. These area-burned studies have spatial domain ranges from hundreds of
square kilometers to continental scale and a temporal range of 10 years to about
100 years. The temporal scale is often limited by the availability of area-burned
data and meteorological data. This section will address both case studies and
area-burned studies. In some studies the area burned is related to indexes such
as components of fire danger rating systems or drought that are derived from
meteorological data.
There have been numerous case studies that address the weather associated
with a fire or an outbreak of fires. Most case studies have documented the
weather prior to and during major fire runs (Stocks and Walker, 1973; Stocks,
1975; Quintilio et al, 1911 \ Flannigan and Harrington, 1987; Hirsch and Flan-
nigan, 1990). These studies have shown that fires spread rapidly when the fu-
els were dry and the weather conditions were warm to hot, dry, and windy.
Turner (1970) studied the effect of hours of sunshine on fire season severity.
The synoptic weather types associated with critical fire weather were studied
by Schroeder (1964). These studies by themselves have limited application be-
cause of the narrow scope in terms of temporal and spatial scales used. How-
ever, they are of value in terms of identifying the most likely meteorological
Chapter 10 Climate, Weather, and Area Burned 353

predictors of area burned that can be used in studies with a larger time and
space domain (area-burned studies).
The most important elements of surface weather with respect to area burned
are cold frontal passage, dry spells, and low relative humidities. These last two
elements influence fuel moisture and the associated fire danger components
which were also found to be important in terms of area-burned activity. Brotak
and Reifsnyder (1977a) studied the synoptic weather conditions associated with
large wildland fires (over 2000 ha) over the eastern half of the United States for
the 1963-1973 period. They found that nearly 80% of these fires were associ-
ated with a cold front, either prior to or following passage of a dry cold front
(Figure 1). It is important to note that a surface wind shift from southwest to
northwest occurs with the passage of a cold front in the northern hemisphere.
This is important because the flank of afirewith a southwest wind, which would
typically be an ellipse in a SW-NE direction, would now become the head of the
fire with a northwest wind which can cause rapid and significant growth in the
fire. Haines et a\. (1983) used over 2500 wildfires from the northeastern United
States to examine the capability of the U.S. National Fire-Danger Rating System
(NFDRS) to predict four measures of fire occurrence: the probability of a fire
day, the number of fires per fire day, the number of fires per day, and the prob-
ability of a large-fire day. Good fits were found for the probability of a fire day
and for the number of fires per fire day, whereas significantly poorer results
were obtained for number of fires per day and the probability of a large fire day.

^ Following dry cold frontal passage

• Prior to cold frontal passage ^
+ Warm sector of low L^'^"^
D Warm sector of high ^ i

0 200 mi U / *
I
A A •

a
•
n

FIGURE 1 Idealized surface map showing relative position of major fires over eastern United
States (from Brotak and Reifsnyder, 1977a).
354 Flannigan and Wotton

Harrington et al. (1983) related the monthly provincial area burned in Canada
to components of the Canadian Fire Weather Index (FWl) System for 1 9 5 3 -
1980. They used mean and extreme values of components of the FWI System
calculated at 41 stations across Canada. Results showed that the monthly means
and extreme maximum values of the Duff Moisture Code (DMC—a model of
the moisture content in the upper portion of the organic layers of the forest
floor) and the Daily Severity Rating (DSR) were the best predictors of area
burned. In western Canada, with the exception of the Yukon and Northwest
Territories, the explained variance (r^) averaged 33%. In the Territories and
eastern Canada, the explained variance averaged 12%. Using the same data set,
Flannigan and Harrington (1988) studied the relation between meteorological
variables and monthly area burned by wildfire from May to August 1953-1980
for nine provincial-sized regions in Canada. They found that severe fire months
were independent of rainfall amount but significantly dependent on rainfall
frequency, temperature, and relative humidity. Results were similar to those
obtained by Harrington et a\. (1983) except that the meteorological variables
did better in the Northwest Territories and eastern Canada than the Canadian
FWI System (Van Wagner, 1987). The most important predictors were long se-
quences of days with less than 1.5 m m of rain and long sequences of days with
relative humidity below 60%. These long sequences were assumed to be asso-
ciated with blocking highs in the upper atmosphere.
The long-term average of area burned across a landscape is determined by a
complex set of variables including the size of the sample area, the period under
consideration, the extent of the forest, the topography, fragmentation of the
landscape (rivers, lakes, roads, agricultural land), fuel characteristics, season,
latitude, fire suppression policies and priorities, fire control, organizational size
and efficiency, fire site accessibility, ignitions (people and lightning), simulta-
neous fires, and the weather. Thus results that explain 30% of the variance over
large portions of the boreal forest are considered statistically significant.

III. WEATHER AND AREA BURNED-

UPPER AIR FEATURES
How do upper air features influence forest fires at the Earth's surface? There are
two ways that the upper atmosphere relates to forest fire activity. First, the
weather observed at the Earth's surface is a strong indicator of what is happen-
ing in the three-dimensional atmosphere. Temperature, precipitation, wind,
cloudiness, and atmospheric pressure depend largely on the horizontal and ver-
tical state of the atmosphere. The strength, location, and movement of surface
highs and lows and associated warm/cold fronts are functions of the three-
Chapter 10 Climate, Weather, and Area Burned 355

dimensional atmosphere. Thus surface weather conditions at the fire site are
greatly influenced by upper air features. Second, many large forest fires are not
constrained to the surface in that these fires may have convection columns that
extend many kilometers into the atmosphere. The interaction between the con-
vection column and the vertical structure of the atmosphere can have a signifi-
cant impact on fire behavior and fire growth (see Chapter 8 in this book).
There are two aspects of this section of relating upper air features to area
burned. The first aspect uses the atmospheric circulation at a specified pressure
level above the Earth's surface, typically the 500-mb level, and relates patterns
in the atmospheric flow to area burned by wildfires. The term long wave is used
with respect to the atmospheric circulation to denote northward or southward
displacements in the major belt of westerlies characterized by large wavelength
and significant amplitude, also known as planetary or Rossby waves. Typically,
three to five long waves can be found encircling the northern hemisphere at any
given time. The second aspect relates to the vertical structure of the atmosphere
and relates this structure to area burned by fire. Vertical profiles of temperature,
atmospheric moisture, and wind speed are commonly used in studies relating
atmospheric structure with area burned.

A. UPPER AIR—CIRCULATION

Upper atmospheric ridging is critical in terms of area burned. Newark (1975)

was among the first to notice that forest fire occurrence in northwestern On-
tario during the summer of 1974 was related to 500-mb-long wave ridging.
Nimchuk (1983) related two episodes of catastrophic burning during the Al-
berta 1981 fire season to the breakdown of the upper ridge over Alberta. These
episodes, which lasted 8 days, accounted for about IfiOOfiOO ha burned. The
breakdown of these upper ridges is often accompanied by increased lightning
activity as upper disturbances (short waves) move along the west side of the
ridge. Additionally, as the ridge breaks down, strong and gusty surface winds
are common. The ridge breakdown is preceded by warm and dry conditions as-
sociated with the upper ridge, which dries the forest fuels. These upper ridges
can persist for weeks to months (Daley, 1991). Brotak and Reifsnyder (1977a)
also studied the upper air conditions associated with 52 major wildland fires
(area burned 2000 ha or more) in the eastern United States during the 1963-
1973 period. They found that the vast majority of the fires were associated with
the eastern portion of a small but intense short wave trough at 500 mb (Fig-
ure 2). Despite the difference in geographical location, the studies by Brotak
and Reifsnyder (1977a) and Nimchuk (1983) may be discussing the same situ-
ation, although the emphasis changes from trough to ridge breakdown from
the former to the latter. Cold fronts are often associated with the breakdown of
356 Flannigan and Wotton

A Following dry cold frontal passage

• Prior to cold frontal passage
+ Warm sector of low
• Warm sector of high

FIGURE 2 Idealized 500-mb map showing relative position of major fires over eastern United
States (from Brotak and Reifsnyder, 1977a).

the ridge or the passing of a short-wave trough, which are also important in
terms of major wildland fires (Brotak and Reifsnyder, 1977a).
Some researchers have used height anomahes in the upper atmosphere to
address the relationship between long waves and area burned. Flannigan and
Harrington (1988) found that the monthly 700-mb height anomaly for the for-
ested regions of their provincial areas was the predictor that was selected the
most when relating meteorological variables to monthly provincial area burned
in Canada 1953-1980. The 700-mb height anomaly is calculated by subtracting
the climatological average 700-mb height, typically a 30-year average, from the
observed 700-mb height. Positive anomalies mean that the observed heights
are higher than average, and, conversely, negative anomalies mean observed
heights are lower than average. The height of the pressure level is a function of
the air column's temperature, with lower heights generally related to lower tem-
peratures and higher heights to warmer temperatures in the total air column
from the surface to the specified pressure level. More recently, some studies
have related height anomalies and blocking ridges to teleconnection patterns
between North America and the surrounding oceans. Johnson and Wowchuk
(1993) found that midtropospheric positive height anomalies (blocking ridges)
were related to large-fire years in the southern Canadian Rocky Mountains
whereas negative anomalies were related to small-fire years. They observed that
these blocking ridges associated with the large-fire years were teleconnected
(spatially and temporally correlated with respect to 500-mb heights) to upper-
Chapter 10 Climate, Weather, and Area Burned 357

level troughs in the North Pacific and eastern North America, which is the
positive mode of the Pacific North America (PNA) pattern (see teleconnection
section). Skinner et al. (1999) found that 500-mb height anomalies were well
correlated with seasonal area burned over various large regions of Canada. They
also found a structure similar to the PNA pattern for the extreme fire seasons
in western and west-central Canada. Current research suggests that blocking
frequency is related to the wave number (the number of long waves in the west-
erlies—typically three to five) with blocking ridges being more frequent with
higher wave numbers (Weeks et al., 1997).

B. VERTICAL STRUCTURE OF THE ATMOSPHERE

Fires can be divided into two groups: convection-column driven and wind
driven (Nelson and Adkins, 1988; Nelson, 1993). Fires in which the convec-
tion column is well developed are more sensitive to the atmospheric lapse rate
because these columns extend farther up into the atmosphere. The lapse rate is
the rate of temperature change with height above ground. Typically, tempera-
ture decreases with height above ground in the lower atmosphere. Inversions
denote those situations where temperature increases with increasing altitude.
Wind-driven fires are influenced by strong winds at the surface and typically
have significant vertical wind shear. Wind shear is the rapid change of speed
and/or direction in any direction; typically there is a significant increase of wind
speed with height above the ground as friction reduces the wind speed near the
Earth's surface. Under such conditions, the convection column is sheared and,
as such, does not play a significant role in fire spread and behavior. Addition-
ally, there is a dynamic interaction between the fire and the atmosphere which
can lead to some erratic fire behavior (see Chapter 8 in this book) such as fire
whirls, horizontal roll vortices, spotting, and the development of pyrocumulus
including thunderstorms generated by the fire (Stocks and Flannigan, 1987).
The vertical structure of the atmosphere influences fire behavior. Dry and
unstable air enhances the growth of forest fires in two ways: first, in the absence
of strong winds, it promotes a well-developed convection column, which may
produce spotting and other erratic fire behavior like fire whirls; second, when
wind speeds are strong near the Earth's surface, the instability allows these high
winds to be mixed down to the surface, promoting fire spread and erratic fire
behavior like horizontal roll vortices (Haines, 1982). An unstable lower atmo-
sphere was a common feature of the large fires that Brotak and Reifsnyder
(1977b) studied.
To address the role of the vertical structure of the atmosphere on fire activ-
ity, Haines (1988) developed a lower atmosphere severity index (LASI) for wild-
land fires to account for the influence of temperature stability and moisture
358 Flannigan and Wotton

content in the lower atmosphere. Unstable air is a layer of air that is character-
ized by a vertical temperature gradient such that air parcels displaced upward
will accelerate upward and away from their original altitude. Haines found that
only 6% of all fire season days fell into the high-index class for the western
United States. However, 45% of days with large and/or erratic wildfire occurred
during those high-index class days. Potter (1995) conducted a regional analysis
of the LASI in the United States and suggested that the Haines LASl might be
modified to improve performance in the southeastern United States. Werth and
Ochoa (1993) used the LASI to evaluate the growth of two forest fires in Idaho.
The LASI performed well at determining the time of the most explosive fire
growth. Brotak (1993) explored whether the LASI could be used in a predictive
sense by using early morning data (1200 GMT). His preliminary results sug-
gested that the LASI calculated using early morning data does not work well in
predicting the fire behavior for that afternoon.
The role of atmospheric stability and vertical moisture profile is a critical as-
pect for fire growth. Potter (1996) examined atmospheric properties associated
with large wildfires (over 400 ha) in the United States from 1971 to 1984. He
compared the lower atmosphere moisture, temperature, wind, and lapse rate for
the 339 large fires in the data set with climatology using the same 14-year pe-
riod. The results showed that the fire-day surface air temperature and moisture
variables (dewpoint depression and relative humidity) at 0000 UTC (Universal
Time Coordinated) differed significantly from climatological averages of these
variables. There was no difference in wind shear between fire days and clima-
tology days. Results from wind speed and lapse rate were inconclusive for this
study. Garcia Diez et al. (1993) found that the stability and the saturation defi-
cit of the lower atmosphere were related to the number of fires occurring in the
following 24 hours in Galicia (Spain). Delgado Martin et al. (1997) defined four
types of days based on stability and moisture levels: dry and unstable, moist and
unstable, dry and stable, and moist and stable. They found for northwestern
Spain that these four types of days and their associated fire risk were related to
area burned, with dry and unstable days having the highest fire risk. Addition-
ally, they suggested that days with a high number of fires are also those that
contribute to more burned area.
Future research relating all components of the weather (surface and upper
air features) should be pursued using a more rigorous spatial approach. For
example, using relatively high-resolution (approx. 50-100 km) gridded area
burned data in the analysis along with finer temporal resolution could yield a
better understanding of the role of weather in area burned and provide insight
into the relationship between surface weather and upper air features from an
area-burned perspective. If or when better relationships between weather and
area burned are obtained, it might be possible to predict area burned for a
season as seasonal climate prediction is beginning to show potential (Uppen-
brink, 1997).
Chapter 10 Climate, Weather, and Area Burned 359

IV. TELECONNECTIONS
Teleconnections are significant temporal correlations between meteorological
variables in widely separated points on the globe. Over the past 100 years of
meteorological study, a number of these correlations have been observed be-
tween various points. Perhaps the most well-known source of teleconnections
is the Southern Oscillation (SO) in the equatorial Pacific Ocean or more spe-
cifically the part of this oscillation known as El Nino (ENSO). El Nifio is char-
acterized by a strong warming of sea surface temperature in the eastern and
central equatorial Pacific Ocean and is accompanied by a weakening of trade
winds in this region. It occurs at irregular intervals from 2 to 7 years, though
its average return period is about 3 or 4 years. During an El Nino (or warm
phase of the Southern Oscillation) there is typically an area of high pressure
over the tropical western Pacific and unusually low pressure in the southeast-
ern Pacific near the coast of South America. The opposite phase of the South-
ern Oscillation, sometimes known as La Niiia, is characterized by higher sur-
face pressures in the eastern Pacific and cooler sea surface temperatures along
the equator.
The direct effects of this persistent weather pattern are quite evident in the
equatorial Pacific. For hundreds of years, the inhabitants of Peru have seen El
Nino's recurrence marked with widespread fish mortality along their coast and
increased rainfall and widespread flooding. In Australia, El Nifio events tend to
bring severe drought across large areas of the eastern part of the continent. The
resulting forest fire situations can be severe as was the case during the 1982/83
El Nino and the losses in forested land and life from the "Ash Wednesday" fires
(Pyne, 1991). The drought and consequent severe wildfire situation in Indone-
sia in 1997 were also a result of the disruption of normal monsoon rains by this
phenomenon.
Effects of this widespread warming of the ocean are not limited however to
the central and western Pacific. There has been a great deal of research into tele-
connections of El Nino and weather patterns farther abroad (Kiladis and Diaz,
1989). Correlations have been found between El Nino occurrence and hurri-
cane activity in the Atlantic (O'Brien et at, 1996) as well as drought occurrence
in regions of Africa (Bekele, 1997). Much work has been done looking for ENSO
teleconnections with weather patterns on the continent of North America as
well. Horel and Wallace (1981) showed that warm waters of the equatorial Pa-
cific were well correlated with positive 700-mb height anomalies over western
Canada and 700-mb height anomalies in the eastern United States. Bunkers
et al.'s (1996) study showed significant correlations between El Nino events
and increased summer precipitation in the northern plains of the United States.
The opposite correlation held for La Nina events. Shabbar and Khandekar
(1996) and Shabbar et at. (1997) described precipitation and temperature pat-
terns across Canada associated with the Southern Oscillation and suggested
360 Flannigan and Wotton

that the significant correlations they found could be used as a long-range fore-
casting technique particularly for the winter season. Such forecasting of tem-
perature and precipitation anomalies would be of importance to fire managers
as these are two important factors in estimating fuel moisture levels and the
consequent levels of forest fire potential.
Studies of fire occurrence in association with El Nino and La Nina events
have been carried out as well. Simard et al. (1985) divided the United States into
five regions and correlated fire occurrence and area burned per year in each of
these regions with a relative El Nino intensity index using about 73 years of his-
torical fire data. Their results imply that any relation between fire activity and
El Nino occurrence in the continental United States will be regional in char-
acter. They found a significant correlation between El Nino occurrence and de-
creased fire activity in the southern states. However, no other statistically sig-
nificant relationships could be found. Swetnam and Betancourt (1990) used fire
scars, tree growth chronologies, and fire statistics from Arizona and New Mex-
ico and correlated these with the Southern Oscillation Index (SOI). The SOI is
defined as the normalized difference in surface pressure between Tahiti, French
Polynesia, and Darwin, Australia. The SOI or its anomaly is often used as an
index of the strength and duration of El Nino and La Nina events. Positive SOI
values indicate a higher pressure in the eastern Pacific and, as such, a La Nina
event. Low SOI values indicate a lower pressure difference and an El Nino event.
Swetnam and Betancourt found that in Arizona and New Mexico large areas
burn after La Nina events and smaller areas burn after El Nifio events. This cor-
responded to drier and wetter springs in the American southwest associated
with La Nina and El Nino events, respectively. Brenner (1991) studied fire oc-
currence and area burned in Florida from 1950 to 1989. He found that a higher
than average area burned in La Nina years and a lower than average area burned
during El Niiio years. It is interesting to note that in this study the highest cor-
relations he found against area burned in Florida were with sea surface tem-
perature anomalies from the central Pacific. This is perhaps not surprising as
one would expect continental North America to be more strongly influenced by
North Pacific atmospheric circulations than those in the tropics. The potential
for North Pacific teleconnections with North American climate and fire activ-
ity will be discussed later in this section.
The temperature and precipitation teleconnection studies of Shabbar and
Khandekar (1996) and Shabbar et al (1997) suggest that a Southern Oscillation
signal might be detectable in Canada. They show significant decreases in win-
ter precipitation across the lower prairies and British Columbia and increases
in the Northwest Territories during El Nino years. La Nina years have a corre-
sponding increase in winter precipitation across the prairies. One might pre-
sume that the area burned during the fire season following such an El Nino or
La Nina winter precipitation anomaly might demonstrate some positive or nega-
Chapter 10 Climate, Weather, and Area Burned 361

4-

h 2
£
o
c
cc

O -2
C/3

-4

50 60 70 80 90 100
FIGURE 3 Southern Oscillation Index (SOI) and central Canada normalized area-burned anom-
aly for 1953 to 1995.

live correlation, respectively. Figure 3 shows normalized area burned anom-

aly in central Canada plotted with smoothed monthly SOI values from 1953 to
1996. There appears, in this graph, to be no strong correlations (negative or
positive) for this period. Correlation of winter levels of SOI and following sum-
mers area burned also yields no significant relationship.
Although some correlations have been found between the Southern Oscilla-
tion and fire occurrence in North America, the relative lack of signal might not
be surprising since El Nino events tend to be strongest in the North American
winter months and typically weaken or disappear as summer (and the fire sea-
son) approach. Indeed the correlation studies of Shabbar and Khandekar (1996)
and Shabbar et a\. (1997) show that links between ENSO and Canadian precipi-
tation and temperature are strongest in the winter months. However, recent re-
search by Bonsai and Lawford (1999) found a relationship between El Nino and
La Niiia events and summer extended dry spells on the Canadian prairies.
The results of Brenner (1991) suggest there may be some value in looking at
North Pacific sea surface temperatures for teleconnection not just with North
American climate but with fire activity. A number of studies have shown a cor-
relation between North Pacific sea surface temperatures and El Nino events
(Reynolds and Rasmusson, 1983; Trenberth, 1990; Deser andBlackmon, 1995).
Studies of the North Pacific have shown Southern Oscillation-like variability in
the sea surface temperature pattern, which imply that a similar North Pacific
Oscillation (NPO) exists (Tanimoto et al, 1993; Trenberth and Hurrell 1993;
362 Flannigan and Wotton

FIGURE 4 Map showing the geopotential height anomahes associated with the positive mode of
the PNA teleconnection (from Horel and Wallace, 1981). The arrows depict the midtropospheric
streamline as distorted by the anomaly pattern. The negative mode of the PNA has the anomalies
with an opposite sign; that is, the low-high-low anomaly sequence (North Pacific-western Canada-
southeast United States) of the positive mode switches to high-low-high anomaly sequence for the
PNA negative mode.

Zhang et al, 1997). A teleconnection has, in fact, been identified between the
North Pacific and North America. Known as the Pacific North American (PNA)
teleconnection (Figure 4), it is really a triple connection between an anti-
cyclonic circulation over the North Pacific, a cyclonic circulation over western
Canada, and a second anticyclonic circulation over the southeastern United
States. Walsh and Richman (1981) found a signature of this triple teleconnec-
tion when they used 31 years of North Pacific sea surface temperature data and
U.S. temperature data to show regional correlations. Their sea surface temper-
ature data correlated well with air temperature fluctuations over the far west-
ern states and the southeastern states but with opposite signs in the two re-
gions. This work also suggested that such correlations could potentially be used
for seasonal predictability in some areas. Johnson and Wowchuk (1993) found
a signature of the PNA triple teleconnection during a study of 35 years of area-
burned data in the southern Canadian Rocky Mountains. They found that, dur-
ing large-fire years, surface blocking high-pressure systems in their study area
were correlated with upper level troughs in the North Pacific and in eastern
North America. Skinner ct al (1999) found strong correlations between area
burned in a number of large regions in Canada and the presence of strong
500-mb-level ridging. During extremely high years of area burned, the PNA
Chapter 10 Climate, Weather, and Area Burned 363

teleconnection pattern was evident with significant correlations between the

500-mb values at the three "triplet" locations.
Teleconnections from sea surface temperatures such as the Pacific North
America pattern and those from the Southern Oscillation result from complex
ocean-atmosphere forcing coupled with large-scale atmospheric circulations.
Climate modeling with coupled atmosphere/ocean models has shown that
tropical flow patterns and consequent rainfall show little sensitivity to initial
conditions of the atmosphere but are determined mainly by boundary condi-
tions of sea surface temperature in the model (Shukla, 1998). The details of
these teleconnections are not as important to us in the context of this chapter
as the fact that the relatively high temporal variability of weather parameters
over land masses can be shown to be correlated with the more slowly varying
oceans. Weather prediction limits based on the intrinsic variability in the at-
mosphere are thought to be on the order of 2 weeks. However, ocean variability
occurs on a much slower time scale, and, as such, teleconnections imply that
seasonal climate prediction may be possible (Uppenbrink, 1997). The use of
such teleconnections as long-range forecasting tools has great appeal to all those
involved in the fire business. Recent research has shown that the phase of the
North Pacific Oscillation has a great influence on climate anomalies in North
America during the Southern Oscillation (Gershunov et al, 1999). They found
correlations between the Southern Oscillation and Pacific sea surface tempera-
tures and precipitation anomalies in North America to be much stronger during
the high phase of the North Pacific Oscillation. This has strong implications for
longer term predictions of North America's weather and climate as the North
Pacific Oscillation varies with a 50 -70-year time period (Minobe, 1997). Being
able to plan a fire season's activity during the winter before, based on current
sea surface temperature levels, could potentially allow fire management agen-
cies greater efficiency in allocating resources. Indeed, coupled atmosphere/
ocean modeling of the strong El Nino in 1997-1998 has shown very good re-
sults in predicting several months in advance seasonal precipitation anomalies
in North America (Kerr, 1998). However, predicting fire activity is a difficult
task. Forest fire occurrence and area burned tend to have a strong dependence
on shorter term local weather, particularly dry spells and wind events, than they
do on longer term conditions. Certainly, while an exceptionally dry or warm
winter would seem to indicate a potential for much drier fuels for the upcoming
fire season, small-scale synoptic events such as showers or thunderstorms tend
to drive day-to-day fire potential. However, the correlations already found be-
tween area burned and the Southern Oscillation and Pacific Ocean sea surface
temperatures show there is potential, with further analysis, that these telecon-
nections can be used to predict qualitatively the fire severity potential of the
upcoming fire season. Preliminary results from Flannigan et at (2000) show
364 Flannigan and Wotton

TABLE 1 Maximum Correlation between Pacific SSTs and Seasonal Provincial

Area Burned 1953-1976 and 1977-1995«

Province 1953-1976 1977-1995

BC 0.53 -0.52
16°S 106°E 20°N 102°W
Yukon/NWT 0.64 0.68
20°S 178°E 20°N 122°E
Alberta -0.72 0.55
36°N 142°W 0°N 162°E
Sask. 0.51 -0.55
16°N 166°W 12°S 162°E
Man. 0.61 0.60
24° N 174°W 12°N 130°E
W. Ont. -0.70 0.59
4°N 150°W 4°N 118°W
E. Ont. -0.70 0.64
20°N 134°W 4°S 146°E
Que. -0.72 0.62
4°N 102°W 20°S 106°E

"The location of the maximum correlation is displayed as well.

significant correlations between the winter season sea surface temperature

(Jan.-Apr.) and provincial seasonal area burned (May-Aug.) in Canada by
segregating the analysis by North Pacific Oscillation phase (Table 1). Figures 5
and 6 show the relationship between the seasonal area burned in eastern On-
tario and the lagged (Jan.-Apr.) sea surface temperature anomaly for the 1977-

2500
2000 • Area burned
1500 11 Seasonal SST

1995
FIGURE 5 Normalized seasonal area burned in Eastern Ontario and lagged (Jan.-Apr.) SST
anomaly for 1977-1995.
 365

1500

-1000-J
Area burned
FIGURE 6 Normalized SST anomaly (Jan.-Apr.) versus normalized seasonal area burned for
eastern Ontario 1977-1995.

1995 period with a time series and a scatter plot. The correlation is .64 (see
Table 1) which explains about 41% of the variation in the data.

V. FUTURE WARMING AND AREA BURNED

There is consensus in the scientific community that human activities are re-
sponsible for recent changes in the climate (Intergovernmental Panel on Cli-
mate Change, IPCC, 1996). Specifically, increases in radiatively active gases
such as carbon dioxide, methane, and the chlorofluorocarbons in the atmo-
sphere are causing a significant warming of the Earth's surface. Significant in-
creases in temperature are anticipated in the next century and are projected to
occur rapidly.
There are many General Circulation Models (GCMs) that enable researchers
to simulate the future climate. These models are three-dimensional representa-
tions of the atmosphere, ocean, cryosphere, and land surface. Transient simu-
lations are available from GCMs which allow examination of the possible rates
of change in the climate. Also, recent GCM simulations include sulfate aero-
sols. The major areas of uncertainty in GCMs include clouds and their radia-
366 Flannigan and Wotton

tive effects, the hydrological balance over land surfaces, and the heat flux at the
ocean surface (IPCC, 1996). Despite these uncertainties, the GCMs provide the
best means available to estimate the impact of changes in the future climate on
the fire regime at larger scales. Most models agree in predicting the greatest
warming at high latitudes in winter. Confidence is lower for estimates of pre-
cipitation fields, but many models suggest an increased moisture deficit partic-
ularly in the center of continents during summer. In addition to temperature,
other weather variables will be altered in a changed climate such as precipita-
tion, wind, and cloudiness. The variability of extreme events may also be al-
tered (Mearns et al, 1989; Solomon and Leemans, 1997). Some studies suggest
universal increases in fire frequency with climatic warming (Overpeck et ah,
1990; IPCC, 1996). The universality of these results is questionable because an
individual fire is a result of the complex set of interactions that include ignition
agents, fuel conditions, topography, and weather including temperature, rela-
tive humidity, wind velocity, and the amount and frequency of precipitation.
Increasing temperature alone does not necessarily translate into greater fire
disturbance.
Flannigan and Van Wagner (1991) compared seasonal fire severity rating
values (SSR—a component of the Canadian Forest Fire Weather Index System)
from a 2xC02 scenario (mid-21st century) versus the IXCO2 scenario (approx.
present day) across Canada. Their study used monthly anomalies from three
GCMs (Geophysical Fluid Dynamics Laboratory, GFDL; Goddard Institute for
Space Studies, GISS; and Oregon State University, OSU). The results suggest
increases in the SSR all across Canada with an average increase of nearly 50%,
which might translate roughly into an increase of area burned by 50%. Stocks
et al. (1998) used monthly data from four GCMs to examine climate change
and forest fire potential in Russian and Canadian boreal forests. Forecast sea-
sonal fire weather severity was similar for the four GCMs, indicating large in-
creases in the areal extent of extreme fire danger in both countries under a
2xC02 scenario. Stocks et al. (1998) also conducted a monthly analysis using
the Canadian GCM, which showed an earlier start to the fire season and sig-
nificant increases in the area experiencing high to extreme fire danger in both
Canada and Russia, particularly during June and July. Flannigan et al. (1998a)
used daily output from the Canadian GCM to model the Fire Weather Index for
both the IXCO2 and 2xC02 scenarios for North America and Europe. The FWl
represents the intensity of fire as energy output per unit length of fire front.
Figure 7 shows the ratio of the 2xC02 to IXCO2 values for both mean FWI and
maximum FWl for the 9 years of simulation for North America and Europe.
There is a great deal of regional variation between areas where FWl decreases in
a 2xC02 scenario (values below 1.00) to areas where the FWl increases greatly
in the warmer climate. Much of eastern Canada and northwestern Canada have
ratios below 1.00, indicating that the FWl decreases despite the warmer tem-
Chapter 10 Climate, Weather, and Area Burned

FIGURE 7 Mean and maximum FWI ratios (IxCOj/lxCOj) for North America (a, b) and Europe
(c, d) (Flannigan et al, 1998a).

peratures associated with a IxCOj climate. Noteworthy is the area of decreased

FWI over western and northwestern sections of Canada where historically
large portions of the landscape have been burned. However, due to the coarse
spatial resolution of the GCM (approx. 400 km), confidence in the results over
complex terrain like mountainous areas is low. In such areas, a Regional Cli-
mate Model (RCM) should be used (Caya et al, 1995) where the finer spatial
resolution (—40 km) can better resolve mountain ranges, and so on. Significant
increases in the FWI are evident over parts of central North America. The ratio
368 Flannigan and Wotton

of extreme maximum values of the FWI for the 9-year period show a similar
pattern, with higher ratios over central continental areas and lower values over
portions of eastern Canada. On the other hand, there are increases in the maxi-
mum FWI over portions of western Canada. Over northern Europe, Figure 7
shows increased mean FWls over the southern half of Sweden and extreme
southeastern Finland for warmer conditions, whereas the remainder of north-
ern Europe shows decreased mean FWIs. Interestingly, according to the simu-
lations, fire danger will decrease in the northern sections of northern Europe
where traditionally forest fires have been large and frequent, and fire danger
will increase over southern Sweden where forest fires are not normally severe.
Results for maximum FWls show a similar pattern when compared with the
mean FWIs, with the exception of southern Norway where FWIs increased.
Consequences of climate change on fire disturbance must be viewed in a spa-
tially dependent context.
Flannigan et al. (1998a) suggest that the reason for the decreased FWI de-
spite the increasing temperature is due primarily to changes in the precipita-
tion regime and, in particular, to increases in precipitation frequency. These
model results (Figure 7) are in good agreement with recent fire history studies
(~ the last 200 years) (Flannigan et al. 1998a), many of which show decreasing
fire activity despite the warming since the end of the Little Ice Age (—1850).
These modeled results are also consistent with the modeled fire weather and
charcoal record anomalies for a warm period during the mid-Holocence about
6000 years BP, which was about 1°C warmer than present for Canada (Flanni-
gan et aL, 1998b).
Other factors such as ignition agents, length of the fire season, and fire man-
agement policies may greatly influence the impact of climate change on the fire
regime. Ignition probabilities may increase in a warming world due to in-
creased cloud-to-ground lightning discharges (Price and Rind, 1994). Price
and Rind suggest that lightning-caused fires would increase by 44% and that the
associated area burned would increase by 78% by the end of the 21st century,
although they assume no changes in fuels, which may greatly influence the
lightning ignitions and area burned. The longer fire season will begin earlier in
the spring and extend longer into the autumn. Wotton and Flannigan (1993)
estimated that the fire season length in Canada on average will increase by 22%,
or 30 days in a 2xC02 world. Fire management policies and effectiveness will
continue to change. Also, research has suggested that the persistence of block-
ing ridges in the upper atmosphere will increase in a 2xC02 climate (Lupo et al,
1997), which could have a significant impact on forest fires as these upper
ridges are associated with dry and warm conditions at the surface and are con-
ducive to forest fires. These are all confounding effects that may dampen or am-
plify the impact of a changing climate on the fire regime.
Future work might include using the improved relationships between
climate/weather and area burned to estimate what the fire regime will be like
Chapter 10 Climate, Weather, and Area Burned 369

in the future. This would assist land managers with long-term planning. Also,
predictions of future fire regimes are critical in terms of the carbon balance
where changes in the natural disturbance regime like fires could be responsible
for the boreal forest becoming a source of carbon rather than a sink (Solomon
and Leemans, 1997).
The important aspect of the impact of climate change on forest fires with re-
spect to the influence on vegetation is that fire may be more important than di-
rect effects of climate change (Weber and Flannigan, 1997). Fire can act as an
agent of change to hasten the modification of the vegetation landscape into a
new equilibrium with the climate if species are able to migrate fast enough.

VI. SUMMARY
Weather is a critical factor in determining the timing and size of fires. We have
seen in this chapter that dry, windy, and warm surface weather conditions are
conducive to area burned. However, these surface weather conditions are a
reflection of the synoptic weather pattern which is a function of the dynamics
of the three-dimensional atmosphere. The breakdown of the upper blocking
ridge is one pattern of the upper atmospheric circulation that is a common fea-
ture of significant area-burned events. These blocking ridges typically bring
7-10 days or longer periods of warm dry weather followed by windy conditions
as the ridge deteriorates. The breakdown of the upper ridge is often associated
with a cold front passage at the surface which has been shown to be a common
factor of large fires. Additionally, because of some large-scale circulation pat-
terns, there are significant relationships between the weather in widely sepa-
rated points on the globe, and often these circulation patterns have a cyclical
nature such that seasonal forecasts may be viable. Last, preliminary indications
from climate-modeling work suggest that climate change may result in in-
creased area burned for many regions of the world.

ACKNOWLEDGMENTS
We thank two anonymous reviewers whose comments and suggestions improved this chapter.
We also thank Brenda Laishley for editing this chapter.

REFERENCES
Bekele, F. (1997). Ethiopian use of ENSO information in its seasonal forecasts. Int. J. African
Stud. 2. http://www.brad.ac.uk/research/ijas/ijasno2/ijasno2.html
Bonsai, B. R., and Lawford, R. G. (1999). Teleconnections between El Nino and La Nifia events and
summer extended dry spells on the Canadian Prairies. Int. J. Climatol. 19, 1445-1458.
370 Flannigan and Wotton

Brenner, J. (1991). Southern Oscillation anomalies and their relationship to wildfire activity in
Florida. Int.J. WildlandFire 1, 73-78.
Brotak, E. A. (1993). Low-level weather conditions preceding major wildfires. Fire Manage. Notes
54,23-26.
Brotak, E. A., and Reifsnyder, W. E. (1977a). An investigation of the synoptic situations associated
with major wildland fires. J. Appl. Meteorol. 16, 867-870.
Brotak, E. A., and Reifsnyder, W. E. (1977b). Predicting major wildland fire occurrence. Fire Man-
age. Notes 38, 5-8.
Bunkers, M. J., Millar, Jr., J. R., and deGaetano, A. T. (1996). An examination of El Nino-La Nina-
related precipitation and temperature anomalies across the northern plains. J. Climatol. 9, 147-
160.
Caya, D., Laprise, R., Giguere, M., Bergeron, G., Blanchet, J. R, Stocks, B. J., Boer, G. J., and McFar-
lane, N. A. (1995). Description of the Canadian regional climate model. Water, Air Soil Pollut.
82,477-482.
Daley, R. (1991). "Atmospheric Data Analysis." Cambridge University Press, Cambridge.
Delgado Martin, L., Garcia Diez, A., Rivas Soriano, L., and Garcia Diez, E. L. (1997). Meteorology
and forest fires: Conditions for ignition and conditions for development. J. Appl. Meteorol. 36,
705-710.
Deser, C , and Blackmon, M. L. (1995). On the relationship between tropical and North Pacific sea
surface temperature variations. J. Climatol. 8, 1677-1680.
Flannigan, M. D., and Harrington, J. B. (1987). Synoptic conditions during the Porter Lake burn-
ing experiment. Climatol. Bull. 21, 19-40.
Flannigan, M. D., and Harrington, J. B. (1988). A study of the relation of meteorological variables
to monthly provincial area burned by wildfire in Canada 1953-80. J. Appl. Meteorol. 27, 4 4 1 -
452.
Flannigan, M. D., and Van Wagner, C. E. (1991). Climate change and wildfire in Canada. Can. J.
For. Res. 21,66-72.
Flannigan, M. D., Bergeron, Y., Engelmark, O., and Wotton, B. M. (1998a). Future wildfire in cir-
cumboreal forests in relation to global warming. J. Veg. Sci. 9, 477-482.
Flannigan, M. D., Todd, B., Wotton, B. M., Skinner, W R., Stocks, B. J., and Martell, D. L. (2000).
Pacific sea surface temperatures and their relation to area burned in Canada. In "Preprints of
Third Symposium on Fire and Forest Meteorology," pp. 151-157. American Meteorological So-
ciety, Boston.
Flannigan, M. D., Wotton, B. M., Richard, P., Carcaillet, C , and Bergeron, Y. (1998b). Fire weather:
Past, present and future. In "Preprints of Ninth Symposium on Global Change Studies," pp. 305 -
309. American Meteorological Society, Boston.
Garcia Diez, E. L., Labajo S., and de Pablo Davila, F. (1993). Some meteorological conditions asso-
ciated with forest fires in Galicia (Spain). Int. J. Biometeorol. 37, 194-199.
Gershunov, A., Barnett, T. P, and Cayan, D. R. (1999). North Pacific interdecadal oscillation seen
as factor in ENSO-related North American climate anomalies. FOS Trans. Am. Geophys. Union
80, 25, 29-30.
Haines, D. A. (1982). Horizontal roll vortices and crown fires. J. Appl. Meteorol. 21, 751-763.
Haines, D. A. (1988). A lower atmosphere severity index for wildland fires. Nat. Weath. Digest 13,
23-27.
Haines, D. A., Main, W A., Frost, J. S., and Simard, A. J. (1983). Fire-danger and wildfire occur-
rence in the northeastern United States. For. Sci. 29, 679-696.
Harrington,]. B., Flannigan, M. D., and Van Wagner, C. E. (1983). "A Study of the Relation of Com-
ponents of the Fire Weather Index System to Monthly Provincial Area Burned by Wildfire in
Canada 1953-80." Inf. Rep. Pl-X-25. Canadian Forest Service, Petawawa National Forestry In-
stitute, Chalk River.
Chapter 10 Climate, Weather, and Area Burned 371

Harvey, D. A., Alexander, M. E., and Janz, B. (1986). A comparison of fire-weather severity in
northern Alberta during the 1980 and 1981 fire seasons. For. Chron. 62, 507-513.
Hirsch, K. G., and Flannigan, M. D. (1990). Meteorological and fire behavior characteristics of the
1989 fire season in Manitoba, Canada. In "International Conference on Forest Fire Research,"
pp. B.06-1-B.06-16, Coimbra.
Horel, J. D., and Wallace, J. M. (1981). Planetary-scale atmospheric phenomena associated with the
southern oscillation. Mon. Weather Rev. 109, 813-829.
Intergovernmental Panel on Climate Change. (1996). "Climate Change 1995 Impacts, Adaptations
and Mitigation of Climate Change: Scientific-Technical Analyses." Cambridge University Press,
Cambridge.
Johnson, E. A. (1992). "Fire and Vegetation Dynamics: Studies from the North American Boreal
Forest." Cambridge University Press, Cambridge.
Johnson, E. A., and Wowchuk, D. R. (1993). Wildfires in the southern Canadian Rocky Mountains
and their relationship to mid-tropospheric anomalies. Can. J. For. Res. 23, 1213-1222.
Kerr, R. A. (1998). Models win big in forecasting El Nino. Science 280, 522-523.
Kiladis, G. N., and Diaz, H. F. (1989). Global climate anomalies associated with extremes in the
southern oscillation. J. Climatol. 2, 1069-1090.
Lupo, A. R., Oglesby, R. J., and Mokhov, 1. 1. (1997). Climatological features of blocking anticy-
clones: A study of Northern Hemisphere CCMl model blocking events in present-day and
double CO2 concentration atmosphere. Clim. Dynam. 13, 181-195.
Mearns, L. O., Schneider, S. H., Thompson, S. L., and McDaniel, L. R. (1989). Climate variability
statistics from General Circulation Models as applied to climate change analysis. In "Natural
Areas Facing Climate Change" (G. P. Malanson, Ed.), pp. 51-73. SPB Academic Publishing,
The Hague.
Minobe, S. (1997). A 50-70 year climatic oscillation over the North Pacific and North America.
Geophys. Res. Let. 24, 683-686.
Nelson, R. M., Jr. (1993). Byram's derivation of the energy criterion for forest and wildland fires.
Int. J. Wildland Fire 3, 131-138.
Nelson, R. M. Jr., and Adkins, C. W (1988). A dimensionless correlation for the spread of wind-
driven fires. Can. J. For. Res. 18, 391-397.
Newark, M. J. (1975). The relationship between forest fire occurrence and 500 mb ridging. Atmo-
sphere 13, 26-33.
Nimchuk, N. (1983). "Wildfire Behavior Associated with Upper Ridge Breakdown." ENR Rep. No.
T/50. Alberta Energy and Natural Resources, Forestry Service, Edmonton.
O'Brien, J. J., Richards, T. S., and Davis, A. C. (1996). The effect of El Nino on US land falling hurri-
canes. Bull. Amer. Meteorol. Soc. 77, 773-774.
Overpeck, J. T., Rind, D., and Goldberg, R. (1990). Climate-induced changes in forest disturbance
and vegetation. Nature 343, 51-53.
Potter, B. E. (1995). Regional analysis of Haines' LAST Fire Manage. Notes 55, 30.
Potter, B. E. (1996). Atmospheric properties associated with large wildfires. Int. J. Wildland Fire 6,
71-76.
Price, C , and Rind, D. (1994). The impact of a 2xC02 climate on lightning-caused fires. J. Clima-
tol. 7, 1484-1494.
Pyne, S. J. (1991). "Burning Bush: A History of Fire in Australia." Henry Holt and Company,
New York.
Quintilio, D., Fahnestock, G. R., and Dube, D. E. (1977). "Fire Behaviour in Upland Jack Pine: The
Darwin Lake Project." Inf. Rep. NOR-X-174. Environment Canada, Canadian Forest Service,
Northern Forestry Research Centre, Edmonton.
Reynolds, R. W, and Rasmusson, E. M. (1983). The North Pacific sea surface temperature associ-
ated with El Niiio events. In "Proceedings of the Seventh Annual Climate Diagnostics Work-
372 Flannigan and Wotton

shop." U.S. Dept. of Commerce, National Oceanic and Atmospheric Administration, Washing-
ton, DC.
Schroeder, M. J. (1964). "Synoptic Weather Types Associated with Critical Fire Weather." USDA
Forest Service, Pacific Southwest Forest and Range Experiment Station, Berkeley.
Shabbar, A., and Khandekar, M. (1996). The impact of El Nino-Southern Oscillation on the tem-
perature field over Canada. Atmosphere-Ocean 34, 401-416.
Shabbar, A., Bonsai, B., and Khandekar, M. (1997). Canadian precipitation patterns associated with
the southern oscillation. J. Climatol. 10, 3016-3027.
Shukla, J. (1998). Predictability in the midst of chaos: A scientific basis for climate forecasting. Sci-
ence 282, 728-731.
Simard, A. J., Haines, D. A., and Main, W. H. (1985). Relations between El Nino/southern oscilla-
tion anomalies and wildland fire activity in the United States. Agric. For. Meteorol. 36, 93-104.
Skinner, W R., Stocks, B. J., Martell, D. L., Bonsai, B., and Shabbar, A. (1999). The association be-
tween circulation anomalies in the mid-troposphere and area burned by wildland fire in Can-
ada. Theor. Appl Climatol. 63, 89-105.
Solomon, A. M., and Leemans, R. (1997). Boreal forest carbon stocks and wood supply: Past, pres-
ent and future responses to changing climate, agriculture and species availability. Agric. For.
Meteorol. 84, 137-151.
Stocks, B.J. (1975). "The 1974 Wildfire Situation in Northwestern Ontario." Inf. Rep. O-X-232.
Canadian Forest Service, Great Lakes Forest Research Centre, Sault Ste. Marie.
Stocks, B. J., and Flannigan, M. D. (1987). Analysis of the behaviour and associated weather for a
1986 Northwestern Ontario wildfire: Red Lake 7. In "Proceedings of the Ninth Conference on
Fire and Forest Meteorology," pp. 94-100. American Meteorological Society, Boston.
Stocks, B. J., and Walker, J. D. (1973). "Climatic Conditions Before and During Four Significant
Forest Fire Situations in Ontario." Inf Rep. O-X-187. Canadian Forest Service, Great Lakes For-
est Research Centre, Sault Ste. Marie.
Stocks, B. J., Fosberg, M. A., Lynham, T. J., Mearns, L., Wotton, B. M., Yang, Q., Jin, J-Z., Lawrence,
K., Hartley, G. R., Mason, J. A., and McKenney, D. W (1998). Climate change and forest fire po-
tential in Russian and Canadian boreal forests. Climatic Change 38, 1-13.
Swetnam, T. W (1993). Fire history and climate change in giant sequoia groves. Science 262,
885-889.
Swetnam, T. W, and Betancourt, J. L. (1990). Fire-southern oscillation relations in the south-
western United States. Science 249, 1017-1020.
Tanimoto, Y., Iwasaka, N., Hanawa, K., and Toba, Y. (1993). Characteristic variations of sea surface
temperature with multiple time scales in the north Pacific. J. Climatol. 6, 1153-1160.
Trenberth, K. E. (1990). Recent observed interdecadal climate changes in the northern hemisphere.
Bull. Amer. Meteorol. Soc. 71, 988-993.
Trenberth, K., and Hurrell, J. W (1993). Decadal atmosphere-ocean variations in the Pacific. Clim.
Dynam. 9,303-319.
Trewartha, G. T., and Horn, L. H. (1980). "An Introduction to Climate," 5th ed. McGraw-Hill,
New York.
Turner, J. A. (1970). Hours of sunshine and fire season severity over the Vancouver Forest District.
For. Chron.46, 106-111.
Uppenbrink, J. (1997). Nota bene: Climate—seasonal climate prediction. Science 277, 1952.
Van Wagner, C. E. (1987). "The Development and Structure of the Canadian Forest Fire Weather
Index System." Tech. Rep. 35. Canadian Forest Service, Ottawa.
Walsh, J. E., and Richman, M. B. (1981). Seasonality in the associations between surface tempera-
ture over the United States and the north Pacific Ocean. Mon. Wea. Rev. 10, 767-782.
Weber, M. G., and Flannigan, M. D. (1997). Canadian boreal forest ecosystem structure and func-
tion in a changing climate: Impact on fire regimes. Environ. Rev. 5, 145-166.
Chapter 10 Climate, Weather, and Area Burned 373

Weeks, E. R., Tian, Y., Urbach, J. S., Ide, K., Swinney, H. L., and Ghil, M. (1997). Transitions be-
tween blocked and zonal flows in a rotating annulus with topography. Science 278,1598-1601.
Werth, P., and Ochoa, R. (1993). The evaluation of Idaho wildfire growth using the Haines Index.
Weather and Forecasting 8, 223-234.
Wotton, B. M., and Flannigan, M. D. (1993). Length of the fire season in a changing climate. For.
Chron. 69, 187-192.
Zhang, Y., Wallace, J. M., and Battisti, D. S. (1997). ENSO-like interdecadal variability: 1900-93.
J. Climatol 10, 1004-1020.
This Page Intentionally Left Blank
 CHAPTER 11
Lightning and Forest Fires
D O N LATHAM
USDA Forest Service, Fire Sciences Laboratory, Rocky Mountain Research Station,
Missoula, Montana

EARLE WILLIAMS
Parsons Laboratory, Massachusetts Institute of Technology, Cambridge, Massachusetts

I. Introduction
11. Lightning
A. The Origin and Characteristics of Lightning
III. Previous Studies of Lightning-Initiated Fire
A. Early Lightning Fire Research
B. Project Skyfire
C. Recent Studies
IV. Interaction between Lightning and Fuels
A. Lightning Paths on Trees
B. Ecological Damage Not Involving Fire
C. Fuels on the Forest Floor
V. How Ignition Occurs
A. Formulation of the Problem
B. Energy Transfer from the Lightning Channel
C. Structure of the Continuing Current Channel
D. Requirements for Fuel Ignition by Lightning
E. Theoretical Heat Flux Calculations
VI. Ignition Experiments with Real Forest Fuels
A. Apparatus and Experimental Design
B. Experimental Results
VII. Generating Models for Operational Use
A. Criteria for Utility
B. Probability Distributions for
Continuing Currents
C. Conditional Probability of Ignition
D. Implementation of Ignition Probability
VIII. Smoke, Lightning, and Cloud Microphysics

Forest Fires
Copyright ® 2001 by Academic Press. All rights of reproduction in any form reserved. 375
376 Latham and Williams

IX. Global Implications of Lightning

Ignition Characteristics
X. Conclusion
References

I. INTRODUCTION
Lightning ignition of wildland fuels plays a major role in the maintenance and
evolution of ecosystems. Lightning not only ignites fire but also weakens trees,
facilitating insect and disease attack, causes physical damage, and kills trees and
groups of trees (Taylor, 1973). Lightning ignition may also play a large part in
forest response to global climate change.
Fire can also alter lightning. Cumulus clouds, called pyrocumulus, can be
formed under the proper conditions by large fires. Pyrocumulus clouds have a
high proportion of positive cloud-to-ground lightning, the opposite of lightning
from a "normal" cumulus (Latham, 1991). One hypothesis for the cause is an
interaction between the earth's electric field and the fire (Vonnegut et a/., 1995).
Smoke from fires can apparently cause inverted storms over a large area (Lyons
etal, 1998a).
In this chapter, we will show how the predominant mechanism for lightning
ignition works and how this knowledge can be used for forest health and fire
management. We will discuss worldwide lightning and fire patterns. Before dis-
cussing ignition and other consequences, a brief introduction to lightning is
supplied; it is a subject not usually discussed in a forestry curriculum.

II. LIGHTNING
A. T H E ORIGIN AND CHARACTERISTICS
OF LIGHTNING

Lightning is an energetic electrical discharge caused by the separation of posi-

tive and negative charge in clouds leading to voltage differences of order 1 0 -
100 MV. The weight of the evidence shows that the process of charge separation
requires the presence of water substance in all three phases—solid, liquid, va-
por (Williams, 1989). Such a condition is called mixed phase. The mixed phase
region of the atmosphere is bounded below by the 0°C isotherm (where ice par-
ticles melt) and above by the — 40°C isotherm (where supercooled liquid water
is spontaneously transformed to ice). Under typical conditions, the mixed phase
region is 4000 -5000 meters thick. Though a wide variety of meteorological
Chapter 11 Lightning and Forest Fires 377

conditions can lead to lightning, for example, isolated air mass thunderstorms,
frontal storms, and snowstorms, the presence of a mixed phase region appears
to be a necessary condition for lightning generation.
Separation of positive and negative charge in regions with both ice particles
and water droplets (mixed phase) is thought to be caused by the collisions of
ice particles prevalent in such conditions: graupel and ice crystals (or smaller
graupel). Graupel are millimeter to centimeter-sized hydrometeors, which grow
by the accretion of supercooled cloud droplets. Ice crystals grow by water vapor
deposition in a mixed phase environment at the expense of the liquid water be-
cause the equilibrium vapor pressure with respect to ice is less than the vapor
pressure with respect to liquid. In ordinary thunderstorms, for reasons that are
still poorly understood, negative charge is selectively transferred to the graupel
particles and positive charge, to the smaller ice crystals. The faster falling grau-
pel particles carry negative charge downward relative to the positive charge at
higher levels, creating the positive thundercloud dipole. The resulting lower
negative charge region is typically in the —5 to — 20°C zone with the positive
charge at higher (colder) levels. In summertime, the lower negative region is
5-7 km above mean sea level, with the positive region at 7-12 km. This posi-
tive dipole is responsible for the two most prevalent lightning types: intracloud
(IC) lightning and cloud-to-ground (CG) lightning.
Intracloud lightning is a discharge bridging the upper positive and lower
negative charge of the thundercloud dipole and is contained entirely within the
cloud, therefore playing no role whatever in fire initiation at the ground. IC
lightning is almost invariably the first lightning to occur in a developing storm;
some storms produce only IC lightning. The IC rate can become very large dur-
ing strong convective surges when deep mixed phase development is underway.
Though some IC flashes can be quite energetic, the majority have smaller peak
currents and smaller charge moments (the charge amount multiplied by the dis-
tance the charge moves) and involve less total energy than CG flashes. One im-
portant reason for the difference is that IC flashes can occur over small scales
(hundreds of meters) whereas ordinary negative ground flashes are required to
bridge the typical gap to ground of 5000-7000 m.
The lightning ground flash, as the active player in forest fire initiation,
deserves more detailed discussion. CG lightning is most commonly initiated
within the cloud, in the vicinity of the lower charge reservoir (usually negative).
An ionized path, the stepped leader, is forged through the air toward ground in
a region of high electric field. This stepped leader carries the large negative po-
tential of the lower charge region toward Earth. As the stepped leader nears the
Earth, an intense electric field develops between leader and ground. The field
promotes electrical streamer propagation upward from elevated points on the
ground that can connect to the approaching leader. When a connection is
made, the bright, high-current (10-100 kA) return stroke is initiated and prop-
378 Latham and Williams

agates upward toward the cloud at a speed approaching that of hght (1-2 X
10^ = m s~^). For reasons not well understood, this leader-return stroke se-
quence is often repeated at intervals of a few tens of milliseconds, causing the
characteristic flicker of lightning. The extraordinary brightness of the return
stroke channel may give the impression that the lightning channel is very broad.
In reality, channel diameters are in the range of millimeters to centimeters, an
important consideration for the modeling discussion of Section V (e.g., Orville
andHelsdon, 1974).
In the majority of ground flashes, the return stroke current peaks, in less
than 1 )Lis, to values in the range of 5 -30 kA and then promptly decays in a few
hundred microseconds. Despite the extraordinary peak power in such events
(upwards of 10^^ W), both observations and simulations have shown that the
short duration of the return stroke is inadequate to raise trees and other flora in
its path to kindling temperature and initiate fire (Taylor, 1969; Darveniza and
Zhou, 1994).
In about 30% of return strokes, a sustained current of low amplitude, a con-
tinuing current, is observed to flow in the channel to ground immediately fol-
lowing the current peak for a period varying from milliseconds to hundreds of
milliseconds (Uman, 1969). The conditions necessary for continuing currents
are still not well resolved, but observations suggest that larger-than-usual reser-
voirs of electric charge in the cloud are necessary to sustain long continuing
currents and that the return strokes initiating them are somewhat smaller in
current amplitude than others. The early ground flashes in a developing storm,
when the charge reservoir is still of modest size, seldom exhibit continuing cur-
rents. The largest and most energetic continuing currents are observed by ELF
methods (Burke and Jones, 1996; Cummer and Inan, 1996; Huang et al, 1999)
beneath the extensive stratiform regions of precipitation common in mesoscale
convective systems (Williams, 1998) and large winter snowstorms. Flashes
having at least one stroke with a continuing current are called hybrid flashes.
Strokes within flashes without a continuing current are called discrete strokes.
Approximately 90% of CG flashes worldwide transfer negative electric
charge to ground. The 10% positive minority, whose physical origins are still
poorly understood, are more prevalent in the dissipating stages of local thunder-
storms, in the broad stratiform regions of mesoscale convective systems, in
winter storms, and in thunderstorms ingesting smoke from fire (Latham, 1991;
Lyons et al, 1998a; Section VIII). The continuing current characteristics of pos-
itive and negative ground flashes are markedly different, and these differences
are important in the context of fire initiation. The great majority of positive
ground flashes are single stroke, and a continuing current follows almost all of
these strokes. The majority of negative flashes are multistroke, and about half
of these have accompanying continuing currents (Uman, 1969).
Chapter 11 Lightning and Forest Fires 379

The development of statistics on parameters relevant to CG lightning has

been facilitated by the availability of networks for ground flash detection,
such as the National Lightning Detection Network currently operational in the
United States (Global Atmospherics, Inc.) (Cummins et al, 1998). This network
can now locate the great majority of CG contact points within the contiguous
United States and Canada to within a few kilometers spatially and within 1 fis
accuracy in time. The polarity, return stroke peak current, and stroke multi-
plicity are now routinely archived for all ground flashes. Unfortunately, due to
the limited low-frequency response of this equipment, and the limited range of
the continuing current electric field change, no information on the continuing
current component of ground flashes is presently available operationally. Most
information on this component comes from local storm studies (Uman, 1969).
Recent studies with both the NLDN (Lyons et al, 1998b) and ELF electro-
magnetic sensors (Boccippio et al, 1995; Burke and Jones, 1996; Cummer and
Inan, 1996; Huang et al, 1999) have focused attention on positive ground
flashes, which are likely the most energetic lightning flashes on the planet. Radar
observations (Ligda, 1956) presented evidence for horizontal extension of light-
ning within the cloud for distances exceeding 100 km. It is now well recognized
that this "spider" or "sheet" lightning ultimately connects to Earth in a positive
ground flash with a long continuing current (Boccippio et al, 1995; Mazur et al,
1998). Such discharges are also capable of lone excitation of the electromag-
netic Schumann resonances of the Earth-ionosphere cavity to levels exceeding
all other lightning combined (Sentman, 1987; Boccippio et al, 1995; Burke and
Jones, 1996; Huang et al, 1999). These spectacular discharges have also been
clearly identified with sprites and elves, newly discovered luminous discharge
phenomena in the mesosphere 60-90 km above the causal positive lightning.
Both the energetic nature of positive flashes and their high probability to ex-
hibit continuing currents have led to popular belief that this lightning type is
the leading initiator of forest fire. This common view has been questioned, how-
ever, by Flannigan and Wotton (1991), who call attention to the large numbers
of negative ground flashes with continuing current.
Lightning's potential to initiate fire has also been judged, incorrectly, on the
basis of its visual appearance—"hot" lightning is bright white or bluish light
and "cold" lightning has a dull red appearance. Quantitative analysis of the op-
tical spectrum of lightning (Orville and Hendersen, 1986) has shown that the
intrinsic spectra of all return stroke channels (the predominant source of light-
ning light) exhibit a very broad optical region, similar to sunlight. The light
from very distant lightning (often referred to as "heat" lightning) is selectively
filtered in the blue end of the spectrum by Rayleigh scattering, much like the
light from the setting sun, leaving a predominance of red light. Lightning at
closer range will suffer less selective loss of blue light and will appear whiter in
380 Latham and Williams

color. As discussed in Section VLB, the most critical parameter for lightning ig-
nition of forest fuels is the presence and duration of the continuing current and
not the temperature of the lightning plasma.

III. PREVIOUS STUDIES OF

LIGHTNING-INITIATED FIRES
In this section, observations and summaries of lightning-initiated fires are dis-
cussed. The studies can be broken arbitrarily into three categories: early. Proj-
ect Skyfire, and recent. The early studies were done before the characteristics of
lightning were discovered and researched. Project Skyfire was designed to study
the possibility of reducing fire numbers by cloud seeding. Recent studies bene-
fit from lightning location data and organized descriptors of forest fuels.

A. EARLY LIGHTNING FIRE RESEARCH

Plummer (1912) observed:

The same flash may strike and blast a number of trees, and the resuUs may be quite
as curious and erratic as the hghtning itself. A tree may be scorched, it may be stripped
of its leaves, it may be cleft longitudinally, or, more rarely, severed horizontally. Pieces
of bark or wood may be torn off in strips. One-half of a tree's crown may be withered,
the other half remaining unharmed. Sometimes the bark is stripped from only one
side, occasionally without a trace of burning; at other times it may be riddled, as by
worms, with a multitude of little holes. The lightning furrow on a tree is usually
single; but it may be double, usually in parallel lines. Furrows may be oblique or spi-
ral, the current in such cases following the grain of the new wood. If the tree is in-
flammable or is rendered very dry by the flash a fire may result. In other cases the dry
duff or humus at the base of the tree is ignited by the flash.

This description is entirely adequate for today's observations.

Plummer's attempt to explain the wide range of visible effects hypothesized
distinctly different "upward-going" and "downward-going" flashes to ground,
causing different kinds of damage to trees. He does not specify what damage is
caused by which direction. As discussed in Section II, the situation is more
complex. In any case, the "direction" of flashes, and their complexity, was not
firmly demonstrated until 1934 (Schonland and Collens, 1934), so there was
no way for Plummer to know of the complexity or to formulate a test for the
hypothesis.
Another hypothesis of the time was that some species of trees were preferen-
tially hit by lightning. Most of the research was done in Europe and on Forest
Service lands in the United States (as summarized in Plummer, 1912). Unfor-
Chapter 11 Lightning and Forest Fires 381

tunately, the data were not corrected for the relative density of tree type, so the
results are questionable. In fact, Plummer's conclusions state, in part, that any
kind of tree is likely to be struck by lightning and that the greatest number
struck in any locality will be of the dominant species, conclusions that hold to-
day. As we shall see, some species do have a higher probability of ignition from
lightning, due mainly to their differing foliage cast.
Another interesting note from Plummer is that "Lightning is extremely rare
in Alaska, and no forest fires are known to have resulted from it" in contrast to
about 200 reported lightning fires per year from modern reports (Bureau of
Land Management, private communication). Lightning fires are currently im-
portant enough in Alaska that it was chosen for the first trials of lightning loca-
tion systems for fire management.
Subsequent lightning ignition studies were done in the northwestern area
of the United States because most lightning-caused fires occur there. The area
includes western Montana, northern Idaho, and parts of eastern Washington,
eastern Oregon, and northwestern Wyoming. For this area, in the years 1906-
1911, about 15% (864 fires) of the fires on the National Forests were started by
lightning (Plummer, 1912). So, for those years, the number of person-caused
fires far exceeded lightning-caused fires.
By 1924, the percentage of lightning-caused fires had apparently increased
—"Lightning causes an average of 1/3 of the fires in this region ..." (Gisborne,
1924). He also noted that " . . . only one-fourth of the lightning storms start
fires ..." and that the storms occurred in seven "waves" (no duration timescale
given) during the months of July and August. From data taken by lookouts
(numbers are not given), the conclusion was drawn that " . . . one-fourth of the
lightning storms ordinarily start fires ..." and that storms could be categorized
on the basis of fire starts into "fire-starting" and "nondangerous" types. Gis-
borne suggested that fire-starting storms had a much higher CG/incloud light-
ning ratio than the nondangerous ones. No numbers or statistics were given to
support this viewpoint.
In a later report, Gisborne (1926) stated that for the years 1907 to 1925, the
average percentage of all fires started by lightning was 39% for the years 1907
to 1925, 51% in 1924, and 80% in 1925. The reason given for this increase was
an increase in the number of storms during 1924 and 1925 based on lookout
reports. He noted that the Weather Service storm-days numbers, determined by
thunder heard at Weather Service stations, were too small by a factor of three
or so, when compared to the data from 170 lookout stations. Since there were
four weather stations covering approximately the same area as the 170 look-
outs, this discrepancy is no surprise.
Testing the hypothesis that storms could be separated into two categories,
fire-starting and nondangerous, Gisborne (1926) found, based on 2186 reports
covering 3 years that stated definitely whether or not the storm was a fire starter.
382 Latham and Williams

"safe" storms (nondangerous) had about 30% CG flashes and the "dangerous"
(fire-starting) storms had about 40% CG flashes. He did state that 1046 addi-
tional reports gave CG percentages, but the storms were not classified, so those
data were not used (!). Besides rejecting data, another problem with the analy-
sis is that more than one lookout can count a given lightning flash.
Gisborne (1931) later found, based on 8128 observations made between
1924 and 1928, that safe storms had 24% CG lightning and dangerous ones had
44%; that is, storms having high percentages of in-cloud lightning caused more
fires than those having low percentages. No mention is made of how many ob-
servations were made but not included in the data used. He does note that
"before the degree of lightning-fire danger can be estimated satisfactorily, fire
weather forecasts must consider other factors—the characteristics of the storms
and probably the seasonal and current moisture content and inflammability of
the forest materials" and that " . . . thunderstorm frequency alone is not a de-
pendable criterion of the probability of lightning-caused fires." The hypothesis
that thunderstorm type alone is sufficient for fire prediction is thus not proved,
and lightning counts are not directly correlated to fire starts. Rainfall accom-
panying safe storms was observed to last 43% longer than that from dangerous
storms, according to Gisborne's data. So, are safe storms safe because ignited
fires are put out by rain or because there are fewer CG flashes in them? Gisborne
does not answer this question.
Morris (1934), although not proposing an ignition mechanism, came to sev-
eral conclusions based on an extensive study of lightning and fire reports by fire
lookouts in Oregon and Washington. Important for our exposition is this sub-
set of conclusions:

1. " . . . No more fires per acre occur in one altitude class than in another";
that is, lightning fires, at least in Oregon and Washington, do not have a
bias for high altitudes;
2. " . . . It is evident that danger zoning for lightning fires, if based solely
on previous [historical] distribution of the fires is likely to be very
inaccurate and unreliable";
3. "No definite lightning storm lanes or frequent 'breeding' spots were
found";
4. Thunderstorm days can be classified to indicate fire-starting potential
(the classes were similar to those of Gisborne (1931); and
5. Storms accompanied by high rainfall led to fewer fires than storms that
had less rain.

Results 1 and 3 are contrary to anecodotal evidence that lightning strikes pref-
erentially on ridges or peaks, rather than in valleys. Result 5 supports the "ob-
vious"; storms often put out, through rainfall, the fires they start. We will have
Chapter 11 Lightning and Forest Fires 383

more to say about Results 2 and 4 later. Morris did not present IC/CG ratios or
percent IC in a way usable for separating storms.
Barrows et a\. (1977) summarized lightning-caused fire data for the years
1931-1973. For the first time, lightning fire statistics are given in terms of a
stated density, rather than by arbitrary classification of land areas or by Forest
Service Forest. Over this period, the average number of lightning-caused fires in
the Region 1 National Forests (Montana, northern and central Idaho, and north-
ern Wyoming) was 0.85 fires per 100 km^/y. The standard deviation was 0.41,
indicating a high variability from year to year. The density also varies over the
landscape. This finding reinforces the earlier ones of this section that the light-
ning fire density is highly variable from year to year and place to place. No light-
ning statistics to compare to these numbers are available because the large look-
out network studies had been discontinued, and remote lightning locations
were yet to come. The percent of total fires caused by lightning is 77% averaged
over the years studied. This percentage has been nearly constant since 1931.
The increase in the percentage of lightning caused fires from 15 to 77% over
the years between 1912 and 1931 is curious. Plausible reasons for this increase
include less and more careful logging activity, a reduction in man-caused fires,
better reporting of fires in remote areas, fuel complexes becoming more fire-
prone to lightning ignition due to suppression, a significant reduction in fires
caused by railroad activity, and/or a general change in climate. Establishing the
relative contributions of these proposed explanations is a difficult task.
All the studies in this section share one major flaw. Except for a passing re-
mark in Gisborne's (1931) study, the tacit assumption is made of a correspon-
dence between number of CG flashes and/or the relative amount of CG and
in-cloud flashes and the number of fires that result. Rainfall is assumed to in-
fluence the number of fires through immediate wetting of the fuels either before
or after ignition. The fuel type is only broadly considered, that is by predomi-
nant tree species, and fuel state is not considered. (Fuel type and fuel state are
loosely defined but useful terms. Fuel type includes the biological and physical
descriptions appropriate to a location. Fuel state is the moisture content of the
various sizes of fuels.) Classification of thunderstorms either by IC/CG ratio or
by air mass or frontal, or other schemes, was perceived to be the most useful in-
formation for prediction of lightning fire incidence.

B. PROJECT SKYFIRE

Project Skyfire began with the speculation that cloud seeding might be used to
reduce lightning-caused fires—either by increasing rain or by influencing the
thunderstorm to alter lightning production (Schaefer, 1949). As the observa-
tions and studies of 1912-1934 showed (see Section II.A), thunderstorms with
384 Latham and Williams

low rainfall and low CG count were thought to be responsible for a majority of
the fires in the northern Rocky Mountains, the northwest forests of Oregon and
Washington, and the forests of northern California. After the cloud-seeding
demonstrations of Project Cirrus (Schaefer, 1953), the Skyfire project was born
in the Northern Rocky Forest and Range Experiment Station (Barrows, 1954).
The aim of the project was to reduce lightning by either overseeding mature
storms or by seeding early in the storm's development to initiate precipitation
in the cloud. Early precipitation would cause cessation of rapid growth, reduc-
ing the amount of lightning. Enhancement of rainfall was not the aim of Skyfire,
although earlier work (e.g., Gisborne, 1931) indicated that storms having high
rainfall amount were accompanied by fewer fires than storms with low rainfall.
The investigation of cloud types and locations led to preliminary cloud seed-
ing in 1956-1957 in Arizona and Montana (Barrows et al, 1958) to test the
effectiveness of ground-based and airborne silver iodide cloud-seeding genera-
tors. Fuquay and Baughman (1969) found, through research carried out in
Montana, that the number of CG lightning flashes might be suppressed by cloud
seeding but that a larger data sample and a properly randomized experiment
were necessary for a definitive conclusion.
By 1976, the final analysis of Skyfire data on randomized seeding from the
1960's was complete (Fuquay, 1974; Baughman et al, 1976). The conclusions
were that cloud seeding could reduce the frequency of CG flashes by more than
one-half and that the average continuing current duration might be reduced by
as much as one-fourth. The possibility that reduced rainfall accompanies re-
duced lightning is mentioned. For lightning reduction, seeding in the early stage
of cloud growth was found most effective.
A good summary of the cloud-seeding and lightning reduction experiments
as well as a critique can be found in Dennis (1980), who takes issue with the
analysis of Baughman et al. (1976). As with the majority of weather modifica-
tion results, there is no broad consensus on the efficacy of cloud seeding for
lightning suppression or rainfall enhancement. No conclusive proof was found
that cloud seeding reduces either the number or severity of wildfires. As hap-
pens often in scientific research, however, the by-products of Skyfire proved at
least as useful as the main product.
One of the important results was to further the conjecture (McEachron and
Hagenguth, 1942; Berger, 1947; Malan, 1963; Loeb, 1966) that the continuing
current might be the part of the lightning discharge responsible for ignitions in
forest fuels. Relating the current behavior of an individual lightning flash to the
incidence of fire at the same location is clearly a challenging task, and, for this
reason, few observations are available. Fuquay et al. (1967, 1972), in two in-
vestigations, reported on 11 CG fire-starting flashes containing continuing cur-
rents varying from 40 to 280 ms in duration. They used triggered photographs
of lightning as well as electric field recordings to determine the location of the
flashes as well as their composition. Spotters in light aircraft searched the light-
Chapter 11 Lightning and Forest Fires 385

ning terminus locations for fires. A ground investigation followed the aircraft
observation. In addition to the fire starters, five nons tar ting strikes, two hybrid
and three discrete, were studied. Every fire considered was started by a flash
with a continuing current longer than 40 ms. These observations led to the con-
clusion that earlier speculations were correct; the continuing current is the ma-
jor cause of lightning-caused fires in forests.
Skyfire results also included measurements of lightning IC/CG ratios (Z val-
ues, Boccippio et al, 2000) for seeded thunderstorms versus nonseeded storms.
The ratios were Z = 4.4 for seeded storms and Z = 3.0 for nonseeded storms.
Both of these values are higher than the Gisborne (1931) data of Z = 2.9 for
non-fire-starting storms and Z = 1.3 for fire-starting storms. The Skyfire data
were taken with electronic recording equipment and cameras rather than by
lookouts, which might explain the difference. Boccippio et al. (2000) found val-
ues Z = 3 to Z = 7 for the same geographic locations. Evidently there is a high
variability in the IC/CG ratio; separation of storm fire-starting effectiveness
based on this ratio needs further study.
Skyfire came to a close in the early 1970s, having established that the cause
of lightning fire ignitions was the continuing current in the lightning flash. Evi-
dence was gathered that cloud seeding, if done properly, could reduce the num-
ber of CG discharges and the incidence and duration of continuing current dis-
charges from a given storm. The research leading to the models discussed later
in this chapter had its origins in Skyfire.

C. RECENT STUDIES

Recent studies of lightning-caused fires benefit from two major data sources not
available to earlier researchers: organized fuel state descriptions and accurate
regional CG lightning location. Fuel type and fuel state, principally moisture
content in specified fuel arrays, have been combined into fire danger indices for
forest managers. Generally, fuels in a given fuel type are distributed in the for-
est by amount of dead biomass stratified by size. Each size adsorbs water from
rain and moist air. The amount of water in each size class is combined into an
index indicating roughly the severity of a fire should one start. The United
States uses the National Fire Danger Rating System (Deeming et ah, 1977), and
Canada uses the Canadian Fire Weather Index System (Van Wagner, 1987).
Although these systems are not identical, they are similar in that each has in-
dices that, for various fuel types, indicate fuel moisture. The indices are rou-
tinely calculated and archived, presenting a uniform database to use with light-
ning locations.
Lightning location is accomplished by detecting low-frequency electromag-
netic radiation emitted by lightning (Section II). Either triangulation or time-
of-arrival techniques can be applied. CG (positive and negative) and IC strokes
3oD Latham and Williams

are distinguished by differences in the radiation patterns. Systems have been in

use since about 1978, first in Alaska and subsequently in the United States and
Canada. Early systems claimed accuracy of location in the range 1-10 km. Re-
cent systems, using Global Positioning Systems (GPS) claim accuracy on the or-
der of 500 m (Cummins et al, 1998). Efficiency of 95% detected CG events or
better is claimed for modern systems. Position accuracy and efficiency depend
on the location of the CG with respect to the receiving stations in the detection
network. CG locations are archived by the company that now operates the net-
works: Global Atmospherics, Inc., of Tucson, AZ. Several recent studies use fuel
indices and lightning location together with fire reports and weather reports
and are summarized later. Flannigan and Wotton (1991) studied the relation-
ship between fires and lightning in northwestern Ontario, Canada. They found,
essentially, that negative lightning ignited more fires than positive lightning and
that the duff moisture code and the multiplicity of the strokes in the lightning
flash, for negative strokes, were important independent variables. Although not
specified in their work, it seems that stroke multiplicity has a connection be-
cause the probability of a continuing current in a flash is weakly connected to
the number of strokes (Uman, 1969).
McRae (1992) found no connection between elevation, slope, aspect, or topo-
graphic unit and lightning fires for the Australian Capital Territory, supporting
the much earlier findings of Morris (1934). But Van Wagtendonk (1991) did
find altitude dependence for lightning-ignited fires in Yosemite National Park
and justified this dependence on the grounds of vegetation type and preferred
lightning storm development. Renkin and Despain (1992) found no altitude de-
pendence for lightning fires in Yellowstone Park but did find a preponderance
of lightning fires in mature stands of Douglas fir, spruce fir, and decadent lodge-
pole, as opposed to successional lodgepole and multiaged lodgepole stands. In
addition, a threshold level of moisture in fine fuels, 13% of dry weight, was
found above which fires would not continue to burn after an ignition.
Meisner (1993) studied lightning-caused fires for the period 1985 to 1991 in
a small area in Southern Idaho. The lightning strikes were randomly distrib-
uted over the terrain, as were the fuel classes. The procedure used by Meisner
was to break down the area by species and then calculate the efficiency on a per-
species basis. Lightning-caused fire efficiencies were calculated for each 100 km^
pixel on the map and then averaged for the fuel classes in the sampled area. The
highest lightning efficiency was for logging slash, at 0.1 fires/CG flash, and the
lowest was 0.003 fires/CG flash for agricultural crops. Efficiencies for mixed firs
and Ponderosa pine were 0.02 fires/CG flash and 0.04 fires/CG flash, respec-
tively. One of the findings of the study was that " . . . on days when there were
100 or more detected lightning strikes, the correlations between fire danger in-
dices and number of fires were typically doubled." This compares nicely with
Gisborne's (1931) findings summarized in Section 111.A; there seem to be storm
Chapter 11 Lightning and Forest Fires j5o /

day types that correspond to high fire incidence. But according to Gisborne's
observations, high CG hghtning amounts resuU in fewer fires, the opposite
conclusion. Current hghtning detection systems do not presently report in-
cloud flashes, so no division of storms or storm days could be made on this ba-
sis to test Gisborne's (1931) statement.
Nash and Johnson (1996) thoroughly analyzed the coupling of synoptic
weather conditions with local scale weather and fuel conditions, the latter ex-
pressed in terms of the Canadian fire indices, and resulting fires in Canadian bo-
real and subalpine fuel types. They found, analyzing 2551 fires and 1,537,624
CG lightning flashes, that the best fuel state predictor for lightning fires was the
Fine Fuel Moisture Code index. Above the moisture index of 87 (equivalent
to a moisture content of 14% of dry weight; higher index value means lower
moisture), very close to the 13% value found by Renkin and Despain (1992) for
fires in Yellowstone National Park, the lightning-caused fire efficiency (number
of fires/number of CG's in the same area) increased rapidly. The highest effi-
ciency found was 0.06 fires/CG flash. No overall average was given, but typi-
cal values were between 0.01 and 0.04 fires/CG flash, in the same range as
the Meisner (1993) values. Nash and Johnson also found that lightning effi-
ciency of fires was higher under synoptic high pressure, when persistent rain-
fall was not expected. This description presents yet another "storm day type"
stratification.
Rorig and Ferguson (1999) generated a discrimination rule for selecting
"dry" vs. "wet" thunderstorm day occurrence in the Pacific Northwest. The
rule uses the dewpoint depression at 85 kPa and the temperature difference be-
tween 85 and 50 kPa. No connection was made to fires.
For this chapter, we gathered fire and lightning locations, as well as fuel
types as data for constructing map layers in a Geographic Information System
(GIS). The data cover western Montana, northwestern Wyoming, and northern
Idaho, as is apparent from the state outlines on Figures 1, 2, and 3 (see color
insert). Summaries of lightning flash position data were obtained from the Bu-
reau of Land Management (BLM) and Global Atmospherics, Inc., and fire data
from the USDA Forest Service and Department of Interior land management
agencies (primarily the BLM). Figures 1,2, and 3 show derived spatial layers for
fires, lightning density, lightning ignition efficiency, and fuel type for the years
1986-1992. In each of these figures, the fire locations are plotted as black dots.
All the pixels shown on these maps are 10 km on each side (Lambert projec-
tion); the data were aggregated upward from 1-km pixels.
From Figure 1, it is evident that lightning density and fires are only loosely
connected. That is, lightning is a necessary (by definition), but not a sufficient
condition for ignition. Other factors, such as fuel type and moisture, must be
considered. Notice that, in the very smaU areas of high strike density, no fires
occurred over the time period covered by the maps. By inspection, roughly half
388 Latham and Williams

0.06

A/"
"5^ r3>

.<S^
Fuel Type
FIGURE 4 Ignition efficiency as a function of fuel type.

the fires occurred in the minimum hghtning density category of 15 - 1 9 9 strikes/

100 km^. The results shown in Figure 4 indicate that the average ignition ef-
ficiency over the 7-yr period is about 0.026 fires/CG flash, combining positive
and negative flashes. Because of the importance of fuel type and state, lightning
density maps cannot provide a surrogate for fire density maps, either for past
fires or for future fires, unless carefully combined with other data. This finding
echoes Gisborne's results of 1931. Unfortunately, Gisborne did not calculate an
efficiency that can be compared with the recent studies. We have not tested the
separation of storms or storm-days into safe or dangerous categories. Fuel type
and fuel state are of more importance than the incidence of lightning in deter-
mining fires (Figure 3). World fire and lightning maps (Figures 18 and 19; see
color insert) support this view. The 7-yr data span covers a wide range of fuel
states and can be considered as integrating over them. Lightning efficiency val-
ues were extracted from the map data on a 1-km pixel size. Figure 4 presents
the lightning efficiency as a function of fuel type. The highest lightning ignition
efficiencies are in the Douglas fir and spruce fir/mixed conifer fuels. Meisner's
(1993) data did not have a Douglas fir fuel type, so no direct comparison for that
fuel type can be made. His value for Ponderosa pine was 0.042 fires/CG flash
as compared to 0.029 fires/CG flash for this analysis. This discrepancy is due
at least in part to a smaller area and smaller time period in the Meisner study;
the analysis done for this chapter spans a m u c h larger area and time period.
Chapter 11 Lightning and Forest Fires 3o9

Nash and Johnson (1996) found ignition efficiency values ranging from 0.01 to
0.04 fires/CG flash, in the range of our analysis.
The lightning-caused fire ignition efficiency for water (Figure 4) is 0.019
fires/CG flash. This is a good example of the misleading effect of small sample
size, as pointed out in Nash and Johnson (1996). The water area in our maps
is 26,247 km^ as compared to a wooded area of 690,596 km^ and grass, shrub,
and cropland area of 6,072,307 km^. The water area is almost all in two large
lakes with wooded surroundings. Reports of fire locations are often not accu-
rate, and a small position error can place a fire on the water instead of on the
shore. All these factors lead to an error that appears large because of its absur-
dity and comparative magnitude but that is actually very small.
More work is needed in the realm of lightning-caused fire efficiency. From
the studies summarized here, the efficiency apparently depends on a multitude
of factors. Among them, in order of spatial scale, are synoptic weather condi-
tions, fuel types, thunderstorm characteristics such as rainfall and lightning
rates, lightning characteristics such as positive, negative, and hybrid, and fuel
state, predominately moisture content of fine fuels. Broad agreement is evident
among the studies that the efficiency values for fuels and fuel conditions of in-
terest are roughly between 0.01 and 0.05 fires/CG lightning flash. The next sec-
tions will explore the details of the interaction of lightning with fuels and the
physics of the ignition process to explain the low efficiencies.

IV. INTERACTION BETWEEN

LIGHTNING AND FUELS
A. LIGHTNING PATHS ON TREES

The concept of "upward" and "downward" discharge directions (Plummer,

1912; Section III.A) was generated by observations of termination of the dis-
charge path at some specific point on the trunk of the tree. This is caused by
flashover, a phenomenon that occurs on towers, electric poles and other ele-
vated structures including trees (Darveniza, 1980). That is, at some position on
the tree or other conductor, the discharge "decides" to go to ground through the
air rather than continue along the conductor. The details of this phenomenon
are not well understood, especially with respect to trees. But the fact that it is
not universal, that sometimes the discharge does proceed all the way down the
tree (or conductor), even out the root structure, could have been the source of
Plummer's (1912) hypothesis of up-going and down-going strikes. The actual
physics of the lightning-tree-ground interaction during a lightning discharge
has not been thoroughly explained.
Plummer (1912) proposed that an increase of tree conductivity due to wet-
390 Latham and Williams

ting by rain had a part in ignition. The Ughtning strike path usually follows the
cambium layer of the tree because it has higher conductivity than either wet bark
or the woody interior of the tree (Defandorf, 1955). Cambium cells transport
the water necessary for the tree's operation and exhibit higher conductivity par-
allel to the direction of growth of the tree as opposed to perpendicular to that
direction (Du Moncel, 1877). Lightning scars on trees almost invariably feature
blown-off strips of bark, leaving a mark on the trunk showing that the path was
under the bark. Also, the path of the strike is very often spiral, following the
spiral long axis of the cambium cells. Sometimes, of course, the lightning path
is straight down the tree and/or on the bark surface (Taylor, 1973).

B. ECOLOGICAL DAMAGE N O T INVOLVING FIRE

Many lightning strikes to trees do not result in fire. In fact, a strike to a tree of-
ten causes only mechanical damage, up to and including complete destruction
of the tree (Plummer, 1912; Taylor, 1969). This damage could be caused by the
rapid gas expansion in a return stroke, by the sudden creation of steam by a re-
turn stroke in damp rotten heartwood from a continuing current, or by both.
Only the aftermath is available to show the effect, and a mechanism has not been
truly identified. No statistics are available for the proportion of strikes to trees
that cause damage or kill the tree without starting a fire. Because it is weakened
by the strike, a lightning-struck tree can also act as a vector for insect and dis-
ease infestation, leading to insect destruction of the struck tree and infestation
of trees surrounding it (McMullen and Atkins, 1962; Schmitz and Taylor, 1969).
Also, there are instances, especially in the southern United States, of group-
kills of trees caused by root destruction from the ground path of a discharge to
one tree. Taylor (1969) provides an excellent summary of lightning effects on
the forest that do not involve fire.

C. FUELS ON THE FOREST FLOOR

Anecdotal evidence implies that lightning ignition of forest fuels takes place al-
most exclusively in the fine fuels on the forest floor. Typically, in the forests of
the northwestern United States and other forested areas of the world, lightning
strikes a tree and, although the connection process for lightning to ground is
complex, the result is the establishment of a path down the trunk toward the
ground (Figure 5). Sometimes, the strike will flash over to ground, at a height
of a meter or two. The reason for this behavior is not known. The breakdown
path to mineral soil, which can be considered "ground," passes, in either case,
Chapter 11 Lightning and Forest Fires 391

FIGURE 5 Lightning path on a tree (Plummer, 1912).

through the fine fuels on the forest floor. Ignitions occur along this path in the
htter and duff layers on the forest floor (Fuquay et al, 1967,1972; Taylor, 1969).
The composition of litter and duff layers varies considerably from place to place
in forested and grassland areas, depending on the local ecosystem. For example,
Ponderosa pine has, as do most pines, a needle very long in comparison to its
diameter. Because of this, the litter layer in ecosystems in which this species is
dominant is relatively thick compared to the duff layer. Firs, on the other hand,
have needles that are much shorter, making for a compact duff layer and a very
thin, almost nonexistent, litter layer.
Occasionally, the stroke will pass through the moist heartwood of a dead tree
(snag) or a live tree. W h e n that happens, the tree may be blown apart (Figure 6),
with or without accompanying fire. Fires often b u r n from the inside of snags
that are not blown apart to the forest floor fuels, causing holdover fires (fires
that smolder for long times before they are discovered; Frandsen, 1987).

V. HOW IGNITION OCCURS

A. FORMULATION OF THE PROBLEM

One of the most important of the Skyfire findings was identification of the
continuing current as the primary cause of lightning ignitions of forest fuels
392 Latham and Williams

^S^§^mhh ML

FIGURE 6 A tree demolished by lightning (Plummer, 1912).

(Section III.B). The field studies of Taylor (1969) verified that burning debris
blown off trees from discrete strokes onto litter or duff fuels does not cause
ignition, and Fuquay (1974) formed the conjecture that the duration of the
continuing current might be important in the ignition efficiency. This informa-
tion formed the basis for a model to predict fire density (fires/100 km^) given
the presence of lightning and the type and state of the fuel. That is, real-time
and predicted data from reliable, convenient, and inexpensive sources, such as
weather nowcasts and forecasts, and lightning location networks, would be
used with algorithms to predict the incidence of fires on the landscape.
The task, then, was to form a construct or set of questions for the model. The
questions are:

• What is the thermal and radiating structure of the continuing current?

• What is the energy required to ignite fine fuels?
• Is radiation or convection the predominant mode of energy transmission
from the continuing current channel to the fuel?
• What experiments are necessary to establish usable parameters?
• If a predictive model is found, what is the set of data needed to imple-
ment the model for practical use?
Chapter 11 Lightning and Forest Fires 393
How should predictions be integrated with existing fire systems such as
the National Fire Danger Rating System?

B. ENERGY TRANSFER FROM

THE LIGHTNING CHANNEL

The basic framework for ignition modeling is shown in Figure 7. A cylindrical

conducting channel of hot gas, whose characteristics such as diameter and tem-
perature are determined by a suitable physical model (Section V.C), passes into
a layer of uniform fuel overlaying an electrically conductive soil layer that is
the electrical ground terminal for the current in the channel. General assump-
tions for this framework are (1) temperature profile consistent with a hot gas arc
channel model, (2) vertical penetration of the fuel, (3) infinite channel height
(or at least very much longer than the depth of the fuel), and (4) no effects from
horizontal current flow in the soil. Once the channel structure, particularly tem-
perature, is established, energy transfer to the fuel depending on the mode of
transfer will be calculated, with surface element A the location of maximum
radiation transfer from the arc column and surface element B the location of
maximum convective transfer (Figure 7). Conductive heat transfer, by diffu-
sion within a medium, does not apply to heat transfer from the hot gas channel
to the fuel (Incropera and DeWitt, 1996). Conduction does play a role in the
structure of the arc channel.

Arc column

Radiating
Element

Surface
element Surface
element A

Mineral soil

FIGURE 7 Schematic of a continuing current path through forest floor fuel.

394 Latham and Williams

C. STRUCTURE OF THE CONTINUING

CURRENT CHANNEL

Details of the thermal and radiation structure of the continuing current chan-
nel were calculated using hot gas models of an electric arc channel. That is, en-
ergy balance and conservation of mass equations, together with Ohm's Law, are
solved with appropriate boundary conditions. These equations were solved nu-
merically by Uman and Voshall (1968) for a channel with no current flow such
as a return stroke decay channel and by Latham (1986) with electric current
flow. The latter model included radiation cooling and full physical hot gas char-
acteristics under cylindrical symmetry. Few measurements are available for long
arcs. Electrode effects dominated most measured arc characteristics (Strom,
1946; King, 1961). As a consequence, channel characteristics are poorly defined
for validating models of arcs in regions away from electrodes, the relevant situa-
tion with lightning continuing currents.
Latham's (1986) model will be used to give the details of the modeled con-
tinuing current channel. The assumptions are relatively simple and have been
justified (Uman and Voshall, 1968; Latham, 1986):

• Radial symmetry, giving a cylindrical arc;

• Unit length of channel, hence no axial dependence; and
• Optically thin gas, that is all radiation from the hot gas escapes the
channel and is not captured and electron and ion temperatures are
the same.

Values for hot gas physical constants as a function of temperature were avail-
able from numerous research studies (see Latham, 1986, for the extensive list).
Using the assumptions given here, the energy conservation equation for an ele-
ment of the gas in cylindrical coordinates is

pCp{dT/dt) + pCpU{dT/dr) = l/r{[d/dr][rk{dT/dr)]} -\- aE^ - S (1)

The conservation of mass requires

(dpldt) + {l/r)[{d/dr)rpu] = 0 (2)

Ohm's Law stated for the gas column is

j = (TE (3)
The equation of state for the hot gas is

p = pRTZ (4)
Chapter 11 Lightning and Forest Fires 395

0
Q.
E
,05

0 1 2 3
Radius (cm)
FIGURE 8 Temperature profiles of the theoretical channel as time progresses.

Boundary conditions are stated at r = 0 and at r = CXD as

r = 0, dT/dr = 0; oo, T= T., (5)

where p is gas density, u is radial velocity, r is channel radius, T is gas temper-

ature, k is thermal conductivity, Cp is specific heat at constant pressure, Z is
compressibility, p is pressure, R is gas constant, a is electrical conductivity, E
is electric field, j is current density, S is radiation density, and t is time.
The initial conditions are given as a radially small (0.5 cm) high-temperature
(12000 K) pulse at t = 0 (Figure 8), simulating the return stroke channel.
Combining Eqs. ( l ) - ( 4 ) describing the channel behavior results in a highly
nonlinear system. Various mathematical attacks to solve the equations in closed
form were not successful. As a result, finite difference computer computation
was applied. As usual, when a finite difference approximation is used, some con-
ditions on the problem cannot be actualized. In the problem at hand, we can-
not integrate the equations to infinite radius. Initially, the radial boundary was
placed at 100 cm for a one cm channel, leading to extensive calculation times.
A boundary condition, derived from the equations, circumvented this problem
(Latham, 1986), allowing more rapid calculation. With the boundary condi-
tions properly handled, temperature profiles at successive times from 0 to 50 ms
for an unbounded channel result (Figures 8 and 9). The results can be sum-
marized qualitatively as follows: energy balance in the interior of the channel
is dominated by the thermal conductivity of the hot air in the channel. On the
396 Latham and Williams

5 12 Radius

Central temperature

20 30 40 50
Time (ms)
FIGURE 9 Channel central temperature as a function of time.

Other hand, the radial growth rate of the channel is determined by the conduc-
tivity contrast between the ambient air and the hot channel because the edge
of the channel moves outward by conductive heating at the channel edge. Ba-
sically, the radial structure at the channel edge is independent of the interior
temperature (Figure 8). Also, if the current ceases, the channel temperature
will decay, but the radius will continue to expand, as in the model of Uman and
Voshall (1968).
Grieg et al. (1985), through observations coupled with calculations, main-
tained that turbulent convective mixing causes much faster (a factor of 2000)
channel collapse than conduction. The expansion and collapse of that arc may
resemble the collapse of the much higher energy return stroke channel. High-
speed images of the arcs used in the experiments referred to in Section VI show
no turbulent breakdown, but the energy densities are lower. Evidently, more
exploration is necessary concerning the transition from stroke channel to con-
tinuing current channel.
Measurements have been made in other studies that support the model, at
least in the prediction of the temperature of the arc. Latham (1984) presented
an analysis showing that the ratio of the Nj ^ first negative radiation, an excited
state of the nitrogen molecule, and the cyanogen (CN) radiation over the range
6000-9000 K can be used to estimate the channel temperature (Figures 10
and 11). Salanave (1965) and Connor (1967) reported strong radiation at these
wavelengths, although the two wavelengths were not differentiated. An example
Chapter 11 Lightning and Forest Fires 397

0.01
9 10
Tx10'(°K)

FIGURE 10 N2^(1N) and CN violet radiation as a function of temperature.

of the ratio measurement is given in Latham (1984); the ratio gives a tempera-
ture of about 6500 K that corresponds to the model (Figure 12). The author has
made no further measurements, but vahdation of the model should be con-
ducted with further spectral measurements during the continuing current phase
of CG flashes.

50-, 25-|
140-^
P40- •B20-
CO
Total radiation
|l2oJ
O o
§30- •>15- c _
o
o z
(D o =5 lOoJ
2 Radiation efficiency
•1 2 0 - |io- "co I
CO
•u
03 •^80-^
DC
10- 5- - N/(1N)/CN violet)
60-^
oJ oJ
0 10 20 30 40 50
Time (ms)
FIGURE 11 Radiation and radiation efficiency as functions of time for the calculated channel.
398 Latham and Williams

-6500 K Calculated Temperature

*«AjvVHy*~«*/v«v'

I 100 ms

FIGURE 12 A measured N2^(1N)/CN intensity ratio for a lightning stroke with continuing
current.

D. REQUIREMENTS FOR FUEL

IGNITION BY LIGHTNING

As stated earlier, forest floor fuels that are most susceptible to lightning igni-
tion fall into three rough categories: loose, unconsolidated fuels with low bulk
densities (for example Ponderosa pine or lodgepole pine litter); tightly packed
small fuel particles (such as Douglas fir); or consolidated rotted fuels (such as
peat). That peat can be ignited by lightning at all might be surprising at first,
considering its water content and density. Not only do these ignitions occur, for
example in the Florida Everglades, but the resulting fires can also smolder for
long periods of time. This behavior should be explained in our model structure.
Burning (combustion) actually takes place as diffusion flames in the complex
hydrocarbon gases that are "cooked" out of the fuel by heating, a process called
pyrolysis. Pyrolysis is generally dependent on the temperature of the fuels (e.g.,
Susott, 1984). Energy transferred to the fuel by heat processes must raise the
temperature to a level sufficient to initiate pyrolysis. Calculation of this tem-
perature requires the construction of an appropriate heat (conservation of en-
ergy) equation for the heated object. These equations are usually very complex,
and almost always nonlinear, because of the geometry involved and because ra-
diation effects are proportional to T^ (Cox, 1995; Incropera and DeWitt, 1996).
Several researchers have proposed ignition criteria for forest fuels (Martin
and Broido, 1963; Simms and Law, 1967; Susott, 1984; Jones et al, 1990). In
general, the criteria are given in two ways: in terms of the temperature of par-
ticles at the time flaming begins and in terms of the energy flux and duration
necessary to generate flaming in the material. Two ignition conditions are pos-
sible—spontaneous and piloted. In spontaneous ignition, a heat pulse is ap-
Chapter 11 Lightning and Forest Fires 399
plied to the material and either ignition occurs or not. In piloted ignition, the
material is subjected to either a steadily increasing or constant heat source for
durations longer than a few seconds, preheating the material, and then sub-
jected to a pulse or increase in heat. Again, ignition either occurs or not. Igni-
tion by lightning is an example of spontaneous ignition, whereas a propagating
fire is an example of piloted ignition (Cox, 1995).
In our analysis, we will rely on measured ignition criteria for short radiation
pulses as a minimum of about 1.67 X 10^ W m"^ for a 500 ms pulse (Martin and
Broido, 1963), supported by measurements made by Simms and Law (1967).
These ignition criteria ignore problems having to do with moisture content,
physical arrangement, and the like (which act to increase the needed ignition
flux) and allow us to explore the possibilities for various modes of heat trans-
fer. The cylindrical model of Section V.C can be used to calculate theoretical
heat fluxes associated with radiation and convection.

E. THEORETICAL HEAT FLUX CALCULATIONS

Referring to Figures 7 and 13, the radiation flux at patch A can be calculated.
Establish a cylindrical coordinate system with its origin in the plane contain-
ing A, with the z-axis vertical and coinciding with the axis of a cylinder of hot
gas with temperature T (K) and radius a (cm). Assuming a "top hat" profile for

Edge of Channel
FIGURE 13 Coordinate layout for radiation and convection calculations.
400 Latham and Williams

the channel, if 5 (W m~^) is the total emission per unit volume from the hot gas
at constant temperature and is zero outside the cylinder, the flux at A is

oo
, , r, Sr
„. COS
_„„ ^il/ ( drd^dz
J 477]
0 0 0

the variables in the problem being shown in Figure 13. With that geometry, the
equation can be restated as

a ITTOO

S [ ( ( rz drdd^dz
^^ ^^
477 JJ JJ JJ(r^
(,2++ a^a^+^^^ -^- 2ar
- ^-cos --
# ) ^^
^^^^^ ^^^
0 0 0

Integrating over z and invoking symmetry on # leaves

•J J (r^
^ ~ 277J J (r^ + a^ - 2 a r c o s # ) ^ ^ '
0 0

This integral has a singularity at r = a. Moving A just a little way radially from
r = a to avoid the singularity gives a value for the integral of F^ = 03aS W m"^.
Proceeding in the same way for a patch at B placed at ^ = 0 gives a flux of

a ITTOO

S ( [ [ r{a - cos ^)drdMz

^^ " 4 ^ J J J {r^ + a^ + z^-larcos-dy^ ^^^
0 0 0

which reduces to Fg = 0.5aS W m~^.

For the present purpose, then, we take the flux from the column to be about
Sal2 W m~^ and assume some values for a growing channel. At a generously
high gas temperature of 8000 K (Figure 9), the radiation heating at a radius
of 1 cm is about 1 X 10^ W m ~ ^ or 5 X 10^ J m"^ for a 500-ms pulse. This
value is less than the very generous required minimum energy for ignition of
8 X 10^ J m~^ for a 500-ms pulse as established in Section V.D. Thus, although
radiation heating is definitely present, it is unlikely to be a dominant part of the
ignition of forest floor fuels by lightning. The relative weakness of radiation as
an ignition source was also seen in the scenario developed for ignition of forest
fuels by meteor fragments in Melosh et a\. (1990).
Convective heating may be estimated for small-diameter fuels, such as pine
needles, that are bathed in the gas flow of the arc. Suppose the gas tempera-
ture is 8000 K as in the previous example (Latham, 1986). The gas density is
0.0235 = k g m " ^ , the viscosity about 1.5 X 10"^ poise (Adams, 1966). The
Chapter 11 Lightning and Forest Fires 401

kinematic viscosity is thus 6.4 X lO""*^ m^ s~\ The diameter, d, of a pine needle
from a long-needled species, such as Ponderosa pine, is about 1 mm. The ve-
locity of the gases in an arc discharge such as a continuing current is about
2000 m s~^ perpendicular to the cross section of the arc (Strachan and Barrault,
1975). The Reynolds number of the gas flow is thus about 30, and the Nusselt
number, Nu, is about 3 (Incropera and DeWitt, 1996). The thermal conductiv-
ity, fe, of hot air at 8000 K is 2.44 WK"^ m~^ (Adams, 1966), and the heat
transferred to the fuel element is about (T^ - T^)*Nu*fe/(i or 5.5 X lO'^ W m~^,
where T^ is the gas temperature and T^ is the ambient temperature. The resul-
tant energy deposited in 500 ms would be 2.5 J m~^. The convective heat trans-
fer is roughly three orders of magnitude larger than the radiation heating cal-
culated earlier.
If a more solid fuel is involved, such as a duff layer, consider the wall of the
fuel to be the wall of a pipe with a hot gas in turbulent flow in the pipe. Calcu-
lation for heat transfer to the wall of a pipe using the same values as for small
fuel heating gives a flux at the fuel surface of about 1 X 10^ J m~^, again larger
than the heat flux from radiation as calculated above. Under the reasonable as-
sumptions made, the importance of the duration of the current rather than its
amplitude is clear.
The scenario for ignition of forest floor fuels can, then, be summarized as
follows. A return stroke, or perhaps a junction process arc establishes a path
through the fuel. The high-current (tens of kiloamperes) initiating arc is fol-
lowed in some cases by a smaller continuing current on the order of 100 A, last-
ing tens to hundreds of milliseconds. The continuing current arc, which can-
not shrink, expands continuously into the fuel. The center portion of the arc,
under any scenario, is hot enough to ablate completely the material in its in-
terior, and only gas is left behind. As the continuing current stops, the cen-
tral temperature in the channel decays, and the channel continues to expand.
The ring of unablated material at the channel edge either has received enough
energy through combined radiation and convection to maintain the ignition
or not.
These approximate calculations serve to show the complexity of heat trans-
fer from lightning continuing currents to forest floor fuels. They also lead to the
conclusion that theoretical solutions of the ignition problem cannot give re-
sults sufficiently accurate for routine operations. Darveniza and Zhou (1994),
having done some impulse and arc heating studies using high-voltage arcs, sug-
gested, "It was found that the power and energy transferred to the fuels from
per centimeter of the arc were much greater than those of the arc itself [sic] and
were current and fuel properties dependent." This statement appears to support
our view that convection transfer is more important than radiation transfer. We
had already turned to experiment in 1988 to lay the foundation for a useful
product, as is described in the next section.
402 Latham and Williams

VI. IGNITION EXPERIMENTS

WITH REAL FOREST FUELS

A. APPARATUS AND EXPERIMENTAL DESIGN

As we have shown, the interaction between the hot gas channel of the hghtning
continuing current and the fuels through which it passes is extremely complex.
Consequently, a series of experiments was carried out to form a model that
would allow calculation of ignition probabilities from easily obtained data.
An arc generator was constructed to simulate, as closely as possible, the dis-
charge of a lightning continuing current. The arc generator used 42 12-V truck
batteries connected in series, controlled by a silicon-controlled rectifier switch
and initiated by either an exploding tungsten wire or by drawing the arc with a
rapidly moving carbon electrode (Latham, 1987, 1989).
In all, eight types of fuel samples (Table 1), taken from the wild and repre-
sentative of most forest floor types, were subjected to a range of arcs. A series
resistance in the arc circuit controlled the current, and a computer-controlled
silicon-controlled rectifier switch controlled arc duration. The moisture con-
tent of the fuels was varied between 5 and 40% of dry weight. In all, 1600 trials
were conducted with a current range of 20-500 A and durations from 10 to

TABLE 1 Coefficients of the Probability Equation for Fuel Types

Used in the Ignition Experiments

Fuel type A B (1/ms)

Ponderosa pine litter 0.97 - 0.19*M/ 0.012

Punky wood -0.59 - 0.15*M/ 0.005
(rotten, chunky)
Punky wood powder 1.2 - 0.12*M/ 0.002
(4.8 cm deep)
Punky wood powder 0.13 - 0.05*M/ 0.005
(2.4 cm deep)
Lodgepole pine duff - 5 . 6 + 0.68*d 0.007
Douglas fir duff - 7 . 1 * 1.4*d 0.006
Englemann spruce mixed 0.79 - 0.081*M/- 8.5*pb 0.011
(high-altitude mixed)
Peat moss 0.42 - 0.12*M/ 0.005
(commercial)

M/is fuel moisture, % of dry weight; d is depth, cm; and pb is bulk density
(g/cm^). The coefficient A is dimensionless, and B (ms~^) multiplies the
continuing current duration in milliseconds.
Chapter 11 Lightning and Forest Fires 403

500 ms; the arc could not be maintained below 20 A, and 10 ms was a lower
limit on the switch and arc initiation apparatus. A maximum current of 500 A
and duration of 500 ms were chosen as high enough that extremely few con-
tinuing currents would exceed them.
Defining ignition was difficult. Pyrolysis of biomass materials unquestion-
ably occurs at the edges of the lightning continuing current path, since pyroly-
sis results from temperatures near and above about 600 K (Susott, 1984), and
the 7 0 0 0 - 8 0 0 0 K temperatures in the body of the arc (Figure 8) ablate the cen-
ter material. Pyrolysis products from the material at the edge of the arc b u r n as
linear diffusion flames, in which combustion depends on local fuel-air mixing
rather than premixing as in a Bunsen burner flame (Cox, 1995).
Operationally, the difference between an ignition and a fire simply depends
on whether or not the fire is seen and/or reported. Practically speaking, an
ignition has taken place if the ignition source is sufficient to produce a self-
sustaining combustion process once the source is gone. For the experimental
results presented here, if the sample totally burned, or if smoldering continued
after 5 min, an ignition was recorded.

B. EXPERIMENTAL RESULTS

The experiments were designed to develop relationships between the indepen-

dent variables and the probability of ignition. Independent variables for each
fuel type for the ignition tests were arc current, arc duration, fuel moisture con-
tent, fuel bulk density, and fuel depth. For each set of conditions for a given fuel
type, the result was either an ignition or not, a binary value. The results were
tested using logistic regression, a systematic way of turning a binary result,
such as the presence or absence of an ignition, into a probability of occurrence
of the phenomenon (Loftsgaarden and Andrews, 1992). Typical results are
plotted in Figures 14 and 15. The former is from one series of tests on Pon-
derosa pine litter; the latter is from Douglas fir duff. The regression process
eliminated unimportant independent variables for each fuel type. One variable,
arc current, was found unimportant, and arc duration was found to be most
important, as conjectured by Fuquay et al. (1972). The unimportance of the
arc current arises from the temperature structure of the arc channel, as devel-
oped in Section V.E. The most significant independent variable, after duration,
is plotted on the ordinate for each of the fuel types. Lines of equal ignition
probability as determined from the regression are also shown.
Figure 14 presents the results of tests on Ponderosa pine. The second most
significant variable for this species was moisture content. Generally, as the
moisture content increased, more energy, that is a greater current duration, was
required to obtain ignition, as the regression probability lines indicate. No data
404 Latham and Williams

50 _, p=0.10

40

30

20

10

o-V-
0 110 220 330 440 550
Duration (ms)
FIGURE 14 Ignition experiment data for Ponderosa pine needle fuel.

could be taken for moistures in excess of roughly 30% of dry weight because
the woody material could not hold more moisture due to fiber saturation. Near
fiber saturation, currents in excess of 400 ms invariably ignited the sample, and
at the lower limit of the experiment, 8% moisture, currents of 50-ms duration
ignited these fuels about 50% of the time.
Figure 15 shows the results of ignition tests on Douglas fir duff. For this spe-
cies, the predominant secondary variable was, surprisingly, the depth of the
fuel. Duff shallower than 2.8 cm did not ignite, and deeper samples ignited at
small current duration, 50 ms.

Burn
No burn

6H
Q.
0)
Q 4

2-\

0-V
110 220 330 440 550
Duration (ms)
FIGURE 15 Ignition experiment data for Douglas fir duff beds.
Chapter 11 Lightning and Forest Fires 405

An examination of samples that did not ignite demonstrated that the mate-
rial in the high-temperature portion of the arc was ablated, as predicted from
the temperatures calculated in Section V.E (Figure 8). Almost all samples with
some soil showed fulgurites; glass was created from the melting of sand grains
(2000 K) in the soil and dispersed through the fuels. The edges of the arc path
through the fuels were charred. Clearly, the important mode of energy transfer
was, as indicated previously by magnitude comparison, convective transport to
fuel particles at the edge of the arc, as surmised in Section V.E. This means that
the energy transferred to the material depended on the temperature structure
of the edge of the arc, roughly the same for all central core temperatures (Fig-
ure 8) and durations.
The difference in the results implies a difference in heating mechanism be-
tween the more compact fuels, such as short-needled species duff (Figure 15),
and the less compact litter of the long-needled species (Figure 14). In fact, the
ignition probability for the former fuels did not depend on the fuel moisture or
bulk density, and the latter depended on the moisture but not on bulk density
or on depth.
Long-needled species litter is loosely arranged with large air spaces com-
pared to the diameter of the needles. Short-needled species such as Douglas fir,
on the other hand, have almost no litter layer, and the air spaces in the duff layer
are small compared to the needle diameter. For this reason, the convection
transfer to long-needled litter takes place to individual particles, and the trans-
fer to short-needled litter takes place more in the form of the less-efficient wall
transfer (Section V.E). Figure 15 shows that for Douglas fir duff less than about
2.8 cm deep, no ignitions occurred, even at fuel moisture content as low as 8%
(a "very dry" duff moisture). So there may be an effect, in this dense fuel, of
needing a "tube" to somehow constrain the arc in order to form a kind of muffle
furnace effect (see also Chapter 13).
Ignition of spruce-fir samples having a mixed structure was dependent on
both moisture and bulk density. No ignitions were observed for current dura-
tions on the order of 10 ms, and only one, in Ponderosa pine litter, at that value
for fuel moisture of about 7%. Equipment limitations prevented current dura-
tions less than 10 ms. It is possible that, at least in some fuel types, currents
shorter in duration than 10 ms can cause ignitions. Certainly many continuing
currents are in the neighborhood of this duration (Thomson, 1980). Trends in
the data for all species do show that as the current duration decreases, ignition
probability decreases, and zero current should give no ignitions.
Frandsen (1987) showed that commercial peat moss, a reasonable surrogate
for duff, could sustain smoldering combustion at moistures of 93-103% of dry
weight. So the moisture content of the short-needled duff seems to be relatively
unimportant in sustained combustion as well as in ignition. Meisner (1993) de-
termined that the ignition efficiency in mixed fir fuels was 0.024 fires/CG flash.
406 Latham and Williams

and the efficiency for Ponderosa pine fuel type was 0.042 fires/CG flash, about
twice as large. This may be explained by the relative efficiency of heat transfer
in the two fuel types. Further explanation would require further experiments.
Application of logistic regression to the existing experimental results does, how-
ever, produce a useful application.
The form of the logistic regression equations obtained from the data is

pci{A, B, 0 = [1 + exp(-A - Bt)]'^ (10)

where pci{A, B, 0 is the probability of ignition. A and B are coefficients depend-

ing on the experimental material, where A is dimensionless and B has dimen-
sion (ms~^), and t is the continuing current duration in milliseconds (Table 1).
Unfortunately, logistic regression implies a finite probability of ignition with
zero continuing current duration, inconsistent with the lightning observations
discussed in Section V.A. There may be a technique or formulation that will al-
low rectification of this shortcoming.
The consequences of the regression artifact are slight. The number of igni-
tions compared to nonignitions in the experiments for current duration near
10 ms is very small (see Figures 14 and 15). Also, the observations in Fuquay
et at. (1972) show no fires caused by continuing currents with durations less
than 40 ms and an average duration of 165 ms for those continuing currents
starting fires. After an ignition event, the probability of fire growth can be ob-
tained from fire data (Latham, 1979).

VII. GENERATING MODELS

FOR OPERATIONAL USE
Using experimental results for operations requires knowledge of not only the
fuel parameters for the locus of a strike but also the presence and duration of a
continuing current in the strike. The fuel parameters are accessible in existing
operational databases, and the location of the flash is known from lightning
network data (to a claimed accuracy of 500 m), at least in the United States and
parts of Canada.

A. CRITERIA FOR UTILITY

Although ignition models can provide ignition probabilities for most of the fu-
els found in forested and grassy landscapes, the models still have to be accom-
modated to reality. According to the models as presented, we can obtain an ig-
nition probability if we know the kind of fuel present, its state (e.g., depth.
Chapter 11 Lightning and Forest Fires 407

moisture content, and bulk density), and the presence and duration of a con-
tinuing current.
Fortunately, lightning location systems now cover the United States and
parts of Canada. Unfortunately, the location data cannot give either the dura-
tion of a continuing current or even the presence of one. ELF methods are well
suited to identifying and quantifying the continuing current (Burke and Jones,
1996; Huang et al, 1999), but these methods are presently not applicable to
operations. There are, however, statistics for the probability distribution of the
duration of continuing currents in both positive and negative discharges, as well
as the probability that a flash has a stroke with a continuing current (Latham
and Schlieter, 1989). The statistics we use are valid for thunderstorms in the
northern Rocky Mountains.

B. PROBABILITY DISTRIBUTIONS
FOR CONTINUING CURRENTS

Application of the ignition equations requires knowledge of continuing cur-

rent durations on a flash-by-flash basis. The statistics of continuing current du-
rations have been generally thought to be lognormal (e.g., Cianos and Pierce,
1972), although Thomson (1980) found that a lognormal distribution assump-
tion for continuing current durations failed a x^ test. Other distributions might
work as well and are more tractable for mathematical manipulation. The Wei-
bull is a two-parameter distribution that is the statistic of choice for failure of
a whole unit contingent on the failure of a part or parts of the whole (Hahn and
Shapiro, 1967). This viewpoint is attractive in that it might correspond to a
continuing current channel fed by subchannels in the cloud. Weibull and log-
normal distributions were each fit to continuing current duration data from the
Skyfire experiments (Fuquay and Baughman, 1969), covering 141 negative hy-
brid flashes (duration between 20 and 520 ms) and 54 positive flashes. The two
resulting distributions were tested according to a method developed by Kap-
penman (1988). The Weibull distribution was found to better describe the mea-
surements. According to the Weibull fit, the distribution
n \ / t ^""1
p(t, n,s) = [-][-] exp (11)

generates probability distribution functions, p, for continuing current duration,

t in milliseconds, with the parameters n = 1.6, s = 207.8 for negative currents,
and n = 2.3, s = 69.3 for positive currents. These coefficients have no physical
interpretation. The distributions are shown in Figure 16 and are valid for thun-
derstorms in the northern Rocky Mountains. Thomson's (1980) results for Port
Moresby show that there are certainly continuing current durations less than
408 Latham and Williams

0.020 - I

p 0.015 J

^o 0.010.

CO
0.005 J
o

0.000 -
1 \ \ \
100 200 300 400
Continuing current duration (ms)
FIGURE 16 Probability distributions for negative (dashed) and positive (solid) continuing
currents.

20 ms but, as the experiments summarized in Section VI show, extremely few

ignitions are observed for durations less than 20 ms and none below 10 ms for
laboratory experiments.

C. CONDITIONAL PROBABILITY OF IGNITION

Because the probability distribution function for continuing currents is known,

ignorance of the characteristics of any given continuing current can be dealt
with using conditional probability. The probability of occurrence of an ignition
is dependent on several parameters. If a statistical description of one of the pa-
rameters, such as current duration, is known, the parameter can be effectively
removed from the parameter list by integrating the probability of occurrence
over the probability distribution of the parameter. In the present case, integra-
tion allows calculation of the probability of ignition per continuing current for
known fuel parameters. The probabilities of ignition were given in Table 1, and
the probability distributions for continuing currents in Section VII.B.
The conditional probability for lightning ignition is calculated using the
integral

Pi{n, s, A,B) = p(t, n, s) pci{A, B, t) dt (12)

where pi{n, 5, A, B) is the probability density for ignition per continuing cur-
rent event (between 0 and 1), A and B are coefficients from Table 1, and t is du-
Chapter 11 Lightning and Forest Fires 409

TABLE 2 Conditional Probabilities for Ignition of Fuel Types

Used in the Ignition Experiments

Fuel Pi neg Pi pos

Ponderosa pine Utter 1.04 exp(-0.054M/) 0.92*exp(-0.087*M/)

Punky wood (rotten, chunky) 0.59*exp(-0.094*M/) 0.44*exp(-0.11*M/)
Punky wood powder 0.9*exp(-0.056*M/) 0.86*exp(-0.06*M/)
(4.8 cm deep)
Punky wood powder 0.73 - 0.011*M/ 0.6 - 0.11*M/
(2.4 cm deep)
Lodgepole pine 1.0/[1 +exp(3.84 - 0.6*d)] 1.0/[1 + exp(5.13 - 0.68*d)]
Douglas fir 1.0/[1 + exp(5.48 - 1.28*d)] 1.0/[1 + exp(6.69 - 1.39*d)]
Englemann spruce 0.8 - 0.014*M/ 0.62*exp(-0.05*M/)
(subalpine)
Peat moss (commercial) 0.84*exp(-0.058*M/) 0.71*exp(-0.72*M/)

Mf is fuel moisture, % of dry weight, and d is depth, cm.

ration in milliseconds. The final form of this ignition probability, in approxi-

mation, is given in Table 2.
Based on similarity of forest floor and tree characteristics, the results of the
conditional probability estimates can be applied to other species (Latham and
Schlieter, 1989).

D. IMPLEMENTATION OF IGNITION PROBABILITY

Figure 17 shows one way to implement the ignition probabilities into an oper-
ational system. In this scheme, a Geographic Information System map layer for
ignition probability per positive and per negative lightning flash is generated
using the algorithms developed in the last section. As the lightning location data
come in, the location is binned into pixels that correspond to the ignition prob-
ability layer. The combination of the two becomes a projected fires layer. This
scheme has not been implemented. A somewhat simpler scheme is presently
employed. Lightning locations are aggregated into 1-hr files. The files are made
available to dispatch users. Since the locations are points, they are used as an
overlay on many different maps, including ignition efficiency, fuel type, terrain,
or any other desired layer. Implementation of more sophisticated schemes is a
goal for future studies.
Ignition probability information can also be used in gaming and forecast-
ing as well as combined with forest growth programs. These uses have yet to be
implemented.
410 Latham and Williams

Vegetation or
fuel type
layer

Weather
Moisture layers(s) temp,
algorithm R.H., rainfall

Ignition Duff depth

moisture processor <-
layer

^r •

Vegetation Ignition Duff depth

to ignition probability layer
type algorithm

i
Ignition
probability
layer

^f
Projected Lightning
fires locations
algorithm layer

^^
Projected
fires
layer

FIGURE 17 Operational diagram for ignition use of probabilities.

VIII. SMOKE, LIGHTNING,

AND CLOUD MICROPHYSICS
Large-scale networks for cloud-to-ground lightning detection have disclosed
an intriguing coupling between forest fires and the electrification of thunder-
storms within their range of influence. As emphasized in Section II, the great
majority of ordinary thunderstorms produce ground flashes with negative po-
larity. Within the last 20 years, several authors (Vonnegut and Orville, 1988;
Latham, 1991; Vonnegut, et al, 1995; Lyons et al, 1998a) have documented a
shift from negative to positive ground flash prevalence in association with for-
est fires and forest fire smoke.
Chapter 11 Lightning and Forest Fires 411

At present, it is not known whether a single mechanism will afford an expla-

nation for these several observations with different attendant conditions. When
aerosol in fire smoke is ingested in a convective storm, complexity rises in areas
of chemistry, cloud microphysics, and electrification. In line with these compli-
cations, multiple explanations have been put forward for the observations cited
in the previous paragraph. Latham (1991) suggested a thunderstorm whose
main dipole is inverted with respect to the usual positive dipole. Vonnegut et a\.
(1995) made a case for an influence-charging mechanism, with ingestion of
negative space charge in the fire smoke (Latham, 1999) leading to an inverted
polarity cloud. Lyons et al. (1998a) also considered an influence mechanism to
explain their observations but discounted it because the anomalous thunder-
storms they observed, though ingesting fire smoke, were more than 1000 km
from the source of the smoke, casting doubt on the persistence of appreciable
space charge. Lyons et al (1998a) did call attention to the effect of aerosol smoke
on cloud condensation and on the size of cloud droplets, with more numerous
smaller droplets accompanying smoke ingestion by the cloud, an effect docu-
mented at the cloud scale in satellite observations (Kaufman and Nakajima,
1993; Rosenfeld and Lensky, 1998), and more recently by in situ observations
(Reid et al, 1999). Recent laboratory simulations (Avila et al, 1998) suggest
that strong positive charging of graupel particles might occur in the presence
of small cloud droplets, resulting in an inverted polarity cloud, perhaps con-
ducive to positive ground flashes. Harvey and Edwards (1991) theorized that
the collection efficiency of smoke by droplets is enhanced if the smoke particles
are charged.
Clearly, in situ measurements of clouds ingesting fire smoke are needed to
shed further light on physical mechanisms. Two interesting feedback effects
associated with the fire smoke/cloud coupling add incentive to such studies.
First, if clouds ingesting smoke produce large numbers of positive ground
flashes, and if these ground flashes exhibit the continuing currents typical of
other positive flashes, then the forest fire threat is exacerbated. In fact, the Red
Lake #7 fire in Canada grew a cloud that produced lightning, starting six addi-
tional fires (Stocks and Flannigan, 1987). Second, given the evidence that smoke
ingestion by deep convection decreases cloud droplet size (Reid et al, 1999), it
is likely that droplet coalescence will be suppressed with the effect of a lower
precipitation yield (Schaefer and Day, 1981) to aid in dousing the fire naturally.

IX. GLOBAL IMPLICATIONS OF LIGHTNING

IGNITION CHARACTERISTICS
Global maps of lightning and forest fire incidence are shown in Figures 18 and
19, respectively (see color insert) (S.Goodman and H. Christian, personal com-
munication, 1999; Dwyer et al, 1999). As emphasized earlier in this chapter.
412 Latham and Williams

both lightning and flammable forest fuel are required for ignition of natural
fire. The areas of most prevalent lightning are the tropical continental zones—
Southeast Asia, Northern Australia, Africa, Central and South America. Forests
are also prevalent in these zones but as rainforests, with precipitation in abun-
dance, flora with high moisture content, and ground beneath the rainforest
canopy often completely inundated. The latter condition is likely to suppress
the most common fire initiation mechanism at midlatitude (described in Sec-
tions IV and V). This claim appears to be supported by the near equatorial zones
of minimum fire activity in Africa and South America, where the intertropical
convergence zone is likely on an annual basis, and where lightning activity is a
maximum. The majority of fires within these tropical zones are slightly dis-
placed from the equatorial regions and are set intentionally in the respective
dry seasons (associated with large-scale subsidence of the equatorial Hadley
circulation) to burn back the flora of the previous wet season and prepare for
crop planting.
The desert areas of the sub tropics, the more permanent areas of large-scale
subsidence from the intertropical convergence zone (e.g., the African Sahara,
the Mexican Sonoran desert, the Chilean desert, the Namibian desert, the cen-
tral area of Australia) lack both forest fuel and the moist convection necessary
for lightning. As a consequence, these areas are largely devoid of naturally
caused fire.
The northern boreal forests, heavily populated with flammable conifers, also
experience a moderately high incidence of lightning. In fact, the observations
with the Optical Transient Detector in space (Figure 18) showed a surprisingly
high incidence of lightning in all mid-to-high latitude land regions [e.g., north-
ern Canada, northern Russia (Siberia)]. Fires in a huge area of northern Russia
are increasingly well documented by satellite (Kasischke et al, 1999). Much of
the lightning in this region had evidently been missed in earlier space-based
observations (e.g., Orville and Henderson, 1986) because of the sustained day-
light into the late evening at this high latitude that interferes with the optical
detection of flashes (S. Goodman, personal communication 1998).
Northwestern North America, including British Columbia, is watered by
ocean air mass coming in off the Pacific Ocean, dropping rain on the mostly fire-
impervious high biomass density on the western slopes of the coastal moun-
tains. The eastern slopes, on the other hand, have a generally lower biomass
density, due to lower overall rainfall, but a higher lightning ignition probabil-
ity, and hence more fires. The Rocky Mountains get winter snowfall that pro-
vides necessary moisture for forests and dry summers that generally cause high
ignition efficiencies (Section IV. C). The forests of the American southwest, up
into Utah, generally receive moisture from air from the Gulf of Mexico during
the thunderstorms that arise in the summertime. The lightning fire season in
this area is in the early summer (May and June), whereas the fire season in the
northern Rockies is late in the summer (July and August).
Chapter 11 Lightning and Forest Fires 413

Suppression of fires started by natural ignitions in this region causes fuel

build-up and eventually ecosystem conversion to unhealthy forest. Natural ig-
nition under these unhealthy conditions will eventually cause large, uncontrol-
lable fires. Application of lightning ignition probabilities can be useful for iden-
tifying those areas that are prone to high lightning ignition probability and help
decide where fires should be started, and where natural ignition can be helpful.
When used in conjunction with lightning location, ignition efficiencies can in-
dicate areas that are likely to have large numbers of holdover fires.
Moving eastward in North America, central Canada and the United States
have vast cropland and very little timber. The ignition efficiency for grass is
lower, and much of the area is irrigated; lightning-caused fire is of little impor-
tance even though lightning densities are high. Further east, deciduous for-
ests have shaded, moist, dense forest floors, with concomitant low ignition ef-
ficiency. Peak fire periods are spring and fall, with few fires in the summer.
The southeast is forested, but there are relatively few lightning-caused fires
because of moist fuels. Lightning fires do occur in the Everglades peat (Sec-
tions V.D and VLB).
Lightning ignition considerations were used to calculate the effect of global
warming on lightning caused fires (Price and Rind, 1994). The results showed
that, under a scenario of COj doubling, lightning fires would increase by 30 -
77%, depending on the region of the United States considered. The calculations
were made under the assumption that the ecosystems did not change, that is,
that the fuels were constant over time. If there were a considerable increase of
fires, there would be a fuel change. As with many calculations of this type, there
are uncontrolled variables and no data. (See Chapter 10 in this book for more
discussion of the climate issue.)

X. CONCLUSION
Lightning ignition of wildland fuels plays a major role in the maintenance and
evolution of ecosystems. In this chapter, we have discussed the predominant
mechanism for lightning ignition and how this knowledge can be used in fire
management. There is little doubt, based on theoretical approaches, laboratory
experiments, and global fire and lightning data, that the fuel type and fuel state
play a much larger role than the lightning density in lightning fire ignitions.
The efficiency of lightning fires seems to be in the range of 0.01-0.04 fires/CG
flash in much of North America; that is, only about one to four flashes in 100
start fires. It remains to be seen whether these efficiencies are larger in other
places. Certainly, in central Africa and South America, efficiencies are much
lower; how much lower we have no way of knowing at present.
As we have seen, lightning fires can act upon the lightning environment
through smoke, even inverting the charge structure of thunderstorms. Again,
414 Latham and Williams

the exact mechanism remains a mystery, and there may be more than one
mechanism at work. Charge separation occurs due to flaming in the presence
of electric fields, and the smoke carrying that charge is carried into local pyro-
cumulus. That charge may not remain on the smoke particles for long, how-
ever, and inverted storms in regions far from the smoke origin in space and
time may be due only to the presence of smoke particles and their interaction
with thunderstorm-charging mechanisms. Finally, lightning ignition of fires
may or may not have a role in ecosystems altered by climate change.

REFERENCES
Adams, P. (1966). High-temperature, equilibrium thermodynamic properties of chemically react-
ing mixture of N2, O2, CO2, A, and Ne. Aerospace Tech. Rept. 9, Hughes Corp., Los Angeles.
Avila, E., Caranti, G., Castellano, N., and Saunders, C. (1998). Laboratory studies of the influence
of cloud droplet size on charge transfer during crystal-graupel collisions. J. Geophys. Res. 103,
8985-8996.
Barrows, J. S. (1954). Lightning fire research in the Rocky Mountains. J. For. 52, 845-847.
Barrows, J. S., Deterich,J. H., Odell, C. A., Kaehn, C. H., Fuquay, D. M., Schaefer, V. J., MacCready
Jr., R B., Colson, D., and Wells, H. J. (1958). Project Skyfire. In "Final Report of the Advisory
Committee on Weather Control," Vol. II, pp. 105-125. U.S. Government Printing Office, Wash-
ington, DC.
Barrows, J. S., Sandbert, D. V., and Hart, J. D. (1977). "Lightning Fires in Northern Rocky Mountain
Forests." Final Report, Cooperative Agreement 16-440-CA. USDA Forest Service, Intermoun-
tain Forest and Range Experiment Station, Ogden.
Baughman, R.G., Fuquay, D. M., and Mielke Jr., P. W (1976). Statistical analysis of a randomized
lightning modification experiment. J. Appl. Meteorol. 15, 790-794.
Berger, K. (1947). Lightning research in Switzerland. Weather 2(8), 231-238.
Boccippio, D. J., Williams, E. R., Heckman, S. J., Lyons, W A., Baker, I. T., and Boldi, R. (1995).
Sprites, ELF transients, and positive ground strokes. Science 269, 1088-1091.
Boccippio, D. J., Cummins, K., Christian, H. J., and Goodman, S.J. (2001). Combined satellite and
surface-based estimation of the intracloud/cloud-to-ground lightning ratio over the continen-
tal United States. Mon. Weath. Rev., in press.
Burke, C. P., and Jones, D. L. (1996). On the polarity and continuing current in unusually large
lightning flashes deduced from ELF events. J. Atmos. Terrestr. Physics 58, 531-548.
Cianos, N., and Pierce, E. T. (1972). "A Ground-lightning Environment for Engineering Usage."
Tech. Rep. 1, SRI Project 1834. Stanford Research Institute, Menlo Park.
Connor, T. R. (1967). "The 1965 ARPA-AEC Joint Lightning Study at Los Alamos." Publ. LA-3754,
Vol. 1. Los Alamos Scientific Laboratory, Los Alamos.
Cox, G., Ed. (1995). "Combustion Fundamentals of Fire." Academic Press, London.
Cummer, S. A., and Inan, U. S. (1996). Sprite-producing lightning using ELF radio atmospherics.
Geophys. Res. Lett. 24, 1731-1734.
Cummins, K. L., Murphy, M. J., Bardo, E. A., Hiscox, W L., Pyle, R. B., and Pifer, A. E. (1998). A
combined TOA/MDF technology upgrade of the U.S. national lightning detection network.
J. Geophys. Res. 103, 9035-9044.
Darveniza, M. (1980). "Electrical Properties of Wood and Line Design." University of Queensland
Press, St. Lucia.
Chapter 11 Lightning and Forest Fires 415

Darveniza, M., and Zhou, Y. (1994). Lightning-initiated fires: Energy absorbed by fibrous materi-
als from impulse current arcs. J. Geophys. Res. 99, 10663-10670.
Deeming, J. E., Burgan, R. E., and Cohen, J. D. (1977). "The National Fire Danger Rating System—
1978." Gen. Tech. Rep. lNT-39. USDA Forest Service, Intermountain Forest and Range Experi-
ment Station, Ogden.
Defandorf, F. M. (1955). A tree from the viewpoint of lightning. J. Wash. Acad. Sci. 45, 333-339.
Dennis, A. S. (1980). "Weather Modification by Cloud Seeding." Academic Press, New York.
Du Moncel, T. (1877). Recherches sur la conductibilite electrique des corps mediocrament con-
ducteurs et les phenomenes qui les accompagnet. Annates de Chimie et dePhysique [vol.?], 471.
Dwyer, E., Pereira, J. M. C , Gregoire, J-M., and DaCamara, C. C. (1999). Characterization of the
spatio-temporal patterns of global fire activity using satellite imagery for the period April 1992
to March 1993. J. Biogeog., in press.
Flannigan, M. D., and Wotton, B. M. (1991). Lightning-ignited fires in northwestern Ontario. Can.
J. For. Res. 21, 277-287.
Frandsen, W. H. (1987). The influence of moisture and mineral soil on the combustion limits of
smoldering forest duff. Can. J. For. Res. 17, 1540-1544.
Fuquay, D. M. (1974). Lightning damage and lightning modification caused by cloud seeding. In
"Weather and Climate Modification" (W N. Hess, Ed.), pp. 604-612. Wfley, New York.
Fuquay, D. M., and Baughman, R. G. (1969). "Project Skyfire Lightning Research." Final Report to
the National Science Foundation, USDA Forest Service, Intermountain Forest and Range Ex-
periment Station, Missoula.
Fuquay, D. M., Baughman, R. G., Taylor, A. R., and Hawe, R. G. (1967). Characteristics of seven
lightning discharges that caused forest fires. J. Geophys. Res. 72, 6371-6373.
Fuquay, D. M., Taylor, A. R., Hawe, R. G., and Schmidt Jr., C. W (1972). Lightning discharges that
started forest fires. J. Geophys. Res. 77, 2156-2158.
Gisborne, H. T. (1924). Lightning fires and storms in the northern Rocky Mountains. In "The Tim-
berman," p. 2. Portland.
Gisborne, H. T. (1926). Lightning and forest fires in the northern Rocky Mountains. Mon. Weath.
Rev. 54, 281-286.
Gisborne, H. T. (1931). A five-year record of lightning storms and forest fires. Mon. Weath. Rev. 59,
139-150.
Greig, J. R., Pechacek, R. E., and Raleigh, M. (1985). Channel cooling by turbulent convective mix-
ing. Physics of Fluids 28, 2357-2364.
Hahn, G. J., and Shapiro, S. S. (1967). "Statistical Models in Engineering." John Wiley and Sons,
New York.
Harvey, T. R, and Edwards, L. L. (1991). A parametric investigation of electrical effects on aerosol
scavenging by droplets over large fires. Atmos. Environ. 25A, 2607-2614.
Huang, E., Williams, E., Boldi, R., Lyons, W, Nelson, T., Taylor, M., Heckman, S., and Wong, C.
(1999). Criteria for sprites and elves based on Schumann resonance measurements. J. Geophys.
Res. 104, 1693-1694.
Incropera, F. P., and DeWitt, D. P. (1996). "Fundamentals of Heat and Mass Transfer," 4th ed. John
Wiley and Sons, New York.
Jones, J. C , Ramahti, H., Fowler, D., Vorasurayakarnt, J., and Bridges, R. G. (1990). The self heat-
ing and thermal ignition propensity of forest floor litter. J. Fire Sci. 8, 207-223.
Kappenman, R. F. (1988). A simple method for choosing between the lognormal and Weibull dis-
tributions. Stat. Prob. Lett. 7 (Sept), 123-126.
Kasischke, E. S., Bergen, K., Fennimore, R., Sotelo, R, Stephens, G., Janetos, A., and Shugart, H. H.
(1999). Satellite imagery gives clear picture of Russia's boreal forest fires. EOS 80, 141.
Kaufman, Y. J., and Nakajima, T. (1993). Effect of Amazon smoke on cloud microphysics and al-
bedo analysis form satellite imagery. J. App. Meteorol. 32, 729-744.
416 Latham and Williams

King, L. A. (1961). "The Voltage Gradient of the Free-burning Arc in Air or Nitrogen." ERA Rep.
G/XT172. The British Electrical and Alhed Industries Research Association, Leatherhead.
Latham, D.J. (1979). "Progress Toward Locating Lightning Fires." Res. Note INT-269. USDA For-
est Service, Intermountain Forest and Range Experiment Station, Ogden.
Latham, D. J. (1984). A numerical simulation of the lightning continuing current channel. In "Pro-
ceedings, Seventh International Conference on Atmospheric Electricity," pp. 417-420. Ameri-
can Meteorological Society, Boston.
Latham, D.J. (1986). Anode column behavior of long vertical air arcs at atmospheric pressure. IEEE
Trans. Plasma Set PS-14, 220-227.
Latham, D.J. (1987). "Design and Construction of an Electric Arc Generator for Fuel Ignition Stud-
ies." Res. Note INT-366. USDA Forest Service, Intermountain Research Station, Ogden.
Latham, D. J. (1989). "Characteristics of Long Vertical DC Arc Discharges." Res. Pap. INT-407.
USDA Forest Service, Intermountain Research Station, Ogden.
Latham, D.J. (1991). Lightning flashes from a prescribed fire-induced cloud. J. Geophys. Res. 96,
17151-17157.
Latham, D. J. (1999). Space charge generated by wind tunnel fires. Atmos. Res. 51, 267-278.
Latham, D. J., and Schlieter, J. A. (1989). "Ignition Probabihties of Wildland Fuels Based on Simu-
lated Lightning Discharges." Res. Pap. lNT-411. USDA Forest Service, Intermountain Research
Station, Ogden.
Ligda, M. G. H. (1956). The radar observation of lightning. J. Atmos. Terr. Phys. 9, 329-346.
Loeb, L. B. (1966). The mechanisms of stepped and dart leaders in cloud-to-ground lightning
strokes. J. Geophys. Res. 71, 4711-4721.
Loftsgaarden, D. O., and Andrews, P. A. (1992). "Constructing and Testing Logistic Regression
Models for Binary Data: Applications to the National Fire Danger Rating System." Gen. Tech.
Rep. INT-286. USDA Forest Service, Intermountain Research Station, Ogden.
Lyons, W. A., Nelson, T. E., Williams, E. R., Cramer, J. A., and Turner, T. R. (1998a). Enhanced posi-
tive cloud-to-ground lightning in thunderstorms ingesting smoke from fires. Science 282, 7 7 -
80.
Lyons, W A., Uliasz, M., and Nelson, T. E. (1998b). A climatology of large peak current cloud-to-
ground lightning flashes in the contiguous United States. Mon. Weath. Rev. 126, 2217.
Malan, D.J. (1963). "Physics of Lightning." English University Press, London.
Martin, S., and Broido, A. (1963). "Thermal Radiation and Fire Effects of Nuclear Detonations."
Rep. AD 422411. U.S. Naval Radiological Defense Laboratory, San Fransisco.
Mazur, V, Shao, X. M., and Krehbiel, P. (1998). "Spider" lightning in intracloud and positive cloud-
to-ground flashes. J. Geophys. Res. D16, 19811-19822.
McEachron, K. B., and Hagenguth, J. H. (1942). Effect of lightning on thin metal surfaces. AIEE
Trans. 61, 559-564.
McMullen, L. H., and Atkins, M. D. (1962). On the flight and host selection of the Douglas-fir beetle
Dendroctomus pseudotsugae Hopk. (Coleoptera: Scolytidae). Can. Entomol. 94, 1309-1325.
McRae, R. H. D. (1992). Prediction of areas prone to lightning ignition. Int. J. Wildl. Fire 2,123-130.
Meisner, B. N. (1993). "Correlation of National Fire Danger Rating System Indices and Weather
Data with Fire Reports." Final Report, Interagency Agreement No. R500A20021. USDA Forest
Service, Forest Fire Laboratory, Pacific Southwest Research Station, Riverside.
Melosh, H. J., Schneider, N. M., Zahnle, K. J., and Latham, D.J. (1990). Ignition of global wildfires
at the Cretaceous/Tertiary boundary. Nature 343, 251-254.
Morris, W G. (1934). "Lightning Storms and Fires on the National Forests of Oregon and Wash-
ington." USDA Forest Service, Pacific Northwest Forest Experiment Station, Portland.
Nash, C. H., and Johnson, E. A. (1996). Synoptic climatology of lightning-caused forest fires in
subalpine and boreal forests. Can. J. For. Res. 26, 1859-1874.
Chapter 11 Lightning and Forest Fires 417

Orville, R. E., and Helsdon Jr., J. H. (1974). Quantitative analysis of a lightning return stroke for
diameter and luminosity changes as a function of time. J. Geophys. Res. 79, 4059-4067.
Orville, R. E., and Henderson, R. W. (1986). Global distribution of midnight lightning: Septem-
ber 1977 to August 1978. Mon. Wea. Rev. 114, 264-2653.
Plummer, F. G. (1912). "Lightning in Relation to Forest Fires." Bulletin 111. USDA Forest Service,
Government Printing Office, Washington, DC.
Price, C , and Rind, D. (1994). The impact of a 2xC02 climate on lightning-caused fires. J. Climate
7, 1484-1494.
Reid, J. S., Hobbs, P. V., Ragno, A. L., and Hegg, D. A. (1999). Relationships between cloud droplet
effective radius, liquid water content, and droplet concentration for warm clouds in Brazil em-
bedded in biomass smoke. J. Geophys. Res. 104, 6145-6153.
Renkin, R. A., and Despain, D. G. (1992). Fuel moisture, forest type, and lightning-caused fire in
Yellowstone National Park. Can. J. For. Res. 22, 37-45.
Rorig, M. L., and Ferguson, S. A. (1999). Characteristics of lightning and wildland fire ignition in
the Pacific Northwest. J. Appl. Met. 38, 1565-1575.
Rosenfeld, D., and Lensky, 1. (1998). Satellite-based insights into precipitation formation processes
in continental and maritime convective clouds. Bull. Amer. Meteorol. Soc. 79, 2457-2476.
Salanave, L. E. (1965). The photographic spectrum of lightning: Determination of channel tem-
perature from slitless spectra. In "Problems of Atmospheric and Space Electricity" (S. C. Cori-
niti, Ed.), pp. 371-384. Elsevier Publishing Company, New York.
Schaefer, V. J. (1949). "The Possibihty of Modifying Lightning Storms in the Northern Rockies."
Pap. No. 19. USDA Forest Service, Northern Rocky Mountain Forest and Range Experiment
Station, Missoula.
Schaefer, V.J. (1953). "Final Report Project Cirrus, Part 1." Report No. RL-785. General Electric
Research Laboratory, Schenectady.
Schaefer, V. J., and Day, J. A. (1981). "A Field Guide to the Atmosphere." Houghton Mifflin and Co.,
Boston.
Schmitz, R. R, and Taylor, A. R. (1969). "An Instance of Lightning Damage and Infestation of Pen
derosa Pines by the Pine Engraver Beetle in Montana." Res. Note INT-88. USDA Forest Service,
Intermountain Forest and Range Experiment Station, Ogden.
Schonland, B. F. J., and Collens, H. (1934). Progressive lightning. Proc. Roy. Soc. (Lond) A143,
654-674.
Sentman, D. D. (1987). EOS 67 (suppl.) 67, 1069.
Simms, D. L., and Law, M. (1967). The ignition of wet and dry wood by radiation. Combust. Flame
11,377-388.
Stocks, B. J., and Flannigan, M. D. (1987). Analysis of the behaviour and associated weather for a
1986 northwestern wildfire: Red Lake #7. In "Proceedings, Ninth Conference on Fire and For-
est Meteorology," pp. 94-99. American Meteorological Society, Boston.
Strachan, D. C , and Barrault, M. R. (1975). "Axial Velocities in the High Current Free Burning Arc."
Rep. ULAP-T34. Department of Electrical Engineering and Electronics, University of Liver-
pool, Liverpool.
Strom, A. R (1946). Long 60-cycle arcs in air. AIEE Trans. 65, 113-117.
Susott, R. A. (1984). "Heat of Preignition of Three Woody Fuels Used in Wfldfire Modeling Re-
search." Res. Note INT-342. USDA Forest Service, Intermountain Forest and Range Experiment
Station, Ogden.
Taylor, A.R. (1969). Lightning effects on the forest complex. In "Proceedings, Annual Tall Timbers
Fire Ecology Conference," pp. 127-150. Tall Timbers Research Station, Tallahassee.
Taylor, A. R. (1973). Ecological aspects of lightning in forests. In "Proceedings, Annual Tall Timbers
Fire Ecology Conference," pp. 455-482. TaU Timbers Research Station, Tallahassee.
418 Latham and Williams

Thomson, E. M. (1980). Characteristics of Port Moresby ground flashes. J. Geophys. Res. 85, 1025-
1036.
Uman, M. A. (1969). "Lightning." McGraw-Hill, New York.
Uman, M. A., and Voshall, R. E. (1968). Time interval between lightning strokes and the initiation
of dart leaders. J. Geophys. Res. 73, 497-506.
Van Wagner, C. E. (1987). "Development and Structure of the Canadian Forest Fire Index System."
Tech. Rep. 35. Canadian Forestry Service, Ottawa.
Van Wagtendonk, J. W (1991). Spatial analysis of lightning strikes in Yosemite National Park. In
"Proceedings, Eleventh Conference on Fire and Forest Meteorology," pp. 605-611. American
Meteorological Society, Boston.
Vonnegut, B., and Orville, R. E. (1988). Evidence of lightning associated with the Yellowstone Park
forest fire (abstract). EOS Trans. AGU 69, 1071.
Vonnegut, B., Latham, D. J., Moore, C. B., and Hunyady, S.J. (1995). An explanation for anomalous
lightning from forest fire clouds. J. Geophys. Res. 100, 5037-5050.
Williams, E. R. (1989). The tripole structure of thunderstorms. J. Geophys. Res. 94, 13151-13167.
Wilhams, E. R. (1998). The positive charge reservoir for sprite-producing lightning. J. Atmos. Terr.
Phys. 60, 689-692.
 CHAPTER

Statistical Inference for

Historical Fire Frequency
Using the Spatial Mosaic
W. J. REED
Department of Mathematics and Statistics, University of Victoria, Victoria, B.C., Canada

I. Introduction
II. Graphical Analysis
III. Statistical Inference with Prespecified Change Points
A. Constructing a Quasi Likelihood for an
Overdispersed Multinomial Model
IV. The Efficiency of Sample vs. Map Data
V. Determining Epochs of Constant Eire Erequency
References

1. INTRODUCTION

In forested regions where the principal agent of stand regeneration is fire (e.g.,
in boreal forests), the spatial mosaic of stands depends on the frequency and ex-
tent of fires in the past. To some extent, knowledge of the spatial mosaic can be
used to make inference about past fire frequency. This chapter deals with sta-
tistical methodology for such inference. Given complete knowledge of fire his-
tory, one could, in principle, reconstruct the spatial mosaic. In contrast, knowl-
edge of the current spatial mosaic is not sufficient for complete knowledge of
past fires. The reason for this is that evidence of fires in the past can be obliter-
ated by subsequent fires burning over all or part of the extent of the earlier fires.
The present spatial mosaic provides complete evidence of only the most recent
fire or fires; it provides decreasing information about fires more distant in time;
and of very early fires, little or no evidence may remain.
A time-since-fire map is closely related to the spatial mosaic and gives the
time since the most recent fire at every point in a study area. In contrast a time-
since-fire sample contains only the time since the most recent fire at a sample of

Forest Fires
Copyright © 2001 by Academic Press. All rights of reproduction in any form reserved. 419
420 W.J. Reed

points (drawn according to some well-defined sampling plan) from the study
area. Usually because of the limitations of techniques of dating stand origins or
past fires, the time-since-fire observations are grouped into classes (often of
width one decade). In this chapter statistical methodologies for analyzing time-
since-fire map and sample data will be discussed. The main issues addressed are:

• methods of graphical data analysis of such data;

• estimation of fire frequency at distinct epochs in the past. Likelihood
methods will be used to obtain point estimates and confidence intervals.
• testing for the significance of changes in fire frequency at specified
times;
• determination of when (possibly multiple) changes in fire frequency
may have occurred.
The methods presented collect and summarize recent research in this area.

II. GRAPHICAL ANALYSIS

A great deal of information about historical fire frequency can be obtained by
simple graphical analysis. For time-since-fire map data, one can plot cumulative
proportional areas, with time since last fire ^ t, on a logarithmic scale, against
time t (on a natural scale). The lower panels of Figures 1, 3 and 4 are examples
of such plots. For sample data, the same procedure can be followed using the
cumulative proportion of sample points with time since last fire ^ t (Fig-
ure 2). If the hazard of burning has been both age-independent and time-
homogeneous, the points on the plot should lie close to a straight line with
negative slope (equal to the hazard of burning), since at any point the proba-
bility that the most recent fire occurred t or more years ago is e~^\ where A is
the hazard of burning. Abrupt changes in the plot suggest temporal changes in
the hazard of burning. For example, if the hazard of burning changed at some
distinct change point (as might possibly have occurred for example at the time
of European intervention), then one would expect a sharp elbow in the plot at
that point. Thus a plot which naturally divides into a number of straight line
segments suggests a number of distinct historical epochs of roughly constant
fire frequency, with the changes in fire frequency occurring at times determined
by the points of intersections of adjacent segments. By comparing such plots for
different subsets of the study area (e.g., broken up spatially or by aspect) one
can get a good idea of whether historical fire frequency was homogeneous or
not with respect to these categories.
Even though the vertices and slopes of line segments determined graphically
provide rough estimates of change points and intervening fire frequencies, it
is desirable to have more objective statistical methodologies for accomplish-
ing this. Subsequent sections of this chapter describe such methodologies.
 Boundary Waters Canoe Area
100000
•D
0
C
60000 H

20000
Lilfl 11 j i Ji 1,1 1
100 200 300
CO
Time since fire

^ i.oooo-f

0.0100-

0.0001 H
o
100 200 300
Tinne since fire
FIGURE 1 Time since last fire for Boundary Waters Canoe Area (Heinselman, 1973). The top
panel shows the areas (on vertical axis in hectares) with given time since last fire (horizontal axis
in years). The bottom panel plots the cumulative proportional area (on logarithmic scale) with time
since fire exceeding the time (in years) on the horizontal axis.

'oT
CO
100-
VN
•••V
^
^ 50- ^^
\
O
V
O \
CD

1 10- ^
^0 •
•
> •
•« 5- •
D •
E •
O •

1- •
h r ^— \ \ \— 1
50 100 150 200 250 300

Time since last fire

FIGURE 2 Cumulative percent of sample points (on logarithmic scale) with time since fire ex-
ceeding the time (in years) on the horizontal axis for time-since-fire sample data of Wood Buffalo
National Park (Larsen, 1996).
422 W.J. Reed

I
CO
CD
^ 10

^
\ ^^ ^
'^^s ]yIL^B ^ J I ^ ill^i illlr^M_, m m\

100 200 300 400

Years since last fire
FIGURE 3 Time since last fire for Kananskis Watershed (Johnson and Larsen, 1991). The top
panel shows the percentage area (on vertical axis) in given time-since-fire class (horizontal axis in
years). The bottom panel plots the cumulative percentage area (on logarithmic scale) with time
since fire exceeding the time (in years) on the horizontal axis.

However, the problem is approached in stages. First, the problem of obtaining

maximum likelihood estimates (MLEs) of the hazards of burning prevailing
between prespecified change points is addressed, along with how to obtain
approximate confidence intervals and tests for the significance of the change
points. This is done for both map and sample data. Following this, the question
of determining change points from the data alone is addressed.

III. STATISTICAL INFERENCE WITH

PRESPECIFIED CHANGE POINTS
We assume that the observations whether for map or sample data are binned
into time-since-fire classes 1, 2, . . . , m, which are of equal width, say T years,
save for the oldest which is open-ended. Specifically, assume that the classes
1, 2 , . . . , m — 1 are defined by time since last fire in the intervals [0, T),
[T, 2T),... , [(m — 2)T, (m — 1)T), respectively and that class m is defined by
time since last fire greater than or equal to (m — 1 )T. In practice, the resolution
Chapter 12 Statistical Inference for Historical Fire Frequency 423

14-|
CO 1 2

s 10.

n iQ

Years since last fire

FIGURE 4 Time since last fire for Glacier National Park (Johnson et al, 1990). The top panel
shows the percentage area (on vertical axis) in given time-since-fire class (horizontal axis in years).
The bottom panel plots the cumulative percentage area (on logarithmic scale) with time since fire
exceeding the time (in years) on the horizontal axis.

of dating fires is no finer than one year, so T will be a positive integer (often
T = 10 or 20).
Under tfie assumption of homogeneity, at any given location in the study
area, the probability that the time since last fire belongs to class j can be de-
termined in terms of the hazards of burning prevailing at and since that last
fire date, by observing that to belong to class j , there must have been a fire
between (j — 1)T and jT years ago, and subsequently no fire. Specifically, it is
-^~\^Tr (1 — e~ ^) = 6j, say, where A^^^ is the instantaneous hazard of burn-
ing between (i — 1)T and iT years ago. With known change points, the pa-
rameters A^^% for i = 1, 2 , . . . , m, can be expressed in terms of parameters rep-
resenting the hazards of burning in the epochs between the change points. For
example, with no change points, all the A^'^ are equal to the assumed constant
hazard AQ, say. If there were one change point pT years ago, with a hazard of
burning A2, say, prevailing before that time and A^, since that time, then
A^'^ = Al for i = 1, 2 , . . . , pT and A^*^ ^ A2 for i = p + 1, p + 2 , . . . , m, etc.
To estimate by maximum likelihood the parameters representing the hazards
of burning in the epochs between the change points, one needs to construct a
424 W.J. Reed

likelihood function [i.e., determine the probability of the observed areas (or
frequencies of sample points) falling in the various classes] in terms of these
parameters. If one could reasonably assume independence between points in
the study area, with respect to time since last fire, the likelihood function could
be obtained directly from the multinomial probabability distribution. How-
ever, such an assumption is not reasonable because of the fact that fires spread
spatially. To reflect this fact, one needs a model for the distribution of areas in
time-since-fire classes which exhibits a contagion effect. Two approaches have
been used to date to accomplish this. One is to use a parametric model (the
Dirichlet distribution) for the areas in time-since-fire classes. This provides a
likelihood directly for map data (Reed, 1994), whereas a likelihood for sample
data can be obtained by computing the (marginal) probability of observing the
observed frequencies in the various classes, assuming that the proportional ar-
eas follow the Dirichlet distribution. This leads to a likelihood of the form of
the Dirichlet compound multinomial distribution, which is the multivariate ana-
logue of the widely used beta-binomial distribution (Reed, 1998).
The other approach, which in many ways is more straightforward, is to in-
corporate the contagion effect by using an overdispersed multinomial model with
corresponding quasi likelihood (Reed et al, 1998). This is based on the as-
sumption that the contagion effect, due to spatial spread of fires, has no effect
on the expected areas (or frequencies) in the time-since-fire classes, but it does
have an effect on the variances-covariances, inflating them by a constant fac-
tor. This second approach is more general in the sense that the Dirichlet dis-
tribution for proportional areas, and the resulting distribution for frequencies
of sample points, are both overdispersed multinomial distributions themselves
(Reed, 1998). Which of the two approaches is superior in any given context is
not known. This issue could probably be resolved by studying the performance
of the two methods using simulated data. However, to date such a study has not
been undertaken. In this chapter, we concentrate on the overdispersed multi-
nomial approach. One advantage is that analytic expressions can be obtained
for maximum likelihood estimates of epochal hazards of burning and for their
standard errors. There is a simple interpretation of the MLEs, and furthermore
they agree well with the graphical estimates. In contrast, using the Dirichlet
model approach, numerical maximization is required to obtain MLEs of the
epochal hazards of burning, and it is not possible to obtain a simple interpre-
tation of the estimates.

A. CONSTRUCTING A QUASI LIKELIHOOD FOR

AN OVERDISPERSED MULTINOMIAL MODEL

For data from an overdispersed multinomial distribution in which there is a

constant inflation factor (overdispersion parameter, a^) for all variances and co-
Chapter 12 Statistical Inference for Historical Fire Frequency 425

variances, one can construct a quasi likelihood function of the form

where for map data yj = Aj/{ 2]!=i Aj), the proportional area in class j , and for
sample data, with frequencies jj, y^ = fj/{^^=ifj), the proportion of the ob-
servations in class j . The overdispersion parameter a^ will differ for map and
sample data. In most respects regarding model parameters (other than a^), the
quasi likelihood behaves like an ordinary log likelihood. Thus, one can find
MLEs by maximizing the quasi likelihood and obtain approximate standard
errors from the observed (quasi) Fisher information matrix (Hessian matrix of
second derivatives of quasi likelihood at the maximum).
Simple analytic expressions are available for the MLEs of the hazards of burn-
ing in distinct epochs. For example if there is only one epoch (i.e., the same
hazard of burning AQ is assumed to have prevailed at all times in the past), then
from differentiation of the quasi-likelihood it is easy to show that the MLE is
Ao = (-1/T)log(qo)
where

2m}=
^0 ^m-l
ym-i
and

i=j+l

is the proportional area (or proportion of the observations, for sample data)
with time since last fire at least jT, for j = 1, 2 , . . . , m — 1. Furthermore, the
approximate variance of the estimator qo is

,(S7=?Sj)(2;=?yj)
var(qo) - (T -—^
( 2 j = i Sj_,y
If it is assumed that there are several epochs with distinct hazards, say Ai be-
tween 0 and piT years ago, Aj between p^T and pjT years ago, etc., then the
MLEs of the separate hazards of burning can be found in a similar way. For ex-
ample, if there are two epochs separated by a single change point pT years ago,
then for the more recent epoch
Ai = (-l/T)log(qO
and for the earlier epoch
X, = {-l/T)log{q2)
426 W.J. Reed

where

Furthermore, these estimates are independent and have approximate vari-

ances given by

var(qi) - (7 .^p 3 ; var(q2) - cr -—n

To use these variance formulas to obtain standard errors and confidence in-
tervals for the hazard of burning, etc., one needs to estimate the overdispersion
parameter a^. This is usually done using the Pearson estimate, which for map
data is

m-r- ipte^{i -e;)

where d^ is the MLE of 6^ and r is the number of epochs. With a single hazard of
burning assumed to have prevailed at all times past, the MLEs of the cell prob-
abilities are 6j = (1 - ^o)^/)"\ j = 1, 2 , . . . , m - 1 and 6^ = ^o""^- With two
epochs, they are ^^ = (1 - q i ) q r \ J = 1, 2 , . . . , p, ^^ = (1 - ^2)^1 q2~^'\ j =
p + 1,2, . . . , m — 1 and 6,^ = qf q^"^~\ Similar formulas hold when there are
more than two epochs.
An approximate 100(1 — a)% confidence interval for the hazard of burn-
ing in epoch r can be obtained from the confidence interval for q,, which is

q, ± Z«/2V(6-^/o-^)var(q,)

where Za/i is a percentage point from the standard normal distribution. The
transformation q —> ( — 1/T) log(q) will convert this to a confidence interval for
the hazard of burning and q ^ ( —T)/(log(q)) will convert it to a confidence in-
terval for the fire cycle.
A test for the significance of a change point between epochs can be con-
structed by computing the liklelihood ratio (LR) statistic. Specifically, one com-
putes the test statistic

A = 24(Qi - Qo)
a
where Qi and QQ are the maximized quasi likelihoods assuming the change
point present and not, respectively, and &l is the Pearson estimate of a^ com-
puted with the change point present. The null distribution of this statistic from
which a P-value can be obtained is approximately Fi^_^_i where v is the num-
Chapter 12 Statistical Inference for Historical Fire Frequency 427

ber of epochs when the change point is present. An important point to note
here is that, because of the problem of selection bias, this procedure is not valid
if the change point is suggested from an examination of the data, e.g., from a
graphical anlysis (Reed et al, 1998). One can also use a LR statistic to construct
confidence intervals for the epochal hazards of burning (or fire cycles) using
the method described in Reed et al. (1997).

Example 1. Map Data. Boundary Waters Canoe Area

Figure 1 shows the time-since-fire distribution for the Boundary Waters Canoe
Area, from the maps obtained by Heinselman (1973) in his pioneering fire his-
tory study. Since fires are dated to the year, the cell width here is T = 1. Euro-
pean settlement in the area first occurred in the 1860s, and there appears to be
a change in the slope of the semi-log cumulative frequency plot around this
time (a little more than 100 years ago). Assuming a single change point in 1865
yields the following maximum likelihood estimates of the hazards of burning
and corresponding fire cycle (with approximate 95% confidence intervals):
Epoch 1 (1865-1971): A^ = 0.00883 p.a., FCi - 113 (73-245) yr
Epoch 2 (before 1865): A2 = 0.0121 p.a., FC2 = 83 (42-3915) yr
The overdispersion parameter is estimated as 0.0593.

Example 2. Sample Data. Wood Buffalo National Park

Larsen (1996) reported a detailed study of the fire history of Wood Buffalo Na-
tional Park, a 45,000 km^ area of boreal forest straddling the border between
Alberta and the Northwest Territories. Larsen determined (to the nearest year)
the time since last fire at 166 randomly selected sites. These are displayed in Fig-
ure 2 with m = 300 classes of width T = 1 year. The graph suggests a change
point at p = 129 years ago (1860). Assuming this to be the case, the maximum
likelihood estimates of the hazards of burning and corresponding fire cycle
(with approximate 95% confidence intervals) are:
Epoch 1 (1861-1989): Ai = 0.0142 p.a., FCi = 70 (58-90) yr
Epoch 2 (1860 and earlier): A2 = 0.0292 p.a., FC2 = 34 (23-69) yr
The overdispersion parameter is estimated as 0.0106.

IV. THE EFFICIENCY OF

SAMPLE VS. MAP DATA
Map data corresponds to a complete census of the study area. Since one can ob-
tain estimates of past fire frequencies using observations on time since the last
428 W.J. Reed

fire at only a random sample of points in the study area, an obvious question is
how much precision is foregone by determining time since fire at a sample of
points rather than everywhere in the study area. Of course, the reasons for col-
lecting time-since-fire map data extend beyond the estimation of historical fire
frequency. These issues are not addresssed here. Rather, attention is confined to
assessing the relative statistical efficiency of using sample vs. map data for such
estimation.
To this end, suppose that the proportional areas over the whole study area
Uj = Aj/{ 2j=-i Aj) follow an overdispersed multinomial distribution with over-
dispersion parameter al < I. Under this assumption,

var(aj) = aldj{l - Oj); cov(ai, Uj) = -crlefij

By first conditioning on the proportional areas a^ one can show that for n time-
since-fire observations at a random sample of points in the study area, the pro-
portional frequencies yj = fj/n in the m classes follow an overdispersed multi-
nomial distribution with overdispersion parameter

crl = ^[l +(n- 1V^]>C7^

The quasi Fisher information matrix I, which is obtained as the expected value
of the Hessian matrix of second derivatives of the quasi likelihood, provides a
measure of the precision of the MLEs. It is easy to show that this information
matrix for map data, JQ say, differs from that for sample data, I^ say, only by the
scalar multiplicative constant, that is.
2
cr,
a 1

The relative efficiency (ratio of variances of estimators) for sample data relative
to map data is, thus,

al nal
Rel. Eff. =-i = , ' .
a\ 1 + (n - l)al
which is easily confirmed to be less than one and converging to one as sample
size n ^ oo (i.e., estimates based on sample data will have lower precision than
those based on map data) but will approach the latter as sample size grows large.
One can substitute an estimate for a^ in this expression to assess the loss in
precision by using sample data.

Example 1 (continued). Boundary Waters Canoe Area

The estimate of the overdispersion parameter for these data was a^ = 0.0593.
Table 1 gives estimates, using this value, of the relative efficiency of sample data
Chapter 12 Statistical Inference for Historical Fire Frequency 429

TABLE 1 Estimated Relative Efficiency of Using Sample Data (sample size n) vs. Map Data
for Boundary Waters Canoe Area and Ratio of Expected Widths of Confidence Intervals

Sample size n 10 20 50 100 200 500 1000

Rel. eff. 0.39 0.56 0.76 0.86 0.93 0.97 0.98
Ratio of C.I. widths 1.61 1.34 1.15 1.08 1.04 1.02 1.01

(for various sample sizes n) vs. map data and the ratio of (expected) widths of
corresponding confidence intervals. It is clear that a sample of moderate size
(100 to 200) would be almost as good as a complete map, in terms of the pre-
cision of estimates. The expected width of confidence intervals would be only
8% greater with n = 100 and only 4% greater with a sample of size n = 200.

Example 2 (continued). Wood Buffalo National Park

Larsen (1996) determined time since last fire at n = 166 sample points in Wood
Buffalo National Park. The overdispersion parameter for a sample of this size
was estimated as aj = 0.0106. Thus, an estimate of the overdispersion parame-
ter for map data is aj = (166 X 0.0106 - 1)/(166 - 1) = 0.0046. Note that
this estimate is smaller by an order of magnitude than the corresponding esti-
mate for the Boundary Waters Canoe Area, suggesting a much smaller contagion
effect (and thus a finer patch mosaic) in Wood Buffalo than in the Boundary
Waters. Using this estimate, one can assess how much precision was foregone
using a sample of points rather than a complete map survey and how the pre-
cision of estimates from sample data depends on sample size. Table 2 gives in-
formation on this.
The results suggest that confidence intervals would be about one third
shorter if map data had been obtained. Unlike the case of the Boundary Waters,
where relatively small samples could provide precision comparable with map
data, for Wood Buffalo such small samples would not be adequate. For example,
for the Boundary Waters, a sample of size 50 would yield confidence intervals

TABLE 2 Estimated Relative Efficiency of Using Sample Data vs. Map Data
for Wood Buffalo National Park and Ratio of Expected Widths of Corresponding
Confidence Intervals (second row)

Sample size n 10 20 50 100 166 200 500 1000

Rel. eff 0.044 0.085 0.19 0.32 0.43 0.48 0.70 0.82
Ratio of C.I. widths 4.76 3.44 2.31 1.78 1.52 1.44 1.20 1.10
C.I. width rel. to actual 3.13 2.27 1.52 1.17 1.00 0.95 0.79 0.73

The last row is the ratio of width of confidence intervals using sample size n and actual sample
size 166.
430 W.J. Reed

only about 15% wider than map data; for Wood Buffalo, such a sample size
would yield confidence intervals 130% wider than map data (and 50% wider
than those obtained with the actual sample size, 166). The reason for the differ-
ence lies with the order of magnitude difference in the overdispersion parame-
ter estimates. With greater contagion, and on average larger patches in the time-
since-fire mosaic in the Boundary Waters, a relatively small number of sample
points can, with high probability, provide information on most patches. On the
other hand, with the smaller contagion parameter and finer patch mosaic in
Wood Buffalo, one needs many more sample points on average to cover most
patches.

V. DETERMINING EPOCHS OF
CONSTANT FIRE FREQUENCY
In the previous sections, it was assumed that the number of epochs, with con-
stant hazard of burning, and the change points dividing them were known. Most
often, this will not be the case, and one will face the problem of determining
the epochs from the data. This section provides a brief description of a method-
ology for accomplishing this.
From the statistical point of view, determining the number and location of
the change points (i.e., determining the epochs) is a problem in model identi-
fication, analagous to deciding which regressor variables should be included in
a regression model. There are a number of ways to approach such problems.
One is to use some sort of iterative procedure in which change points can be
added or removed from the model (analagous to forward selection and step-
wise procedures in regression). This approach has been used for identifying fire
epochs (Reed, 1998), but there are a number of difficulties associated with it.
An alternative approach which avoids these difficulties is to use the Bayes Infor-
mation Criterion to decide on the best single (or several) models.
From one point of view, the Bayes Information Criterion (BIC) can be viewed
as a log likelihood, adjusted for the number of parameters in a model (here the
number of change points and intervening hazards of burning). It can also be in-
terpreted in a Bayesian context (see later discussion).
Consider a hieararchy of models:

HQ-. NO change points (constant hazard of burning at all times in the past).
Hi'. One change point (separating two epochs with distinct hazards of
burning),
Hj. Two change points (separating three epochs with distinct hazards of
burning),
etc.
Chapter 12 Statistical Inference for Historical Fire Frequency 431

Hjj! k change points (separating fe + 1 epochs with distinct hazards of

burning).
Under any model H^ there are 2r + 1 undetermined parameters (r change
points separating r + 1 hazards of burning A^ or parameters qi = e~^'^) in ad-
dition to the overdispersion parameter. MLEs of the change points can be found
by comparing the quasi hkehhoods, maximized over the (r + 1)^^ parameters,
for all {^~^) possible choices of r change points. A comparison of models HQ,
HI, ... , etc., can be achieved by comparing the BIC for each model. It can be
shown (Reed, 2000) that an approximation for the BIC is

BIQ-D,-(df),log(^j (2)

where D^ is the minimum scaled quasi deviance for model H^ i.e.,

2
^yj logyj - E l 2 ^j log * + ^ yj] iog(i - ^0

minimized over the choice of r change points po, pi,... , Pr (with po — 0 and
p^+i = m). The terms q^ are the MLEs of the q^ parameters for epochs i =
1, 2 , . . . , r + 1). Precisely,
+ 1 5j

^J-Pi^i + l ^J-1

In Eq. (2), the term (df)^ is the residual degrees of freedom for model H^ [i.e.,
df^ = (m — 1) — (2r + 1) and n = 2 ] l i Sj_J. The estimate a^ is computed
under the "biggest" model contemplated, H}^. Precisely,

.,_ 1 f(yj-^Ojf
m-2r-2pie^(l - 0.)
where Oj is the MLE of dj under H,^.
The model with the smallest BIC can be thought of as the one with the best
fit, adjusting for the number of parameters. Also one can give a Bayesian inter-
pretation to the BIC (see Raftery, 1995). If one associates prior probabilities
TTi, 772, ... , TTjj to the modcls Hi, H2,... , H;,, respectively, then the posterior
probabilities, after incorporating the data, are
exp(—^BIC.ITT.
P(H.|data)= - , r=l,...,fe
Zi=o exp(-2BICi)7ri
In particular for a uniform prior (i.e., all models having same credibility a pri-
ori), the 7TS drop out of this expression, and then clearly the model with the
smallest BIC is the one with the largest posterior probability.
432 W.J. Reed

In implementing this procedure, there is the problem of specifying a maxi-

mum possible number of change points, k. While, in principle, one could set
k = m — 1, corresponding to different hazard of burning in each period, there
are difficulties with doing this. First, it would lead to a considerable computa-
tional load, since under a given model H^, the MLEs of the r change points are
found by direct search. Second, there is the problem of estimating the overdis-
persion parameter cr^, which is estimated under the largest model contemplated.
If k is set too large, there will be few degrees of freedom for estimating a^. In
the following examples, a maximum of fe = 6 change points was used. In nei-
ther example was the BIC minimum with either 5 or 6 change points and the
relative values of the BICs did not change much when k was reduced from 6 to
5, giving some comfort to the assumption that the models HQ, ... , Hg cover all
realistic possibilities.

Example 3. Kananaskis River Watershed

Johnson and Larsen (1991) present results of a fire history study of the 495 km^
area of the Kananaskis watershed on the eastern side of the southern Rocky
Mountains in Alberta, with a climate "transitional between plains and cordil-
leran types." Attempts by Johnson and Larsen to divide the map into spatial
subunits with distinct fire hazard rates were unsuccessful. The time-since-fire
distribution for the whole study area is displayed in Figure 3. There are m = 40
age classes of width T = 10 yr. The lower panel (a logarithmic plot of cumula-
tive frequency against time since fire) suggests a number of possible change
points (e.g., at 40, 60, 130, 230 and 280 years ago). Table 3 presents the MLEs
of change points for models HQ, ... , Hg, along with the BICs and posterior prob-
abilities assuming a uniform prior on HQ, ... , Hg. The overdispersion parameter
was estimated under Hg.
The only plausible models appear to be H2, H3, and H4, with H3 being by far

TABLE 3 Maximum Likelihood Estimates of Change Points in Various Models, Associated

BICs, and Posterior Probabilities of the Various Models, Assuming a priori That All Seven
Models Are Equally Probable (for Kananaskis Watershed time-since-fire data)

Posterior
Model MLEs of change points BIC probability

Ho -55.52 0.000
Hi 4 -149.06 0.004
H2 4,6 -156.35 0.171
H3 4, 6, 24 -159.39 0.780
H4 4,6,13,23 -153.35 0.038
H5 4, 6, 7, 13, 23 -149.42 0.005
tie 4 , 6 , 7 , 13, 19, 27 -136.71 0.000
Chapter 12 Statistical Inference for Historical Fire Frequency 433

TABLE 4 Maximum Likelihood Estimates and 95% LR Confidence

Intervals for the Fire Cycle in the Four Epochs between the Three
Estimated Change Points in Model H3 (for Kananaskis Watershed
time-since-fire data)

Fire cycle (yr)

Epoch
i Date MLE 95% Con. Int.

1 1940-1980 6409 969-715,000

2 1920-1940 49 34-73
3 1750-1920 136 101-189
4 pre-1750 48 30-85

the most plausible. Since the change point at 4 (1940) and 6 (1920) appear as
MLEs for all these models, one can conclude with a very high degree of cer-
tainty that there were indeed changes in fire frequency at around those times.
Furthermore, there is very strong support of an additional change at 23 (1750).
Of course by varying the prior probabilities of the various models, one can
change the posteriors. However, a substantial skewness in the prior is required
to shift the highest posterior probability from H3. Thus, one can conclude that
the data contain strong support for the three-change-points model, with the es-
timated changes occurring around 1940, 1920, and 1750.
The MLEs of the fire cycle (inverse of the hazard rate) and 95% likelihood
ratio confidence intervals (Reed et ah, 1998) in the four epochs separated by the
three identified change points are displayed in Table 4. The MLEs are also shown
as line segments superimposed on the semilog cumulative frequency plot in
Figure 5 (top panel). As one would expect, the confidence intervals for the fire
cycle in adjacent epochs do not overlap. The estimated hazard for the post-
1940 epoch is negligible. In fact, less than one tenth of one percent of the whole
study area burned in the 40-yr period (1940-1980).
Johnson and Larsen (1991) graphically identified a change point around
1730 and estimated the pre-1730 fire cycle at about 50 years, which agrees well
with the preceding results. However, they failed to identify the more recent
change points identified here.

Example 4. Glacier National Park

Figure 4 presents data obtained by Johnson et al. (1990) from stand-origin
maps for Glacier National Park (600 km^ of forested land) in the Rocky Moun-
tains of British Columbia. The vegetation is classified as being of the Interior
Wet Belt Forest type. The data were presented in 20-yr age classes (T = 20,
m = 21). Table 5 gives details of MLEs, BICs, and posterior probabilities under
a uniform prior for models HQ, Ui,... , U^.
434 W.J. Reed

- 1001 Kananaskis Watershed

8 50

10-1
5 1
13

o
100 200 300 400
Years since last fire
•g100- •""•"••"-*—*^, Glacier National Park
S 50-
CD
Q.
0
N. "
|io-
i 5-
3
O iJ1—1 , 1 X • 1
100 200 1
300 "n—'
400
Years since last fire
FIGURE 5 Epochs of constant hazard of burning as determined by the analysis in Section IV for
Kananskis Watershed and Glacier National Park. The points are the cumulative percentage areas
(on logarithmic scale) with time since fire exceeding the time (in years) on the horizontal axis. The
ranges of the line segments correspond to distinct epochs of constant hazard of burning; the slopes
of the segments correspond to the maximimum likelihood estimates of the hazards.

The data provide support only for models with four or more change points,
with the model H^ standing out with a very large posterior probability. To shift
the posterior mode from H^ requires a very substantial skewness in the prior
distribution. Consequently the four-change-point model seems by far the most
plausible, with estimated change points at 2, 5, 10, and 16 (i.e., around 1940,

TABLE 5 Maximum Likelihood Estimates of Change Points in Various Models, Associated

BICs, and Posterior Probabilities of the Various Models, Assuming a priori That All Seven
Models Are Equally Probable (for Glacier National Park time-since-fire data)

Posterior
Model MLEs of change points BIG probability

Ho _ 201.63 0.000
Hi 16 74.38 0.000
H2 12, 16 42.79 0.000
H3 5, 10, 16 28.08 0.000
H4 2,5, 10, 16 -9.27 0.797
H5 2,5, 10, 12, 16 -6.98 0.202
He 2,5, 10, 12, 16,18 1.76 0.002
Chapter 12 Statistical Inference for Historical Fire Frequency 435

TABLE 6 Maximum Likelihood Estimates and 95% LR Confidence Inter-

vals for the Fire Cycles in the Five Epochs between the Four Estimated
Change Points in Model H4 (for Glacier National Park time-since-fire data)

Epoch Fire cycle (yr)

i Date MLE 95% Con. Int.

1 1940-1980 1980 565-16700

2 1880-1940 156 40-181
3 1780-1880 1827 673-8102
4 1660-1780 151 106-224
5 pre-1660 25 17-42

1880, 1780, and 1660). There is some possibility of a fifth change point, esti-
mated at 1740. MLEs and 95% Ukelihood ratio confidence intervals for the fire
cycle (inverse of the hazard rate) in the five epochs separated by the four es-
timated change points under H4 are displayed in Table 6. The MLEs are also
shown as line segments superimposed on the semilog cumulative frequency
plot in Figure 5 (bottom panel).
It is worth noting that a very similar model was identified for these data us-
ing backward elimination methods (Reed, 1998), the only difference being
the change point in the 18 th century was estimated at 12 (1740) rather than
10 (1780).

REFERENCES
Heinselman, M. L. (1973). Fire in the virgin forests of the Boundary Waters Canoe Area. Quat. Res.
3, 329-382.
Johnson, E. A., and Larsen, C. P. S. (1991). Climatically induced change in fire frequency in the
southern Canadian Rockies. Ecology 72, 194-201.
Johnson, E. A., Fryer, G. I., and Heathcott, M.J. (1990). The influence of man on frequency of fire
in the interior wet belt forest, British Columbia. J. Ecol. 78, 403-412.
Larsen, C. P. S. (1996). Spatial and temporal variations in boreal forest fire frequency in northern
Alberta. J. Biogeog. 24, 663-673.
Reed, W.J. (1994). Estimating the historic probability of stand-replacement fire using the age-class
distribution of old-growth forest. For. Sci. 40, 104-119.
Reed, W.J. (1997). Estimating historical forest fire frequency from time-since-fire sample data. IMA
J. Appl Med. Biol. 14,71-83.
Reed, W J. (1998). Determining changes in historical forest fire frequency from a time-since-fire
map. J. Agri. Biol. Environ. Stat. 3, 430-450.
Reed, W J. (2000). Reconstructing the history of forest fire frequency—Identifying hazard rate
change points using the Bayes' Information Criterion. Can. J. Stat. 28, 353-365.
Reed, W J., Larsen C. P. S., Johnson, E. A. and MacDonald, G. M. (1998). Estimation of temporal
variations in fire frequency from time-since-fire map data. For. Sci. 44, 465-475.
This Page Intentionally Left Blank
 CHAPTER

Duff Consumption
K. MiYANISHI
Department of Geography, University of Guelph, Guelph, Ontario, Canada

I. Introduction
II. Characteristics of Duff
III. Empirical Studies of Duff Consumption
IV. Flaming Combustion
V. Smoldering Combustion and Pyrolysis
VI. Models of Smoldering Combustion
VII. Contribution of Smoldering Combustion Models
to Understanding of Duff Consumption
Notation
References

I. INTRODUCTION
Forest soils (e.g., spodosols) are often characterized by a distinct layer of par-
tially to well-decomposed organic matter lying above the A horizon of the min-
eral soil. This organic layer consists of three distinguishable horizons: the top L
(litter) horizon, composed of the undecomposed litter in which the materials
are readily discernible as leaves, twigs, etc.; the middle F (fermentation) hori-
zon, characterized by partially decomposed litter which is still recognizable as
to its origins; and the bottom H (humus) horizon of well-decomposed organic
matter in which the original structures are no longer discernible (Canada Soil
Survey Committee, 1978). The boundary between the F and H horizons can
sometimes be quite distinct due to a concentration of white fungal hyphae al-
though in some cases the two horizons are almost indistinguishable from each
other (Berg and Ekbohm, 1991). In the forestry and fire ecology literature, the
F and H horizons together are generally referred to as duff (e.g.. Van Wagner,
1972; Dyrness and Norum, 1983; Merrill and Alexander, 1987), whereas the
term forest floor generally includes all three horizons (i.e., the litter and duff)
(Merrill and Alexander, 1987), but see also Chapter 4, Section Il.A.

Forest Fires
Copyright © 2001 by Academic Press. All rights of reproduction in any form reserved. 437
438 K. Miyanishi

Consumption of litter and duff by fire has been of interest for some time to
foresters and ecologists (e.g., Chrosciewicz, 1959, 1967, 1974; Adams, 1966;
Ahlgren, 1970), particularly those working in conifer-dominated forest in cool
to cold climates (e.g., boreal and subalpine forest, Douglas fir forests of the Pa-
cific Northwest) for a number of different reasons. Both litter and duff, espe-
cially the upper F layer, dry out more rapidly than mineral soil (Van Wagner,
1987). As a result, it has long been observed that seedlings establishing on such
substrate are more subject to drought mortality than seedlings establishing on
mineral soil. Charron (1998) found that, in the boreal mixedwood forest in Sas-
katchewan, first-year tree seedling mortality was high enough on litter and the
F layer to virtually preclude successful seedling recruitment on these substrates.
Successful post-fire tree regeneration in these forests is therefore determined to
a large degree by the extent to which duff is removed by fire and suitable seed-
bed is made available (Chrosciewicz, 1974, 1976; Zasada et al, 1983; Thomas
and Wein, 1985; Weber et al, 1987). The extent to which duff is consumed by
fire would also influence understory plant mortality due to lethal heating (if
not combustion) of roots, rhizomes, and perennating buds of shrubs and her-
baceous perennials that are found within the duff (Flinn and Wein, 1977; Hun-
gerford et al, 1995; Schimmel and Granstrom 1996; see also Chapter 14 in this
book). Furthermore, as duff is consumed by fire, any dormant seeds that had
accumulated within the duff would also be consumed. Therefore, the extent of
duff consumption can influence the density and species composition of trees,
shrubs and herbs regenerating after a fire. Duff also acts as a reservoir of nutri-
ents for recycling (Weber, 1985; Weber et al, 1985); therefore, post-fire nutri-
ent cycling would be influenced by the presence or absence of duff. Finally, con-
sumption of duff by fire and subsequent mineral soil exposure can leave sites
(particularly on steep slopes) vulnerable to soil erosion.
This phenomenon of organic matter consumption in forest fires is not re-
stricted to mid- to high-latitude coniferous forests. Some humid tropical forests
develop root mats which are thick layers of litter impregnated with roots (Stark
and Jordan, 1978); these forests include high-altitude mountain forests and
tropical lowland forests growing on highly weathered acid soils (Kingsbury and
Kellman, 1997; Zech et al, 1997). Development of these root mats has been in-
terpreted as an adaptation for nutrient retention (Jordan, 1982; Jordan and Her-
rera, 1981) as well as a response to aluminum toxicity of the mineral soil (Kings-
bury and Kellman, 1997). Litter decomposition, mineralization, and most of
the plant uptake of the released mineral nutrients are concentrated in this or-
ganic root mat layer above the mineral soil (Stark and Jordan, 1978). The gen-
erally larger seeds of tropical tree species, resulting in relatively large seedlings
(Richards, 1952), and the abundance of precipitation in the humid tropics
suggest that removal of this organic layer may not be a major factor in seedling
survivorship as it is in the higher latitude conifer forests. Thus, Kellman and
Chapter 13 Duff Consumption 439

Meave (1997) found that neither germination nor early seedhng survivorship
was significantly higher in burned plots than in control plots within the tropi-
cal gallery forests in Belize. However, they did find higher seedling densities in
plots from which the litter layer had been manually removed and bare soil ex-
posed. Also, given the abundance and concentration of roots in these organic
layers above the mineral soil, consumption of this layer by fire would be ex-
pected to have a significant impact on plant mortality (both canopy and under-
story) as well as on nutrient cycling (Stark and Jordan, 1978). Although fire may
have been a relatively uncommon phenomenon in undisturbed humid tropical
forests (Uhl et al., 1988), it is becoming a more widespread disturbance in the
tropics due to changes in land use (Uhl and Kauffman, 1990; Kauffman, 1991).
Kauffman et al. (1995) noted rootmat consumption in forest clearance fires in
the Brazilian Amazon. These human-set fires as well as lightning-caused wild-
fires from savannas may spread to adjacent undisturbed forests, particularly dur-
ing unusually dry periods (Leighton and Wirawan, 1986). Extensive rootmat
consumption by wildfires has been observed recently in Venezuela (M. Kell-
man, personal communication) and in the cloud forests of Mexico (J. Meave,
personal communication).
The physical/chemical processes involved in the consumption of organic
soil horizons by fire are the same regardless of whether they are located in bo-
real, temperate or tropical forests. This chapter will explain these processes and
present some of the models proposed in the combustion literature. However, in
order to investigate combustion of any fuel, it is necessary to know something
about its chemical composition and physical characteristics.

II. CHARACTERISTICS OF DUFF

Duff is derived from plant litter (leaves, twigs, etc.) which is largely composed
of cellulose, hemicellulose, and lignin (Mason, 1976). Both cellulose and hemi-
cellulose decompose much more rapidly than lignin (Alexander, 1977; Berg and
Staaf, 1981; Berg et al., 1982); lignin has been described as "the most recalci-
trant among organic compounds in litter" (Aber and Melillo, 1991). Thus, the
chemical composition of duff shows a gradient with depth of decreasing propor-
tions of cellulose/hemicellulose and increasing proportions of lignin (Berg et al.,
1982). In other words, the F layer would have a greater proportion of cellulose/
hemicellulose and a smaller proportion of lignin than the H layer. Unlike cellu-
lose, lignin is a complex phenolic macromolecule whose units are not linked in
an organized repeating manner; thus the precise structure of lignin is not known,
and each lignin molecule may be different (Taiz and Zeiger, 1991). Further-
more, within the plant tissues, lignin is covalently bound to cellulose and other
polysaccharides. Although thermal degradation (pyrolysis) and combustion
440 K. Miyanishi

chemistry of cellulose have been well studied (e.g., Lewellen et al, 1976; Moussa
et al, 1976; Bradbury et al, 1979; Shafizadeh and Sekiguchi, 1984; Ohlemiller,
1985; 1990a; 1990b), there is much less known about lignin combustion.
Plant litter also contains compounds such as cutin, suberin, and associated
waxes plus a host of secondary compounds, generally grouped as terpenes/
terpenoids, phenolic compounds and alkaloids. These secondary compounds
are generally specific to a plant species or a related group of species (Taiz and
Zeiger, 1991). Thus, the litter and the resulting duff layer would consist of a
mixture of compounds that varies spatially both within and between sites, de-
pending on the species composition of the plant community (Berg and Ekbohm,
1991). This heterogeneity in the chemical composition of duff could present
challenges in describing its combustion chemistry and attempting to develop
general models of its combustion.
The F and H layers also differ in physical characteristics. The H layer is com-
posed of smaller particles and has a higher bulk density than the F layer (Clay-
ton et al, 1977). These differences would influence their thermophysical prop-
erties such as conductivity and diffusivity and thus play a role in the rates of
combustion propagation in the two layers.
Although very great depths of organic matter can accumulate in waterlogged
sites, often creating large peat deposits in mid- to high-latitude wetlands, the
thickness of the duff layer in better-drained upland boreal forest generally varies
from 5 to 30 cm (Chrosciewicz, 1974, 1989; Dyrness and Norum, 1983; Schim-
mel and Granstrom, 1996). Green et al (1993) similarly described F + H hori-
zons in various types of coniferous forests on the west coast of North America
ranging in depth from 4 to 25 cm and in central Siberia ranging from 8 to
10 cm. Organic root mats in tropical forests have been reported ranging from
0.2 to 10.6 cm (Kingsbury and Kellman, 1997) and from 15 to 40 cm (Jordan
and Herrera, 1981). Thus, it would seem that the organic layer in non-wetland
forests over a wide range of climatic conditions generally ranges from a few cen-
timeters to 3 0 - 4 0 cm in thickness. The thickness of organic matter accumula-
tion in any particular site depends primarily on the decomposition rate which
is influenced by temperature and moisture as well as litter quality characteris-
tics such as pH, base content, and presence of polyphenols/ tannins (Williams
and Gray, 1974; Mason, 1976; Delaney et al, 1996; Krause, 1998). Between-site
differences in microclimate and hydrology and between species differences in
litter characteristics thus result in differences in duff thickness between stand
types within the same region (Potts et al, 1983; Krause, 1998). For example, in
boreal upland sites, Picea mariana stands tend to have thicker duff with distinct
F and H layers, whereas Pinus banksiana stands have thinner duff often with
an indistinct H layer (Van Wagner, 1972; Dyrness and Norum, 1983; Miyanishi
et al 1999). Also, duff depth depends on time since disturbance (primarily fire
or logging). The high-intensity wildfires typical of the boreal forest generally
Chapter 13 DufF Consumption 441

result in some consumption of duff (Wein, 1983;Johnson, 1992). Krause (1998)

also reported a significant decrease in duff following logging of a Pinus bank-
siana stand, with minimum levels occurring 6 - 9 yr after cutting. This post-
logging decrease was attributed to the decreased litter input as well as increased
decomposition rate due to higher ground temperatures. In the absence of dis-
turbance, duff depths do not continue to accumulate indefinitely; eventually an
equilibrium is reached between the rates of litterfall and decomposition. Krause
(1998) found that duff accumulation leveled off after 12 years and that the
amounts of duff in 12 to 16-year-old stands of Pinus hanksiana in New Bruns-
wick were not different from those found by Weber et al. (1985) in a 53-year-
old P. hanksiana stand in eastern Ontario. Also, Perala and Alban (1982) re-
ported no significant differences in mean duff depths in stands ranging from 40
to 250 years old. Thus, duff depths in the boreal forest appear to reach an equi-
librium within the expected interval between fires (Johnson, 1992).

III. EMPIRICAL STUDIES

OF DUFF CONSUMPTION
Most studies of duff consumption have focused on measuring and attempting
to predict mean values of duff consumption, either in terms of a mean decrease
in depth or a mean loss of duff mass (e.g.. Blackball and Auclair, 1982; Dyrness
andNorum, 1983; Little etal., 1986; Weber etal., 1987; Brown etaL, 1985,1991;
Reinhardt et al, 1989). However, duff consumption within burned forest sites
is extremely variable; as Dyrness and Norum (1983) observed after a fire in
Alaska, "[t]he sites give the impression of many deeply burned holes scattered
among broad areas of unburned or scarcely burned forest floor" (Figure 1).
Similar observations have been noted by others (e.g., Zasada et ah, 1983). This
patchiness in duff consumption appears to be more pronounced as one goes
further north in the boreal forest. In terms of regeneration, the important factor
is obviously not so much the overall or mean decrease in duff depth but the area
of suitable seedbed exposed (Chrosciewicz, 1974, 1976). And since successful
seedling recruitment is largely confined to the burned patches or holes in the
duff (Charron, 1998), the total area and the spatial patterns of duff burnout
would play a significant role in both the density and spatial distribution of tree
seedlings in the postfire regenerating stands. These patterns of burned out duff
also play a role in the spatial patterns of postfire understory species distribu-
tion and composition due to mortality of vegetatively reproducing species that
are rooted in the duff as well as in the mineral soil beneath the duff (Flinn and
Wein, 1977; Steward et al, 1990; Schimmel and Granstrom, 1996).
To attempt to understand the areal extent and spatial patterns of duff con-
sumption both within and between stands and to determine the factors that
442 K. Miyanishi

FIGURE 1 Photograph of the forest floor following a fire in a boreal forest site in central Sas-
katchewan. The burned holes in the duff are clearly defined from the remaining unburned duff.

influence or determine these, it is necessary to understand the processes by

which duff is consumed during forest fires. Fuels are consumed by either flam-
ing combustion or smoldering combustion. Flaming combustion is the rapid
oxidation of volatiles producing considerable visible radiation (see Chapter 2
in this book for a more detailed discussion). The volatiles in forest fires include
terpenes, fats, oils and waxes; these are low-molecular-weight components of
fuels that are readily volatilized at low temperatures. More importantly, the
flames consume the gaseous pyrolysis products of the principal polymeric con-
stituents of the woody fuels. Smoldering combustion is a nonflaming "slow
combustion process in a porous medium in which heat is released by oxidation
of the solid" (Schult etal, 1995); it is characterized by thermal degradation and
charring of the fuel (Ortiz-Molina et al, 1978). Both forms of combustion in-
volve the coupling of heat generation and heat transfer processes (Ohlemiller,
1985); however, the heat-generating process in flaming combustion is gas-phase
oxidation, but it is solid-phase oxidation in smoldering combustion.
Many of the empirical studies on duff consumption make little or no men-
tion of which combustion process is assumed to be operating. The approach in
these studies is to produce strictly empirical regression models that attempt to
predict duff consumption (decrease in depth or mass) from an assortment of
Chapter 13 Duff Consumption 443

measurable variables such as preburn duff depth and various measures and in-
dices of duff moisture (e.g., Chrosciewicz, 1978a, 1978b; Sandberg, 1980; Black-
hall and Auclair, 1982; Brown et al, 1985; Little et al, 1986). The major prob-
lem with this approach is that selection of the independent variables is rarely,
if ever, justified on the basis of any combustion process by which duff is con-
sumed. Even though many of these studies obtain statistically significant regres-
sions, their empirical models are specific to the site, species, and weather con-
ditions under which their data were collected. For example, Little et al. (1986)
produced different predictive equations for slash burns within Pacific-North-
west Douglas fir clear-cuts depending on whether the number of days since rain
was greater or less than 25 days and also on the presence or absence of a dis-
tinct dry layer of duff. Not only do the coefficients differ between these regres-
sion models but the predictive variables included in the models also differ. Thus,
these studies and models provide little basis for improving our understanding
of how duff is consumed or for developing more generalizable models of duff
consumption. Furthermore, the lack of any process-based selection of variables
in regression models can sometimes lead to trivial results and conclusions. For
example, the best fit regression model obtained by Reinhardt et al. (1989) had
only one independent variable, preburn duff depth. Since duff consumption
was fairly complete in their experimental burns, they concluded that "[t]he
amount of duff consumed depended mostly on how much was available for
consumption."
Since duff consumption can potentially occur by both flaming and smolder-
ing combustion, both processes will be addressed with respect to their role in
duff consumption in wildfires.

IV. FLAMING COMBUSTION

The flaming front of a forest fire involves the flaming combustion of fine fuels
(Johnson, 1992); flaming combustion is also discussed in Chapters 2, 3, and 5
in this book. Flaming combustion occurs when pyrolytic decomposition of the
fuel supplies sufficient volatiles to fuel the flame. Lewellen et al. (1976) found
that, if cellulose is heated rapidly and the initial products of pyrolysis do not
have a chance to react and polymerize (such as if the fuel is a very thin layer of
cellulose), it is completely volatilized. Thus, such rapid heating of cellulose
would be expected to result in flaming combustion. However, when cellulose
is heated slowly, the slow release of volatiles allows them to remain within the
fuel matrix for an extended period; repolymerization of the volatiles occurs,
producing a solid substance called char. While there may be some initial flam-
ing under such circumstances, the rate of production of volatiles is insufficient
to sustain flaming combustion and smoldering of the char residual occurs.
444 K. Miyanishi

While thin layers of duff or the upper layers of deep duff, if very dry, can be
consumed in the flaming front, deeper layers of more decomposed, more com-
pacted duff are unlikely to undergo self-sustained flaming combustion in a wild-
fire due to their chemical and physical properties. Lignin is less readily vola-
tilized and has a substantially higher char yield than cellulose when heated
(Rothermel, 1976; Shafizadeh and DeGroot, 1976; Shafizadeh and Sekiguchi,
1984). The higher lignin:cellulose ratio of duff compared with undecomposed
litter would mean a slower rate of release of volatiles. Therefore, heating of
deep duff layers in a forest fire is unlikely to produce a sufficient concentration
of volatiles to sustain flaming combustion. If flames from other burning objects
are impinging on it, the duff can flame initially; however, it will not retain these
flames once the external flames vanish.
The degree of volatilization is also influenced by the rate of heating. The rate
of heating of duff in a forest fire is influenced by its moisture content. Duff dries
out more slowly and therefore generally contains more moisture than litter (ex-
cept shortly after a light rain that occurs following a long dry period). Thus,
often when the litter has dried sufficiently to allow propagation of flaming com-
bustion, the duff may still be quite moist. The latent heat of vaporization re-
quired to drive off the moisture in the duff would act as a large heat sink, slow-
ing the rate of its heating by the flaming front.
Finally, according to Rothermel (1972), the rate of heat release of spreading
fire per unit area decreases to a very low value as the packing ratio of the fuel ap-
proaches 10%. Packing ratio is the ratio of the fuel volume to the total volume of
the fuel bed. Since duff has packing ratios greater than 10% (Frandsen, 1991a),
the rate of heat generation by its oxidation is generally insufficient to sustain
flaming combustion. Thus, as noted in Chapter 6 in this book, combustion of
such "thermally thick" fuels has little, if any, impact on the behavior of the ad-
vancing firefront.
However, while the duff's contribution of fuel and heat to the flaming front
may be relatively insignificant, heat from the flaming front can be the major fac-
tor in the smoldering combustion of duff (Van Wagner, 1972), particularly in
transient (non-self-sustaining or dependent) smoldering that assumes that the
duff is both ignited and consumed by heat transferred downward from the flam-
ing front (Hawkes, 1993).
Van Wagner (1972) presented a heat budget-based model of duff consump-
tion in upland pine stands which assumed such dependent smoldering. Before
the temperature of the duff can be raised to its ignition temperature, heat from
the fire must first drive off any moisture in the duff. Thus, the model assumes
that the processes of heat transfer from the flaming front to the duff, evapora-
tion of any moisture in the duff, and heating of the duff to the ignition point de-
termine the amount of duff consumed.
Since the model assumes that the heat supply to the duff is derived only from
Chapter 13 Duff Consumption 445

downward radiation from the flaming front, this heat flux can be obtained us-
ing the Stefan-Bokzmann law:
qji = EjiaT} (1)
where qji is the flux of energy radiated by the flame (W cm"^), EJI is the emis-
sivity of the flame, a is the Stefan-Bokzmann constant (5.669 X 10"^^
W cm"^ K""^), and T^ is the temperature (Kelvin) of the flame. Emissivity is the
ratio of the radiative flux emitted by a body to that emitted by a black body at
the same temperature, a black body being one that radiates at aT^. Thus, the
emissivity of virtually all heat sources would be some value less than one.
Since the duff is also radiating energy to the flame
^d = ^dO-Td (2)

the net interchange of radiative energy from the flame to the duff (q) is

^ = ^d^lji - (^ji^d (3)

where a^ is the absorptivity of the duff and Uji is the absorptivity of the flame.
Absorptivity is the proportion of incident radiant flux that is absorbed by a body;
a black body is a perfect absorber of radiation and has an absorptivity of one.
Substituting Eqs. (1) and (2) for qji and q^, Eq. (3) becomes
q = a^SjaTJ - ajiS^rJl (4)
By Kirchhoff's law, s^ = a^ and s^ = a^; therefore, Eq. (4) can be rewritten as
q = a^sjiam - T^) (5)
Van Wagner assumed that duff absorptivity (fl^) was 1, duff ambient tempera-
ture (Td) was 20°C (293 K), and flame temperature (Tji) was 800°C (1073 K).
Substituting these values and, by convention, putting the 10~^^ of the Stefan-
Bokzmann constant into the denominator of the temperature terms, Eq. (5)
becomes
1073 Y _ ( 293 Y 7.473 sji (6)
5.669 Eji
1000/ Vioooy
For the duff to burn, the temperature of the duff must be raised to its igni-
tion temperature. The quantity of heat required for this, H (measured in J g~^)
is given by
H = Mc^(1000 - 20) + IM + Q ( 3 0 0 - 20) + 50.23 = 2585M + 460
(7)
The first term represents the energy required to heat the moisture in the duff
from its initial temperature (20°C) to boiling point (100°C) where M is the
moisture content (% dry w t ) and c^ is the specific heat of water (4.18 J g"^).
446 K. Miyanishi

The second term is the energy required to vaporize this water where I is the la-
tent heat of vaporization (2.25 X 10^ J g~^). The third term gives the energy re-
quired to heat the duff from 20°C to its ignition temperature of 300°C, where
Q is the specific heat of duff (1.47 J g"^ °C"^). The final term is the heat of de-
sorption of bound water in the duff (50.23 J g"^).
If the net energy from the flame given by Eq. (6) is equal to the energy re-
quired to burn the duff given by Eq. (7), the resulting energy balance can be
written as

DVH
q = (8)
w
where D is the mass of duff consumed (g cm~^), Vis the rate of advance (cm s"^)
of the flaming front, and w is the horizontal thickness or width of the flaming
front (cm). Solving for D gives
qw
D = ^^ (9)
VH

Since w/V = r, the residence time (s) of the flaming front, Eq. (9) can be re-
written as
rq
D=-^ (10)
H
Although residence time (r) varies with the fire behavior, for his experimental
burns of pine stands Van Wagner assumed a value of 60 s. Substituting this
value and Eqs. (6) and (7) into Eq. (10) gives
7A73Ea • 60
D= (11)
2585M + 460
In order to be able to predict duff consumption (D) from duff moisture con-
tent (M), it is necessary to obtain an expression for e^. In lieu of determining Sji
from actual measurements of downward heat flux, Van Wagner used measured
values of D and M to obtain estimates of Sji and then plotted these estimates of
Sji against M, obtaining a straight line relationship:

sji = 0.770 - 0.00543M (12)

Einally, substituting Eq. (12) for e^ in Eq. (11) gives

7.473(0.770 - 0.00543M) • 60 345.253 - 2.435M

D = ^ = (13)
2585M + 460 2585M + 460
The model showed a reasonable fit to data from 12 experimental fires, 6 in Fi-
nns resinosa and P. strohus stands and 6 in P. banksiana stands (Figure 2). How-
Chapter 13 DufF Consumption 447

3.0-

2.5-
^^0.941(141.8-m) .
17.74+ m
^ 2.0

§ 1.5
C/)
c
o A Jack pine
o
3= 1.0 • Red and white pine
a
0.5

0.0 1 \ \ \ 1 \—^-*^ 1 • I
0 20 40 60 80 100 120 140 160 180
Moisture content (%)
FIGURE 2 Empirical results and model prediction of duff consumption (kg m~^) as a function
of percent moisture content. From Van Wagner (1972).

ever, as Van Wagner noted, this fit of tfie empirical data to tlie model is not a
test of model validity since the estimate of Sji was obtained empirically from
measured values of D and M and then this estimate was used to obtain the final
equation for determining D from M.
One limitation of the model is its prediction of 7.55 kg m~^ as the maximum
amount of duff that could be consumed (even at 0% duff moisture which would
almost never occur in most forests even under the driest weather conditions).
This value converts to a maximum depth of duff that could be consumed of
10.6 cm, assuming a duff bulk density of 0.071 g m~^ (the standard Duff Mois-
ture Code layer in the Canadian Fire Weather Behaviour System is 7 cm deep
and 5 kg m~^ in weight). This maximum depth exceeds the duff depths found
in the Pinus stands of the Great Lakes-St. Lawrence forest for which the model
was developed and, therefore, may not be unreasonable. However, without some
modification, the model would obviously not be generalizable to boreal forest
stand types with deeper duff depths, such as Picea mariana stands. This limit
also suggests that heat transfer from the flaming front may not be the main pro-
cess driving duff consumption observed in stands with deep duff layers.
Despite some of the questionable assumptions and limitations, Van Wagner's
(1972) model is notable for its process approach to modeling duff consumption,
in contrast to almost all of the previous and subsequent studies in the forestry
literature that have taken a strictly empirical approach.
Hawkes (1993) noted that this model did not take into account any heat gen-
erated by the combustion of the duff itselL Therefore, he modified Van Wagner's
448 K. Miyanishi

model by adding heat transfer from the combustion zone in the duff. Based on
six temperature histories of smoldering in peat, he assumed an average tem-
perature of 400°C for the duff combustion zone and used the Stefan-Boltzmann
law to estimate this second heat source. As in the Van Wagner model, Hawkes'
model assumes that all heat transfer is by radiation and does not consider con-
duction or convection. In testing this revised model in the lab using a propane
burner as the heat source and peat as the fuel, Hawkes obtained a reasonably
good fit between the predicted and measured depths of burn (converted from
the weight of duff consumed). However, because he was interested only in de-
pendent smoldering (such as might occur in slash burns), he limited the mois-
ture levels of his samples to those that would not allow self-sustaining smolder-
ing combustion (i.e., a minimum duff moisture content of 220%). Van Wagner's
(1972) model predicted no duff consumption when the moisture content is
greater than 134% because beyond that moisture level, the surface litter fuels
would not carry a burn. Thus, the usefulness of Hawkes' model to more realis-
tic conditions in fires (whether natural or prescription set) may be questionable.
The problem was that at lower fuel moistures, self-sustained smoldering oc-
curred in the deep fuel beds. This suggests that in most forest fires duff would
not be consumed only during passage of the flaming front. In fact, studies have
found that duff consumption more typically occurs by self-sustained smolder-
ing combustion after passage of the flaming front (e.g., Wein, 1981; 1983; Hun-
gerford et al, 1995). This may also partially explain why studies of experimen-
tal burns have not been able to find a strong relationship between fire intensity
and duff consumption (e.g., Alexander, 1982).
Since the physical and chemical characteristics of duff, in fact, make smol-
dering combustion more likely than flaming combustion (as discussed earlier),
and since smoldering is now generally recognized as the major process respon-
sible for duff consumption in forest fires (Frandsen, 1991b; Johnson, 1992;
Hungerford et al, 1995), the next section will explain the process of smolder-
ing and discuss some of the models that have been proposed for this process.

V. SMOLDERING COMBUSTION
AND PYROLYSIS
Smoldering has been defined as "a self-sustaining, low temperature combustion
process involving pyrolysis of the substrate ahead of a solid-phase combustion
front" (Shafizadeh et al, 1982). It is characterized by thermal degradation and
charring of the solid material with evolution of smoke (Moussa et al, 1976). Be-
cause it sometimes involves emission of visible glow, smoldering has also been
referred to by some as glowing combustion (Williams, 1977; Johnson, 1992).
Chapter 13 Duff Consumption 449

However, Drysdale (1985) makes a distinction between glowing combustion

and smoldering and, although glowing combustion "is associated with the sur-
face oxidation of carbonaceous materials," it differs from smoldering in that
"thermal degradation of the parent fuel does not occur, nor is it required." On
the other hand, if glowing combustion specifically refers only to the process of
surface oxidation of a solid, it may be viewed as the final stage of the smolder-
ing process (Simmons, 1995). Only porous materials which form a solid carbo-
naceous char upon thermal degradation are capable of self-sustained smolder-
ing combustion (Drysdale, 1985). Duff as well as organic peat soils are capable
of such sustained smoldering (Frandsen, 1987, 1991a,b; Hawkes, 1993; Hun-
gerford et al., 1995).
A large number of studies have been conducted on smoldering in various
materials such as dust/sawdust (Cohen and Luft, 1955; Palmer, 1957), ciga-
rettes (Egerton et al, 1963; Muramatsu et al, 1979; Yi and Kim, 1996), rolled
cardboard/paper and incense sticks (Kinbara et al, 1967; Sato and Sega, 1985),
polyurethane foam (Ohlemiller et al, 1979) and cellulose (Lewellen et al, 1976;
Moussa et al, 1976; Ohlemiller, 1985, 1990a,b). Few, if any, studies appear to
have been conducted on smoldering in duff, although one study was conducted
on packed beds of plant litter (Jones et al, 1994). Therefore, any understand-
ing of smoldering combustion of duff must come from empirical studies and
theoretical models of smoldering in other materials that have been better stud-
ied. A number of lab studies on smoldering have used peat as a substitute for
forest duff with the justification that peat has similar particle sizes and bulk den-
sities as duff (Frandsen, 1987, 1991a,c; Hartford, 1989; Hawkes, 1993). Dried
commercial peat generally consists of 95% organic matter (Hawkes, 1993). Us-
ing Curie-point pyrolysis-mass spectrometry, Bracewell and Robertson (1987)
showed that the chemical composition of peat was very similar to that of the
F duff layer. Unfortunately, most of the literature on duff consumption (cited
in an earlier section of this chapter) as well as the studies of smoldering in peat
mentioned here indicate little awareness of the numerous studies of the smol-
dering process published in the combustion literature. What follows is an in-
troduction to this literature and is not meant to be a comprehensive review.
Smoldering differs from flaming combustion in its less complete oxidation
of the fuel, its lower temperature, and its slower rate of propagation (Ohlemil-
ler, 1985). Ohlemiller noted that, despite wide variation in fuel type and fuel
configuration, smoldering velocities do not vary greatly, suggesting that the pro-
cess is limited by the oxygen supply rate rather than by the fuel oxidation ki-
netics. Moussa et al (1976) and Williams (1977) had also concluded that smol-
dering velocities were generally governed by the rate of oxygen diffusion rather
than by reaction kinetics. If the rate of oxygen supply is increased sufficiently,
smoldering can undergo a transition to flaming combustion (e.g., the use of bel-
lows to revive a dying fire).
450 K. Miyanishi

As with flaming combustion, smoldering occurs through the couphng of heat

release and heat transfer mechanisms. Smoldering is initiated by heating of a
porous solid such as cellulosic polymers, resulting in its decomposition and
yielding char and volatiles. Char is "the black solid residue of a variable aro-
matic nature remaining after the end of the initial rapid weight loss upon heat-
ing of a polymer" (Ohlemiller, 1985). Pyrolytic degradation of polymers such
as cellulose is exothermic if oxygen is available or endothermic in the absence
of oxygen; both processes produce char which then reacts with oxygen in an
exothermic reaction. Sustained propagation of smoldering depends on suffi-
cient heat from the exothermic reactions being transferred ahead of the smol-
dering zone to result in pyrolysis of the virgin fuel and exposure of fresh char
which can then oxidize (Drysdale, 1985). Thus, pyrolysis and char production
are essential steps in smoldering combustion.
Pyrolysis temperatures for the three main constituents of duff differ: hemi-
cellulose 2 0 0 - 2 6 0 ° C , cellulose 2 4 0 - 3 5 0 ° C , and lignin 2 8 0 - 5 0 0 ° C (Roberts,
1970). Lignin requires higher temperatures and yields less volatiles and more
char than cellulose (Shafizadeh and DeGroot, 1976; Drysdale, 1985). The py-
rolysis of cellulosic materials has been extensively studied (e.g., Broido, 1976;
Lewellen et al, 1976; Shafizadeh and DeGroot, 1976; Bradbury et al, 1979;
Shafizadeh et al, 1982; Shafizadeh and Sekiguchi, 1984); therefore, most of the
following section on pyrolysis will come from studies of cellulose pyrolysis. Py-
rolysis is also discussed in Chapters 2 and 3 in this book.
Low temperature pyrolysis of cellulose produces a large variety of end prod-
ucts including CO2, CO, CH4, C2H4, H2, H2O, low-molecular-weight organic
acids, and levoglucosan (Shafizadeh and DeGroot, 1976). Lewellen et al (1976)
proposed that, with decreased heating rate, these products of pyrolysis have a
longer residence time within the pyrolyzing matrix. This extended period of
contact between the primary products allows their repolymerization, cracking
or cross-linking to produce char. Thus, Lewellen et al (1976) and Bradbury et al
(1979) found that char yields decreased with increased heating rate (e.g., from
20% of original cellulose weight at ~ 3 0 0 ° C s~^ to 2% at ~1500°C s"^). With
an appropriate heating rate, thin samples of cellulose can be completely vola-
tilized with no production of residual char. Besides rate of heating, char forma-
tion is also a function of particle size and the presence of moisture and inor-
ganics; larger particle size and the presence of moisture and inorganics increase
the yield of char (Shafizadeh and DeGroot, 1976). Thus, the relatively large
particle sizes in duff, the presence of duff moisture, and the relatively slow heat-
ing rate in a forest fire would be expected to result in high char yields from duff.
Although studies of pyrolysis of cellulosic materials indicate the extreme
complexity of the process, Bradbury et al (1979) found that the pyrolytic gasi-
fication of cellulose could be described by either a two- or three-step reaction
scheme, depending on the temperature range (Figure 3). At higher tempera-
Chapter 13 Duff Consumption 451
a) Two-step volatlles

cellulose
W ^^ K

char + gases

b) Three-step
K^ ^ volatlles
k ^x-^^
cellulose ^ L ^ active cellulose i.
^ char + gases

FIGURE 3 (a) Two-step and (b) three-step reaction schemes for pyrolytic gasification of cellulose
where W refers to the normalized weights of the respective materials—cellulose (cell), activated
cellulose (A), volatlles (v), char (c), and gases (g)—and k refers to the rate constants. From Brad-
bury et a\. (1979). A kinetic model for pyrolysis of cellulose. J. A^ipl. Polm. Sci. 23, 3271-3280.
Copyright © 1979 John Wiley & Sons, Inc. Reprinted by permission of John Wiley & Sons, Inc.

tures, pyrolysis of cellulose can be modeled as two competitive first-order re

actions (Figure 3a) with the following rate equations:

= {K + K)[w,cell} (14)
dt
dW,
035k,[W, cell] (15)

where W^^n, W^, W^, and Wg are the normalized weights of cellulose, volatlles,
char, and gases respectively (i.e., W^^n + W^ + W^ + Wg = 1). The rate con-
stants, k^ and fee can be expressed by the following Arrhenius relationships:
47,300
1.9 X lO^^exp mm (16)
RT
36,600 \
K = 7.9 X lO^'exp - (17)
RT J m m
where R is the universal gas constant and T is temperature. The numerical val-
ues are the preexponential factor and activation energy for each reaction and
would be specific to the reaction. Temperature (T) at time t is given by

T = Tf-{Tf- To) e x p ( - 2 . 9 2 0 (18)

where Tj is the final pyrolysis temperature and TQ is the initial temperature.

452 K, Miyanishi

For lower temperatures where an initial period of accelerating weight loss is

observed, Bradbury et ah (1979) proposed a model involving three reactions
(Figure 3b); first, an active cellulose is formed by an initiation reaction; then the
active cellulose decomposes by two competitive first-order reactions, the first
yielding volatiles and the second yielding char and a gaseous fraction. The rate
equations for this model are

dt
djWj,)
= H'^ceii] - (K + K)[WA] (20)
dt

dm = 035K[Wj,] (21)
dt
where W^ is the normalized weight of active cellulose.
The rate constant k^ can be expressed by the following Arrhenius relationship:
/ 58,000 \ . , ,
ki = 1.7 X 10^^ exp( 1 min~^ (22)

Shafizadeh and Sekiguchi (1984) also found that the chemical composition
and combustion behavior of cellulosic chars varied with the temperature at
which the fuel was heated. Chars formed at lower temperatures had higher con-
centrations of aliphatic carbons, whereas chars formed at higher temperatures
had higher concentrations of aromatic carbons. Aliphatic carbons are more re-
active and burn at lower temperatures than the more stable aromatic carbons.
There is some empirical evidence for distinct endothermic pyrolysis and exo-
thermic oxidation zones in smoldering (Muramatsu et at., 1979) which have
been incorporated into some smoldering models (e.g., Moussa et al, 1976;
Leisch, 1983; Peter, 1992). Separation of pyrolysis and oxidation zones depends
on the relative movement of the oxygen supply gas and solid reaction zones.
However, Ohlemiller (1985) also cited evidence for both endothermic nonoxi-
dative pyrolysis and exothermic oxidative pyrolysis and proposed that the two
processes compete for the virgin fuel (Figure 4). Ohlemiller concluded that "the
available evidence points to the likelihood of a significant role for oxidative py-
rolysis as an appreciable heat source in many [but not all] smoldering pro-
cesses." Thus, Kashiwagi and Nambu (1992) and Di Blasi (1995) separated the
degradation of cellulosic fuel to char into an endothermic nonoxidative reac-
tion and an exothermic oxidation reaction.
Following pyrolysis of the virgin fuel and the production of char, the next
major process involved in smoldering is the exothermic oxidation of the char,
producing ash and gases. This process is the primary heat source for the propa-
Chapter 13 DufF Consumption 453

Char' + Gases'

exo.
Ash + Gases"
exo.

Char + Gases
FIGURE 4 Scheme of thermal behavior in smoldering showing the two competing processes by
which fuel is pyrolyzed—endothermic pyrolysis in the absence of oxygen and exothermic pyroly-
sis in the presence of oxygen. Subsequent oxidation of char is exothermic. Reprinted from Prog.
Energy Combust. Scl, Ohlemiller, T. J., Modeling of smoldering combustion propagation, pp. 277-
310, 1985, with permission from Elsevier Science.

gation of smoldering; therefore, an understanding of smoldering propagation

requires an understanding of the factors determining the rate pf char oxidation.
To identify the mechanisms controlling the propagation of smoldering, Ohle-
miller (1990a) looked at the structure (spatial distribution of temperature, oxy-
gen mole fraction, and residual organic mass fraction) of the smoldering wave
as it moved laterally through a thick (18-cm) permeable horizontal layer of cel-
lulose insulation; this is an inherently two-dimensional propagation process.
Smoldering was initiated on one end face of the fuel layer, and the movement
of the smoldering zone was followed by three sets of thermocouples embedded
at two depths and at three positions along the fuel as well as by a movable oxy-
gen probe. The distribution of the residual organic mass fraction was obtained
by rapidly extinguishing smoldering by smothering with nitrogen and then sec-
tioning the fuel layer. Figure 5 shows the profiles of temperature, oxygen mole
fraction and residual organic mass fraction. The length of the smoldering wave
was found to be about two to three times greater than the fuel depth (which was
18 cm). Since Palmer (1957) found a comparable ratio of wavelength to fuel
thickness for sawdust layers only a few centimeters deep, Ohlemiller (1990a)
concluded that similar factors, mainly the rate of oxygen supply, control the
rate of smoldering over this range of fuel thicknesses. Figure 6 shows the oxy-
gen inflow paths and heat release zones in a smoldering wave that Ohlemiller
inferred from his analysis of the wave structure. He presented a further argu-
ment that the shape of the smoldering wave is determined by the rate of oxy-
gen diffusion from above. Leisch (1983) also showed that combustion wave
velocity generally decreased with depth in smoldering beds of grain dust and
wood dust and concluded that the velocity of smoldering was determined by
the net effect of heat generation (determined by oxygen diffusion rate from the
surface) and heat loss from the surface.
A distinction is made in the smoldering literature between forward and re-
verse smoldering, the two limiting cases of one-dimensional smoldering propa-
gation. The reactants in forward (or co-flow) smoldering enter the reaction zone
454 K. Miyanishi

Top surface profile

Isotherms (°C)

Profiles of Constant
Oxygen Mole Fraction

Top surface profile Dashed lines are profiles

of constant fraction of
Min.=0.11 initial organic material
=0.05

Relative distance (cm)

FIGURE 5 Steady state profiles of temperature, oxygen mole fraction, and fraction of remaining
organic material superimposed on a cross-sectional view of the smolder wave. Reprinted by permis-
sion of Elsevier Science from Smoldering combustion propagation through a permeable horizontal
fuel layer, Ohlemiller, T. J., Combustion & Flame 81, 341-353, Copyright 1990 by The Combustion
Institute.

from opposite sides, somewhat analogous to diffusion flames (Figure 7). The
reactants in reverse (or opposed flow) smoldering enter the reaction zone from
the same side, somewhat analogous to a premixed flame (Buckmaster and Lo-
zinski, 1996). The horizontal propagation process (Figure 6) has a mixed char-
acter. Smoldering studies also distinguish between situations of free flow or
forced flow of air/oxygen to the fuel. Smoldering of duff following a forest fire
would thus be typically categorized as a free flow forward smoldering process,
except near the outward spreading front where it approaches a reverse smol-
dering character.
Chapter 13 Duff Consumption 455

Buoyant plume

Oxygen
diffusion

Direction of
smolder propagation
FIGURE 6 Oxygen inflow paths and heat release zones in a spreading smolder wave. Reprinted
by permission of Elsevier Science from Smoldering combustion propagation through a permeable
horizontal fuel layer, Ohlemiller, T. J., Combustion & Flame 81, 341-353, Copyright 1990 by The
Combustion Institute.

In summary, the major heat sources involved in smoldering combustion are

the processes of oxidative pyrolysis (polymer degradation) and char oxidation,
whereas the major heat sinks are the processes of nonoxidative pyrolysis and
water movement (Ohlemiller, 1985). Any modeling of smoldering combustion
would thus require rate expressions for these reactions as well as for the trans-
fer of heat between the sources and sinks.

x=0

M. M.
flux of solid gas flow

unburned solid and air and char

oxygen free gas

oxidation zone

FIGURE 7 Illustration of forward smoldering in which the reactants enter the reaction zone from
opposite sides, analogous to a diffusion flame. Reprinted by permission of Elsevier Science from An
elementary discussion of forward smolder, Buckmaster, J. and Lozinski, D., Combustion & Flame,
104, 300 -310, Copyright 1996 by The Combustion Institute.
456 K. Miyanishi

VI. MODELS OF SMOLDERING COMBUSTION

Various heat transfer models have been proposed for the propagation of smol-
dering; Drysdale (1985) gives a highly readable description of a couple of the
models (Moussa et aL, 1976; Williams, 1977), and Ohlemiller (1985) provides
an extensive review and evaluation of a large number of models. The simplest
models are ones that are based on a greatly simplified heat balance and that de-
rive a simple algebraic equation showing the dependence of smoldering ve-
locity on various parameters (e.g., Cohen and Luft, 1955; Kinbara et aL, 1967;
Egerton et al, 1963).
As a model of smoldering, Williams (1977) used the fundamental equation
for horizontal fire spread through any porous fuel layer, based on application
of the energy-conservation principle:
q = pVAh (23)
where q is the net energy flux (W cm"^) across the surface of fire inception, p
is the bulk density of the fuel (g cm~^), V is the rate of spread of smoldering
(cm s~^), and Ah is the thermal enthalpy change (J g~^) in raising unit mass of
fuel from ambient temperature to the ignition temperature. To determine the
rate of smoldering, this equation can be rewritten as

V= ^ (24)
pAh
Obviously, in this model, there is no separate consideration of the endothermic/
exothermic production of char and the exothermic oxidation of char, nor any
consideration of the relative flow direction of air and the reaction zone. The
model only considers the net energy flux from these processes. As the net energy
flux (q) decreases, the rate of propagation of smoldering decreases.
Assuming that the ignition temperature is not too different from the maxi-
m u m temperature in the exothermic oxidation zone (T^^), then
Ah - q , ( T , ^ - To) (25)
where Cj^^ is the heat capacity of the fuel (J g~^ °C~^) and TQ is the ambient tem-
perature (°C). Also, assuming that heat transfer from the exothermic oxida-
tion zone to the endothermic pyrolysis zone is by conduction and that a quasi
steady state exists, Eq. (24) becomes

K^max ~ ^o) 1
(26)

The term (T^^ — TQ) cancels out, leaving:

A 1
(27)
pCfu ^
Chapter 13 Duff Consumption 457

unaffected
fuel char/ash

^ PROPAGATION

FIGURE 8 Simple heat transfer model for the propagation of smoldering combustion. From
Drysdale (1985). Copyright John Wiley & Sons. Reproduced by permission of John Wiley & Sons
Limited.

where A is thermal conductivity of the fuel (W cm~^ °C~^) and x is the distance
(cm) over which heat is being transferred (Figure 8), which has been found
by Palmer (1957) to be on the order of 0.01 m or 1 cm. Thus, according to this
model, the rate of propagation is independent of the maximum temperature in
the oxidation zone. Furthermore, since thermal diffusivity (a) is defined as

a = (28)
pC
Eq. (27) can be rewritten simply as

V^ a (29)
X

Both Ohlemiller (1985) and Drysdale (1985) note that this model gives very
crude estimates and can only be used to give an order of magnitude estimate
of the rate of propagation of a smoldering wave; for example, for dry duff, a ^
10~^ cm^ s~^ (Marshall and Holmes, 1979). Thus, the estimated rate of propa-
gation of smoldering within dry duff would be of the order of 10~^ cm s~^
(10~^ m s~^). Despite this criticism of the model, Jones et al. (1994) applied it
to smoldering in packed beds of Casuarina needles and found that the rate of
propagation estimated by the model (4 X 10""^ m s~^) was remarkably close to
the mean propagation rate obtained from six experimental burning trials (4.7 X
10""^ m s"^). Even though this value is an order of magnitude greater than those
reported for smoldering of other lignocellulosic materials, Jones et al. attrib-
uted this difference to the configuration of the material and the effect of this
on the width of the burning zone which they found to be approximately two
needle diameters, about 0.001 m or 1 mm, in contrast to the 0.01 m given by
Palmer (1957).
A second smoldering model is that presented by Kinbara et al. (1967); this
458 K. Miyanishi

model considered downward propagation of smoldering on vertical cylindrical

samples (e.g., circular rods of rolled paper or incense sticks) that are exposed
to air on all sides. Their model has three assumptions:

1. The heat produced by combustion per unit time is proportional to the

rate of air supplied by diffusion which is proportional to (T — TQ)
where T is the temperature of the sample and TQ is the ambient tempera-
ture (°C);
2. This heat is inversely proportional to (Tjg — TQ), where Tjg is the ignition
temperature of the sample;
3. A stagnant layer of combustion gases covers the surface of the sample,
and both the fresh-air supply to the sample and the heat dissipation
from the sample to the air take place at the surface of this layer.

Given these assumptions, Kinbara et al. proposed an equation for the prop-
agation velocity (V) of smoldering:

(30)
.cVJKs Ti^ To

where A is thermal conductivity, c is specific heat, p is density of the fuel, S is

cross-sectional area of the smoldering solid, p is the periphery of the stagnant
layer, h is the heat transmission coefficient, and q is the heat produced by com-
bustion of the solid. As with Williams' model, this model uses only the net heat
flux from the endothermic and exothermic reactions involved in smoldering.
Kinbara et al. found that their model gave a good fit to data obtained from
downward smoldering of small (less than 20 m m diameter) fuels (rolled paper
and incense sticks). This model was specifically developed to describe smolder-
ing in these very small cylindrical fuels and was not intended to be generalized
to other very different fuel configurations such as that found on the forest floor.
Thus, the area of the smoldering zone in this model is constrained by the diam-
eter of the fuel, and duff on the forest floor would represent a fuel of virtually
infinite diameter. Also, due to convection currents bringing air upward around
the periphery of the cylindrical fuel, a cone-shaped smoldering zone develops;
this would not be the case for downward smoldering of a semi-infinite fuelbed.
However, Kinbara et al.'s model represents another relatively simple model of
smoldering based on energy conservation principles.
A third model reviewed by Drysdale (1985) and Ohlemiller (1985) is an ex-
ample of a model that does not use a net heat flux but deals separately with the
endothermic pyrolysis of cellulose and exothermic char oxidation. This steady
state model of smoldering (again in cellulose cylinders) developed by Moussa
et al. (1976) assumes that these processes occur in two zones separated by an
interface with suitable matching requirements (Figure 9). Also, unlike the pre-
Chapter 13 Duff Consumption 459
Interface Oxygen Char-
Oxidation
Zone
Pyrolysis
Zone Free-Convective
Heat Loss

Virgin
Material
(V)

-X 0 X
FIGURE 9 Idealized model of smoldering combustion showing heat and mass transports to and
from the pyrolysis and char-oxidation zone. From Moussa et al. (1976).

vious two models which do not deal with extinguishment, this model can de-
termine the extinguishment limit.
In the first step of this model, the temperature and speed of a steadily prop-
agating pyrolysis zone are determined for a given heat flux at the interface. The
major process is conductive heat transfer within the fuel which is undergoing
thermal degradation (assumed to be an endothermic process). The governing
equations for the cellulose pyrolysis model can be found in the appendix of
Moussa et al. (1976). The following equation is an approximate closed-form so-
lution that they present for their pyrolysis model:

Ei
IPooClnnL A RTsJ F, \ RT, RT^ Fc V RTJ\}
(31)
where V is propagation speed, A is thermal conductivity, p is density, c is spe-
cific heat of the fuel, F^ and Fg are the frequency factors for the char and volatile
forming reactions (assumed to occur in parallel), O is the ratio of virgin fuel
density at the interface over that at x ^ — oo, F^ and Fg are the activation ener-
gies of the char and volatile formation reactions, R is the universal gas constant,
and the interface temperature T^ = T^ -^ ^JipoJ^^)- Fi{x) is the exponential in-
tegral function whose values can be obtained from books of standard mathe-
matical tables (e.g., Beyer, 1991). The first term (outside the square brackets)
represents heat transfer which is assumed to be solid-phase conduction within
a semi-infinite slab and is dependent on the thermophysical properties of ther-
mal conductivity (A), density (p), and specific heat (c) of the fuel (see further
discussion of conduction in Chapter 14 in this book). The terms within the
square brackets are a result of the pyrolysis kinetics.
460 K. Miyanishi

In the next step of the model, the heat flux generated by char oxidation is
determined for specified ambient oxygen mole fractions and partial pressures.
The reaction rate (fe) for char oxidation is represented by

k=FP}ilexp(-^^ (32)

where F is the frequency factor for char oxidation, PQ^^ is the partial pressure of
oxygen at the char surface, E^ is the activation energy for char oxidation, R is
the universal gas constant, and T^ is the temperature at the char surface.
The oxygen flux at the surface {JQJ is governed by diffusion through the
free-convective boundary layer surrounding the char and is given by

Jo. = 2 . 2 X 1 0 - ^ ^ ^ ^ ^ ^ - ^ (33)

where T^ is the mean temperature at the boundary layer, 8 is the average thick-
ness of the boundary layer, PQ and PQ are the partial pressures of oxygen at
X - ^ —GO and x = 0, respectively, and P is pressure.
The steady partial pressure of oxygen at the char surface can be obtained
by equating its rate of arrival [Eq. (33)] with its rate of consumption, which
is given by Eq. (32) times a stoichiometric ratio, i^, according to the reaction
C + O2 ^ COj. To obtain the heat flux generated by char oxidation, the rate of
char reaction [Eq. (32)] is multiplied by its heat of combustion (32 kj g~^).
Finally, the heat flux (qj available for pyrolysis is the heat flux generated by
char oxidation (q) minus heat loss by radiation and free convection to the sur-
roundings through the circumferential area and by conduction to the residual
char through the cross-sectional area:

q,7T^ = h^TTd". (q - salt - h{T, - T^))) (34)

where d is the diameter of the fuel sample, T^ is the temperature at the oxidizing
surface (x = 0), T^Q is the temperature at x —> — 00, and h is the free-convection
heat transfer coefficient (McAdams, 1942; see also Chapter 14 in this book).
The heat loss by conduction to the residual char is assumed to equal the feed-
back heat flux available for pyrolysis. By matching the available heat flux with
that required for pyrolysis, the steady smoldering speed and temperature and
the extinguishment limit can be determined as shown in Figure 10. When the
feedback heat flux does not intersect the curve for the heat flux required for
propagation of pyrolysis, extinguishment occurs (as shown with the inverted
U-shaped dashed curve). When the two curves are tangential, the extinguish-
ment limit is predicted. The intersections of the two curves give two solu-
tions; only the upper solution is stable. Thus, the usefulness of this model is
Chapter 13 DufF Consumption 461

Temperature, °K
1600 1400 1200 1000 900 800 700

\ Kinetic Limit
(Po^=15.7cmHG)
Diffusion Limit
(Xo=0.21)
Generated Heat Flux
200

100

40
CO
E
CD
O

CO
20
0

I
g
03
DC
I 10

0.60 0.80 1.00 1.20 1.40

lO'/T, °K^
FIGURE 10 Heat fluxes versus inverse temperature where XQ^ is oxygen mole fraction and PQ^ is
oxygen partial pressure (cm Hg). The feedback heat fluxes are calculated for the case of standard
atmospheric conditions (XQ^ = 0.21, PQ^ = 15.7) and for the extinguishment case (XQ^ = 0.21,
PQ^ = 5.7). Also shown is the approximate solution for the required heat flux for Aldrich's kinetics.
From Moussa et al. (1976).
462 K. Miyanishi

that, unlike the previous two models that only determine velocity of the smol-
dering wave, this one can be used to determine the extinguishment limit.
However, as with the Kinbara et al (1967) model, the direct applicability
of the model by Moussa et al. (1976) for studies of duff consumption is con-
strained by the fact that these models were developed for small vertical cylin-
drical fuels. Smoldering combustion in horizontal dust layers, as modeled by
Leisch (1983), would be closer to the situation of smoldering in duff. In Leisch's
model, velocity of smoldering is given by

pC T, - To

where A, p, and C are the thermal conductivity, density, and heat capacity of
the fuel; T^^ is the maximum temperature in the oxidation zone; T^ is charring
temperature, and TQ is ambient temperature. The value of y^x is given by

(Surface Area) ,,^

Tox = ^ -PoK (36)
Mass
The rate of char oxidation, k^^, is determined by

Kx = Pox ^^A~^f~) ^^'^^

where F^^ = Poxa + ^oxg^ ^oxa is the preexponential factor for the oxidation of
char to ashes, F^^^ is the preexponential factor for the oxidation of char to gases,
F^^ = Eg^a + F^^^g (the activation energies for the oxidation of char to ashes and
gas), and T^,^ = (T, + T^^)/2.
A limitation of Leisch's (1983) model in attempting to understand the de-
velopment of burned holes in duff is that it is a one-dimensional model. Di Blasi
(1995) presented a two-dimensional model of smoldering propagation through
horizontal layers of cellulosic particles. In this model, the chemical processes
involved in smoldering are modeled as one-step reactions for each main chemi-
cal pathway: pyrolytic degradation of cellulose to char and gases, oxidative deg-
radation of cellulose to char and gases, and oxidation of char to ash and gases.
The model does not include any gas phase reactions since they are generally not
considered of importance in smoldering (Ohlemiller, 1985). The physical pro-
cesses are then described by energy, momentum, and mass balance equations.
The model essentially consists of ten governing equations that are solved nu-
merically (a method of solution is given by Di Blasi, 1994). Simulation results
of the model (Di Blasi, 1995) gave profiles of the structure of the smoldering
wave in terms of temperature and char density, oxygen mass fraction and solid
density, and gas overpressure (Figure 11). These profiles indicated that both
Chapter 13 Duff Consumption 463

37.5

37.5

37.5
X X100[A77]
FIGURE 11 Simulated structure of the smoldering wave after 1280 s from ignition: (a) soHd lines
are contours of temperature (K) in steps of 50 K and dashed lines are contours of char density
(kg m~^) in steps of 0.75 kg m~^; (b) solid lines are contours of oxygen mass fraction in steps of
0.023 and dashed lines are contours of solid density (kg m"^) in steps of 6.8 kg m"^; (c) contours
of gas overpressure (atm) in steps of 1.5 X 10~^. From Di Blasi (1995). Copyright Overseas Pub-
lishers Association N.V. with permission of Gordon and Breach Publishers.

pyrolytic and oxidative degradation of the fuel occur down to a depth of about
1 cm. Below that depth, only endothermic pyrolysis occurs due to depletion
of oxygen. Thus, the solid degradation front advances more rapidly near the
surface. However, the maximum reaction rates (and temperatures) occur just
beneath the surface where significant amounts of oxygen are still available for
both oxidation reactions but where heat loss from the surface is less than at the
surface.
464 K. Miyanishi

Numerous other one- and two-dimensional models of smoldering combus-

tion can be found in the literature (e.g., Egerton et al., 1963; Ohlemiller et ah,
1979; Moallemi et al, 1993; Buckmaster and Lozinski, 1996; Yi et al, 1998).
Ohlemiller (1985) presented a general model of the thermophysics of propaga-
tion which he used to compare and evaluate various models of smoldering com-
bustion propagation. He concluded that all of the models he reviewed, includ-
ing that of Moussa et al (1976), represented great simplifications of the general
model as is evident by the lists of assumptions accompanying each of the mod-
els. The usefulness of these models, according to Ohlemiller, is not so much
in being good predictive models but "in rationalizing certain experimentally-
observed trends." Furthermore, they provide a basis for determining and testing
the important factors that influence the rate of smoldering and the extinguish-
ment limits. As noted earlier, without such a physical process basis, empirical
studies of duff consumption can provide little justification for selection of vari-
ables to measure and can lead to spurious or trivial relationships that do not
advance our understanding of duff consumption.

VII. CONTRIBUTION OF SMOLDERING

COMBUSTION MODELS TO UNDERSTANDING
OF DUFF CONSUMPTION

The purpose of this chapter has been to introduce the reader to some of the com-
bustion literature that would be useful for developing an understanding (and
potentially models) of duff consumption by fire. It was not intended to be a
comprehensive review or evaluation of the smoldering literature (see Ohlemil-
ler, 1985). There has also been no discussion here of ignition of duff (see Chap-
ters 2 and 11 in this book for more on this topic), and the focus has been on
the propagation/extinction of smoldering. It bears repeating that, with few ex-
ceptions, empirical studies of duff consumption reported in the forestry and
ecological literature indicate no awareness of this extensive body of knowledge
on smoldering, the major process by which duff is consumed by fire. Since the
typical purely empirical approach results in very site-specific predictive models
of duff consumption, the justification for some of these studies appears to be
simply that a particular forest type in a particular region has not yet been stud-
ied. As mentioned in an earlier section, rarely, if ever, can the empirically de-
rived predictive equations developed for a particular site, forest type, and set
of fire weather conditions be used elsewhere and under different conditions.
Therefore, rather than more empirical studies of duff consumption using a more
or less arbitrary assortment of variables, what is required now is to make use of
the available models of the smoldering process as a guide to select more ap-
propriate variables for study and to work on further developments/adaptations
Chapter 13 Duff Consumption 465

of existing models of smoldering to make them useful for the study of duff con-
sumption in forest fires.
Heat transfer and energy-balance-based models all indicate that smoldering
can only be propagated and sustained if sufficient heat for pyrolysis is trans-
ferred to the virgin fuel from the oxidation zone. Thus, factors that increase
heat loss between the oxidation zone and the pyrolysis zone would be impor-
tant in determining the extinguishment limit. One such factor is the thickness
of the fuel layer (Bakhman, 1993). The layer of duff on mineral soil may be seen
as a semiopen system consisting of a flat layer of fuel on an inert bed with the
other surface (the top) bordering gas (air). For continued propagation of smol-
dering in any system, heat losses from the reaction zone (char oxidation) must
not be too high since sufficient heat for the endothermic pyrolysis of the next
layer of duff must be supplied from this reaction zone. The fuel itself acts as in-
sulation, and the thinner the fuel layer, the greater the heat loss from the sys-
tem. Thus, Bakhman (1993) found that the temperature at which a smolder-
ing wave was initiated decreased with increasing thickness of the fuel. Palmer
(1957) found that, at depths greater than the critical minimum depth necessary
for sustained smoldering, the rate of smoldering varied with the depth of the
fuel layer. He also found that the minimum depth for sustained smoldering in-
creased with increasing mean particle diameter. For cork dust with a mean par-
ticle diameter of 2.0 mm, the minimum depth for sustained smoldering was
47 mm. The mean particle size of duff collected from a Pinus contorta-Picea en-
gelmannii stand in Alberta was found to be 2.15 mm diameter (Hawkes, 1993).
Thus, smoldering may not be sustainable in such duff with depths less than
about 50 mm. Jones et al. (1994) also reported minimum depths of packed
needle fuels that were required for sustained smoldering. This heat loss expla-
nation might also partially account for the finding in some empirical studies that
preburn duff depth is a significant factor in duff consumption (Blackball and
Auclair, 1982; Brown et al., 1985; Chrosciewicz, 1978a,b); however, see the ear-
lier discussion of this with regard to the Reinhardt et al. (1989) study. Further-
more, it may help to explain the results of Miyanishi et al. (1999) who found
within the same fire significantly larger burned holes in the duff of Picea mari-
ana stands with thicker preburn duff layers than in adjacent Pinus banksiana
stands with thinner duff. If thickness of the fuel and the subsequent effect on
heat loss is a critical factor as proposed by Bakhman (1993), we might expect
propagation of smoldering to be more likely in stands with deeper duff, result-
ing in larger patches of burned out duff. The question then arises: to what ex-
tent can within-stand variation in duff depth account for extinguishment of
smoldering and thus contribute to the limits of the burned patches?
All the smoldering models require as inputs parameter values for the particu-
lar fuel (e.g., density, thermal conductivity, heat capacity, heats of reaction).
While values of these properties are available for a similar fuel, peat, and for
466 K. Miyanishi

some of the constituents of duff (particularly cellulose but also lignin to some
extent), little information is available on values for duff from various forest
types. How much do these properties of duff vary spatially and at what scales
is there significant variation? Is within-stand variation as great as between-
stand variation? Furthermore, how does spatial variation in these fuel proper-
ties compare with the spatial and temporal variation in other significant factors
such as fuel moisture? It may turn out that, despite the variation in these fuel
properties, the variation in fuel moisture may be so much greater that, for all
intents and purposes, some constant values (e.g., mean values for particular for-
est types) can be used in the models (as is the case with values for fuel chemistry
in fire behavior models).
Most of the models also indicate that the rate of oxygen supply (i.e., diffu-
sion of oxygen through the fuelbed from the surface) is the rate-limiting factor
for smoldering (Ohlemiller, 1985,1990a). Also, Moussa et al (1976) concluded
that the extinguishment limit would be set by the rate of oxygen diffusion to
the char zone. In the case of duff smoldering in a forest fire situation, ignition
of the duff would normally occur on the surface (unlike in many of the lab sit-
uations described in this chapter where the whole surface of one end of the
fuelbed is ignited). Thus, in a forest fire, the smoldering zone would progress
both downward and laterally in the duff, creating a hole that extends in size as
smoldering continues (Figure 12). The distance between the oxygen source at
the surface and the char oxidation zone might thus remain relatively constant,
and we would not generally expect smoldering to be extinguished due to a lack
of oxygen. One situation in which oxygen might become limiting would be
as the smoldering zone approached a region of saturated (water-logged) duff.
However, in this situation, it is also unlikely that the heat generated by the char

Drying Zone

Organic Matter

FIGURE 12 Cross-sectional diagram of duff showing the development of a burned hole from
smoldering combustion. From Hungerford et al. (1993).
Chapter 13 Duff Consumption 467

oxidation zone would be sufficient to evaporate all the water and raise the tem-
perature of the duff high enough to produce pyrolysis and char formation.
Thus, as has long been recognized, most likely the key factor that determines
the extinguishment limit of smoldering in the forest fire situation is duff mois-
ture. Moisture in the duff would result in latent heat flux for evaporation of the
water. As noted by Frandsen (1987), the large heat of vaporization of water
(2.25 X l O ^ J g ' ^ ) provides a very effective heat sink. Sufficient latent heat flux
could result in insufficient heat available for endothermic pyrolysis (charring)
of duff, thus resulting in extinguishment of smoldering. According to Yi and
Kim (1996), pyrolysis in the region where water evaporation is occurring can
be considered negligible. In most experimental situations, the moisture content
of the fuel is typically uniform throughout the fuel. However, duff moisture
content varies spatially, both within stands and between stands. Thus, further
study and modeling of the spatial variation in duff moisture at various scales
would be useful.
Bridge and Johnson (2000) showed that the organization of plant commu-
nities along moisture and nutrient gradients could be used to relate the spatial
distribution of tree species (stand types) to hillslope position and surficial geol-
ogy. Thus, on glaciofluvial substrate in the mixedwood boreal forest of central
Saskatchewan, the ridge tops (which are drier and more nutrient poor) are domi-
nated by Pinus banksiana, and the mid to bottom slopes (which are wetter and
more nutrient rich) are dominated by Picea mariana. Furthermore, duff depths
in stands dominated by Pinus banksiana are typically significantly less than duff
depths in stands dominated by Picea mariana (Johnson, 1981; Dyrness and No-
rum, 1983). Thus, we might expect some correlation between duff depth and
hillslope position (i.e., distance from the ridge line). Furthermore, studies have
shown that a significant proportion of duff moisture is due to upward trans-
fer of water from the mineral soil to the duff (e.g., Samran et ah, 1995). Since
soil moisture is influenced by substrate (e.g., glacial till has a higher field capac-
ity than glaciofluvial substrate) and since, within substrate type, soil moisture
shows a gradient along hillslopes, it may be possible to use a hydrologic model
to predict between-stand variation in duff moisture. With information on be-
tween-stand variation in both duff depth and duff moisture, it may be possible
to begin to develop a model of variation in duff consumption at the hillslope
scale.
At the stand level, Bajtala (1999) found significantly drier duff beneath com-
pared to beyond tree crowns within stands of Pinus banksiana and Picea mari-
ana. Also, Miyanishi et ah (1999) reported a significant spatial correlation be-
tween burned holes in the duff and standing boles of trees killed by fire. If duff
moisture is the major factor determining extinction of smoldering in forest
fires, the patchiness of duff consumption within stands often observed in for-
est fires would indicate significant variation in duff moisture at this scale. Such
468 K. Miyanishi

variation could potentially be explained by differences in precipitation input

due to interception by tree crowns, differences in evaporation as well as dew
formation due to crown interception of radiation, and differences in bulk den-
sity and moisture-holding capacity of the ground cover (herbaceous/moss cover
versus needle litter). Thus, a hydrologically based model of within-stand spatial
variation in duff moisture could also be developed. Since such a model would
involve the role of the tree canopy in duff moisture dynamics, it could indicate
a significant relationship between the areal extent of duff consumption within
burned forests and density of trees, particularly those that have a higher inter-
ception efficiency (e.g., Picea spp. compared with Pinus spp.). Finally, the pro-
posed models for duff moisture at the two scales could be combined.
Another aspect that bears further investigation is the interaction between the
effects of duff depth and moisture. Recall from the earlier discussion that there
exist critical minimum depths of fuelbed required for smoldering propagation
resulting from the balance between heat generation from the char oxidation
zone and convective heat loss from the surface. Since the process of evapora-
tion of water from the duff is also a heat sink, we might expect the critical depth
of fuelbed to increase with increasing moisture content. Thus, McMahon et a\.
(1980) reported sustained smoldering in organic soil blocks with moisture con-
tents as high as 135%. Some preliminary lab studies using varying depths of a
peat fuelbed with varying moisture content have, in fact, shown a significant
positive relationship as predicted (Miyanishi and Johnson, unpublished data).
Although not exhaustive, the preceding discussions illustrate how an ap-
preciation and understanding of the processes involved in duff consumption
and use of smoldering models available in the combustion literature can lead to
a somewhat different approach to the study of duff consumption than has been
used in the past. This physical process-based approach provides a means of go-
ing beyond empirical case studies, leads to more appropriate research ques-
tions, and ultimately should result in a better understanding of the phenome-
non of duff consumption.

NOTATION
ROMAN LETTERS

C heat capacity J kg~^ K"^

D duff consumption kg m~^
E activation energy J mol~^
F frequency factor s~^
H heat of reaction J kg~^
Chapter 13 Duff Consumption 469

J mass flux kgm-'s-^

M moisture content % dry weight
P pressure Jm"^
R universal gas constant 8.314Jmor^K~^
S cross-sectional area m^
T temperature °CorK
V rate of flaming front advance m s~^
V wave speed m s~^

w normalized weight
X mole fraction
a absorptivity dimensionless
c specific heat J kg-^ K-i
d sample diameter m
h free-convection heat transfer coefficient Wm~^K~^
k specific rate of reaction s~^ or min~^
I latent heat of vaporization 2.25 X l O ^ J k g ' ^
V periphery m
^ heat flux Wm"^
t time s
w width of combustion zone m
X spatial coordinate m

GREEK LETTERS

a thermal diffusivity m^s-^

8 boundary layer thickness m
E emissivity dimensionless
A thermal conductivity Wm'^K'^
P density g cm~^ or kg m~^
a Stefan-Boltzmann constant 5.668 X 1 0 " ^ W m " ^ K
T residence time s
a ratio of the virgin cellulose density at
the interface over that at x - ^ -- c o ,
taken to be 10~^
470 K. Miyanishi

SUBSCRIPTS

c char
cell cellulose
A active cellulose
d duff
f final
fu fuel
fl flame
g gas
i initial
ig ignition
nax maximum
V condensable volatiles
w water
s conditions at x = 0 (surface)
0 ambient
oo conditions at x ^ — oo

REFERENCES
Aber, J. D., and Melillo, J. M. (1991). "Terrestrial Ecosystems." Saunders College Publishing, Phila-
delphia.
Adams, J. L. (1966). Prescribed burning techniques for site preparation in cutover jack pine in
southeastern Manitoba. Pulp Pap. Mag. Can. 67 WR: 574-584.
Ahlgren, C. E. (1970). "Some Effects of Prescribed Burning on Jack Pine Reproduction in North-
eastern Minnesota." Miscellaneous Report No. 94. Agricultural Experiment Station, University
of Minnesota, St. Paul.
Alexander, M. (1977). "Introduction to Soil Microbiology." John Wiley and Sons, New York.
Alexander, M. E. (1982). Calculating and interpreting forest fire intensities. Can. J. For. Res. 60,
349-357.
Bajtala, M. J. (1999). "Spatial Patterns of Duff Consumption in Black Spruce and Jack Pine Stands
in the Boreal Mixedwood Forest." M.Sc. thesis, University of Guelph, Guelph.
Bakhman, N. N. (1993). Smoldering wave propagation mechanism. 1. Critical conditions. Combus-
tion, Explosion and Shock Waves 29, 14-17.
Berg, B., and Ekbohm, G. 1991. Litter mass-loss rates and decomposition patterns in some needle
and leaf htter types. Long-term decomposition in a Scots pine forest. VII. Can.j. Bot. 69, 1449-
1456.
Berg, B., and Staaf, H. (1981). Leaching, accumulation and release of nitrogen in decomposing for-
est litter. Ecol. Bull. (Stockholm) 33, 163-178.
Berg, B., Hannus, K., Popoff, T., and Theander, O. (1982). Changes in organic chemical components
Chapter 13 DufF Consumption 471

of needle litter during decomposition. Long-term decomposition in a Scots pine forest. I. Can.
J. Bot. 60, 1310-1319.
Beyer, W. H. (1991). "CRC Standard Mathematical Tables and Formulae," 29th ed. CRC Press, Boca
Raton.
Blackball, J. W., and Auclair, A. D. (1982). Best solution models of prescribed fire impact in sub-
alpine Picea glauca-Ahies lasiocarpa forests of British Columbia. In "Problem Analysis of Pre-
scribed Burning: A Synthesis Text" (A. D. Auclair, Ed.), p. 32. Department of Fisheries and
Forestry, Canadian Forest Service, Pacific Forest Research Centre, Victoria.
Bracewell, J. M., and Robertson, G. W. (1987). Characteristics of soil organic matter in temperate
soils by Curie-point pyrolysis-mass spectrometry, III. Transformations occurring in surface or-
ganic horizons. Geoderma 40, 333-344.
Bradbury, A., Sakai, Y., and Shafizadeh, F. (1979). A kinetic model for pyrolysis of cellulose. J. Appl.
Polym. Sci. 23, 3271-3280.
Bridge, S. R. J., and Johnson, E. A. (2000). Geomorphic principles of terrain organization and vege-
tation gradients. J. Veg. Sci. 11, 57-70.
Broido, A. (1976). Kinetics of solid-phase cellulose pyrolysis. In "Thermal Uses and Properties of
Carbohydrates and Lignins" (F. Shafizadeh, K. V. Sarkanen and D. A. Tillman, Eds.), pp. 1 9 -
35. Academic Press, New York.
Brown, J. K., Marsden, M. A., Ryan, K. C , and Reinhardt, E. D. (1985). "Predicting Duff and Woody
Fuel Consumed by Prescribed Fire in the Northern Rocky Mountains." Research Paper INT-337.
USDA Forest Service, Intermountain Forest and Range Experiment Station, Ogden.
Brown, J. K., Reinhardt, E. D., and Fischer, W. C. (1991). Predicting duff and woody fuel consump-
tion in northern Idaho prescribed fires. For. Sci. 37, 1550-1566.
Buckmaster, J., and Lozinski, D. (1996). An elementary discussion of forward smolder. Combust.
F/ame 104, 300-310.
Canada Soil Survey Committee. (1978). "The Canadian System of Soil Classification." Publication
1646. Research Branch, Canadian Department of Agriculture, Ottawa.
Charron, I. (1998). "Sexual recruitment of trees following fire in the southern mixedwood boreal
forest of Canada." M.Sc. thesis, Concordia University, Montreal.
Chrosciewicz, Z. (1959). "Controlled Burning Experiments on Jack Pine Sites." Technical Note
No. 72. Canada Department of Northern Affairs and National Resources, Forestry Branch, For-
est Research Division, Ottawa.
Chrosciewicz, Z. (1967). "Experimental Burning for Humus Disposal on Clear-Cut Jack Pine Sites
in Central Ontario." Publication No. 1181. Canada Department of Forestry and Rural Develop-
ment, Forestry Branch, Ottawa.
Chrosciewicz, Z. (1974). Evaluation of fire-produced seedbeds for jack pine regeneration in cen-
tral Ontario. Can. J. For. Res. 4, 455-457.
Chrosciewicz, Z. (1976). Burning for black spruce regeneration on a lowland cutover site in south-
eastern Manitoba. Can. J. For. Res. 6, 179-186.
Chrosciewicz, Z. (1978a). "Slash and Duff Reduction by Burning on Clear-Cut Jack Pine Sites in
Southeastern Manitoba." Information Report NOR-X-199. Northern Forest Research Centre,
Canadian Forest Service, Edmonton.
Chrosciewicz, Z. (1978b). "Slash and Duff Reduction by Burning on Clear-Cut Jack Pine Sites
in Central Saskatchewan." Information Report NOR-X-200. Northern Forest Research Centre,
Canadian Forest Service, Edmonton.
Chrosciewicz, Z. (1989). Prediction of forest-floor moisture content under diverse jack pine can-
opy conditions. Can. J. For. Res. 19, 1483-1487.
Clayton, J. S., Ehrlich, W. A., Cann, D. B., Day, H. H., and Marshall, I. B. (1977). "Soils of Canada.
Volume 1—Soil Report." Canada Department of Agriculture Research Branch, Ottawa.
Cohen, L., and Luft, N. W. (1955). Combustion of dust layers in still air. Fuel 34, 154-163.
Delaney, M. T., Fernandez, I. J., Simmons, J. A., and Briggs, R. D. (1996). "Red Maple and White
472 K. Miyanishi

Pine Litter Quality: Initial Changes with Decomposition." Technical Bulletin 162. Maine Agri-
cultural and Forest Experiment Station, Orono.
Di Blasi, C. (1994). Smolder spread through thin horizontal fuel layers. In "Advanced computa-
tional methods in heat transfer III. Proceedings of the Third International Conference" (L. C.
Wrobel, C. A. Brebbia and A. J. Nowak, Eds.), pp. 323-330. Computational Mechanics Publica-
tions, Southampton.
Di Blasi, C. (1995). Mechanisms of two-dimensional smoldering propagation through packed fuel
beds. Combust. Sci. Tech. 106, 103-124.
Drysdale, D. (1985). "An Introduction to Eire Dynamics." John Wiley and Sons, Chichester.
Dyrness, C. T., and Norum, R. A. (1983). The effects of experimental fires on black spruce forest
floors in interior Alaska. Can. J. For. Res. 13, 879-893.
Egerton, A., Gugan, K., and Weinberg, E. (1963). The mechanism of smoldering in cigarettes. Com-
bust. Flame 7, 63-78.
Flinn, M. A., and Wein, R. W. (1977). Depth of underground plant organs and theoretical survival
during fire. Can. J. Bot. 55, 2550-2554.
Erandsen, W H. (1987). The influence of moisture and mineral soil on the combustion limits of
smoldering forest duff. Can. J. For. Res. 17, 1540-1544.
Erandsen, W H. (1991a). Burning rate of smoldering peat. Northwest Sci., 65, 166-172.
Erandsen, W H. (1991b). Smoldering spread rate: A preliminary estimate. In "Proceedings of the
Eleventh Conference on Eire and Eorest Meteorology" (P. L. Andrews and D. F. Potts, Eds.),
pp. 168-172. Society of American Foresters, Bethesda.
Erandsen, W H. (1991c). Heat evolved from smoldering peat. Int. J. Wildland Fire 1, 197-204.
Green, R. N., Trowbridge, R. L., and Klinka, K. (1993). "Towards a Taxonomic Classification of Hu-
mus Forms." Eorest Science Monograph 29. Society of American Foresters, Bethesda.
Hartford, R. A. (1989). Smoldering combustion limits in peat as influenced by moisture, mineral
content, and organic bulk density. In "Proceedings of the Tenth Conference on Eire and Forest
Meteorology" (D. C. Maclver, H. Auld, and R. Whitewood, Eds.), pp. 282-286. Forestry Can-
ada, Environment Canada, Ottawa.
Hawkes, B. C. (1993). Factors that influence peat consumption under dependent burning condi-
tions: A laboratory study. Ph.D. dissertation. University of Montana, Missoula.
Hungerford, R. D., Erandsen, W H., and Ryan, K. C. (1995). Ignition and burning characteristics of
organic soils. In "Proceedings of the Tall Timbers Eire Ecology Conference in Wetlands, a Man-
agement Perspective" (S. I. Cerulean and R. T. Engstrom, Eds.), pp. 78-91. Tall Timbers Re-
search Station, Tallahassee.
Johnson, E. A. (1981). Vegetation organization and dynamics of lichen woodland communities in
the Northwest Territories, Canada. Ecology 62, 200-215.
Johnson, E. A. (1992). "Fire and Vegetation Dynamics: Studies from the North American Boreal
Forest." Cambridge University Press, Cambridge.
Jones, J. C , Goh, T. P. T., and Dijanosic, M. J. (1994). Smoldering and flaming combustion in
packed beds of Casuarina needles. J. Fire Sci. 12, 442-451.
Jordan, C. E (1982). Amazon rain forests. Am. Scientist 70, 394-401.
Jordan, C. E, and Herrera, R. (1981). Tropical rain forests: Are nutrients really critical? Am. Nat.
117, 167-180.
Kashiwagi, T., and Nambu, H. (1992). Global kinetic constant for thermal oxidative degradation of
a cellulosic paper. Combustion and Flame 88, 345-368.
Kauffman, J. B. (1991). Survival by sprouting following fire in tropical forests of the eastern Ama-
zon. Biotropica 23, 219-224.
Kauffman, J. B., Cummings, D. L., Ward, D. E., and Babbitt, R. (1995). Eire in the Brazilian Ama-
zon: 1. Biomass, nutrient pools, and losses in slashed primary forests. Oecologia 104, 397-
408.
Chapter 13 Duff Consumption 473

Kellman, M., and Meave, J. (1997). Fire in the tropical gallery forests of Belize. J. Biogeog. 24, 2 3 -
34.
Kinbara, T., Endo, H., and Sega, S. (1967). Downward propagation of smoldering combustion
through solid materials. In "Eleventh Symposium (International) on Combustion," pp. 5 2 5 -
531. The Combustion Institute, Pittsburgh.
Kingsbury, N., and Kellman, M. (1997). Root mat depths and surface soil chemistry in southeast-
ern Venezuela. J. Tropical Ecol. 13, 475-479.
Krause, H. (1998). Forest floor mass and nutrients in two chronosequences of plantations: Jack
pine vs. black spruce. Can. J. Soil Sci. 78, 77-83.
Leighton, M., and Wirawan, N. (1986). Catastrophic drought and fire in Borneo tropical rain for-
est associated with the 1982-1983 El Nirio Southern Oscillation Event. In "Tropical Rain For-
ests and the World Atmosphere" (G. T. Prance, Ed.), pp. 75-102. AAAS Selected Symposium
101, Westview, Boulder.
Leisch, S. O. (1983). "Smoldering Combustion in Horizontal Dust Layers." PhD dissertation. Uni-
versity of Michigan, Ann Arbor.
Lewellen, P. C , Peters, W. A., and Howard, J. B. (1976). In "Sixteenth Symposium (International)
on Combustion," pp. 1471-1480. The Combustion Institute, Pittsburgh.
Little, S. N., Ottmar, R. D., and Ohmann, J. L. (1986). "Predicting Duff Consumption from Pre-
scribed Burns on Conifer Clearcuts in Western Oregon and Western Washington." Research Pa-
per PNW-362. USDA Forest Service, Pacific Northwest Forest and Range Experiment Station,
Portland.
Marshall, T. J., and Holmes, J. W (1979). "Soil Physics." Cambridge University Press, New York.
Mason, C. F. (1976). "Decomposition." Edward Arnold (Publishers) Ltd., London.
McAdams, W H. (1942). "Heat Transmission," 2nd ed. McGraw-Hill, New York.
McMahon, C.K., Wade, D. D., and Tsoukalas, S. N. (1980). Combustion characteristics and emis-
sions from burning organic soils. In "Proceedings of the 73rd Annual Meeting of the Air Pollu-
tion Control Association," Paper No. 80-15.5.
Merrill, D. F., and Alexander, M. E. (1987). "Glossary of Forest Fire Management Terms." Canadian
Committee on Forest Fire Management, National Research Council of Canada, Ottawa.
Miyanishi, K., Bajtala, M. J., and Johnson, E. A. (1999). Patterns of duff consumption in Pinus hank-
siana and Picea mariana stands. In "Proceedings of Sustainable Forest Management Network
Conference, Science and Practice: Sustaining the Boreal Forest," pp. 112-115. Sustainable For-
est Management Network, Edmonton.
Moallemi, M. K., Zhang, H., and Kumar, S. (1993). Numerical modeling of two-dimensional smol-
dering process. Combust. Flame 95, 170-182.
Moussa, N. A., Toong, T. Y., and Garris, C. A. (1976). Mechanism of smoldering of cellulosic ma-
terials. In "Sixteenth Symposium (International) on Combustion," pp. 1447-1457. The Com-
bustion Institute, Pittsburgh.
Muramatsu, M., Umemura, S., and Okada, T. (1979). A mathematical model of evaporation-
pyrolysis processes inside a naturally smoldering cigarette. Combust. Flame 36, 245-262.
Ohlemiller, T.J. (1985). Modeling of smoldering combustion propagation. Prog. Energy Combust.
Sci. 11,277-310.
Ohlemiller, T.J. (1990a). Smoldering combustion propagation through a permeable horizontal fuel
layer. Combust. Flame 81, 341-353.
Ohlemiller, T. J. (1990b). Forced smolder propagation and the transition to flaming in cellulosic
insulation. Combust. Flame 81, 354-365.
Ohlemiller, T. J., Bellan, J., and Rogers, F. (1979). A model of smoldering combustion applied to
flexible polyurethane foams. Combust. Flame 36, 197-215.
Ortiz-Molina, M. G., Toong, T-Y., Moussa, N. A., and Tesoro, G. C. (1978). Smoldering combus-
tion of flexible polyurethane foams and its transition to flaming or extinguishment. In "Seven-
474 K. Miyanishi

teenth Symposium (International) on Combustion," pp. 1191-1200. The Combustion Institute,

Pittsburgh.
Palmer, K. N. (1957). Smoldering combustion in dusts and fibrous materials. Combust. Flame 1,
129-154.
Perala, D. A., and Alban, D. H. (1982). "Rates of Forest Floor Decomposition and Nutrient Turn-
over in Aspen, Pine, and Spruce Stands on Two Different Soils." Research Paper NC-227. USDA
Forest Service, North Central Forest Experiment Station, St. Paul.
Peter, S. (1992). "Heat Transfer in Soils Beneath a Spreading Fire." Ph.D. dissertation, University
of New Brunswick, Fredericton.
Potts, D. F., Zuuring, H., and Hillhouse, M. (1983). Spatial analysis of duff moisture and structure
variability. In "Proceedings of the Seventh Conference on Fire and Forest Meteorology," pp. 1 8 -
21. American Meteorology Society, Boston.
Reinhardt, E. D., Brown, J. K., and Fischer, W. C. (1989). Fuel consumption from prescribed fire
in northern Idaho logging slash. In "Proceedings of the Tenth Conference on Fire and Forest
Meteorology" (D. C. Maclver, H. Auld, and R. Whitewood, Eds.), pp. 155-160. Forestry Can-
ada, Environment Canada, Ottawa.
Richards, P. W. (1952). "The Tropical Rain Forest: An Ecological Study." Cambridge University
Press, Cambridge.
Roberts, A. F. (1970). A review of kinetics data for the pyrolysis of wood and related substances.
Combust. Flame 14, 261-272.
Rothermel, R. C. (1972). "A Mathematical Model for Predicting Fire Spread in Wildland Fuels."
Research Paper INT-115. USDA Forest Service, Intermountain Forest and Range Experiment
Station, Ogden.
Rothermel, R. C. (1976). Forest fires and the chemistry of forest fuels. In "Thermal Uses and Prop-
erties of Carbohydrates and Lignins." (F. Shafizadeh, K. V. Sarkanen, and D. A. Tillman, Eds.),
pp. 245-259. Academic Press, New York.
Samran, S., Woodard, P. M., and Rothwell, R. L. (1995). The effect of soil water on ground fuel avail-
abiUty. Forest Sci. 41, 255-267.
Sandberg, D. V. (1980). Duff reduction by prescribed underburning in Douglas-fir. Research Pa-
per PNW-272. USDA Forest Service, Pacific Northwest Forest and Range Experiment Station,
Pordand.
Sato, K., and Sega, S. (1985). The mode of burning zone spread along cylindrical cellulosic mate-
rial. J. Fire Sd. 3,26-34.
Schimmel, J., and Granstrom, A. (1996). Fire severity and vegetation response in the boreal Swed-
ish forest. Ecology 77, 1436-1450.
Schult, D. A., Matkowsky, B. J., Volpert, V. A., and Fernandez-Pello, A. C. (1995). Propagation and
extinction of forced opposed flow smolder waves. Combust. Flame 101, 471-490.
Shafizadeh, R, and DeGroot, W. F. (1976). Combustion characteristics of cellulosic fuels. In "Ther-
mal Uses and Properties of Carbohydrates and Lignins." (F. Shafizadeh, K. V. Sarkanen, and
D. A. Tillman, Eds.), pp. 1-17. Academic Press, New York.
Shafizadeh, R, and Sekiguchi, Y. (1984). Oxidation of chars during smoldering combustion of cel-
lulosic materials. Combust. Flame 55, 171-179.
Shafizadeh, R, Bradbury, A. G. W., DeGroot, W. R, and Aanerud, T. W. (1982). Role of inorganic
additives in the smoldering combustion of cotton cellulose. Ind. Eng. Chem. Prod. Res. Dev. 21,
97-101.
Simmons, R. R (1995). Fire chemistry. In "Combustion Fundamentals of Fire" (G. Cox, Ed.),
pp. 405-473. Academic Press, London.
Stark, N. M., and Jordan, C. R (1978). Nutrient retention by the root mat of an Amazonian rain for-
est. Ecology 59, 434-437.
Chapter 13 Duff Consumption 475

Steward, F. R., Peter, S., and Richon, J. B. (1990). A method for predicting the depth of lethal heat
penetration into mineral soils exposed to fires of various intensities. Can. J. For. Res. 20, 9 1 9 -
926.
Taiz, L., and Zeiger, E. (1991). "Plant Physiology." Benjamin/Cummings Publishing Company, Red-
wood City.
Thomas, P. A., and Wein, R. W. (1985). The influence of shelter and the hypothetical effect of fire
severity on the postfire establishment of conifers from seed. Can. J. For. Res. 15, 148-155.
Uhl, C , and Kauffman, J. B. (1990). Deforestation, fire susceptibility, and potential tree responses
to fire in the eastern Amazon. Ecology 71, 437-449.
Uhl, C , Kauffman, J. B., and Cummings, D. L. (1988). Fire in the Venezuelan Amazon 2. Environ-
mental conditions necessary for forest fires in the evergreen rainforest of Venezuela. Oikos 53,
176-184.
Van Wagner, C. E. (1972). Duff consumption by fire in eastern pine stands. Can. J. For. Res. 2, 3 4 -
39.
Van Wagner, C. E. (1987). "Development and Structure of the Canadian Forest Fire Weather In-
dex System." Forestry Technical Report 35. Canadian Forestry Service, Ottawa.
Weber, M. G. (1985). Forest soil respiration in eastern Ontario jack pine ecosystems. Can. J. For.
Res. 15, 1069-1073.
Weber, M. G., Hummel, M., and Van Wagner, C. E. (1987). Selected parameters of fire behavior
and Pinus hanksiana Lamb, regeneration in eastern Ontario. For. Chron. 63, 340-346.
Weber, M. G., Methven, I. R., and Van Wagner, C. E. (1985). The effect of forest floor manipulation
on nitrogen status and tree growth in an eastern Ontario jack pine ecosystem. Can. J. For. Res.
15,313-318.
Wein, R. W. (1981). Characteristics and suppression of fires in organic terrain in Australia. Aust.
For. 44, 162-169.
Wein, R. W. (1983). Fire behavior and ecological effects in organic terrain. In "The Role of Fire in
Northern Circumpolar Ecosystems" (R. W Wein and D. A. MacLean, Eds.), pp. 81-95. John
Wiley, New York.
Williams, F. (1977). Mechanisms of fire spread. In "Sixteenth Symposium (International) on Com-
bustion," pp. 1281-1294. The Combustion Institute, Pittsburgh.
Williams, S. T., and Gray, T. R. G. (1974). Decomposition of litter on the soil surface. In "Biology
of Plant Litter Decomposition" (C. H. Dickinson and G. J. F. Pugh, Eds.), pp. 611-632. Aca-
demic Press, London.
Yi, S-C, and Kim, Y. C. (1996). Modeling of smoldering cellulosic materials—Evaporation and
pyrolysis processes. Preprints of the Division of Fuel Chemistry of the American Chemical Society
41, 1183-1186.
Yi, S-C, Bae, S. Y., and Kim, S. H. (1998). Heat transfer of a smoldering flammable substrate. Part
2. A theoretical model and its application. J. Fire Sci. 16, 32-45.
Zasada, J. C , Norum, R. A., Van Veldhuizen, R. M., and Teutsch, C. E. (1983). Artificial regenera-
tion of trees and tall shrubs in experimentally burned upland black spruce/feather moss stands
in Alaska. Can.f. For. Res. 13, 903-913.
Zech, W, Senesi, N., Guggenberger, G., Kaiser, K., Lehmann, J., Miano, T. M., Miltner, A., and
Schroth, G. (1997). Factors controlling humification and mineralization of soil organic matter
in the tropics. Geoderma 79, 117-161.
This Page Intentionally Left Blank
 CHAPTER 14
Fire Effects on Trees
M. B. DICKINSON AND E. A.JOHNSON
Kananashis Field Stations and Department of Biological Sciences,
University of Calgary, Calgary, Alberta, Canada

I. Introduction
A. Heat Transfer Mechanisms
II. Effects of Fire on the Tree Bole
A. Flame Characteristics
B. Heat Transfer into the Bole
III. Effects of Fire on Canopy Components
A. Plume Characteristics
B. Heat Transfer into Canopy Components
IV. Root Necrosis
V. Tree Mortality
VI. Discussion
Notation
Additional Readings
References

I. INTRODUCTION

There has been considerable interest among ecologists and foresters in the ef-
fects fires have on populations, communities, and ecosystems. However, rela-
tively little attention has been paid to understanding how wildfires cause these
effects. Needed is a more thorough understanding of the physical processes of
heat transfer between the fire and the organism and in turn how the resulting
patterns of individual damage affect population and ecosystem processes. To
achieve such insight, ecologists and foresters must borrow liberally from the
work of physicists and engineers. The preceding chapters in this book will be
of considerable help.
In this chapter, we will explain some of the ways to model heat transfer be-
tween fires and trees. While our focus is primarily on trees, the discussion ap-
plies with suitable modification to herbs and shrubs. We also have used rela-
tively simple models so that both their underlying mechanisms and predictive
Forest Fires
Copyright © 2001 by Academic Press. All rights of reproduction in any form reserved. 477
478 Dickinson and Johnson

Fuel bed ^
FIGURE 1 Cross-sectional view of a fire line burning in surface fuels (e.g., leaf litter, small-
diameter downed-woody material, herbs, and shrubs). The flame gives rise to the buoyant plume
and both are characterized by a turbulent flow regime. Smoldering combustion may proceed at low
rates in organic soil layers and decomposed downed-woody material after the flame has passed.

powers will be intuitively clear. The heat transfer models introduced in this
chapter are most applicable to trees damaged in surface fires with relatively low
fire-line intensities (kW m~^). The heat sources in the models are the flame, the
buoyant plume arising from the flame, and the smoldering combustion of or-
ganic material (Figure 1).
Plants can be divided into components that have similar heat transfer char-
acteristics (Figure 2). Since the components are primarily constructed of cellu-
lose, their heating is determined mainly by their physical dimensions and spatial
position relative to the heat sources. Plants must fulfill four principal func-
tions: mechanical support, photosynthesis, reproduction, and hydraulics (Nik-
las, 1992). As a consequence, trees have a rather stereotyped spatial arrangement
of parts. The stem and branches position thin photosynthetic structures high
enough to intercept light. The sizes of the branches are scaled by the weight of
the foliage they support (e.g.. Corner, 1949; Ackerly and Donoghue, 1998). The
trunk, which elevates the canopy, will be larger in diameter and scaled roughly
as a power of height, 1.5 fordicotsand 1.1 for conifers (Niklas, 1994). The whole
structure is attached to the ground by roots scaled by their functions of attach-
ment (Ennos, 1993) and soil-resource acquisition (Russell, 1977).
Thus, crown components will generally be small in diameter, meaning that
they will have little internal resistance to heating by the flame or buoyant plume
and can be killed and combusted easily. Trunks as a whole will have consider-
able resistance to heating by the flame but can be damaged because their vas-
cular cambium is near the outside of their diameter. Trunks, except when small
Chapter 14 Fire Effects on Trees 479

Heat source ^M|M|MBB|^^^M Tree component

Plume flllj^^ Canopy

-Foliage
-Buds
-Twigs

Flame /i Bole

Flaming and glowing i*"V^^V i!^-^ Roots

combustion in fuel bed
Smoldering combustion in
organic soil
FIGURE 2 The spatial arrangement of tree components and the sources of heat which affect them
in surface fires. The fohage, buds, and twigs are heated by the plume and the live bark and vascular
cambium in the bole are heated by the flame. The base of the bole and roots may be heated by the
downward heat flux from flaming and glowing combustion in the fuelbed or by smoldering com-
bustion in organic soil layers after the fire line has passed.

in diameter, are not consumed by the fire. The roots are insulated from the heat
of the flame but are heated by smoldering combustion after the fire line has
passed. Roots are vulnerable to smoldering combustion because heating is of-
ten of long duration and the vascular cambium is near the surface.
In this chapter, we will begin by discussing tissue necrosis in the crown,
bole, and roots. Then, we will introduce an approach by which patterns of tis-
sue necrosis in these components might be used to understand the death of in-
dividual trees. As background for this chapter. Chapter 2 describes flames,
Chapters 5 and 6 describe spreading fires, and Chapters 7 and 8 describe
plumes above fires. Chapter 13 addresses in more detail smoldering combus-
tion processes and their effects on tree recruitment.

A. HEAT TRANSFER MECHANISMS

Heat transfer occurs by conduction, convection, and radiation. Heat transfer

texts include an introduction to these heat transfer mechanisms, providing
models that can be applied to the heating of trees (e.g., Sucec, 1985; Holman,
480 Dickinson and Johnson

1986). The introductory material presented in this section will serve as a basis
for understanding the more complex models used in later sections to describe
necrosis of individual plant components.

1. Conduction
Conduction heat transfer occurs along a gradient from high to low tempera-
tures (as required by the second law of thermodynamics). The rate of heat trans-
fer is determined by the driving force (the temperature difference) and the heat-
ing properties of the material. Conduction in solids occurs by collisions between
molecules and by the migration of free electrons. In a fluid (such as air), con-
duction is due to collisions of molecules that are in constant, random motion.
The simplest case of conduction is when heat flow occurs along one space
coordinate (i.e., is one dimensional). For instance, consider the central region
of a semi-infinite slab (i.e., a slab that is thin relative to its height and width)
that is homogeneous in terms of its temperature and thermal properties (Fig-
ure 3A). If the temperature of one of the slab's surfaces is raised, heat will be
transferred inward, perpendicular to the surface. If the surface temperature is
held constant for a long enough time, heat transfer is steady state and is de-
scribed by Fourier's law of conduction
dJ
q = -kA (1)
dx

Temperature

A B
FIGURE 3 One-dimensional steady state conduction through a semi-infinite slab. (A) A homo-
geneous slab, the left surface of which (1) is maintained at a higher temperature than the right (2).
(B) A slab composed of two layers, the left layer. A, having a higher thermal conductivity than the
right [B, see Eq. (4)]. In both cases, the temperature difference has been maintained for a sufficiently
long period that the rate of heat flow q (W) into the left surface is equal to the rate of heat flow out
of the right surface.
chapter 14 Fire Effects on Trees 481

which upon integration gives

q = -kA (2)
Ax
where q is the heat transfer rate (W), k is the thermal conductivity (W m~^ • °C),
A is area (m^), Ax is the thickness of the slab (m), and T^ and Tj (°C) are the
temperatures of the front (heated) and back side of the slab, respectively. The
negative sign is inserted in Eq. (1) because, by definition, the heat flow q is posi-
tive when the temperature derivative dT/dx is negative (i.e., Tj — T^ is nega-
tive). The heat flux (i.e., q/A, the heat flow per unit area) is thus a result of the
gradient in temperature through the slab (dT/dx or [Tj - T j / A x ) and the ther-
mal conductivity k. Equation 1 defines thermal conductivity and, given the heat
flux and the temperature gradient, can be used to derive estimates of thermal
conductivity for a material.
Thermal conductivity of wood, bark, and foliage varies with their density,
moisture content, and temperature (e.g., Byram et al, 1952a; Martin, 1963; Reif-
snyder et al, 1967). If a material does not have homogeneous conductivity val-
ues (e.g., bark and wood) but can be divided into parts connected in series, the
heat transfer equations can be solved simultaneously. For instance, for steady
state heat transfer through a material composed of two layers of different con-
ductivity (Figure 3), the rate of heat flow is equal across each layer:

q = -M-^7 = -hA^—^ (3)

AXA AXB

where Ax^ and Axg are the thicknesses of layers A and B, respectively. These
equations can be solved simultaneously resulting in

(4)
AXA _^ AXB

wherein the resistances to conduction heat transfer for each layer are added to-
gether in the denominator. Notice that the temperature change is constant
within each layer, confirming that the rate of heat transfer q is also constant.

2. Convection

Convection heat transfer between solid objects (e.g., foliage, buds, and twigs)
and fluids (e.g., air) occurs as a result of relative motion. In the heat transfer lit-
erature, free convection occurs when the motion of the fluid is driven by gra-
dients in temperature (and hence density) within the fluid near the surface of
a solid caused by heat transfer between the solid and fluid. For example, heat
waves rising off deserts during windless and sunny days are a manifestation of
482 Dickinson and Johnson

free convection of heat from the warm ground to the overlying and cooler air.
When the relative motion driving convection heat transfer is not caused by lo-
calized temperature gradients at the surface of an object, the process is termed
forced convection. For instance, forced convection heat transfer occurs when
wind flowing through a canopy carries heat away from the surface of radiatively
heated foliage (e.g., Gates, 1980). In this chapter, we are primarily interested
in forced convection, wherein the causes of relative motion are buoyant flames
and plumes.
Convection heat transfer is inherently more complicated than conduction
because it depends on complex processes occurring within the boundary layer
at the surface of a structure. Convection heat transfer is proportional to (1) the
temperature difference between the free-moving fluid and the surface of a struc-
ture and (2) the characteristics of the boundary layer as described by the con-
vection heat transfer coefficient. Steady state convection heat transfer is defined
by Newton's law of cooling
q = -hA{T, - TJ (5)
where q is the rate of heat transfer (W), h is the convection heat transfer coeffi-
cient (W m~^ • °C), A is the surface area over which convection is occurring
(m^), Tj is the temperature at the surface of the structure, and T^ is the tem-
perature of the free-moving fluid. The negative sign indicates that heat flow (q)
into the structure is positive when the temperature of the free-moving fluid
is greater than that of the structure's surface (i.e., T^ — T^ is negative). Equa-
tion (5) is the defining equation for the convection heat transfer coefficient.

3. Thermal Radiation
All bodies above absolute zero emit thermal radiation according to the Stefan-
Boltzmann law:
E = saT"^ (6)
where E is the emitted radiant energy flux (W m~^) integrated over all wave-
lengths, e is the emissivity (dimensionless), a is the Stefan-Boltzman constant
(5.669 X 10"^ W m~^ • °K^), and T is the temperature (°K). Emissivity is the
ratio of the emissive power of a body to that of an ideal emitter (blackbody) and
is assumed to be a constant over all temperatures and wavelengths.
Because the emitted radiant flux is directly proportional to temperature
[Eq. (6)], a structure that is heated above a fire line reradiates a portion of the
heat it gains from the flame or plume causing a proportional reduction in its
temperature. Under the assumption that a structure is being heated only by ra-
diation, net heat transfer into a structure is proportional to the temperature dif-
ference between the structure and its surroundings:
q„ ex (T^ - T:) (7)
Chapter 14 Fire Effects on Trees 483

where q„ is the net heat transfer rate (W), T is the temperature of the structure,
and Tj is the temperature of the surroundings.

II. EFFECTS OF FIRE ON THE TREE BOLE

In low-intensity surface fires, the fire line passes the bole of a tree, bathing it in
flame (Figure 2). Heat is conducted through the bark into the underlying vas-
cular cambium and, if the temperatures are high enough and flame residence
time is long enough, live tissues within the bole are killed. Low-intensity fires
will often have little effect on the canopy because the temperatures in the plume
at the height of the canopy are too low. However, as fire-line intensity increases,
vascular cambium damage will be accompanied by necrosis of canopy compo-
nents. Later in this chapter, we discuss flame characteristics and model the pro-
cess of heat transfer from the flame into the bark at the base of a tree. Then, we
discuss how one would predict the extent of vascular cambium necrosis around
the bole of a tree. To make such a prediction, one must consider the increase in
flame residence times and temperatures that occur on the leeward sides of trees
in the presence of wind, a process that often causes fire scars in trees that sur-
vive fires.

A. FLAME CHARACTERISTICS

Raising the temperature of the bole involves convection and radiation heat
transfer between the flame and the surface of the bark and conduction through
the bark and underlying wood (e.g., Fahnestock and Hare, 1964). The relatively
simple heat transfer model we present is based on the assumption that the main
source of heat will be flame gases, not radiation, and that the convection heat
transfer coefficient between the flame and the surface of the bark will be very
large because of the steep gradient in temperatures at the bark's surface (see dis-
cussion that follows). With these simplifying assumptions, flame gas tempera-
tures become the primary focus. Unfortunately, predicting and measuring flame
gas temperatures are not straightforward tasks.
Flames propagate through a natural fuelbed primarily by convective and ra-
diative heating of the fuel particles ahead of the fire (e.g.. Chapters 5 and 6 in
this book; Thomas, 1967; Drysdale, 1990). As the flames approach, fuel particle
temperatures increase, moisture is driven off, and solid fuel is converted to gas
(see review in Albini, 1980, and Chapter 3 in this book). Piloted ignition occurs
when the rate of gas liberation is sufficiently high. The rate of mixing of fuel
with ambient air limits the rate of combustion in the fuelbed. The remaining
gaseous fuel enters the zone above the fuelbed, is mixed with air by turbulence
and diffusion, and combusts.
484 Dickinson and Johnson

Flame temperatures are dependent on several processes. The amount of heat

produced by combustion is dependent on the chemical composition of the gas-
eous fuel and the completeness of combustion. Part of the heat from combus-
tion goes to raising the temperatures of the gases in the mix (including air and
water vapor), and another part is lost through radiation and mixing with am-
bient air. Because of the number of variables involved (e.g., Albini, 1980), it is
not yet practical to predict the temperatures of flames burning in natural fuels.
The incorporation of chemical kinetics in fire behavior models is expected to
lead to advances in predicting flame temperatures and other aspects of fire be-
havior (Weber, 1991).
The alternative to predicting flame temperatures from models is to measure
them. However, flame temperatures reported in the literature should be treated
with caution (Martin et al, 1969; Vines, 1981; GiU and Knight, 1991). The tem-
perature measured is that of the measuring device and not of the flame gases.
Correctly estimating gas temperatures requires balancing heat gained by the
measuring device (e.g., a thermocouple) with heat lost:

^cv + ^r = qrr + ^cd (8)

where q^^ is heat gained by convection, q^ is heat gained by radiation, q^^ is heat
lost by re-radiation, and q^^ is heat lost by conduction along the leads of the
thermocouple sensor away from the sensing junction (Martin et al, 1969). Be-
cause it generally has a small effect on the heat budget of the thermocouple, con-
duction along the leads can be ignored. Shielded-aspirated thermocouples are
used to estimate flame gas temperatures because they maximize heat gain by
convection, minimize heat gain by radiation from the flame, and minimize the
effect of radiative heat loss from the thermocouple. Convection heat transfer
scales with the velocity of air flow and is increased as flame gases are drawn past
the thermocouple. Shielding the thermocouple minimizes heat gain from ra-
diation from the flame. To minimize net radiative heat transfer [Eq. (7)] be-
tween the thermocouple and the shield, the flame gases are drawn through con-
centric layers of material surrounding the thermocouple, heating the material
and thereby increasing q^ to a value comparable to q^^.
Martin et al. (1969) report temperatures of flames above a burning crib of
white fir sticks measured either by a shielded-aspirated thermocouple or a ther-
mocouple that was neither shielded nor aspirated. Maximum temperatures mea-
sured were approximately 1100 and 800°C, respectively, the lower value being
one often reported for natural fires.
Because temperatures vary vertically through a flame, it is not immediately
clear at which height temperatures should be estimated. Vertical temperature
distributions through the centerline of the flame and plume (i.e., maximum
temperatures) have been described by Weber et al. (1995); see also Chapter 5.
Maximum temperatures are highest in the fuelbed. Above the fuelbed, tempera-
tures decline exponentially with height through the flame and plume because of
Chapter 14 Fire Effects on Trees 485

Gaussian

FIGURE 4 Top-hat and Gaussian temperature profiles in a plume above a fire line. Similar pro-
files could be fit to the fire line. The top-hat profile can be thought of as an average temperature.

turbulent mixing. Weber et al. (1995) closely described the vertical temperature
profile with semiempirical equations that required minimal field measurements.
To predict heat transfer into the bole, not only must one consider maximum
temperatures but also the time course of temperatures as the flame passes the
bole. Turbulent mixing at the margin of flames leads to a Gaussian rise and fall
in temperatures as a fire passes a given point (Figure 4; see Thomas, 1963; Mc-
Caffrey, 1979). The limit of the visible flame occurs where the temperature rise
above ambient falls below about 500°C (e.g., Thomas, 1963). Flame residence
time Tj (s) is the duration of flaming combustion. Flame residence time can be
calculated from flame width Xj (m) and rate of spread R (m s"^) by the follow-
ing formula:

(9)
R
One can measure residence time, flame width, and rate of spread directly (e.g.,
Rothermel and Deeming, 1980; Simard et al, 1984; McMahon et al; 1986, GiU
and Knight, 1991) or estimate one or more of them from fire-behavior models
(see review in Weber, 1991).
Most interest in fire modeling has been on rate of spread and heat output of
the fire line (i.e., fire-line intensity); as a consequence, models of the residence
time of flaming combustion are not as well developed. Peterson and Ryan (1986)
give an empirical model showing that the residence time of flaming combus-
tion increases with fuel moisture content and decreases with the average fuel
486 Dickinson and Johnson

particle surface area to volume ratio. For a given fuel moisture content, fuelbeds
in which high surface area to volume fuels predominate (e.g., grass) will exhibit
the shortest residence times.
The highest rates of heat transfer into the bark of a tree will occur where
flame temperatures are highest and last the longest. Flames are roughly triangu-
lar in shape, decreasing in width toward the tip (e.g., Thomas, 1963; Nelson,
1980). Because of the greater width and higher maximum temperatures, one
will generally be interested in the lower portion of the flame. As discussed later,
however, bark thickens toward the base of the bole leading to increases in resi-
dence times required to cause vascular cambium necrosis.

B. HEAT TRANSFER INTO THE BOLE

The vascular cambium is killed when subjected to a heat pulse of sufficiently

high temperatures and duration. The vascular cambium is the meristematic tis-
sue that gives rise to the secondary vascular tissues (xylem and phloem). The
bark is usually defined as all the tissue, live and dead, outside the vascular cam-
bium (i.e., including phloem). As a fire line passes a tree, it will rapidly heat the
exterior of the bark. Since wood has low thermal conductivity and the vascular
cambium is relatively near the surface, we assume that the important heating of
the vascular cambium occurs through the overlying bark and not through the
bole itself. The heating will not occur over a long enough time period for a con-
stant temperature gradient to be established in a tree bole, and the bark will
cool after the fire has passed. The lack of a constant temperature gradient char-
acterizes unsteady state heating.
For this general situation, the one-dimensional heat transfer equation (e.g.,
Holman, 1986) is

^ = ^ ^ (10)

where 6 is the temperature rise from ambient (°C), x is depth of the vascular
cambium (m), a is the thermal diffusivity of the bark and underlying wood
(m^ s~^), and r^ is the duration of heating (s). As opposed to the steady state
case where only position x within the material is required [Eq. (1)], both time
T and position must be considered in unsteady state conduction [Eq. (10)].
Equation (10) is one-dimensional in the sense that the temperature gradient
driving heat transfer is directed into the bole from the bark's surface and at right
angles to the bark's surface. The model is applicable to boles to the extent that
bark surface temperatures are even over the surface of the bole and that bole
curvature has minimal effects. Trees with fissured bark (see discussion that fol-
lows) or small diameters may prove to require the consideration of more than
one dimension of heat flow.
Chapter 14 Fire Effects on Trees 487

Equation (10) can be solved by the Laplace transform technique for the fol-
lowing boundary conditions (e.g., Holman, 1986): — hO = —kdd/dxatx = Oand
T > 0, 6 remains finite as x —> CXD and r > 0, and 6 = 6Q = Tj — TQ at r = 0
and 0 < X < oo where h is the convection heat transfer coefficient (W m"^ • °C)
and k is the thermal conductivity (W m~^ • °C). Because the temperature rise is
minimal at the center of the bole, boles heated by flames can be considered to
have infinite depth x (i.e., 0 ^ x < CXD is reasonable). The result is

e
(11)
• 1 - erf j= +

where erf is the Gauss error function whose argument can be found in mathe-
matical tables (e.g., Abramowitz and Stegun, 1964) for a given value of the ex-
cess temperature ratio, defined as

- = (12)

where T is the temperature at depth x (i.e., at the vascular cambium), Tj is the

average flame temperature, and TQ is the ambient temperature. Equation (11) is
called the semi-infinite solid model of transient conduction.
Equation (11) can be simplified by assuming that the convection heat trans-
fer coefficient, h, is infinitely large (i.e., the bark surface temperature is equal to
the flame temperature); it then reduces to

^ = erff ^ = . ) (13)
The vertical velocity of flames and their turbulence ensure that convective heat
transfer rates are high (see discussion that follows), making this assumption
reasonable as a first approximation. Future work on both convection heat trans-
fer between the flame and bole and bark combustion (e.g., Gill and Ashton,
1968; Vines, 1968; Uhl and Kaufmann, 1990; Pinard and Huffman, 1997) wifl
clarify how bark surface temperatures should be treated.
The semi-infinite solid model of transient conduction is a special case of un-
steady state conduction wherein an object's surface temperature increases in-
stantaneously and then remains constant for the duration of heating (this is the
transient temperature trace). In the case of bole heating in fires, the transient
temperature trace has been approximated by an estimate of the average flame
temperature. The temperature profile below the surface of an object at different
times after exposure to a transient temperature trace is shown in Figure 5. After
the flame has passed the bole of a tree, the temperature of the vascular cambium
will continue to rise until the gradient in temperatures at the vascular cambium
488 Dickinson and Johnson

500 - |

400 H

O
o

^ 300 H

0
Q.
E 200 H
,05

100 H

0-1 1 1 r
0 5 10 15
r 25
20
Depth (mm)
FIGURE 5 Temperature profile within a semi-infinite solid (such as a tree bole) 0, 30, 60, and
120 s after being exposed to a transient temperature pulse (i.e., a sudden increase in surface tem-
perature that is sustained indefinitely). Equation (14) was used to calculate temperature profiles for
a bark surface temperature of 500°C, an ambient temperature of 25°C, and a thermal diffusivity of
1.5 X 10"^m^s~\

is reversed [i.e., dT/dx in Eq. (1) becomes positive]. This cooling process can
be included in more complicated models (e.g., Carslaw and Jaeger, 1959; Costa
etal, 1991).
Along with a transient temperature trace, the semi-infinite solid model of
transient conduction [Eq. (13)] assumes that the thermal diffusivities of the
outer and inner bark and underlying wood are equal. Whole bark (inner and
outer) thermal diffusivities appear to be very similar to wood for the species
that have been tested (Reifsnyder et al, 1967). Unsteady state heat transfer mod-
els for layered composites can be used to determine the effects of variability in
thermal diffusivity (e.g., see Carslaw and Jaeger, 1959, for analytical models and
heat transfer texts, Holman, 1986, for numerical approaches).
Equation (13) has been used to explain heating of the vascular cambium in
fires (e.g., see Spalt and Reifsnyder, 1962; Martin, 1963; Vines, 1968; Peterson
and Ryan, 1986; Brown and DeByle, 1987; Outsell and Johnson, 1996). The
model can be used to predict the time required to cause necrosis of the vascular
cambium in a tree of a given bark thickness. To illustrate, we will set T to the
temperature at which vascular cambium necrosis would occur over a short pe-
riod (e.g., 65°C), the ambient temperature to 20°C, and the average flame tem-
perature to 500°C. The excess temperature ratio becomes
Chapter 14 Fire Effects on Trees 489

^ 6 5 - 5 0 0
— = = 0.8958 (14)
^0 20 - 500

and, from the tables for this value of the error function,

-- 1.15 (15)

where r^ is the flame residence time that would cause vascular cambium necro-
sis (i.e., the critical residence time). After inserting a value for the thermal dif-
fusivity, (e.g., 1.35 X 10~^ m^ s"^; Spalt and Reifsnyder, 1962) and rearrang-
ing, Eq. (15) reduces to

T, = 1.4 X l O V (16)

From Eq. (16), one would expect that the critical residence time r^ would
scale as the square of bark thickness. Hare (1965), Vines (1968), and Gill and
Ashton (1968) present vascular cambium temperature data from heating trials
that can be roughly compared with the model's prediction (see Figure 6 for de-
tails). Vines (1968) used a radiant heater and attempted to prevent bark com-
bustion; the averaged data from five species yielded a scaling exponent of 1.2.
Gill and Ashton (1968) used a radiant heater and allowed bark combustion to
occur for three species; the average scaling exponent was 1.1 (range 1.0-1.2).
Hare (1965) used a propane torch to heat the bark of a number of species; the
average scaling exponent was 2.3 (range 2.1-2.5; Figure 6). Hare's (1965) study
best fits the transient temperature trace assumption of the semi-infinite solid
model; thus, it is not entirely surprising that his scaling exponents are closest to
the model's prediction. Costa et al. (1991) related Finns ^pinaster vascular cam-
bium temperatures during surface fires to predictions from a numerical model
that used bark surface temperature traces and included the effect of bole cur-
vature. Their model suggests that the time to reach a given temperature scales
as the 3.2 power of bark thickness.

1. Thermal Tolerance of Cells and Tissues

Tissue and cell necrosis occur at exponentially increasing rates as temperatures
rise. These rates have been described empirically using the Arrhenius equation
of physical chemistry (e.g., Johnson et al., 1974; Castellan, 1983):

da
—— = A exp (17)
dr diT
where 11 is a damage index (dimensionless), r is time (s), A (s~^) is called the
frequency factor, E (J mol"^) is termed the activation energy, St is the gas con-
stant (8.314 J m o F ^ • K), and T is temperature (K). The activation energy E and
490 Dickinson and Johnson

200

(0
150
<D
E
•t—

Q)
O 100
r
0
TD
cn
0
DC
50 i

- T — I I I
0.0 0.2 0.4 0.6 0.8 1.0

X^

Longleaf and slash pine

Loblolly pine and bald cypress
Southern magnolia and sweetbay
Red maple, water oak, dogwood, tupelo, and birch
^ ^ Sweetgum, American holly, and black cherry
FIGURE 6 Heating times required to raise vascular cambium temperatures to 40°C scaled by the
square of bark thickness as suggested by the semi-infinite solid model of transient conduction
[Eq. (17); data from Hare (1965), reprinted from the Journal of Forestry 63 (4): 248-251, published
by the Society of American Foresters, 5400 Grosvenor Lane, Bethesda, MD 20814-2198. Not for
further reproduction]. Heating was done with a propane torch and species whose bark heated at
similar rates were grouped. The near linearity of the relationships suggests that, for this situation,
the model captures the most important aspects of the heat transfer problem.

the frequency factor A are typically estimated from regression analyses and have
been found to be nearly constant within limited temperature ranges (e.g., from
ambient to ~70°C). The group E/?flT is dimensionless and is called the Arrhe-
nius number. Damage CI to plants from exposure to elevated temperatures has
been expressed as cell death (e.g., Lorenz, 1939; Dewey et at, 1977; Levitt,
1980), reduction in cellular respiration (Caldwell, 1993), cessation of cytoplas-
mic streaming (Alexandrov, 1964), and protein coagulation resulting from pro-
tein denaturation (Levitt, 1980). When applied to complex systems such as
cells and tissues, the meaning of the activation energy E and the frequency fac-
tor A are not entirely clear (e.g., Levitt, 1980), though, for animal cell death,
these parameters, when suitably scaled, correspond to those expected for pro-
tein denaturation (Rosenberg et ah, 1961).
The Arrhenius equation [Eq. (17)] can be integrated over a given tempera-
ture profile to provide predictions of damage H (see Martin et al, 1969, and
Chapter 14 Fire Effects on Trees 491

Chapter 7 in this book for a related approach). Predicting the effects of un-
steady state temperature traces in this way have been of particular interest in
the study of human skin burns (e.g., Henriques, 1947; Takata, 1974; Diller and
Hayes, 1983; Diller et al, 1991). When suitably parameterized, Eq. (17) could
serve as the basis for predicting such things as the proportion of phloem cells
surviving in the live bark or the extent to which cellular respiration rates are
reduced by a heat pulse. Equation (17) provides a basis for the finding that the
rates of damage rise approximately exponentially with temperature to the point
at which tissue death is effectively instantaneous (e.g.. Nelson, 1952; Dewey et
al, 1977; Levitt, 1980). It is this critical temperature that justifies the use of the
relatively simple semi-infinite solid model of transient conduction [Eq. (13)] to
predict vascular cambium necrosis.

2. Bark Thickness
Bark thickness varies among species, with tree age, with height along the bole,
and with bole diameter. Bark thickness measured at breast height is often
nearly linearly related to stem diameter and, consequently, can be expressed as
a percentage of stem diameter (Table 1; see also Spalt and Reifsnyder, 1962;
Hengst and Dawson, 1994; Pinard and Huffman, 1997). Age has been found to
affect bark thickness with older stems having thicker bark than younger stems
for a given diameter (Hale, 1955; Spalt and Reifsnyder, 1962).
Bark near the base of trees may often be thicker than one would predict from
a linear relationship between diameter and bark thickness (e.g.. Hale, 1955;
Glasby et al, 1988). Because of the power-law scaling between the critical time
for vascular cambium necrosis and bark thickness (see earlier discussion),
basal thickening of the bark leads to substantially longer critical residence
times (Figure 7; see also Brown and DeByle, 1987).
The one-dimensional, semi-infinite solid model of transient conduction as-
sumes that bark fissuring has minimal effects. During fires, lower bark surface
temperatures have been measured in fissures than on bark plates (e.g., Fahne-
stock and Hare, 1964). It follows that fissuring must lead to deeper boundary
layers and, thus, lower convection heat transfer rates [Eq. (5)]. Consequently,
lower bark surface temperatures in fissures counteract the effects of the thinner
bark. Two- or three-dimensional numerical methods could be used to explore
the effects of bark fissuring (e.g., Holman, 1986), allowing one to determine
when it must be included in models of the bole-heating process.

3. Thermal Diffusivity
Thermal diffusivity quantifies the rate at which a temperature wave penetrates
a material during transient conduction or, in other words, the ease with which
a material absorbs heat from its surroundings. When an object is subjected to
492 Dickinson and Johnson

TABLE 1 Bark Thickness as a Percentage of Diameter at Breast Height

for Selected Species from Two Regions

Region Species Bark (%)

Northern Rocky Mountains, Larix occidentalis 7.4

United States" Pseudotsuga menziesii 6.7
Populus halsamifera 5.9
Abies grandis 4.3
Pinus ponderosa 4.1
Thuja plicata 2.5
Pinus contorta 1.6
Ahies lasiocarpa 1.5
Picea engelmannii 0.7
Eastern Amazonia, Brazil'' Manilkara huberi 9.1
Lecythis lurida 8.5
Lecythis idatimon 7.7
Cordia sericalyx 7.6
Cecropia sciacophylla 5.8
Xylopia aromatica 5.6
Inga alba 4.8
Dipopyros duckei 4.7
Macrolobium angustifolium 4.4
Metrodorea flavida 4.4
Tetragastris altissima 4.3
Jacaranda copaia 4.1
Pourouma guianensis 3.9
Inga sp. 3.5
Ecclinusa sp. 2.3

"Finch (1948) in Spalt and Reifsnyder (1962).

''Uhl and Kauffman (1990).

a rapid increase in temperature (transient conduction), it is necessary to quan-

tify not only the temperature gradient and thermal conductivity [as in the steady
state, Eq. (1)] but also the heat required to raise the temperature of the mate-
rial. This is determined by the material's heat capacity c (J kg~^ • °C) and den-
sity (kg m~^), where heat capacity c is the amount of heat required to raise the
temperature of a unit mass of material by one degree. Thus, thermal diffusivity
(m^s-^)is

(18)
pc
where the denominator is referred to as the volumetric heat capacity (J m~^ • °C).
For tree boles, thermal diffusivity in the semi-infinite solid model of transient
conduction [Eq. (13)] is that of the bark and underlying wood. Wood and bark
Chapter 14 Fire Effects on Trees 493
3-1 | 5 0 - 2 5 0 \1000s
'- \

\
\
- 2H 100.500 \ 2000 s
o
c
o

I 1 H

V500 -1500 \ 8000 s

1 "T 1 1
0 10 20 30 40 50
Bark thickness (mm)

8.5 cm d.b.h., c. 10 yr. old

15 cm d.b.h., c. 25 yr. old
108cmd.b.h.,c. lOOyr. old
FIGURE 7 Predicted flame residence times required to cause necrosis of tlie vascular cambium
at increasing heights along the boles of three Eucalyptus oreades trees of different sizes and ages.
Bark thickness data from Glasby et al. (1988). Critical residence times were calculated from Eq. (14)
for a bark surface temperature of 500°C, a critical temperature for tissue necrosis of 65°C, an am-
bient temperature of 25°C, and a thermal diffusivity of 1.5 X 10"^ m^ s"\

have been found to be similar in their thermal properties (Martin 1963; Reif-
snyder et al, 1967); as such, we will focus on bark. Bark is a mixture of mate-
rials including outer bark, inner bark, air, and water. The different effects of
these components on bark thermal conductivity, density, and heat capacity
must be accounted for in calculations of thermal diffusivity (see Martin, 1963;
Reifsnyder et al, 1967). Table 2 contains values of the properties of barks and
other materials. Air has a relatively low thermal conductivity but a high thermal
diffusivity because of its low density. On the other hand, water and bark have
similarly low thermal diffusivities, water because of its high volumetric heat ca-
pacity and bark because of its low thermal conductivity.
Standard methods for measuring thermal diffusivity and its components are
given in the American Society for Testing and Materials handbooks (e.g., ASTM,
1998). Thermal diffusivity can be estimated by the flash method in which one
side of a thin sample is subjected to a short pulse of radiant energy (e.g., Taylor,
1979). The ensuing temperature rise on the back side of the sample is measured
494 Dickinson and Johnson

TABLE 2 Thermal Conductivity fe, Heat Capacity c, Density p, and Thermal Diffusivity a
of Various Materials at Ambient Temperatures (~20°C)''

k c P a
Material (Wm''-°C) (Jkg-^-°C) (kgm-^) (m^ s"' X 10^)

Water 0.614 4179 998 1.47

Bark
Shortleaf pine 0.100 1348 497 1.5
Longleaf pine 0.105 1268 520 1.6
Martin (1963) — — — 1.2-1.6
Reifsnyder et al. (1967) — — — 0.6-1.6
Wood
Longleaf pine 0.133 1411 540 1.7
Shortleaf pine 0.116 1369 458 1.8
Moist sand 0.52 — — 4.81
Moist silty clay loam 0.82 1720 1250 3.81
Iron 73 452 7897 204
Air 0.026 1006 1.18 219

''Single values for shortleaf and longleaf pine bark and wood are averages for ovendried condi-
tions (Reifsnyder et al, 1967). The ranges in thermal diffusivity of bark are for moisture contents
from < 1 to about 30% moisture. Martin's (1963) range in thermal diffusivities is for samples from
a group of temperate zone species, and Reifsnyder et aVs (1967) is again for shortleaf and longleaf
pine samples. Sand properties are from Van Wijk and Bruijn (1964), and silty loam properties are
from Asrar and Kanemasu (1983). Values for other materials are from Holman (1986).

and, with these data, an unsteady state heat transfer equation is solved for
thermal diffusivity. A benefit of the flash method is that the size of the sample
can be small. As far as we are aware, this convenient method has not been used
for bark.
Alternatively, thermal conductivity and heat capacity can be estimated sep-
arately, and thermal diffusivity can be calculated from Eq. (18). Thermal con-
ductivity can be measured by steady state or transient methods. Martin (1963)
and Reifsnyder et al. (1967) used the transient "hot wire" method in which the
temperature rise in an electrically heated wire imbedded within the sample is
greatest when the thermal conductivity is least. The value of thermal conduc-
tivity is calculated by solving a transient heat transfer model of radial heat flow
containing the thermal properties of the probe and sample. Steady state meth-
ods include the use of a guarded-hot-plate apparatus in which a given heat flux
produces a measured temperature difference across a sample (ASTM, 1998). A
steady state conduction equation based on Fourier's Law [e.g., Eq. (1)] is then
solved to determine the thermal conductivity. Heat capacity is generally mea-
sured by means of a differential scanning calorimeter (ASTM, 1998) wherein a
sample and a reference material are heated at a constant rate by using separately
controlled resistance heaters. The differential heat flow into or out of a sample
Chapter 14 Fire Effects on Trees 495

is compared with the reference material to determine the heat capacity or the
heat capacity plus any applicable latent heats.
In studies of a group of temperate conifer and deciduous species, Martin
(1963) and Reifsnyder et ah (1967) found that bark density and moisture con-
tent are the primary correlates of bark thermal diffusivity. Martin's (1963) equa-
tion for thermal conductivity is based on data from the inner and outer bark of
ten species (both deciduous and conifer) over a range in moisture contents from
0 to 114% of dry weight. Tests were done at 25°C. The equation (converted to
Wm"i-°C)is

k = -0.00846 + O.OOOllOp + 0.000554p^ (19)

where p is bulk density (kg m~^) calculated as the oven dry weight divided by
the volume at the moisture content at which the test was done (so as to account
for shrinkage upon drying). Martin (1963) assumed a 13.6% expansion in vol-
ume between dry and wet bark. The moisture density p^ (kg m~^) is the por-
tion of bark density accounted for by moisture. Equation (19) accounted for
98% of the variation in thermal conductivity. Reifsnyder et al.'s (1967) equation
for the whole bark of three Pinus species (longleaf, shortleaf, and red pine) gives
very similar results. Thermal conductivity within bark is nearly equal along the
radial and transverse planes (Martin, 1963).
Heat capacity of bark is affected by moisture content because of the high heat
capacity of water (Table 2) and because, when a sample is heated, energy is
absorbed as the chemical bonds between water and cellulose are broken (this
is the heat of desorption, Byram et al, 1952b; Martin, 1963; Reifsnyder et al,
1967). Thus, bark heat capacity (J kg~^ • °C) is composed of the following com-
ponents (Martin, 1963):

c = Qb + Mc^ + Ac (20)
where QJ, is the heat capacity of dry bark, M is the moisture content as a fraction
of dry weight, c^ is the heat capacity of water (from tables), and Ac is the ele-
vation of heat capacity from the desorption of bound water. The heat capacity
of dry bark was similar to that of wood and was estimated by Martin (1963) as

c^i, = 1105 + 4.856T (21)

where T is temperature (°C). The elevation in heat capacity was given as follows:

Ac = 1277M, 0 < M < 0.27 (22)

Ac = 348, M > 0.27 (23)
Martin (1963) found no significant differences among species in heat capacity
of dry bark.
496 Dickinson and Johnson

• >

It

Moisture content (%)

FIGURE 8 The effect of moisture and density p (kg m~^) on the thermal diffusivity of shortleaf
and longleaf pine bark. From Reifsnyder et al. (1967).

Bark thermal diffusivity decreases as moisture and density increase (Fig-

ure 8). Bark moisture varies among species depending on the relative amount
of inner and outer bark and the drying characteristics of the outer bark (e.g.,
Reifsnyder et al, 1967). Bark densities can vary considerably, for example, Mar-
tin and Crist (1967) give a range of 280-1290 kg m~^ for individual oven dry
samples from 19 species. Variation in moisture content has less of an effect on
thermal diffusivity than might be imagined. Bark thermal conductivity and vol-
umetric heat capacity both increase with moisture content, but volumetric heat
capacity increases at a faster rate resulting in a small net decrease in thermal
diffusivity (Figure 8).
In our discussion of thermal diffusivity, we have implicitly assumed that it
will be constant as temperatures rise over the course of heating [this is also an
assumption of the semi-infinite solid model of transient conduction, Eq. (13)].
This is clearly not true; for instance, a substantial heat sink is inherent in both
the desorption of bound water and the evaporation of free water (e.g., Byram
et al, 1952b). In addition, thermal conductivities and heat capacities of bark,
water, and air increase while their densities decrease with temperature. A first
Chapter 14 Fire Effects on Trees 497
approximation of the heat sink associated with the evaporation of free water is
to cap the surface temperature to which an object is exposed at 100°C (Mercer
et al, 1994). In fact, bound water undergoes desorption, and free water is evapo-
rated continuously as temperatures rise (e.g., Byram et al, 1952b). Modehng
continuous variation in thermal diffusivity requires more complicated heat
transfer models than Eq. (13) and would appear to yield relatively little predic-
tive benefit at present.

4. Predicting the Extent of Vascular Cambium Necrosis around the Bole

Given that the assumptions of the semi-infinite solid model of transient con-
duction [Eq. (13)] are approximately met, it is possible to estimate the residence
time of the flame that would cause necrosis of the vascular cambium around the
entire circumference of the bole (i.e., girdling). Girdling is predicted if the resi-
dence time of flaming combustion is greater than the critical residence time for
vascular cambium necrosis. In Figure 9, residence times [Eq. (9)] were calcu-
lated from flame width and rate of spread for 50 experimental fires in Australian

5 10 15 20 m
25 "1
^ i 1 \
20 H

15 A

CD
-Q
E
10 4
3

5-1 h
M
r\ 1 •
U 1 J_j.j_j j_
•. ,
^
IL
m ^
^ 1
0 100 200 300 400 500 600
Residence time (s)
FIGURE 9 Comparison of fire residence times and critical residence times for bole vascular cam-
bium necrosis. The frequency distribution gives the residence times of 50 experimental fires in
Eucalyptus sieheri forest. Data from Cheney et al. (1992). The arrows indicate the critical residence
times that would girdle trees with bark of the indicated thickness. Critical residence times were cal-
culated from Eq. (14) for a bark surface temperature of 500°C, a critical temperature for cambial
necrosis of 65°C, an ambient temperature of 20°C, and a thermal diffusivity of 1.35 X 10"'^ m^ s~^
498 Dickinson and Johnson

Eucalyptus sieberi forests (data from Cheney et al, 1992). Given our model of
vascular cambium necrosis [e.g., Eq. (13)] and our chosen values of flame tem-
peratures and thermal diffusivities (Figure 9), only two of these fires had resi-
dence times sufficient to have girdled trees with bark thickness > 1 cm.
If the residence time of a given fire is insufficient to girdle a tree, partial vas-
cular cambium necrosis may still occur. When a fire is burning in wind, heating
on the downwind (leeward) side of a tree bole is either equal to or greater than
heating on the upwind (windward) side. If the unequal heating is sufficient to
cause vascular cambium necrosis on the leeward side, but not on the windward
side, and if the tree survives the fire, a fire scar results. Outsell and Johnson
(1996) provide a mechanistic hypothesis of fire-scar formation based on fluid
dynamics and heat transfer processes that fits various observations about stand-
ing leeward flames and fire scars.
A model of fire-scar formation must account for three observations about fire
scars: (1) fire scars are found only on the leeward sides of trees; (2) small trees
rarely have fire scars; and (3) fire scars are usually triangular in shape, becom-
ing narrower with height. Fire scars are created when fires burn in an ambient
wind because, it is hypothesized, the airflow around the bole of a tree creates a
pair of vortices. In the presence of the vortices, a standing leeward flame devel-
ops as the free-moving flame passes a tree. The standing leeward flame has a
longer residence time and higher temperatures than the free-moving flame, ac-
counting for the greater heat transfer into the leeward bark. Below, leeward vor-
tices and the standing leeward flame are discussed in sequence.
As air flows around the bole of a tree, a reverse flow occurs producing a pair
of vortices (Figure 10a-10c). The formation of the vortices depends on the
magnitude of the bole Reynolds number
uDp ^ ^
Re - — ^ (24)

where D is the characteristic dimension (tree diameter, m), u is the horizontal

wind speed (m s~^), p is the density of the air (kg m~^), and JUL is the dynamic
viscosity of the air (kg m~^ • s). The dynamic viscosity of air increases with
temperature as the fire approaches the tree, but, because they vary over a much
larger range, the wind speed and tree diameter have much larger effects on the
Reynolds number. Note that, with a constant wind speed, trees with smaller di-
ameters will have lower Reynolds numbers than larger trees.
When the Reynolds number is less than about 5, the flow around the tree is
laminar and unseparated from the bole (Figure 10a; see, e.g., Swanson, 1970).
At higher Reynolds numbers, the flow separates from the bole and a pair of ad-
jacent vortices forms against the leeward side of the tree (Figure 10b). The
points of separation define the maximum size of the fire scar. At a Reynolds
Chapter 14 Fire Effects on Trees 499

Re < 5 laminar unseparated flow

5 < Re < 40 pair of attached vortices

40 < Re < 150 laminar vortex street

150 < Re < 300 transition from laminar to

turbulent vortex street
y 300 < Re < 3x10' fully turbulent vortex street

2==C3 3x10' < Re < 3.5x10' transition from laminar to

turbulent boundary layer

3x10^ < Re re-establishment of

turbulent vortex street

FIGURE 10 Flow regime around a cylinder as a function of Reynolds number. From Outsell and
Johnson (1996).

number of about 40, a waviness in the wake behind the bole increases in am-
phtude until it becomes unstable and breaks. The result is a "vortex street"
wherein each vortex of the pair is alternately shed into the wake behind the tree
to be reformed anew against the bole (Figure 10c). The frequency of the shed-
ding is dependent on the Reynolds number. As the Reynolds number increases
further, there are other transitions, for example to turbulent flow (Figures lOd
and lOe) and to a more downwind point at which the flow separates from the
bole (Figure lOf).
The patterns of flow in each leeward vortex are fundamentally the same as
those in fire whirls, tornados, and dust devils (Emmons and Ying, 1967). Ide-
ally, each vortex consists of two zones of flow (Figure 11): an outer zone of ir-
rotational flow and an inner zone, or core, of rotational flow (see, e.g., Eskinazi,
1962). In the outer zone, the flow is translated around the core with no change
in the orientation of the air particles. For this orientation to be maintained, the
500 Dickinson and Johnson

rotational

irrotational

FIGURE 11 Irrotational and rotational zones of flow in a vortex. In the flow, particles change ori-
entation in the rotational zone, while their orientation does not change in the irrotational zone.
From Outsell and Johnson (1996).

tangential velocity of the air particles farthest away from the core must be lower
than that of the air particles closest to the core (Figure 12). For irrotational flow
outside the vortex core, the radius r and tangential velocity u^ yield a constant
(i.e., ru,).
In the rotating core, the tangential velocity and radius have the opposite re-
lationship. Velocity increases linearly with radius and ujr is a constant. Conse-
quently, particles in the rotational zone change their orientation as they rotate
(Figure 11). Rotation in the core is termed solid body rotation because, as in a
rotating solid, the relative distances among particles do not change with time.
The pattern of rotational and irrotational flow in a vortex play an important role
in explaining the development of the standing leeward flame as a free-moving
flame passes a tree.
The residence time of the standing leeward flame is approximately double
that of the free-moving flame. Figures 13a-e shows the development of the
standing leeward flame as the free-moving flame passes by a tree. As soon as the
front of the flame reaches the center of the periphery of the tree (Figure 13b),
the flame is drawn horizontally into the cores of the leeward vortices, increasing
the residence time by approximately 0.5 tree diameters. The standing leeward
flame persists until the trailing edge of the fire leaves the region of the vortices
Chapter 14 Fire Effects on Trees 501
Rotational Irrotatlonal
flow flow

100 200 300 400

Radial distance (mm)
FIGURE 12 The tangential velocity (uj) distribution in a vortex. Tangential velocity increases
from the center of the vortex to the outer boundary of the rotational zone and then decreases
through the zone of irrotatlonal flow. Reprinted from Chigier et ah, Combustion and Flame 14,
171-180, Copyright 1970, with permission from Elsevier Science.

on the leeward side of the tree (Figure 13e), increasing flame residence time by
a further 1.5 tree diameters. Consequently, the residence time of the standing
leeward flame is given by

T,= ? + ? (25)
R R
where 2D/R is the increase in residence time accounted for by the effects of the
leeward vortices [cf. Eq. (9)].
The horizontal draw that increases the residence time of the standing lee-
ward flame results from a "centrifugal pump effect" caused by the relatively low
pressures at the axes of each vortex core (see Chapter 5). The low pressure
within a vortex core is produced by the increase in angular momentum from
the axis to the edge of the rotating core. Friction within the boundary layer
slows the rotation at the base of the vortex allowing boundary-layer air to be
drawn into the relatively buoyant vortex cores.
The centrifugal pump effect is augmented by two additional sources of buoy-
ancy. The first source is the warm gases drawn into the core through the bound-
502 Dickinson and Johnson

a) b)
windward leeward

standing
free- tree bole leeward flame
moving
flame

back front region

of fire of fire of leeward
vortices

standing standing
leeward flame leeward flame

e)

^ 4
region
of leeward
vortices
FIGURE 13 The effect of leeward vortices on a free-moving flame (from Gill, 1974, reprinted
from Forest Science 20 (3): 198-205, published by the Society of American Foresters, 5400 Gros-
venor Lane, Bethesda, MD 20814-2198. Not for further reproduction). When the leading edge of a
free-moving flame reaches the center of the periphery of a tree (a), the flame is drawn into the lee-
ward vortices, producing a standing leeward flame (b). The standing leeward flame continues to
increase in height (c) and then recedes as the free-moving flame passes the tree (d). Once the trail-
ing edge of the free-moving flame is beyond the leeward vortices, the standing leeward flame has
completely receded (e).

ary layer as the free-moving flame approaches the tree. Second, after the flame
is drawn into the vortices, combustion adds substantially to the buoyancy within
the vortex cores; this mechanism has been called the chimney effect (Fahne-
stock and Hare, 1964).
Chapter 14 Fire Effects on Trees 503
The chimney effect is made possible by a reduction in turbulent mixing be-
tween the core and outer irrotational zone. Angular momentum increases from
the core's axis to its outer edge, dampening turbulence within the core. Con-
comitantly, horizontal movement of air into the core is opposed. As a conse-
quence, the buoyancy force within the core is not rapidly decreased by mixing,
resulting in a considerable increase in flame height. In addition, flame height
increases because the flow of gaseous fuel in the vortex cores is greater than the
rate of mixing with the surrounding air, and, hence, there is a greater height
along which combustion can take place.
Differential heating around the base of a tree results not only from the in-
creased residence time of the standing leeward flame relative to the free-moving
flame but also from its elevated temperatures (Figure 14). The cause of the
elevated temperatures appears to be that the rotation of the fuel in the vortex
cores increases the combustion rate of fuel and air.
The typical triangular shape of fire scars can be explained by considering the
three-dimensional temperature profile of the standing leeward flame. Based on

20-

15H

E
o
D
O)
"CD • D
I
• D
5H • D
• D
• D
• D
• D

—r-^—^ \ 1
200 400 600
Temperature (°C)

• Standing leeward flame

• Free-moving flame
FIGURE 14 Vertical temperature profiles of a standing leeward flame and a free-moving flame.
Both flames are of the same average height (indicated by the arrows). Average temperatures in the
standing leeward flame are higher than those of the free-moving flame at any given height. From
Outsell and Johnson (1996).
504 Dickinson and Johnson

Radius (mm)
FIGURE 15 The two-dimensional temperature distribution (°C) in a cross section of a standing
leeward flame. The outer margin of the flame extends approximately to the outermost isotherm.
The temperature is highest at the center of the flame and decreases toward the sides and top. From
GutseU and Johnson (1996).

experimental results in a wind tunnel (GutseU and Johnson, 1996), the tem-
peratures at the surface of a cylinder were found to be highest at the middle of
the standing leeward flame (Figure 15). The temperatures decreased with height
after reaching a peak at around 40% of flame height. This resulted in triangu-
lar isotherms, wide near the base of the cylinder and decreasing in width towards
the flame tip.
Fire scars will not occur on small trees and at low wind speeds because the
Reynolds numbers are too low for the formation of leeward vortices [Eq. (24)].
Unequal heating between the windward and leeward sides of trees will also not
occur if the rate of spread of the free-moving flame is high. This can be appre-
ciated by considering Eq. (25). As the rate of spread increases, the difference
between the residence time of the free-moving flame and that of the standing
leeward flame disappears because 2D/R (the increase in residence time on the
leeward side of the tree because of vortex formation) becomes small.
Chapter 14 Fire Effects on Trees 505

Ultimately, it may prove that relatively simple fluid dynamics and heat trans-
fer models will be adequate to predict the extent of vascular cambium necrosis
around tree boles during wildfire. Thermal tolerance models [e.g., Eq. (17)] will
also be required when estimates of such things as cell survivorship and physi-
ologial impairment are desired.

III. EFFECTS OF FIRE ON

CANOPY COMPONENTS
To understand how fires cause tissue necrosis in tree canopies, one needs to
consider not only the process of heat transfer into canopy components (foliage,
buds, twigs, fruits, and seeds) but also the characteristics of the plume (Fig-
ure 2). Modeling necrosis of canopy components is more involved than mod-
eling bole cambial necrosis because it is not reasonable to assume, as we did for
tree boles being heated by flames, that convection heat transfer rates are high.
Convection heat transfer coefficients are expected to vary considerably with the
characteristics of the air flow at different heights within a plume and with the
size and shape of the canopy component. Predicting convection heat transfer
into canopy components requires estimates of plume temperatures and vertical
velocities at different heights and the residence times of those temperatures.

A. PLUME CHARACTERISTICS

Buoyant plumes are wedge-shaped, turbulent columns of hot, low-density air

that rise above heat sources. As plumes rise, they mix with the cooler and
denser ambient air (e.g., Morton et al, 1956; Taylor, 1961). Because of this tur-
bulent mixing, plume width increases with height, and plume temperatures
and average gas velocities decrease with height. Ambient winds bend plumes,
reducing the temperatures reached at a given height above ground while, at the
same time, increasing the residence times of those temperatures. Plumes above
fire lines burning in surface fuels are assumed to approximate line sources of
heat at sufficient heights within the canopy (e.g.. Van Wagner, 1973). Two-
dimensional plume models form the basis of the following discussion and are
more thoroughly described in Chapter 7. Three dimensions are required to de-
scribe more complex plume behaviors such as vorticity that are key to under-
standing the dynamics of blowup fires (see Chapters 7 and 8).
Steady state, two-dimensional plumes can be approximately described by a
top-hat profile wherein plume characteristics are averaged over the width of the
plume (Figure 4; see also Morton et al, 1956). More realistically, average tem-
peratures and vertical velocities rise gradually from the margin of the plume to
506 Dickinson and Johnson

the centerline (the Gaussian profile). The Gaussian profile results from the lat-
eral entrainment of ambient air into the rising plume.
In the relatively simple heat transfer model we introduce to describe heating
of canopy components, a steady state, top-hat temperature profile drives heat-
ing, and the associated vertical velocity profile is used to estimate convection
heat transfer coefficients. The top-hat plume structure has appeal for heat trans-
fer modeling because it is mathematically simpler to treat top-hat profiles than
Gaussian ones. For plumes in an ambient wind, Gaussian and top-hat mod-
els have been shown to give very similar predictions of plume characteristics
(Davidson, 1986).
Mercer and Weber (1994; see Chapter 5 in this book) provide models of a
line-source, top-hat plume in both still air and a cross wind. Their steady state,
two-dimensional model employs coupled partial differential equations that bal-
ance the fluxes of mass, momentum, and heat between the plume and the am-
bient air. The equations are called conservation equations because the dynamic
and thermodynamic properties of a rising, entraining plume are governed by
conservation of mass, momentum, and energy. These equations can be solved
to derive estimates of plume temperatures, velocities, and widths at different
heights which are then used to evaluate the heating of canopy components.

B. HEAT TRANSFER INTO CANOPY COMPONENTS

The height at which foliage necrosis (i.e., crown scorch) occurs can be predicted
from an equation that estimates the height within the canopy at which a lethal
temperature for foliage necrosis is reached (see Van Wagner, 1973). The lethal
temperature was defined as that at which foliage necrosis would occur at short
exposure times (e.g., 1 minute). No heat transfer model is employed in this ap-
proach. The assumption, of course, is that foliage has little resistance to convec-
tion and conduction and, thus, that the plume temperature is a good approxi-
mation of the temperature of the foliage.
This approach has been surprisingly useful despite differences in foliage
shape, size, display, and, consequently, convection heat transfer coefficients (e.g,
Peterson and Ryan, 1986; Reinhardt and Ryan, 1988; Finney and Martin, 1993;
Gould et al, 1997). For example. Van Wagner (1973) found that the follow-
ing equation (see Taylor, 1961; Thomas, 1964) closely predicted scorch height
Z (m) for a broadleaf and several pine species (see Figure 16):
.j2/3

(26)

where I is fire-line intensity (kW m~^), j is a proportionality constant that must

be estimated horn experimental data, and T^ and Tg are the plume and ambient
Chapter 14 Fire Effects on Trees 507

ZD-

z = 0.094 r y«

15- R' = 0.98

•>
•
•
10-
X

5-

n -
u 1 1 1 1

0 50 100 150 200

I "' (kW m")
FIGURE 16 The height of crown scorch ^ as a function of the two-thirds power of fire-line in-
tensity [see Eq. (27)]. The plume temperature was set to 60°C, the temperature assumed to cause
foliage necrosis after approximately one minute of heating. Data from Van Wagner (1973).

temperatures, respectively. Often, the temperature difference is implicitly in-

cluded in the proportionality constant j (e.g.. Figure 16).
Because no heat transfer model is employed, using Van Wagner's (1973) ap-
proach has several limitations. First, plume temperature alone does not provide
any understanding of how the plume is causing necrosis of the canopy compo-
nent. All the biology is in the heat transfer coefficients of conduction and con-
vection and in the response of the tissue to the temperature pulse! Second, vali-
dation is difficult because the proportionality constant incorporates both heat
transfer processes and plume behavior. Third, several kinds of canopy compo-
nents are more resistant to heating than foliage (e.g., buds, branches, and fruits).
Let us now formulate a simple heat balance for a canopy component so that
we can use it to determine the time required for necrosis in a wildfire plume.
We have specifically chosen a simple model which should be a good first ap-
proximation for small canopy components (e.g., twigs, foliage, buds, flowers,
fruits, and seeds). In the heat balance, heat is convected onto the component's
surface at a rate equal to the increase in internal energy of the component. The
convection (i.e., Newton's law of cooling) and volumetric heat capacity terms
in the heat budget are, respectively.

-hA{T - TJ = cpV— (27)

where c is the heat capacity (kj kg ^ • °C), p is the density (kg m ^), V is the
volume to be heated (m^), r is time (s), h is the convection heat transfer co-
508 Dickinson and Johnson

efficient (W m~^ • °C),A is the surface area over which convection occurs (m^),
T is the temperature of the structure, and T^ is the plume temperature (°C).
Defining 6 = T — T^ and dT/dr = dOldr and then rearranging gives

dr pcV
a first order, ordinary differential equation that can be solved for ^ as a function
of time for the initial condition T = TQ at r = 0. Separating variables and inte-
grating gives
hA
^ = B exp T (29)
pcV
where B is a constant of integration. At r = 0,B must be equal to the difference
between the ambient temperature TQ and the plume temperature T^. Thus, the
equation can be expressed as

hA
(30)
^0 ~ ^oo ^0 pcV J
The equation can be solved for the time required for the canopy component to
reach the critical temperature for tissue necrosis
pcV
T = Wln^o-ln^) (31)

when estimates of the dimensions and heating properties of the component are
available, the plume and ambient temperatures are known, and T is set to a
critical temperature for tissue necrosis (e.g., 65°C).
The coefficient hA/pcV in Eq. (30) is the thermal time constant (s~^) because
it gives the time required for 0 to reach 37% of OQ. The thermal time constant
will vary depending on the biological and physical properties of the canopy
component. For example, the heat capacity and density depend on the kind of
material and its water content, the surface area and volume depend on the size
and shape of the component, and the convection heat transfer coefficient de-
pends on the size, shape, and orientation of the component along with its ar-
rangement relative to other components. Ecologists thus will be interested in
how these values change relative to each other because they describe the means
by which individuals and species differ in how they are affected by fire.
How good is such a simple model? As far as we are aware, there are no tests
of the model for canopy components in fires. However, using results derived
from laboratory heating trials, Johnson and Outsell (1993) used the model to ex-
plain patterns of cone opening in two serotinous-coned Pinus species. In cones,
resin bonds between scales hold the cones closed until they are heated, gener-
ally by fire. W h e n the bonds are broken, the scales reflex allowing seeds to be
Chapter 14 Fire Effects on Trees 509
Dimensionless temperature

.68 .85 .91 .93 .95 .96 .96 .97

1.5 T
I

O Pinus contorta
O # Pinus banksiana
Q
E o
(0 1.0 -
CO
CD
c o
o o

0.5

o od ^ ^ ^ ^ ^
0.0
100 200 300 400 500 600 700 800 900

Temperature (°C)
FIGURE 17 Dimensionless time to cone opening as a function of temperature for two serotinous-
coned conifer species. From Johnson and Outsell (1993).

released. The resins have a meking point (slightly higher than 100°C) that is
substituted for T in Eq. (30). Other parameter values in Eq. (30) were also esti-
mated from the literature. The dimensionless time to cone opening QiA/pcV • r)
at different temperatures in a furnace follov^ed an exponential relationship as
suggested by the model (Figure 17).
The simplicity of Eq. (30) derives from the fact that only the time depen-
dence of temperature is considered and the dependence of temperature on po-
sition within the object is ignored. Because temperature is treated as uniform,
the thermal conductivity is not included in Eq. (30), and the model is called
the lumped-heat-capacity model. Foliage and small branches, twigs, buds, flow-
ers and seeds are thermally thin (i.e., their surface temperatures are close to
their interior temperatures under heating). Accordingly, the lumped-parameter
method may often be a useful approximation considering errors in measure-
ment and in estimation of some of the parameters, particularly the convection
heat transfer coefficient (h).
510 Dickinson and Johnson

As an object becomes larger, it becomes more resistant to being heated by

conduction. We can follow a formal procedure for deciding if the temperature
gradient within an object is small enough to ignore and, consequently, whether
Eq. (30) is applicable. For there to be a minimal internal temperature gradient,
the resistance to conduction must be small relative to the resistance to convec-
tion (e.g., Sucec, 1985); that is,

— «: 1 (32)
K
where R^ is the internal resistance to heating by conduction and R^ is the resis-
tance to convection heat transfer from the outside. These resistances can be cal-
culated by the following equations:

L 1
RiI = — R„ = — (33)
M' " hA.
where L is an average path length over which conduction occurs from the cen-
ter of the object to the surface, and A is the surface area over which convection
occurs (not necessarily the entire surface area A^). The term hA^ determines the
convection heat transfer rate and is familiar from Eq. (5). Substituting Eq. (33)
into Eq. (32) gives

where D has the dimension of length and is called the characteristic dimension
of the canopy component. For our purposes, we define D as the maximum lin-
ear distance along which heat is conducted. For example, for a twig, D would
be bark thickness (i.e., depth of the vascular cambium), and, for a bud, D would
be the depth of the meristem.
The expression on the left of the inequality in Eq. (34) is called the Biot num-
ber (Bi) and is dimensionless. How much less than one should the Biot number
be? W h e n Bi < 0.1, the difference in temperature between a given depth within
the object of interest and its surface is always < 5 % percent (e.g., Sucec, 1985).
When the temperature difference is so small, there is little accuracy lost in us-
ing Eq. (30).

1. Convection Heat Transfer Coefficients

Modeling convection heat transfer [e.g., Eqs. (5) and (31)] requires estimates
of convection heat transfer coefficients h for the canopy components of inter-
est. For simple geometries and flow characteristics, analytical and semianalyti-
cal calculations of h can be made. For the somewhat irregular canopy compo-
Chapter 14 Fire Effects on Trees 511

nents in the turbulent flow of the plume, the use of correlations derived from
extensive experimentation would be expected to give more reliable results. We
show one approximate analytical solution and then introduce the experimen-
tal approach.
As air flows past an object, its velocity approaches zero at the object's sur-
face. Accordingly, the convection heat transfer rate [Newton's law of cooling,
Eq. (5)] must equal the rate of conduction [Fourier's law, Eq. (1)] through this
still layer:

-hA{T, - T^) = -kA^ (35)

where A is the surface area over which heat transfer is occurring, T^ is the sur-
face temperature, T^ is the free-stream temperature, k (W m~^ • C) is the ther-
mal conductivity of the air, and dT/dx is the temperature gradient at the surface
of the object (i.e., the depth x within the object is zero). Rearranging and sim-
plifying gives

-k—
dx
n= ' (36)
^ oo ^ s

when the gradient in temperature dT/dx is steep at the surface of the object, h
is large. The steepness of the temperature gradient dT/dx depends on conduc-
tion heat transfer and fluid dynamics within the boundary layer.
Equation (36) can be solved after a treatment of the boundary-layer condi-
tions (e.g., see Holman, 1986). For instance, the solution for the average con-
vection heat transfer coefficient over the length of a flat plate when flow within
the boundary layer is laminar is given in dimensionless form as
Nu = 0.664 Re^^^ Pr^^^ (37)
where Nu is the average Nusselt number. Re is the Reynolds number, and Pr is
the Prandtl number (all three variables are dimensionless). The Nusselt num-
ber contains the average convection heat transfer coefficient h (W m~^ • °C):

Nu = ^ (38)
k
where D (m) is the characteristic dimension relative to the flow of air (e.g, length
of the plate in this case) and k is the thermal conductivity of air (W m~^ • °C).
The Reynolds number [Eq. (24)] gives the ratio of inertial forces to viscous
forces. At high Reynolds numbers, inertial forces predominate and fluid flow in
the boundary layer is turbulent, resulting in steep temperature gradients and
high convection heat transfer rates.
512 Dickinson and Johnson

The Prandtl number relates the thickness of the fluid dynamic and thermal
boundary layers:

Pr = - = - ^ = ^ (39)

pc
where v is the kinematic viscosity (/x/p, m^ s~^ and a is the thermal diffusiv-
ity of the fluid Qi/pc, m^ s~^). High kinematic viscosities (v) mean that viscous
forces will be felt farther away from the surface of an object. At the same time,
high thermal diffusivities {a) mean that temperature influences from the mo-
lecular transport of heat will be felt farther out in the flow field. The Prandtl
number is thus the connecting link between the velocity field and the temper-
ature field above the surface of an object.
Equation (37) can be solved for the convection heat transfer coefficient by

h = 0 . 6 6 4 ( - j Re^^'Pr^^^ (40)

Because Eq. (40) includes the dimensionless variables that are most important
in determining convection heat transfer rates, it serves as the point of departure
for experimental determinations of convection heat transfer coefficients. For
example. Gates (1980) uses Eq. (40) and experimental determinations of the
proportionality constants to estimate convection heat transfer coefficients for
small leaves
Nu = 1.86Re^^^Pr^/^ (41)
and large leaves
Nu = l.lBRe^^^Pr^/^ (42)
Although the equations fit data from a variety of sources (including field data),
there is some disagreement on the extent to which the equations are valid over
the range of wind turbulence seen within canopies (Schuepp, 1993). Wind tur-
bulence can increase boundary-layer turbulence and, thus, momentum and
convection heat transfer rates.
Few studies have been conducted on convection heat transfer for canopy
components other than broad leaves. However, branches and conifer needles
approach a cylindrical shape, and buds and fruits approach a spherical shape.
Experimental correlations for smooth cylinders and spheres are available in the
literature (e.g., see references in Sucec, 1985). The equations are similar in
form to Eq. (37).
Chapter 14 Fire Effects on Trees 513

Mutual sheltering by aggregated canopy components would tend to reduce

convection heat transfer coefficients for individual leaves, needles, twigs, and
buds. For example, long leaf pines, Pinus palustris (e.g., Andrews, 1917; Wahl-
enberg, 1946; Byram, 1948) and Australian grass trees, a branching, upright
monocot, Xanthorrhoea australis (Gill and Ingwersen, 1976), are famous for
the protection of their meristems by dense foliage. An empirical solution to this
problem has been suggested for convection heat transfer between wind flowing
through the canopy and conifer needles (Monteith and Unsworth, 1990; Niko-
lovetal., 1995):

Nu, = (43)
' 2.1
where the subscript s refers to sheltered and i to isolated needles. The denom-
inator is termed the shelter factor and is an index of the reduction in flow ve-
locity experienced by foliage for forced convection in laminar flow.

2. Predicting Height of Foliage, Bud, and Twig Necrosis

To predict the height of foliage, bud, and twig necrosis, one must take into ac-
count the fact that the fire moves through the fuelbed resulting in a certain resi-
dence time of plume temperatures. The residence time of a top-hat plume in
still air is proportional to the width of the plume (Xp) and the rate of spread of
the fire line [cf. Eq. (9)]:

r,{z) = ^ (44)

where r^ is plume residence time (s) at height z (m) and R is the rate of spread
of the fire line (m s"^). See Chapter 7 for a discussion of residence times for
plumes in a cross wind.
Necrosis of a given canopy component will depend on plume temperature,
the velocity of the plume (through its effect on convection heat transfer coeffi-
cients) , and plume residence time at a given height above a fire line. Plume mod-
els (see Chapter 7) can be used in conjunction with heat transfer and thermal-
tolerance models to calculate the height within the canopy below which necrosis
of a given canopy component would occur. The dimensions of canopy compo-
nents will follow some distribution (e.g.. Corner, 1949; Ackerly and Donoghue,
1998), and, as a first approximation, the average dimension of the canopy com-
ponent may suffice in calculations of the height of necrosis.
Vascular cambium necrosis in the bole is expected to co-vary with damage
514 Dickinson and Johnson

to canopy components. Within species, the thickness of the bark often scales
positively with the height of the canopy. This, along with the scaling of flame
and plume characteristics, would be expected to result in a positive relationship
between the extent of damage to the vascular cambium and the extent of dam-
age to canopy components. The relationship between vascular cambium and
canopy damage would be expected to differ among species because species vary
considerably in (1) the heating properties of their bark and canopy components
and (2) the allometric relationships among tree size, bark thickness, and the
distribution of canopy components.

IV. ROOT NECROSIS

Because soils have low thermal diffusivities (Table 2) relative to the duration
of flaming, glowing, and smoldering combustion, soil heating, like bark heat-
ing, has been modeled as an unsteady state process (e.g., Aston and Gill, 1976;
Steward et al, 1990; Peter, 1992; Oliveira et al, 1997). To predict soil heating,
one must first describe heat flux at the soil surface. Where smoldering does not
occur, this flux is driven by flaming and glowing combustion (Peter, 1992; Oli-
veira et al., 1997). From an analysis of convection and radiation heat transfer,
Oliveira et al. (1997) conclude that radiation will dominate downward heat flux.
Van Wagner (1972) estimated downward radiation heat flux from the Stefan-
Boltzman law and assumptions about flame and ambient surface temperatures
[e.g., Eqs. (6) and (7)]. After the flame has passed, the soil surface cools by con-
vection and radiation (Peter, 1992; Oliveira et al, 1997).
Because of their porosity, heat is transferred through dry soils by conduction,
convection, and radiation. In saturated soils, heat is transferred by conduction
alone (De Vries, 1963; Peter, 1992). The thermal conductivity of dry and satu-
rated soils can be approximated by averaging the effects of minerals, air, and
water (De Vries, 1963). In moist soils (i.e., soils that are not saturated), down-
ward heat transfer from the soil surface occurs not only by conduction but also
by transport of latent and sensible heat through mass transfer of water in both
its liquid and vapor phases (e.g., De Vries, 1958; Westcot and Wierenga, 1974;
Aston and Gill, 1976). Water vapor diffusion through porous soils can account
for a considerable fraction of the downward heat flux (e.g., Westcot and Wie-
renga, 1974). Near the surface in moist soils, the latent heat of vaporization con-
strains soil temperature rise to 80-100°C until liquid water is evaporated (As-
ton and Gifl, 1976; Peter, 1992).
Where an organic layer is absent or, where present, does not smolder, em-
pirical measurements and mechanistic models indicate that maximum soil tem-
peratures typically do not rise above 60°C below 5 cm depth (Aston and Gill,
1976; Raison et al, 1986; Hungerford, 1989; Steward et al, 1990; Bradstock and
Chapter 14 Fire Effects on Trees 515

Auld, 1995; Oliveira et al, 1997). An organic layer that does not combust insu-
lates the underlying mineral soil, but smoldering increases the downward heat
flux (e.g., Swezy and Agee, 1991; Hartford and Frandsen, 1992; Peter, 1992).
Where organic soil layers are present, predictive models of root and basal cam-
bium necrosis will depend on an understanding of the processes that determine
the rates of spread and patchiness of organic layer combustion (see Chapter 13).
There are few studies on necrosis of roots and basal vascular cambium from
soil heating and organic layer combustion. Using soil heat transfer models.
Steward et al. (1990) and Peter (1992) simulated the depth of lethal heat pene-
tration by assuming that roots would die instantaneously if the temperatures in
the surrounding soil reached 60°C. Swezy and Agee (1991) found extensive
fine-root necrosis in surface soils after smoldering combustion, although it is
not clear how much of the necrosis was from smoldering combustion itself and
how much was an indirect response to bud necrosis in the canopy or necrosis
of live bark and sapwood tissues. Long durations of elevated temperatures at the
bases of Ponderosa pines accompanied smoldering combustion of duff, but the
heating did not lead to extensive vascular cambium necrosis (Ryan and Frand-
sen, 1991). Ryan and Frandsen hypothesized that temperatures at the vascular
cambium were constrained by the removal of heat in the mass flow of water in
the sapwood (see also Vines, 1968).

V. TREE MORTALITY
Even though mechanistic models are being developed to link fire behavior and
tissue necrosis (e.g., Peterson and Ryan, 1986; Outsell and Johnson, 1996), little
advancement has been made on drawing mechanistic links between patterns of
tissue necrosis and tree death. Current models are typified by logistic regres-
sions in which a continuous probability function is derived from binary tree
death data (i.e., trees are either dead or alive at some time after fire). The equa-
tions have the following general form:

p = I (-45)
1 + e x p [ —(bo + biXi + bjXi + ••• -h bjXi)]

where P^ is the probability of tree death, the x/s are independent variables, and
the bj's are empirically derived coefficients. Generally, independent variables are
descriptors or surrogates of fire behavior (e.g., fire-line intensity, fuel consump-
tion) and fire effects (e.g., crown scorch, bark char). Equation (45) is not a
mechanistic model because the independent variables and the relationships
among independent variables are chosen with little or no regard to the actual
processes by which either fire causes tissue necrosis or tissue necrosis causes
tree death. Because they do not include processes, current models lack gen-
516 Dickinson and Johnson

erality and cannot be applied beyond the specific conditions on which they
were based.
To understand the process of tree death, we expect that a useful starting
point will be to consider the effects of fires on tree carbon budgets (not a new
suggestion, e.g., Ryan et at, 1988; Harrington, 1993; Glitzenstein et al, 1995).
Trees die when they are unable to acquire or mobilize sufficient resources to
maintain normal function in the face of injury and stress (see Waring, 1987).
It is hypothesized that trees allocate available carbohydrates in a hierarchical
manner: new foliage, buds, and roots have the highest priority while remaining
resources are allocated to stored reserves, growth, and defensive compounds
(Waring and Pitman, 1985; Christiansen et al, 1987; Waring, 1987). Trees that
are likely to die often show departures in carbon allocation patterns relative to
healthy trees.
Imbalances in carbon budgets have been found either to cause death directly
or to precipitate cascades of events that ultimately result in death. In a direct
manner, Douglas fir death following tussock moth defoliation was shown to de-
pend on the extent and duration of the defoliation and the amount of stored
carbohydrate reserves that could be mobilized to renew lost foliage (e.g., Webb,
1981). Imbalances in a tree's carbon budget caused by such stresses as prolonged
drought, defoliation, and suppression may often predispose a tree to being killed
(e.g., Pedersen, 1998; Waring and Pitman, 1985; Christiansen et al, 1987;
Dewar, 1993; Gottschalk et al, 1998; Davidson et al, 1999). For example,
through effects on carbon budgets, defoliation by Gypsy moths makes trees
more susceptible to stem boring insects and root pathogens, albeit by mecha-
nisms that are not entirely straightforward (see review in Davidson et al, 1999).
As with tree death resulting from causes other than fire, tree death from fire-
caused necrosis in the canopy, roots, and bole can be considered from a carbon-
budget perspective. Foliage, bud, and branch necrosis, in analogy to insect
defoliation, would be expected to reduce carbon fixation rates and thereby in-
crease the likelihood that a tree will die. Indeed, the degree of crown scorch
emerges from statistical models [e.g., Eq. (45)] as one of the more consistent
predictors of tree death (e.g.. Van Wagner, 1973; Ryan et al, 1988; Ryan and
Reinhardt, 1988; Finney and Martin, 1993; Harrington, 1993; Regelbrugge and
Conard, 1993). Root necrosis from fire would be expected to affect rates of soil-
resource acquisition and postfire carbon allocation patterns. Phloem and vascu-
lar cambium necrosis in the bole of a tree would be expected to impede trans-
port processes. For example, fire scars were found to be related to crown die
back in mature giant Sequoia trees (Sequoiadendron giganteum), presumably be-
cause fire scars reduced the cross-sectional area of sapwood (Rundel, 1972).
Various features of fire-caused tree death, including season-of-burn effects (e.g.,
Ryan et al, 1988; Harrington, 1993; Ghtzenstein et al, 1995) and differences
among taxa in susceptibility to fire, may prove to be quantitatively predictable
by linking heat transfer, tissue necrosis, and carbon-budget models.
Chapter 14 Fire Effects on Trees 517

VI. DISCUSSION
In this chapter, we have illustrated with relatively simple models the kinds of
mechanistic linkages that we believe are required to develop an understanding
of tissue necrosis and tree death from fires. In the following, we will look at a
series of papers from the mainstream ecological literature that we believe would
have profited from a more mechanistic approach. If anything, we hope to reiter-
ate that simple mechanistic models help to clarify the study of fire effects.
Mutch (1970) suggested that " . . . fire-dependent plant communities burn
more readily than non-fire-dependent communities because natural selection
has favored development of characteristics that make them more flammable."
Testing the Mutch hypothesis requires an understanding of flammability. Flam-
mability has a precise definition for premixed flames (i.e., flames in which the
fuel and air are intimately mixed before ignition; see Chapter 2 in this book).
For premixed flames, flammability refers to successful ignition of the fuel-air
mixture by a pilot flame, ignition being dependent on pressure, temperature,
and the ratio of fuel to air. Flammability is not defined for diffusion flames, such
as those of wildfires, where the fuel and air are initially separated. In the con-
text of forest fires, flammability is a vague term that encompasses most aspects
of fire dynamics (see discussion in Anderson, 1970). Being so broad, it has little
practical use and appears only to generate confusion.
Williamson and Black (1981) and Rebertus and Williamson (1989) found
that, after surface fires, oak stem and whole-plant mortality were greater adja-
cent to canopy pines compared with areas away from pines where the canopy
was dominated by oaks. They attributed this result to the higher maximum tem-
peratures measured near pines during fires. Their conclusions are not strongly
supported because their maximum temperatures are ambiguous, and maximum
temperatures constitute a weak link between fire behavior and tree-stem and
whole-plant mortality.
It is not clear which of several possible maximum temperatures Williamson
and Black (1981) and Rebertus and Williamson (1989) intended to measure.
Accurately estimating temperatures requires that one balance a heat budget for
the measuring device [see Eq. (8) and Chapter 2 in this book]. Maximum tem-
peratures were estimated from waxes of a range of melting points applied to the
surface of ceramic tiles strung at different heights above the ground. There was
no attempt to account for the heat budget of the tiles, consequently; the tem-
peratures have an unknown relationship to flame and plume temperatures. If
one were interested in the maximum temperatures of canopy components (e.g.,
foliage and buds), one would have to match the heating properties of the mea-
suring device with those of the canopy component. However, the heating prop-
erties of the tiles do not appear to mimic any tree component.
Maximum temperatures, even if accurately measured, have limited meaning
in terms of heat transfer into trees. Given the tacit assumption by Williamson
518 Dickinson and Johnson

and Black (1981) and Rebertus and Williamson (1989) that bole heating was
not important, we must determine heat transfer from the plume into the foliage,
buds, and branches. The lumped-heat-capacity model [Eq. (30)] suggests that
average temperatures, vertical velocities, and residence times determine heat
transfer from plumes into canopy components. Maximum temperatures con-
stitute an incomplete description of plumes relative to heat transfer and, con-
sequently, tell us little about the temperatures experienced by canopy compo-
nents. With data of unknown relevance to tissue necrosis, the papers provide a
weak link between fire behavior and tree-stem and whole-plant mortality.
Jackson et al. (1999) explore the relationship between bark thickness and tree
survival after fires. In ecosystems characterized by short-interval, low-intensity
fires, species tended to develop thicker bark early in their ontogeny relative to
species occupying ecosystems in which fires occurred at long intervals. As sug-
gested by a simulation model, the explanation for this result was that higher in-
vestment in bark where fires were of low intensity would increase tree survival
but would not affect tree survival where fires were intense because, in intense
fires, crown necrosis would render bole tissue necrosis irrelevant. A key com-
ponent of the simulation model was the relationship between bark thickness
and the probability that a tree would survive a fire. Heat transfer models would
appear to have allowed Jackson et al. (1999) to constrain the number of pos-
sible relationships. Power-law scaling between bark thickness and critical resi-
dence time (e.g., Figure 6) suggests that minimal investment in bark results in
large increases in protection.
Using mechanistic models requires that ecologists and foresters acquire
skills outside the traditional realm of biology. However, by looking more closely
at, for example, heat transfer into the vascular cambium around the boles of
trees during fires, we find that the biology is in the thickness of the bark, the
shape of the bark surface, the combustion properties of the bark, and the ther-
mal properties of the bark and underlying wood. When mechanistic models
guide our research, we can hope to pose clear and testable hypotheses regard-
ing fire effects.

NOTATION
ROMAN LETTERS

A area m^
A frequency factor s~^
h regression coefficient undefined
c heat capacity Jkg~^«°C
D characteristic dimension m
Chapter 14 Fire Effects on Trees 519

E radiant energy flux Wm~^

E activation energy J mol-^
h heat transfer coefficient Wm"^-°C
I fire-line intensity kWn
k Wm -I or
thermal conductivity
L length m
M moisture fraction dimensionless
Pd probability of death dimensionless
q heat transfer rate W
R fire-line rate of spread m s~^
Ri resistance to conduction
K resistance to convection
T temperature °C, °K
u velocity m s~^
V volume m^
X depth or width m
z height m

GREEK LETTERS

a thermal diffusivity m^s-^

s emmisivity dimensionless
temperature difference
e dynamic viscosity kg m~^ • s
density kgm~^
p moisture density kgm"^
Pm time s
T kinematic viscosity m^s-i
V damage index dimensionless
n
DIMENSIONLESS GROUPS

Bi Biot number
Nu Nusselt number
5 20 Dickinson and Johnson

Pr Prandtl number
Re Reynolds number

CONSTANTS

B integration constant
J proportionality constant
gas constant 8.314Jmor^-K
a Stefan-Boltzman constant 5.668 X 10"^Wm"^-K^

SUBSCRIPTS

c critical value
cd conduction
cv convection
db dry bark
f flame condition
i isolated condition
n net
P plume condition
r radiation
rr re-radiation
s surface or sheltered condition
w water
0 ambient condition
oo free stream condition

ADDITIONAL READINGS
Monteith, J. L., and Unsworth, M. H. (1990). "Principles of Environmental Physics." Edward Ar-
nold Publishers, London.
Sucec, J. (1985). "Heat Transfer." Wm. C. Brown Publishers, Dubuque.
Swanson, W. M. (1970). "Fluid Mechanics." Holt, Rinehart, and Winston, New York.
Chapter 14 Fire Effects on Trees 521

REFERENCES
Abramowitz, M., and Stegun, I. A. (1964). "Handbook of Mathematical Functions." Applied Mathe-
matics Series No. 55. U.S. Department of Commerce, National Bureau of Standards, Washing-
ton, DC.
Ackerly, D. D., and Donoghue, M.J. (1998). Leaf size, sapling allometry, and Corner's Rules: phy-
logeny and correlated evolution in maples (Acer). Am. Naturalist 152, 767-791.
Albini, F. A. (1980). "Thermochemical Properties of Flame Gases from Fine Wildland Fuels." Re-
search Paper INT-243. USDA Forest Service, Intermountain Forest and Range Experiment Sta-
tion, Ogden.
Alexandrov, V. Y. (1964). Cytophysiological and cytoecological investigations of heat resistance of
plant cells toward the action of high and low temperature. Quart. Rev. Biol. 39, 35-77.
Anderson, H. E. (1970). Forest fuel ignitibility. Fire Technol. 6, 312-319.
Andrews, E. F. (1917). Agency of fire in propagation of longleaf pines. Bot. Gaz. 64, 497-508.
Asrar, G., and Kanemasu, E. T. (1983). Estimating thermal diffusivity near the soil surface using
Laplace transform: Uniform initial conditions. Soil Sci. Soc. Amer.J. 47, 397-401.
ASTM. (1998). "Annual Book of ASTM Standards." American Society for Testing and Materials,
West Conshohocken, PA.
Aston, A. R., and Gill, A. M. (1976). Coupled soil moisture, heat and water vapour transfers under
simulated fire conditions. Aus.J. Soil Res. 14, 55-66.
Bradstock, R. A., and Auld, T. D. (1995). Soil temperatures during experimental bushfires in relation
to fire intensity: Consequences for legume germination and fire management in south-eastern
Australia. J. Appl. Ecol. 32, 76-84.
Brown, J. K., and DeByle, N. V. (1987). Fire damage, mortality, and suckering in aspen. Can. J. For.
Res. 17, 1100-1109.
Byram, G. M. (1948). Vegetation temperature and fire damage in the southern pines. Fire Control
Notes 9, 34-36.
Byram, G. M., Eons, W L., Sauer, F. M., and Arnold, R. K. (1952a). "Thermal Conductivity of Some
Common Forest Fuels." USDA Forest Service, Division of Fire Research, California Forest and
Range Experiment Station, Berkeley.
Byram, G. M., Eons, W. L., Sauer, F. M., and Arnold, R. K. (1952b). "Thermal Properties of Forest
Fuels." USDA Forest Service, Division of Fire Research, California Forest and Range Experiment
Station, Berkeley.
Caldwell, C. R. (1993). Estimation and analysis of cucumber (Cucumis sativus L.) leaf cellular heat
sensitivity. Plant Physiol. 101, 939-945.
Carslaw, H. S., and Jaeger, J. C- (1959). "Conduction of Heat in Solids." Clarendon Press, Oxford.
Castellan, G. W (1983). "Physical Chemistry," 3rd ed. Addison-Wesley, London.
Cheney, N. P., Gould, J. S., and Knight, 1. (1992). "A Prescribed Burning Guide for Young Regrowth
Forests of Silvertop Ash." Research Paper No. 16. Forestry Commission of New South Wales,
Research Branch, Sydney.
Chigier, N. A., Beer, J. M., Grecov, D., and Bassindale, K. (1970). Jet flames in rotating flow fields.
Combustion and Flame 14, 171-180.
Christiansen, E., Waring, R. H., and Berryman, A. A. (1987). Resistance of conifers to bark beetle
attack: Searching for general relationships. For. Ecol. Manage. 22, 89-106.
Corner, E.J. H. (1949). The Durian theory of the origin of the modern tree. Ann. Bot. 13, 367-414.
Costa, J. J., Oliveira, L. A., Viegas, D. X., and Neto, L. P. (1991). On the temperature distribution
inside a tree under fire conditions. Int. J. Wildland Fire 1, 87-96.
Davidson, G. A. (1986). Gaussian versus top-hat profile assumptions in integral plume models.
Atmos. Environ. 20, 471-478.
Davidson, C. B., Gottschalk, K. W, and Johnson, J. E. (1999). Tree mortality following defoliation
522 Dickinson and Johnson

by the European gypsy moth (Lymantria dispar L.) in the United States: A review. For. Sci. 45,
74-84.
De Vries, D. A. (1958). Simultaneous transfer of heat and moisture in porous media. Trans. Amer.
Geophys. Union 39, 909-916.
De Vries, D. A. (1963). Thermal properties of soils. In "Physics of Plant Environment" (W. R. Van
Wijk, Ed.), pp. 210-235. North-Holland Publishing Company, Amsterdam.
Dewar, R. C. (1993). A mechanistic analysis of self-thinning in terms of the carbon balance of trees.
Ann.Bot. 71, 147-159.
Dewey, W. C, Hopwood, L. E., Sapareto, S. A., Gerweck, M. S., and Gerweck, L. E. (1977). Cellular
responses to combinations of hyperthermia and radiation. Radiology 123, 463-474.
Diller, K. R., and Hayes, L.J. (1983). A finite element model of burn injury in blood-perfused skin.
J. Biomech. Eng. 105, 300-307.
Diller, K. R., Hayes, L. J., and Blake, G. K. (1991). Analysis of alternate models for simulating ther-
mal burns. J. Burn Care Rehah. 12, 177-189.
Drysdale, D. (1990). "An Introduction to Fire Dynamics." John Wiley and Sons, New York.
Emmons, H. W., and Ying, S.J. (1967). The fire whirl. In "Eleventh Symposium (International) on
Combustion," pp. 475-488. The Combustion Institute, Pittsburgh.
Ennos, A. R. (1993). The scaling of root anchorage. J. Theoret. Biol. 161, 61-75.
Eskinazi, S. (1962). "Principles of Fluid Mechanics." Allyn and Bacon, Boston.
Fahnestock, G. R., and Hare, R. C. (1964). Heating of tree trunks in surface fires. J. For. 62, 799-
805.
Finch, T. L. (1948). Effect of bark growth in measurement of periodic growth of individual trees.
Research Note 60. USDA Forest Service, Northern Rocky Mountain Forest and Range Experi-
ment Center, Missoula.
Finney, M. A., and Martin, R. E. (1993). Modeling effects of prescribed fire on young-growth coast
redwood trees. Can. J. For. Res. 23, 1125-1135.
Gates, D. M. (1980). "Biophysical Ecology." Springer Verlag, New York.
Gill, A. M. (1974). Toward an understanding of fire-scar formation: Field observation and labora-
tory simulation. For. Sci. 20, 198-205.
Gill, A. M., and Ashton, D. H. (1968). The role of bark type in relative tolerance to fire of three cen-
tral Victoria eucalypts. Aus.J. Bot. 16, 491-498.
Gill, A. M., and Ingwersen, F. (1976). Growth oi Xanthorrhoea australis R. Br. in relation to fire.
J.Appl.Fcol. 13, 195-203.
Gill, A. M., and Knight, 1. (1991). Fire measurement. In "Conference on Bushfire Modeling and Fire
Danger Rating Systems" (N. P. Cheney and A. M. Gill, Eds), pp. 137-146. CSIRO Division of
Forestry, Yarralumla, ACT.
Glasby, P, Selkirk, P M., Adamson, D., Downing, A. J., and Selkirk, D. R. (1988). Blue Mountains
ash (Eucalyptus oreades R. T. Baker) in the western Blue Mountains. Proc. Linnean Soc. New
South Wales 110, 141-158.
Ghtzenstein, J. S., Piatt, W. J., and Streng, D. R. (1995). Effects of fire regime and habitat on tree
dynamics in north Florida longleaf pine savannas. Ecol. Monogr. 65, 441-476.
Gottschalk, K. W., Colbert, J. J., and Feicht, D. L. (1998). Tree mortality risk of oak due to gypsy
moth. Eur. J. For. Pathol. 28, 121-132.
Gould, J. S., Knight, I., and Sullivan, A. L. (1997). Physical modeling of leaf scorch height from pre-
scribed fires in young Eucalyptus sieheri regrowth forests in south-eastern Australia. Int. J. Wild-
land Fire 7, 7-20.
Gutsell, S. L., and Johnson, E. A. (1996). How fire scars are formed: Coupling a disturbance pro-
cess to its ecological effect. Can.]. For. Res. 26, 166-174.
Hale, J. D. (1955). Thickness and density of bark: Trends of variation for six pulpwood species.
Pulp Paper Mag. Can. 56, 113-117.
Chapter 14 Fire Effects on Trees 523

Hare, R. C. (1965). Contribution of bark to fire resistance of southern trees. J. For. 63, 248-251.
Harrington, M. G. (1993). Predicting Pinus ponderosa mortahty from dormant season and growing
season fire injury. Int. J. Wildland Fire 3, 65-72.
Hartford, R. A., and Frandsen, W. H. (1992). When it's hot, it's h o t . . . or maybe it's not! (Surface
flaming may not portend extensive soil heating). Int. J. Wildland Fire 2, 139-144.
Hengst, G. E., and Dawson, J. O. (1994). Bark properties and fire resistance of selected tree species
from the central hardwood region of North America. Can. J. For. Res. 24, 688-696.
Henriques, F. C, Jr. (1947). Studies of thermal injury V: The predictability and the significance of
thermally induced rate processes leading to irreversible epidermal injury. Arch. Pathol. 4 3 , 4 8 9 -
502.
Holman, J. R (1986). "Heat Transfer." McGraw-Hill Book Company, New York.
Hungerford, R. D. (1989). Modeling the downward heat pulse from fire in soils and in plant tissue.
In "Proceedings of the 10th Conference on Fire and Forest Meteorology," pp. 148-154. Fores-
try Canada, Ottawa.
Jackson, J. F., Adams, D. C , and Jackson, U. B. (1999). Allometry of constitutive defense: A model
and a comparative test with tree bark and fire regime. The American Naturalist 153, 6 1 4 -
632.
Johnson, E. A., and Gutsell, S. L. (1993). Heat budget and fire behavior associated with the open-
ing of serotinous cones in two Pinus species. J. Veg. Sci. 4, 745-750.
Johnson, F. H., Eyring, H., and Stover, B. J. (1974). "The Theory of Rate Processes in Biology and
Medicine." John Wiley and Sons, New York.
Levitt, J. (1980). "Responses of Plants to Environmental Stresses, Volume 1, Chilling, Freezing, and
High Temperature Stresses," 2nd ed. Academic Press, New York.
Lorenz, R. W. (1939). "High Temperature Tolerance of Forest Trees." Technical Bulletin 141. Min-
nesota Agricultural Experiment Station, St. Paul.
Martin, R. E. (1963). Thermal properties of bark. For. Prod. J. 13, 419-426.
Martin, R. E., and Crist, J. B. (1967). Selected physical-mechanical properties of eastern tree barks.
For. Prod. J. 18,54-60.
Martin, R. E., Cushwa, C. T., and Miller, R. L. (1969). Fire as a physical factor in wildland manage-
ment. In "Proceedings of the Ninth Annual Tall Timbers Fires Ecology Conference," pp. 2 7 1 -
288. Tall Timbers Research Station, Tallahassee.
McCaffrey, B.J. (1979). "Purely Buoyant Diffusion Flames: Some Experimental Results." NBSIR 79-
1910. United States Department of Commerce, National Bureau of Standards, Washington, DC.
McMahon, C. K., Adkins, C. W, and Rodgers, S. L. (1986). A video mage analysis system for mea-
suring fire behavior. Fire Manage. Notes 47, 10-15.
Mercer, G. N., and Weber, R. O. (1994). Plumes above line fires in a cross wind. Int. J. Wildland Fire
4, 201-207.
Mercer, G. N., Gill, A. M., and Weber, R. O. (1994). A time-dependent model of fire impact on seed
survival in woody fruits. Aus.J. Bot. 42, 71-81.
Monteith, J. L., and Unsworth, M. H. (1990). "Principles of Environmental Physics." Edward Ar-
nold, London.
Morton, B. R., Taylor, G. L, and Turner, J. S. (1956). Turbulent gravitational convection from main-
tained and instantaneous sources. Proc. Roy. Soc. Lond. A 234, 1-23.
Mutch, R. W (1970). Wildland fires and ecosystems—A hypothesis. Ecology 51, 1046-1051.
Nelson, R. M. (1952). "Observations on Heat Tolerance of Southern Pine Needles." Research Paper
14. USDA Forest Service, Southeast Forest Experiment Station, Asheville.
Nelson, R. M., Jr. (1980). "Flame Characteristics for Fires in Southern Fuels." Research Paper SE-
205. USDA Forest Service, Southeast Forest Experiment Station, Asheville.
Niklas, K. J. (1992). "Plant Biomechanics: An Engineering Approach to Plant Form and Function."
The University of Chicago Press, Chicago.
524 Dickinson and Johnson

Niklas, K.J. (1994). The scahng of plant and animal body mass, length, and diameter. Evolution 48,
44-54.
Nikolov, N. T., Massman, W. J., and Schoettle, A. W. (1995). Coupling biochemical and biophysi-
cal processes at the leaf level: An equilibrium photosynthesis model for leaves of C3 plants. Eco-
logical Modeling 80, 205-235.
Oliveira, L. A., Viegas, D. X., and Raimundo, A. M. (1997). Numerical predictions on the soil ther-
mal effect under surface fire conditions. Int. J. Wildland Fire 7, 51-63.
Pedersen, B. S. (1998). The role of stress in the mortality of midwestern oaks as indicated by growth
prior to death. Ecology 79, 79-93.
Peter, S. J. (1992). "Heat Transfer in Soils Beneath a Spreading Fire." PhD dissertation. University
of New Brunswick, Fredericton.
Peterson, D. L., and Ryan, K. C. (1986). Modeling postfire conifer mortality for long-range plan-
ning. Environ. Manage. 10, 797-808.
Pinard, M. A., and Huffman, J. (1997). Fire resistance and bark properties of trees in a seasonally
dry forest in eastern Bolivia. J. Tropical Ecol. 13, 727-740.
Raison, R. J., Woods, P. V., Jakobsen, B. F., and Bary, G. A. V. (1986). Soil temperatures during and
following low-intensity prescribed burning in a Eucalyptus pauciflora forest. Aus.J. Soil Res. 24,
33-47.
Rebertus, A. J., and Williamson, G. B. (1989). Longleaf pine pyrogenicity and turkey oak mortal-
ity in Florida xeric sandhills. Ecology 70, 60-70.
Regelbrugge, J. C., and Conard, S. G. (1993). Modeling tree mortality following wildfire in Pinus
ponderosa forests in the central Sierra Nevada. Int. J. Wildland Fire 3, 139-148.
Reifsnyder, W E., Herrington, L. P., and Spalt, K. W. (1967). "Thermophysical Properties of Bark
of Shortleaf, Longleaf, and Red Pine." Bulletin No. 70. Yale University School of Forestry, New
Haven.
Reinhardt, E. D., and Ryan, K. C. (1988). How to estimate tree mortality resulting from underburn-
ing. Fire Management Notes 49, 30-36.
Rosenberg, B., Kemeny, G., Switzer, R. C., and Hamilton, T. C. (1961). Quantitative evidence for
protein denaturation as the cause of thermal death. Nature 232, 471-473.
Rothermel, R. C., and Deeming, J. E. (1980). "Measuring and Interpreting Fire Behavior for Cor-
relation with Fire Effects." General Technical Report INT-93. USDA Forest Service, Intermoun-
tain Forest and Range Experiment Station, Ogden.
Rundel, P. W (1972). The relationship between basal fire scars and crown damage in giant sequoia.
Ecology 54, 210-213.
Russell, R. S. (1977). "Plant Root Systems: Their Function and Interaction with the Soil." McGraw-
Hill, London.
Ryan, K. C , and Frandsen, W H. (1991). Basal injury from smoldering fires in mature Pinus pon-
derosa Laws. Int. J. Wildland Fire 1, 107-118.
Ryan, K. C , and Reinhardt, E. D. (1988). Predicting postfire mortality of seven western conifers.
Can. J. For. Res. 18, 1291-1297.
Ryan, K. C , Peterson, D. L., and Reinhardt, E. D. (1988). Modeling long-term fire-caused mortality
of Douglas-fir. For. Sci. 34, 190-199.
Schuepp, P. H. (1993). Tansley review no. 59: Leaf boundary layers. New Phytologist 125, 477-507.
Simard, A. J., Eenigenburg, J. E., Adams, K. B., Nissen, R. L., Jr., and Deacon, A. G. (1984). A gen-
eral procedure for sampling and analyzing wildland fire spread. For. Sci. 30, 51-64.
Spalt, K. W, and Reifsnyder, W E. (1962). "Bark Characteristics and Fire Resistance: A Literature
Survey." Occasional Paper No. 193. USDA Forest Service, Southeastern Forest Experiment Sta-
tion, Asheville.
Steward, F. R., Peter, S., and Richon, J. B. (1990). A method for predicting the depth of lethal heat
penetration into mineral soils exposed to fires of various intensities. Can. J. For. Res. 20, 9 1 9 -
926.
Chapter 14 Fire Effects on Trees 525

Sucec, J. (1985). "Heat Transfer." Wm. C. Brown Publishers, Dubuque.

Swanson, W. M. (1970). "Fluid Mechanics." Holt, Rinehart, and Winston, New York.
Swezy, D. M., and Agee, J. K. (1991). Prescribed-fire effects on fine-root and tree mortality in old-
growth ponderosa pine. Can. J. For. Res. 21, 626-634.
Takata, A. (1974). Development of criterion for skin burns. Aerospace Med. 45, 634-637.
Taylor, G. I. (1961). Fire under the influence of natural convection. In "The Use of Models in Fire
Research" (W. G. Berl, Ed.), pp. 10-31. Publication 786. National Academy of Sciences, Na-
tional Research Council, Washington, DC.
Taylor, R. E. (1979). Heat-pulse thermal diffusivity measurements. High Temperatures-High Pres-
sures 11,43-58.
Thomas, P. H. (1963). The size of flames from natural fires. In "Ninth Symposium (International)
on Combustion," pp. 844-859. Academic Press, New York.
Thomas, P. H. (1964). "The Effect of Wind on Plumes from a Line Heat Source." Fire Research Note
572. Joint Fire Research Organization, Fire Research Station, Boreham Wood.
Thomas, P. H. (1967). Some aspects of the growth and spread of fire in the open. Bihliog. Agric. Sec.
E For. 40, 139-164.
Uhl, C , and Kauffman, J. B. (1990). Deforestation, fire susceptibility, and potental tree responses
to fire in the eastern Amazon. Ecology 71, 437-449.
Van Wagner, C. E. (1972). Duff consumption by fire in eastern pine stands. Can. J. For. Res. 2, 3 4 -
39.
Van Wagner, C. E. (1973). Height of crown scorch in forest fires. Can. J. For. Res. 3, 373-378.
Van Wijk, W R., and Bruijn, P. J. (1964). Determination of thermal conductivity and volumetric
heat capacity of soils near the surface. Soil Sci. Soc. Amer. Proc. 28, 461-464.
Vines, R. G. (1968). Heat transfer through bark, and the resistance of trees to fires. Aus.J. Bot. 16,
499-514.
Vines, R. G. (1981). Physics and chemistry of rural fires. In "Fire and the Australian Biota,"
(A. M. Gill, R. H. Groves, and I. R. Noble, Eds), pp. 129-149. Australian Academy of Science,
Canberra.
Wahlenberg, W G. (1946). "Longleaf Pine: Its Use, Ecology, Regeneration, Protection, Growth and
Management." Charles Lathrop Pack Forest Foundation, Washington, DC.
Waring, R. H. (1987). Characteristics of trees predisposed to die. BioScience 37, 569-574.
Waring, R. H., and Pitman, G. B. (1985). Modifying lodgepole pine stands to change susceptibility
to mountain pine beetle attack. Ecology 66, 889-897.
Webb, W L. (1981). Relation of starch content to conifer mortality and growth loss after defolia-
tion by the Douglas-fir tussock moth. For. Sci. 27, 224-232.
Weber, R. O. (1991). Modeling fire spread through fuel beds. Prog. Energy Combustion Sci. 17, 6 7 -
82.
Weber, R. O., GiU, A. M., Lyons, P R. A., Moore, R H. R., Bradstock, R. A., and Mercer, G. N. (1995).
Modeling wildland fire temperatures. CALMScience Suppl. 4, 23-26.
Westcot, D. W, and Wierenga, P.J. (1974). Transfer of heat by conduction and vapor movement in
a closed soil system. Soil Sci. Soc. Amer. Proc. 38, 9-14.
Williamson, G. B., and Black, E. M. (1981). High temperature of forest fires under pines as a selec-
tive advantage over oaks. Nature 293, 643-644.
This Page Intentionally Left Blank
 CHAPTER

Forest Fire Management

DAVID L . M A R T E L L
Faculty of Forestry, University of Toronto, Toronto, Ontario, Canada

I. Introduction
II. The Relationship between Fire and Forest Land
Management Objectives
III. Assessing Fire Impacts
A. Least Cost Plus Damage Model of
Fire Economics
B. Assessing the Impact of Fire on Timber Supply
IV. Forest Fire Management Organizations
A. A Simple Fire Suppression Model
B. Fire Load Management
C. Fire Suppression
V Level of Fire Protection Planning
VI. Some Challenges
A. Relating Outcomes to Strategies
B. Dealing with Uncertainty
C. Moving beyond Fire Exclusion
Further Reading
References

I. INTRODUCTION
Societies that are confronted with potentially destructive forest or wildland fires
develop fire management organizations to modify fire's impact on people, the
things they value, and the ecosystems about which they are concerned. A fire
management organization's objectives are determined by the social, economic,
and political institutions that control it. The extent to which it achieves its ob-
jectives depend upon (1) how well the fire and ecosystem processes and the im-
pact of fire management are understood; (2) the degree to which the social and
economic impacts of fire are understood; (3) the technology and resources so-
ciety puts at the organization's disposal; (4) the knowledge, skills, and experi-
ence of the people in the organization; and (5) the challenges posed by nature.
Forest Fires
Copyright © 2001 by Academic Press. All rights of reproduction in any form reserved. 527
528 David L. Martell

Most of the earlier chapters of this book have dealt with the physics, chem-
istry, and biology of wildland fire and its ecological impact on forest stands and
landscapes. The focus has been largely what wildland fire specialists describe as
fire science. Fire scientists carry out basic research (e.g., fire spread), applied re-
search (fire danger rating system development and implementation), and a great
deal of research that cannot readily be classified on the basic/applied research
spectrum. Previous chapters illustrated the importance of moving beyond the
development of relatively simple empirical curve-fitting models and furthering
our basic understanding of fire processes and their impact on ecosystems. In this
chapter, we turn our attention to the ways in which societies attempt to shape
wildland fire and influence its impact on them and surrounding ecosystems.
We study wildland fire from a management perspective and explore what fire
managers do, why they do what they do, how they decide to do what they do,
and the short, intermediate, and long-term social, economic, and ecological
consequences of their actions. We focus on what is commonly referred to as/ire
management.

Fire Management: The Canadian Glossary of Forest Fire Management Terms (Mer-
rill and Alexander, 1987) defines/orestjire management as the "activities concerned
with the protection of people, property, and forest areas from wildfire and the use of
prescribed burning for the attainment of forest management and other land use ob-
jectives, all conducted in a manner that considers environmental, social, and eco-
nomic criteria."

Most North American wildland fire agencies were developed to combat fire
in the early 1900s and for many decades they had names that included terms like
fire control and/ire protection. They were unabashedly fire suppression organi-
zations, and their names reflected their view of wildland fire and its impact on
the world. The current widespread use of the term fire management reflects a
more enlightened view that wildland fire is a natural phenomenon that has a
range of social, economic, and ecological impacts that should be considered
when it is managed.
Fire management programs typically include prevention measures to reduce
the number of people-caused fires that occur, detection systems to find fires
while they are small, initial attack systems to contain fires before they burn over
large areas, and large fire management systems that are designed to minimize the
damage that results from large fires that are not controlled by the initial attack
system. They also include fuel modification measures to mitigate the impact of
fires that do occur and the use of prescribed fire to fulfill silviculture, wildlife
habitat management, and other land management objectives. Fire management
also calls for (albeit very infrequently in most jurisdictions) conscious decisions
to allow some wildfires to burn freely or to be subjected to limited suppression
action if and when the net benefit of doing so is thought to be positive.
Chapter 15 Forest Fire Management 529

Fire management researchers carry out basic research to further our under-
standing of fire management processes (e.g., the influence of suppression ac-
tion on fire growth) and apphed research to develop decision support systems
(e.g., initial attack system models) that fire managers can use to help predict
the consequences of their actions and to evaluate alternative policies, strate-
gies, and tactics. Many fire management systems researchers use Operations
Research/Management Science (OR/MS) methodologies which entail the use
of mathematical models and information technology. Pollock and Maltz (1994,
p. 4) characterize Operations Research "as a process of developing mathematical
models in order to understand (and possibly influence) complex operational
systems." Martell (1982) and Martell et al. (1998) review many of the fire man-
agement operations research publications that have appeared over the last four
decades.
Physical and biological scientists can and often do study systems that are
relatively isolated from human intervention, but fire management researchers
must deal with large complex systems composed of natural processes, people,
and machines. Fire managers must resolve decision-making problems (e.g., ini-
tial attack dispatching) as circumstances dictate and, even though they base
their decisions in part on the understanding of basic physical and biological
processes developed by scientists, they frequently have to deal with processes
that are, at best, poorly understood. Although they can postpone some decisions
(e.g., the purchase of airtankers) while they gather more information and re-
flect upon the possible consequences of their actions, they seldom have the lux-
ury of postponing their decisions until researchers have developed a thorough
understanding of the pertinent processes. Fire management researchers that de-
velop decision support systems to meet their needs must therefore incorporate
in their systems models what is currently known about physical, biological, so-
cial, and economic processes, empirical curve-fitting models, and the subjec-
tive assessments of designated experts. In this chapter, we will describe how
they do so.
A forest fire can burn over a large area during a short period of time. When
it does so, it can leave tremendous impacts on public safety, property, forest re-
sources, and forest ecosystems in its wake. Although many fire management ac-
tivities are directed at specific fires, fire management programs are designed to
modify the impact of fire across vast landscapes and administrative regions over
very long planning horizons that are embedded in larger ecoclimatic and ad-
ministrative hierarchies that span many forest management units, districts, re-
gions, provinces, nations, and even continents. There is no definitive spatial and
temporal scale at which fire can or should be managed. One can focus on a por-
tion of an active fire's current perimeter or one can deal with fire regimes across
very large spatial and temporal scales. We will approach fire and its management
530 David L. Martell

from a range of spatial and temporal scales, and we will study the impact of fire
and fire management on fire regimes at landscape levels embedded within po-
litical and administrative structures that span many such landscapes. We begin
by clarifying our use of the term jire regime.

Fire Regime: V/halen (1995) described afire regime in terms of the following five
basic elements: (1) fire intensity, (2) the season during which burning takes place,
(3) fire size or extent, (4) fire type, and (5) fire frequency.

Fire management programs are designed to modify one or more of those five
basic elements of a fire regime, either directly or indirectly. Prevention mea-
sures, for example, reduce fire frequency, whereas fire detection and suppres-
sion activities reduce final fire size. Prescribed burning can either increase or
decrease fire intensity, size, and frequency.
In Section II we describe the relationship between fire management objec-
tives and higher level forest land management objectives, and in Section III we
deal with fire economics and the assessment of fire impacts. Section IV deals
with fire management organizations, and in Section V we describe how fire
managers carry out level of fire protection planning to balance the benefits of
mitigating the detrimental impacts of fire and reaping its benefits with the cost
of doing so. We conclude with Section VI in which we describe some of the more
significant challenges that will need to be addressed as fire management agen-
cies increasingly move away from fire exclusion to the adoption of policies that
call for modified fire suppression and the use of prescribed fire.
Since we are primarily interested in fire management and its impact on fire
regimes, we describe the activities associated with each subsystem, their po-
tential impact on the basic elements of a fire regime, and their potential social,
economic, and ecological impacts. We discuss the decision-making problems
associated with each fire management subsystem and the management infor-
mation systems and decision support systems that are designed to enhance fire
management decision-making.
Figure 1 is a hierarchical view of fire management decision making which
will be used to structure our investigation of fire management systems. The list
of decisions included in Figure 1 is not exhaustive but is characteristic of most
large North American wildland fire management agencies. Most of those agen-
cies are still very heavily committed to fire control, but the use of prescribed
fire to achieve ecological objectives is a major concern of many agencies, par-
ticularly those responsible for fire management in national parks and wilder-
ness areas in Canada and the United States.
Sound fire management is based, in part, on the extent to which the deci-
sions associated with any particular level in the hierarchy are compatible with
the decisions taken above and below. Consider daily airtanker deployment, for
example. The number of airtankers to be deployed on a particular day will be
Chapter 15 Forest Fire Management 531

Level of Protection Planning

Decide upon the fire regime or bum rate targets by fire management
unit and what fiinds will be allocated to fire management.

Strategic Resource Acquisition and Home Basing

Decide upon the number and location of bases and other fixed
facilities, what resources will be acquired, and where they will be
home based.

Daily Resource Deployment

Decide upon prevention measures, the temporary deployment of

detection, initial attack, and prescribed burning resources.

1
Hourly Detection, Suppression, and Prescribed Burn Ignition
Decide upon the timing and routing of detection patrol aircrafi;, the
dispatch of resources to new and existing wildfires, and the ignition
and control of prescribed fires.

Fire Control
Decide upon prescribed bum ignition pattems and the allocation of
fire suppression resources to prescribed bum and wildfire perimeters.

FIGURE 1 Fire management decision hierarchy.

influenced by the number of airtankers the agency owns or leases for the fire
season (a consequence of higher level strategic decision making) and the cost
of temporarily borrowing more costly airtankers from other agencies. High bor-
rowing costs could deter fire managers from borrowing airtankers from others,
which could produce daily airtanker shortages, longer initial attack response
times, and more escaped fires in the lower initial attack level of the hierarchy.
Fire management poses many such challenges.
We draw heavily on our own experience with the Ontario Ministry of Nat-
ural Resources (OMNR) in the province of Ontario in Canada, and what we
know about other North American wildland fire management agencies to il-
lustrate our discussion. Most wildland fire management agencies are, for the
most part, emergency response systems; however, they vary from jurisdiction
532 David L. Martell

to jurisdiction around the globe, and we do not attempt to capture the rich di-
versity of approaches, philosophies, and practices that are influenced by local
ecosystems, climate, culture, economics, and political systems. For comprehen-
sive historical accounts of the development of forest fire management in some
North American jurisdictions, see, for example, Holbrook (1943), Lambert and
Pross (1967), and Pyne (1982). Pyne (1997) reviews fire and fire management
from a more global perspective.
It is important to note that, although private enterprise plays a significant
role in wildland fire management in North America (e.g., many fire management
agencies contract flying services from private companies and public corpora-
tions) , fire managem^ent is ultimately a public sector endeavor and management
and decision making in the public sector differs in many ways from management
in the private sector. Gass (1994) describes some of the many ways in which
public sector OR/MS differs from its practice in the private sector including the
relative importance of political concerns, the difficulties associated with evalu-
ating costs and benefits, and the "freedom of information" environment that
makes public sector policy analysis and decision-making processes very open.
He points out that, unlike their private sector counterparts that tend to focus on
the "bottom line," public sector decision makers must address efficiency (the
proper use of resources), effectiveness (the attaining of specified goals and ob-
jectives), and equity (the extent to which all citizens are treated alike).

II. THE RELATIONSHIP BETWEEN FIRE AND

FOREST LAND MANAGEMENT OBJECTIVES
Afiremanagement program is usually but one of many components of a broader
forest land management program. Societies develop and implement land man-
agement plans to increase the likelihood that the human activities which take
place on or near designated parcels of land will be compatible with the needs
and desires of the people and institutions that have jurisdiction over that land.
Forest land management plans might, for example, call for some areas to be
treated as wilderness in which the impact of humans is minimized to protect
natural ecosystem processes, others to be designated as recreational parks in
which heavy human use is tempered to reduce its impact on natural processes,
and yet others that are managed primarily for industrial fiber production. Land
management plans should reflect the diverse needs and interests of the many
groups of people who have interests in the forest. Consequently, even though
forest land management is frequently a source of bitter conflict in modern in-
dustrial societies, we will assume that societies can and ultimately do develop
land management plans that reflect the way they want their forest lands to be
Chapter 15 Forest Fire Management 533

Forest Land Management Planning

Objectives

Level of Fire
Protection Planning

Costs &
Impacts

Fire Management System

Fire Load Management Fire Suppression

Subsystem Subsystem

Fuel Management Suppression Resource

Acquisition
Fire Prevention
Dally Deployment
Detection
Initial Attack
Project Fire Management

FIGURE 2 The relationship between fire and forest management planning, a refinement of the
fire management systems framework presented in Martell et oX. (1999).

managed. The fire manager's role is to develop and implement fire management
plans that are compatible with and contribute to the achievement of the objec-
tives of the higher level forest land management plans.
Figure 2 is a refinement of the fire management systems framework pre-
sented in Martell et a\. (1999) and illustrates the hierarchical relationship be-
tween fire management and forest land management. Forest land management
planners should begin by identifying fire management objectives such as aver-
age annual burned area targets and transmitting them to the fire organization.
Fire managers should then determine if they can achieve those targets and de-
velop strategies that will minimize the cost of doing so. They should then trans-
mit their predictions concerning the cost of achieving the specified fire man-
agement targets and their assessment of the social, economic, and ecological
impacts of the fire behavior that might ensue, back up to the forest land man-
agement planners. If the forest land management planners and their clients are
not satisfied with those predicted costs and impacts, they should revise their
fire management objectives and send the revised targets back down to the fire
534 David L. Martell

organization. Area burned is not simply the fortunate or unfortunate conse-

quence of a chance process but rather the resuk of expUcitly planned human
intervention in complex natural processes. The iterative process of balancing
forest land management fire needs with fire management costs and impacts to
achieve a workable solution is what is known as level offireprotection planning,
a topic we will address in Section V.
Just as forest management plans must vary across the landscape, so too must
fire management objectives vary from time to time and place to place. Clough
(1963) said that "management has been defined as the art of 'guiding the activi-
ties of a group of people toward the achievement of a common goal.'" One can
think of forest fire management as the art and science of guiding people in efforts
to modify fire regimes that vary across spatial and temporal landscapes in accor-
dance with forest land management objectives, in a cost-effective manner. This
might, for example, call for some fires to be allowed to burn reasonably freely in
remote wilderness areas, the use of prescribed fire to fulfill ecological and other
land management needs, the aggressive suppression of fire in areas where they
pose significant threats to people and their property, and the attempts to limit
fire losses in timber production areas. In Section IV, we will study the structure
and functions of fire management organizations that are designed to achieve
such objectives, but, before we do so, we will investigate how the social, eco-
nomic, and ecological impacts of fire can be assessed.

III. ASSESSING FIRE IMPACTS

Our discussion of forest fire management has, to this point, been predicated on
an assumption that forest fire managers and their land management clients can
assess the impact of fire and fire management programs on people, property,
forest resources, and ecosystems, and earlier chapters addressed the impact of
fire on ecosystems. Assessing the social and economic impact of fire poses many
complex, largely unresolved problems and is the subject of what forest fire man-
agement specialists refer to SiSfireeconomics. Gorte and Gorte (1979) and Baum-
gartner and Simard (1982) review much of the work that has been carried out
in this area. In this section we describe how such impacts are currently assessed,
and we identify some of the many significant gaps that remain in this area.
Consider a forest fire management agency that strives to minimize the net de-
structive impact of fire subject to constraints on the resources that society puts
at its disposal and the ways in which they are to be used. Funds are allocated to
fire management on the assumption that the subsequent benefits will exceed
the value of the money spent. Fire management costs can be assessed and ex-
pressed in monetary terms by using standard accounting procedures. Salaries
Chapter 15 Forest Fire Management 535

and equipment costs, for example, are easily assessed as is the cost of support-
ing fire management personnel working in the field. Standard accounting prac-
tices will produce reasonably accurate estimates of the annual cost of fixed as-
sets such as aircraft, fire suppression equipment, buildings, communications,
and information systems. The benefits of fire management activities include the
reduced losses that result from limiting the number and size of destructive wild-
fires and the increased forest resource productivity and enhanced environment
that result from the use of prescribed fire and successfully monitoring and
modifying the suppression of beneficial wildfires.
Each year forest fire managers and the governments and other organizations
that fund them must decide how much money to allocate to fire management.
Many attempts have been made to develop formal procedures for determining
optimal, or appropriate, levels of fire protection. Fire management specialists
use what is referred to as the least cost plus damage or LCD fire economics
model, first developed by Sparhawk (1925), to explore such issues. We will use
the simplest variant of the basic LCD model to explain the basic principles of
fire economics.

A. LEAST C O S T PLUS DAMAGE

MODEL OF FIRE ECONOMICS

Forest fire managers use the term presuppression to describe the activities that
take place before fires actually occur; for example, the hiring of fire fighters and
the commitment to pay their regular salaries for the entire fire season regard-
less of the number of fires that actually occur. The extra costs, for example, the
overtime wages earned once a fire has been reported and is being fought, are re-
ferred to as suppression expenditures.
Let us assume that a forest fire management agency must decide how much
money to allocate to presuppression efforts prior to the start of the fire season.
Those funds could be used to maintain fire lookout towers and other infrastruc-
ture and fire suppression equipment such as portable power pumps, hose, and
hand tools. They could also be used to pay the regular salaries of fire fighters
and other staff; to charter detection aircraft, airtankers, and transport helicop-
ters; and to lease or purchase trucks and other resources that can be used to
combat fires once the season begins. The more fire fighters the agency hires, the
less likely shortages will occur. That will make it easier to deploy initial attack
crews at bases close to areas where fires are expected to occur, in advance, be-
fore the fires are reported. The reduced response times that would result should
decrease the likelihood that fires will escape initial attack and the net result
should be a reduction in fire damage and suppression costs. It is reasonable to
assume that suppression costs will decrease as presuppression expenditures
536 David L. Martell

T(p) = p =Presuppression Cost

Cost + Loss

L(p) = Fire Loss

(Damage + Suppression)

Presuppression Budget
FIGURE 3 Sparhawk's (1925) least cost plus damage fire economics model.

increase but, like most human endeavors, at a decreasing rate due to decreas-
ing marginal returns.
Figure 3 depicts the decision-making problem faced by fire managers. Let p
denote the presuppression budget and L{p) the fire loss which includes the sup-
pression cost (e.g., airtanker flying costs and the overtime wages of fire fight-
ers) and the fire damage. The total cost plus loss (T) is p + L(p) which is the
sum of the L(p) curve and the presuppression cost p. The optimum presuppres-
sion budget (p*) is the value of p that minimizes the total cost plus loss. Since
fire is natural and sometimes beneficial, some writers allow L(p) to be positive
(a loss) or negative (a benefit) and refer to the LCD model as the least cost plus
net value change model, but the result will be the same if benefits are treated as
negative costs. The LCD principle, in its simplest form, is that a forest fire man-
agement system should be operated at such a level that the total cost of operat-
ing the system plus the fire loss (or more precisely, the net loss) is minimized
(i.e., minimize cost plus loss).
In the simple case depicted in Figure 3, the total cost plus loss (T) is a con-
vex function of the presuppression cost so one can easily identify its minimum
by taking its derivative with respect to p, equating the result to zero, and solv-
ing for p*, the optimum presuppression budget. Real fire management systems
are far more complicated.

1. LCD Limitations
The LCD model illustrates the basic principles of fire economics but has many
drawbacks that limit its practical use for fire management planning. It is diffi-
Chapter 15 Forest Fire Management 537

cult to specify a production function that relates the annual area burned to
the presuppression budget. Analysis of the impact of fire management on area
burned is confounded by the fact that the historical data may no longer be rele-
vant due to changes in land use, the forest fuel complex, or even climate and
fire suppression technology. Furthermore, the production function is not a uni-
variate function of the presuppression budget as area burned depends not only
on the amount of money spent on presuppression but also on how those funds
are allocated to the many presuppression activities that can take place.
Consider, for example, a very simple fire management organization that has
no prevention program and no airtankers. It has a fixed presuppression budget
and must decide how much money to use to hire fire fighters and how much
money to spend on helicopters and trucks to transport fire fighters to fires. If it
chooses to charter a large number of helicopters and hire a small number of fire
fighters, then fire fighters will be able to fly to fires as soon as they are reported.
However, on days when many fires are reported, the supply of fire fighters will
be quickly exhausted, and the fires that occur later in the day will burn freely for
long periods of time while they wait for fire fighters to be released from the ear-
lier fires or borrowed and flown in from other agencies at increased cost. If the
agency hires many fire fighters and charters few helicopters, then the first few
fires reported each day will be helitacked quickly, but the later fires will wait and
burn freely for long periods of time as crews travel to them by truck or wait for
helicopters to be released from earlier fires.
What is needed, of course, is a multidimensional fire management produc-
tion function that relates the annual area burned to, for example, prevention
efforts, the number and location of fire lookout towers, the number and type of
detection aircraft and detection observers hired, the number and type of air-
tankers chartered, the number and type of transport helicopters chartered, and
the number of fire fighters hired. It is not possible to develop such functions to
be used in the LCD optimization framework at the present time, and it is doubt-
ful that it will be possible to do so in the foreseeable future.
Another critical limitation of the basic LCD model is that its deterministic
structure does not capture the stochastic elements that are so characteristic of
forest fire management systems. Fire management success varies significantly
from year to year, not only due to changes implemented by managers but more
often due to variability in weather, fire ignition, and fire behavior processes.
The most significant problem that limits the practical significance of the LCD
model is one that extends far beyond fire management and that is the difficulty
of assessing fire impacts and expressing them in monetary terms so that the net
impact of alternative fire management strategies can be assessed in monetary
terms. Most forest fire management agencies have a hierarchy of objectives that
guide their activities. For example, the Ontario Ministry of Natural Resources'
fire management objectives are, in order of priority: (1) to enhance public safety,
(2) to reduce the detrimental impact of fire on property and forest resources,
538 David L. Martell

and (3) to utilize the beneficial impacts of fire which may include the use of pre-
scribed fire for silviculture purposes and modified suppression of fires in some
remote areas to benefit from the impact of fire on natural ecosystems.
Historical data clearly indicate that fire management saves lives, but it is dif-
ficult to assess how it contributes to public safety. Even if it were possible to do
so, it would be neither possible nor socially acceptable to assign a monetary
value to the lives saved. Property loss reductions are also difficult to assess as
that would entail identifying physical structures such as houses and transpor-
tation facilities that might have burned had fires not been suppressed, and us-
ing their monetary value to assess the damage averted. Even if that could be
done, the human cost incurred when, for example, homes are burned exceeds
the assessed cost of the structures destroyed, and it would not be possible to as-
sess such costs.
Growing recognition that fire is a natural component of many ecosystems
poses even greater challenges. Previous chapters make it eminently clear that it
is difficult to assess the impact of fire on ecosystems and that such problems
will persist for the foreseeable future. Add to that the need to express such im-
pacts in monetary terms, and it is clear that such problems will not be resolved
for a very long time, if ever.
Given such limitations, the LCD model provides valuable insight into the ba-
sic principles of fire economics but is of little or no practical significance. Fire
managers and the political institutions that support them, therefore, rely on
heuristic approaches to fire management budgeting. In the past, they typically
implemented slight changes from year to year, particularly after "bad" fire sea-
sons when people reacted to severe losses and funding levels were increased or
when governments grew complacent after a few "good" years and attributed
small fire losses to an "overabundance" of fire suppression resources (rather
than some fortuitous combination of adequate resources, sound fire manage-
ment, good weather, and good luck) and diverted funds from fire programs to
other pressing needs. They relied on their intuition and allocated what they
thought might be the best amount of money to spend on presuppression based
on their experience and understanding. In Section V, we describe how level of
fire protection planning is currently practiced, but first we will describe one fire
impact that can be assessed in both physical and monetary terms—the impact
of fire on industrial fiber production or timber supply.

B. ASSESSING THE IMPACT

OF FIRE ON TIMBER SUPPLY

North American forests have been, and continue to be, used extensively for in-
dustrial timber production, and most North American forest fire management
Chapter 15 Forest Fire Management 539

agencies were developed to protect people and timber from fire damage. It is
therefore not surprising that a great deal of effort has been devoted to assessing
the impact of fire on timber supply and that is the impact that is perhaps best
understood. Forests are now viewed as far more than industrial fiber sources
and that will become even more evident in the future, but we do not know how
to measure or assess the impact of fire on those other "nontimber" values. Since
the methodologies developed to assess the impact of fire on timber supply might
ultimately be extended to some nontimber values, this subsection is devoted to
a detailed analysis of the impact of fire on timber supply.
We will use a simple hypothetical 100,000-ha jack pine forest that is used
solely for industrial timber production to illustrate how one can assess the im-
pact of fire on timber supply. Assume our hypothetical forest has the stand age
class structure shown in columns 1 and 2 of Table 1 and in Figure 4.
We will assume that the merchantable volumes produced when stands are
harvested vary with age as shown in column 3 of Table 1, which contains the
gross merchantable volumes for Site Class II jack pine stands in the province of
Ontario taken from Plonski (1974).
Suppose a fire burned 4000 ha of age class 7, 3000 ha of age class 8, 2000 ha
of age class 9, and 1000 ha of age class 10. Using the gross merchantable volume

TABLE 1 Age Class Distribution and Yield of Merchantable

Timber Volume for the Hypothetical Forest

Gross
merchantable
Age class Area volume
(years) (ha) (m^/ha)

0-10 18,127 0
10-20 14,841 0
20-30 12,151 23
30-40 9,948 68
40-50 8,145 110
50-60 6,669 147
60-70 5,460 174
70-80 4,470 190
80-90 3,660 199
90-100 2,996 203
100-110 2,453 204
110-120 2,009 204
120-130 1,644 204
130-140 1,346 204
140-150 1,102 204
> 150 4,979 204
Total 100,000
540 David L. Martell

5 6 7 8 9 10 11 12 13 14 15 16
Age class (10 year intervals)
FIGURE 4 Age class distribution of the hypothetical forest.

figures in column 3 of Table 1, we find that the volume burned would be

1,867,000 m^. Our first task is to assess the impact of such a fire.
Suppose the company that harvests wood from the hypothetical forest sells
it at a rate of $5.00/m^. A superficial burn site assessment of the fire could be
obtained by multiplying the merchantable volume of the wood consumed by
the fire by the selling price of wood and that would produce an estimated cost
of $9,335,000 for the 10,000-ha fire. The true economic cost of the timber de-
stroyed by the fire can differ significantly from this figure as we explain later.
The harvesting activities that take place in a forest that is managed for tim-
ber production will be governed by a timber harvest scheduling plan that stip-
ulates when each stand is to be cut. The economic value of the timber is the
present net worth of all the economic returns associated with the harvesting ac-
tivities that are scheduled to occur over the planning horizon. Suppose future
costs and revenues are discounted at an annual rate of 3%. The present value of
a cubic meter of wood that is to be sold 5 years from now is not $5 but rather
$5/(1.03)^ or $4.31. The economic value of the timber in a forest is therefore
determined by the harvest schedule, but when fire burns stands that are sched-
uled to be harvested, that schedule will have to be revised. The cost of a fire is
therefore the economic value of the timber as determined by the preburn har-
vest schedule minus the economic value of the timber determined by the revised
postburn harvest schedule.
Chapter 15 Forest Fire Management 541

Van Wagner (1979) was one of the first to focus attention on forest-level fire
and timber supply issues, and he used computer simulation models to do so.
Forest managers have developed and used a myriad of techniques to develop
timber harvest schedules, and linear programming optimization models are of-
ten used for such purposes. Johnson and Scheurman (1977) describe what they
refer to as the standard Model 1 and Model 11 forms of the timber harvest sched-
uling model. Timber harvest scheduling linear programming (LP) models are
deterministic models, but the need to account for the variability of fire leads
to stochastic forest level problems that are complex, largely mathematically
intractable problems. Fortunately, Reed and Errico (1986) developed a mean
value formulation of the forest-level problem in which one assumes some con-
stant average fire loss occurs each year. Although this approach ignores the vari-
ability that complicates forest management under uncertainty, their mean value
formulation of the problem produces timber harvest scheduling solutions that
are close to those solutions which more detailed stochastic models would pro-
duce for sufficiently large forest management units. The Reed and Errico (1986)
model is similar in structure to Model II, but it is different enough to merit rec-
ognition as a new form. Model III. One of the most appealing aspects of Model
III is that it is well suited for dealing with flammable forests.
The form of Model III we use here is identical to the model presented in
Martell (1994) which is a variant of the basic Reed and Errico (1986) model.
A 300-yr planning horizon is partitioned into 30 10-yr periods, and the forest
stands are aggregated into 10-yr age classes. Let x^ denote the area of the forest
in age class i at the start of period t, and let hj ^ denote the area of age class i har-
vested during period t. Assume that some constant fraction of the forest (0.2%)
burns each year, that harvesting takes place before burning in each period, and
that the burned area is uniformly distributed over the forest. Stands are assumed
to regenerate naturally to jack pine after harvesting or burning. The harvest flow
is constrained to be constant and a terminal constraint that the merchantable
volume growing in the forest at the end of the 300-yr planning horizon average
40.2 m^ ha"Ms imposed to ensure the entire forest is not clear cut at the end of
the planning horizon.
The Model III LP model can be used to assess the impact of a fire (and fire
management) on our hypothetical forest. The economic impact of a fire on the
timber value of a forest is the timber value of the forest given the prefire timber
harvest schedule less the postfire value under a revised plan. Neither plan has
to be optimal, but it makes little sense to evaluate fire impacts using subopti-
mal timber harvest schedules. The optimal preburn Model III timber harvest
schedule for our 100,000-ha hypothetical forest produces an annual allowable
cut of 2.41 m^ ha~^ yr"^ and a present net worth of $407/ha. The optimal post-
burn timber harvest schedule produced by running Model III with a revised age
542 David L. Martell

class structure (all the burned area is moved into age class 1) produced an an-
nual allowable cut of 2.37 m^ ha~^ yr"^ and a present net worth of $399/ha, a
loss of 2%. Thus a fire that burns 10% of the area and 2 1 % of the volume reduces
the value of the forest by only 2%. It is important to note, however, that this nu-
merical result is specific to this particular hypothetical forest and the relative
sizes of the burn fraction and that loss cannot be generalized to other fires and
forests with different age class structures, growth and yield functions, and har-
vest flow constraints.
Timber harvest scheduling models can also be used to help evaluate and set
initial and extended attack priorities. Suppose that two fires were burning out
of control and threatening to grow even larger in two distinctly different parts
of a forest. One could assess the pre- and postburn values of the forest for both
fires and determine which of the two fires would have a greater impact on the
economic value of the timber in the forest. Such information could be used to
inform decisions concerning the allocation of scarce suppression resources to
escaped fires.
Timber supply models can also be used to evaluate fire management pro-
grams. Figure 5 is a graph that relates the annual allowable cut to the average
annual burn fraction in our hypothetical forest. Suppose that the average an-
nual burn rate of the natural fire regime in our hypothetical forest was 2% and
that fire management reduced that to 0.2%/yr. That would increase the an-

3.00

2.50
=3

0 2.00

i
M 1.50

<
0.50

0.00
0.000 0.010 0.020 0.030 0.040
Annual burn fraction
FIGURE 5 The impact of the annual burn fraction on the annual allowable cut in the hypotheti-
cal forest.
Chapter 15 Forest Fire Management 543

nual allowable cut from 1.40 m^ ha~^ yr"^ (a present net worth of $236/ha) to
241 m^ ha~^ yr~^ (with a present net worth of $407), an increase of 72% in the
timber production value of the forest. Since that present net saving of $171/ha
would be realized over a 300-yr planning horizon, it can be transformed (using
a 3% discount rate) into an equivalent annual saving of $5 ha"^ yr~\
Timber is but one of many forest-based values that benefit people, but some
of the lessons we have learned about evaluating the impact of fire on timber
supply maybe applicable to other values. Boxall et al. (1996), for example, have
studied the impact of fire on the attractiveness of recreational canoe routes.
When a fire burns through an area through which a canoe route passes, that fire
will alter the vegetation of some portion of the larger network of canoe routes
in ways that are not evident by looking at the burn site itself. To assess the im-
pact of such a fire, one would have to evaluate the canoe route network before
and after the fire. Clearly, scale is important when assessing fire impacts and
stand- or burn-site specific assessments are not appropriate for assessing the
impact of fire on timber supply. The management unit is an appropriate scale
at which to assess the impact of fire, but it is important to recognize that it is
based on an assumption that management units are independent entities. That
may well be the case when they are owned or managed by competing enter-
prises, but companies or governments can pool management units for risk man-
agement purposes, and that possibility means that there is nothing inherently
sacrosanct about management boundaries.
If one wishes to assess the ecological impact of fire, it is clear that one must
look beyond the burn itself, but it is not obvious how far one must look. One
can, as is the case with timber production and management units, identify the
habitat utilized by some species of interest and attempt to identify a scale which
captures the true impact of fire on those populations. Clearly, the appropriate
scale will vary by species and the communities of interest.
It is also important to note that, when the impact of a specific fire is assessed,
assessment should be carried out within a context of the underlying on-going
fire regime. Using a Model III LP model pays explicit recognition to the fact
that, were you to save some forest stand from destruction due to fire today, your
"saving" will be measured with respect to the fact that the stand might well
burn in the near future despite your most recent intervention. The fire regime
embedded in Model III accounts for that factor. The larger the average annual
burn fraction, the smaller the economic impact of a specific fire, since increased
annual burn fractions decrease the values of both the pre- and postburn forests.
The ecological impact of fire must also be assessed in such contexts.
Finally, it is important to note that fire impacts need not be expressed in
monetary terms. The impact of fire on timber supply can be evaluated in terms
of its impact on projected harvest flow or dollars, and both are legitimate mea-
sures that may be of interest to fire management clients. Since it would be
544 David L. Martell

difficult, if not impossible, to assess the ecological impact of fire or its impact
on environmental services in monetary terms, fire managers and their clients
need to identify what measure or measures to use and then develop multi-
attribute evaluation procedures for evaluating fire management programs, a
challenge that lies well beyond the scope of this chapter.

IV. FOREST FIRE MANAGEMENT

ORGANIZATIONS
Forest fire management organizations are emergency response systems but, un-
like urban fire, police, and ambulance services, they deal with natural processes
that can have both beneficial and detrimental impacts on people, property, for-
est resources, and ecosystems. Since it is virtually impossible to exclude fire from
most forests and attempts to do so can be very costly, fire management agencies
use the resources allocated to them to minimize what they and their land man-
agement clients collectively judge to be the net detrimental impact of fire.
A fire regime will produce a spatial and temporal distribution of b u r n patches
embedded in a larger forest landscape composed of many patches that have been
shaped and influenced by both human and natural processes. Consider the
simplest case in which we focus on one very simple result of a fire regime, the
average annual fraction of the fire management area or fire region burned each
year. That burn fraction should be compatible with the forest management ob-
jectives, and the role of the fire manager is to develop and implement a fire man-
agement plan that will achieve that average annual burn fraction at a reasonable
cost. Fire managers often partition the area under their jurisdiction (the fire
region) into zones or compartments that are reasonably homogeneous with re-
spect to ecosystem processes, land use activities, and values at risk and then
specify their fire management objectives in terms of fire regimes or the level of
fire protection they will attempt to achieve in each zone.
The forest fire management planning process can then be viewed as com-
prising four basic steps:

1. Partition the fire region into zones or compartments that are reasonably
homogeneous with respect to forest ecosystems, land use patterns, and
values at risk;
2. Assess the potential beneficial and detrimental impacts of fire in each
zone;
3. Select an appropriate level of protection or fire regime for each zone and
develop a plan to minimize the cost of achieving that objective;
4. Implement, monitor, and revise the fire management plan over time.
Chapter 15 Forest Fire Management 545
Forest fire management objectives vary from agency to agency, a reflec-
tion of the variability of both societies and the ways in which they interact with
the ecosystems that envelop them. The Aviation, Flood, and Fire Management
Branch of the Ontario Ministry of Natural Resources, the agency responsible for
forest fire management in the province of Ontario in Canada, is representative
of many large North American forest fire management agencies. The OMNR is
responsible for fire management on more than 85 million hectares and spends
roughly $85 million a year on its activities. A permanent staff of 220 is aug-
mented by 640 fire fighters that are hired each fire season. The agency also op-
erates an aircraft fleet that includes 9 large CL-415 airtankers and 5 smaller Twin
Otters that can serve as airtankers. Approximately 14 helicopters, 15 detection
aircraft, and 7 bird-dog aircraft (that carry air attack officers that direct airtanker
operations over the fire) are hired each season, but those figures vary from year
to year. The OMNR participates in national and international mutual aid agree-
ments that make it possible for it to share its fire suppression resources with
other North American fire management agencies and to borrow resources from
other agencies when the need arises.
During the years 1976-1999, the number of forest fires in Ontario ranged
from 735 to 3970 per year with an average of 1713 fires per year and a median
of 1588 fires per year. The area burned each year during that period ranged
from 9444 to 611,939 ha with an average of 241,734 ha/yr and a median of
176,462 ha/yr. The year-to-year variation in the number of fires and area burned
are illustrated in Figures 6 and 7.

1975 1980 1985 1990 1995 2000

Year
FIGURE 6 Number of forest fires in Ontario, 1976 -1999.
 David L. Martell

100,000 A

1975 1980 1985 1990 1995 2000

Year
FIGURE 7 Area burned by forest fires in Ontario, 1976-1999.

A. A SIMPLE FIRE SUPPRESSION MODEL

Societies allocate funds to fire management on the assumption that the alloca-
tion will reduce the damage that results from wildfire and enable them to reap
some of fire's benefits. In this section, we present a simple conceptual model of
the fire suppression process which will illustrate how such activities can influ-
ence final fire size that, in turn, influences the economic impact of fires.
Forest fires are ignited by people or some natural agent such as lightning, and
they smolder or burn with an active open flame and emit smoke until they are
either extinguished due to a lack of combustible fuel or detected and reported
to the duty officer or initial attack dispatcher who decides what suppression re-
sources will be dispatched to the fire. The initial attack crew travels to the fire
by truck or by air, sets up its suppression equipment at or near the fire, and be-
gins suppression action as soon as possible. The ground crew may or may not
be assisted by airtankers that drop water or fire retardant on, or just ahead of,
the active fire front.
The effectiveness of the initial attack force will depend upon its composi-
tion, how soon it arrives at the fire, the fuel and terrain in which the fire is burn-
ing, and thefire'sbehavior. Most North American forest fires are contained when
they are relatively small, but some escape initial attack and can grow to very
large sizes. Such fires are usually classed as escaped fires and, when they become
large, they can seldom be controlled until favorable weather conditions arrive
Chapter 15 Forest Fire Management 547

and persist long enough for the fire fighters to estabhsh control lines that are
strong enough to contain the fire if and when weather conditions subsequently
deteriorate.
Forest fire management agencies use fire control status classification systems
to identify a fire's control status and important points and intervals of time in
its life cycle. We combined the fire control status classification system used by
the Ontario Ministry of Natural Resources with the formal mathematical model
and graphical representation of the life cycle of a fire developed by Parks (1964)
and modified the combined system slightly to produce the formal description
of the life cycle of a wildland fire presented in Table 2 and Figure 8.
The sooner a fire is detected and reported, the larger the initial attack force;
and the quicker it arrives at the fire, the more likely the fire will be contained
at a small size. Some fires, however, are so large and/or intense that they can-
not be contained regardless of the size of the initial attack force, and they can
burn freely until a significant change in fuel, weather, topography, or some
combination of these factors occurs and the fire intensity decreases to levels
where fire fighters can work safely to contain their spread.

TABLE 2 Fire Control Status Classification System

Event Description Time Time interval

Ignition Fire is ignited. T^IGN —

Detection Fire is detected by the public or the fire T^DET Detection interval
management agency. ~ 'TDET ~" MGN

Report Fire is reported to the duty officer or the T^REP Report interval
initial attack dispatcher. ~ 'TREP "~ T^DET

Dispatch Initial attack resources dispatched to T^DISP Dispatch interval

the fire. ~ T^DISP ~ T^REP

Arrive Initial attack forces arrive at or near the T^ARR Travel time
fire perimeter. ~ 'TARR ~ 'TDISP

Attack Initial attack suppression action begins. T^ATT Response time

~ TATT ~ TREP

Being Held Fire classed as being held (BHE). T h e fire ^BHE Time to containment
is n o longer spreading b u t has the ~ T^BHE ~ 'TATT
potential to resume spreading in the
near future.
Under Control Fire classed as u n d e r control (UCO). T h e T^uco Control interval
fire is n o longer spreading and is n o t ~ T^UCO ~ '^AJT
expected to resume spreading in the
future.
Out Fire is declared out (OUT). TOUT Mop u p
~ T^ouj ~ T^uco
548 David L. Martell

,.. t
Ignition Detection
! ! ! t
Report Dispatch Arrive Start
7^
Being Under Out
TiGN TDET TREP TDIS at Fire Suppression
HelcT Control TOTTT
TARR TATT (BHE) (UCO)
TRHF TT i p o

Time
FIGURE 8 The life cycle of a forest fire.

Parks (1964) developed a simple single fire suppression model that can be
used to illustrate how the components of a fire management system interact to
determine the final area burned. He assumed that fires were reported as soon
as they were detected, and he defined the detection interval r^ = Tj^^j — TJ^N
as the time between ignition and detection. He assumed that the initial attack
force begins suppression action as soon as it arrives at the fire, and he defined
the attack interval, r^ = T^j^ — Tj^^p, as the elapsed time from detection (re-
port) until fire suppression action begins. He defined the control interval, r^ =
'^BHE ~ T^ATT^ as the time to contain the spread of the fire.
Parks began by ignoring fire geometry and developed a very simple aspatial
fire spread model. He assumed that, in the absence of suppression activity, the
growth of a free-burning fire (area/unit time) at the time it is detected (T^^j)
can be represented by the following differential equation:
dy
dt d)

y{o) = yo
where y is the area burned, dy/dt is the rate of fire growth, t is the elapsed time
measured from the time the fire is detected, y(0) is the size of the fire when it
Chapter 15 Forest Fire Management 549

is detected, G^ is the fire's growth rate when it is detected, and H is the accel-
eration component of fire growth.
Let y^ denote the size of the fire at time Tj\jj when suppression action be-
gins. Then

and at that time the fire will continue to grow at a rate

GA = y'{T^) = GD + HT^ (3)

Consider, for example, the simple case in which a fire burns freely on a flat
area in the absence of wind in the shape of a circle with a radius that increases
at a constant rate which depends on the current condition of the forest fuel
complex and the weather. Let R = rr^) + rt denote the radius of the fire at time
t where r is its constant radial rate of spread and t denotes the elapsed time from
the time the fire was detected (and reported). If A(0 denotes the area of the fire
at time t, then

A{t) = Trr\l + 27rrVc,t + irrh^ (4)

and
dA{t)
= lTTr\j) + Irrrh (5)
dt
so

3/D - TTT^l (6)

Gc, = ITTV^TJ) (7)

and

H = iTrr^ (8)

Parks (1964) modeled the growth of a fire that is being fought as follows. Let
X denote the number of units of fire suppression resource allocated to the fire,
and assume that those resources decelerate the growth of the fire. Let E denote
the efficiency factor for the fire suppression resources. Parks assumed that the
effect of the suppression force is to decelerate the rate of growth of the fire at
some rate which is a linear function of the size of the force and the time it
works. The growth of the fire at time t after T^jj, the time suppression begins,
is, therefore,

dy
--- = G^ + Ht-E{t- T^)x t ^ Tj,
dt (9)
= G^- {Ex ~ H){t - T^)
550 David L. Martell

Let Tc = TgfjE — T^jj denote the elapsed time interval from the start of initial
attack until the fire is declared as being held or contained (BHE). It can be
shown that the area burned from the time suppression action begins until the
fire ceases to spread is

The growth of the fire is composed of two parabolic portions, one when the fire
is burning freely and accelerating until suppression action begins, and a second
after suppression action begins when it begins to decelerate. Note that the num-
ber of suppression units allocated to the fire must be greater than WE to bring
the fire under control. Parks' model is illustrated in Figure 8. In his case, there
is no delay between detection and report so Tj^^j — Tj^^p and no delay between
when the crew arrives at the fire and starts suppression so T^^i^ = T^ATT-
Others have studied the fire suppression process and developed mathemati-
cal models to predict the final size and shape of a fire depending upon its be-
havior, the productivity of the suppression force, and their interaction. Fried
and Fried (1996) reviewed some of the many models that have been developed
and developed a containment model that overcomes many of the limitations in-
herent in earlier models. They showed how their model can be applied and
solved numerically to predict the final size and shape of an elliptical fire that is
subjected to direct attack by ground crews.
One of the most appealing features of the Parks (1964) model is that the final
size of the fire can be expressed as an explicit function of fire behavior and fire
suppression productivity parameters. It indicates that the more resources de-
voted to detection, the faster the fire is detected, the smaller TJ^ is, and the smaller
the final size of the fire is. The more money allocated to initial attack transport
(the use of helicopters rather than trucks, for example), the smaller r^ is and the
smaller the final fire size is, assuming that there are always fire fighters avail-
able to be dispatched whenever a fire is reported (an issue we address later in
this chapter). The more suppression resources dispatched to the fire, the shorter
Tc is and the smaller the final fire size is. The more resources devoted to mop
up, the less likely the fire will escape once it has been declared BHE. Such pre-
dictions are compatible with what fire managers believe about real fire suppres-
sion processes.
The next step is to explore the economics of fire suppression, and Parks
(1964) did just that. He defined the following components of the cost of a fire:
Cp — the fixed cost of maintaining the fire suppression organization
($ per fire)
C5 = cost of dispatching a unit of fire suppression resource to a fire (e.g.,
the cost of transporting the resources to the fire) ($ per unit of sup-
pression resource)
Chapter 15 Forest Fire Management 551

C j = cost incurred by the need to increase the fire organization's readiness

while the fire is burning out of control ($ per unit time)
Cg = cost of fire damage incurred per unit area burned ($ per unit area)
Cx = suppression cost incurred while the fire is burning out of control
($ per unit resource per unit time)
Let XQ = H/E denote the minimum suppression force required to contain the
fire, X be the size of the suppression force, and z == x — XQ denote the number
of suppression units above XQ sent to the fire.
The fire will be contained and declared BHE or reach its final size when it
stops growing or dy/dt = 0. Given Eq. (9), it can be shown that the fire stops
growing when r^^ = t — r^ is given by the following expression:

(11)
^ (Ex - H)
The total cost of the fire is, therefore,

C = Cp + C5X + CjTc + Ceyc + CxTcX (12)

or

C = Cp + C5X + Cx X + Cz yA +
Gl
(Ex - H) 2(Ex - H)
(13)
GA
+ C, (Ex - H)

If we let

CT
CR Cg +

then

(CxG^H/E^) + (QG1/2E)
C{z) = Co + CsZ + (14)

where

C5H CXGA
Co = Cp + —- + — - — + CBJA
t, E
Parks (1964) showed that the optimal number of suppression resources to al-
locate to a fire is given by the following expression:

C R G ^ E + 2CxG^H
(15)
552 David L. Martell

1. A Numerical Example
Consider a fire for which G^ = 2.0 ha hr~\ H = 1.5 ha hr~^, E = 0.5 ha hr~^
per suppression unit, y^ = 0.25 ha, and r^ = 0.57 hr. Suppose that Cp = 0, C5 =
100, Cx = 3.0, Cg = 500, and Cj = 0. Using the expression for the optimum
suppression effort given above, XQ = 3.0, z* = 7.0, and the optimum suppres-
sion effort for this fire is 10 units. Figure 9 illustrates how the total cost plus loss
for this fire varies as the fire suppression effort varies.
Parks (1964) applied his model to some specific fire shapes (e.g., direct and
indirect attack on rectangular fires that burn freely on one side like a fire burn-
ing up a canyon). He also extended his model to deal with the optimization of
two simultaneous fires under certainty, and he solved a simple stochastic fire
suppression problem which he modeled as a Markov decision process.
Fire managers must weigh the costs and potential benefits of deciding how
many resources to allocate to each activity and each fire, and, when they do so,
they must consider other fires that are burning or might be reported in the near
future. Given the number of factors that must be considered and the uncertainty
involved, this clearly is a daunting task. Although the Parks (1964) single and
multiple fire suppression process models provide valuable insight into the be-
havior of fire suppression systems, they are not robust enough to serve the prac-

7000

10 15 20 25
Suppression effort dispatched to the fire
FIGURE 9 Numerical example of Parks' (1964) single fire suppression model.
Chapter 15 Forest Fire Management 553

tical needs of fire managers. We will now describe the basic components of a
modern forest fire management system and study how fire managers deal with
such problems. We begin with the fire load management subsystem depicted in
Figure 2.

B. FIRE LOAD MANAGEMENT

The term fire load refers to the magnitude of the suppression task associated
with fires that occur in a designated area during some specified time interval,
typically a day. The more fires that occur and the larger and more intense they
are when detected, the greater the fire load is. The objective of the fire load man-
agement subsystem is to reduce the fire load to be managed. This can be ac-
complished by preventing fires from occurring, modifying the forest fuel com-
plex to temper the behavior of fires that do occur, and detecting and reporting
wildfires soon after they are ignited so that they are small when the initial at-
tack fire suppression force arrives.

1. Fuel Management
The objective of the fuel management subsystem is to modify forest vegetation
or fuel complexes to reduce the likelihood that fires will occur and to stop or
slow the spread of fires and thereby reduce the social, economic, and biologi-
cal impacts of fires that do occur. It includes the construction and maintenance
of fuel breaks, extensive understory thinning or fuel modification, and the use
of prescribed fire to modify fuels to reduce their flammability, for silvicultural
purposes, or to enhance wildlife habitat. Fire is a natural component of many
forest ecosystems, and agencies that are responsible for fire management in
some parks and wilderness areas may use prescribed fire in an attempt to sus-
tain natural ecosystem processes. Fuel management influences the type and in-
tensity aspects of a fire regime directly, but it also indirectly influences its fire
size or extent. Martell (1982) described some of the decision support systems
that have been developed to enhance fuel management decision making, but our
understanding of fuel management is ultimately based on the extent to which
we understand how fire spreads through heterogeneous fuel complexes that in-
clude patches and strips that have been modified to influence fire behavior.

2. Fire Prevention
Since the objective of the fire prevention program is to reduce the number of
people-caused fires that occur in designated areas, it can reduce thefirefrequency
554 David L. Martell

characteristics of a fire regime. Fire prevention specialists must decide how to

encourage people to refrain from starting fires intentionally or accidentally.
They use media advertising campaigns to transmit prevention messages to the
public, invoke site-specific land use restrictions such as forest closures that pro-
hibit travel in designated areas, or ban the use of campfires by recreationists to
reduce the likelihood that accidental fires will occur. They carry out fire investi-
gations to determine how specific fires were started and sometimes invoke law
enforcement measures to sanction people who start fires. Martell (1982) de-
scribed some of the prevention studies that have been carried out in the past.
However, human behavior is influenced by media advertising that reaches them
throughout their lives, beginning with fire prevention programs directed at
young school children as well as localized special advertising and land use re-
strictions invoked during hazardous fire weather conditions. Therefore, it is
very difficult to assess the effectiveness of prevention measures. As a conse-
quence, fire prevention processes are not well understood, and there are few de-
cision support systems to enhance the cost effectiveness of prevention programs.

3. Predicting Daily Fire Occurrence

Prevention planners need to know when and where people-caused fires might
occur so that they can focus their efforts where they will have their greatest im-
pact. Detection planners need to know when and where both people and light-
ning-caused fires are likely to occur so that they can direct their detection pa-
trols to fly near those fires soon after they ignite. Later in this chapter, we will
illustrate the importance of daily fire occurrence predictions to fire managers
who must deploy their initial attack resources close to fires before they are re-
ported to minimize initial attack response times. In this section, we will discuss
daily fire occurrence prediction, one of the key components of a fire manage-
ment information system. We will deal primarily with people-caused fires to
explore the basic principles of daily fire occurrence prediction, but we will also
summarize very briefly some of the lightning-caused fire occurrence and fire
arrival prediction systems that have been developed.

a. Predicting Uncertain Events

The daily occurrence of forest fires can be characterized as a random or
chance process. Just as we are uncertain whether heads or tails will result when
a coin is tossed, we are almost always uncertain how many fires will occur
each day. On some days fire occurrence predictions can be made with greater
certainty than on others; for example, that no fires will occur on a rainy day
with no lightning. Nevertheless, there is almost always some degree of uncer-
tainty involved.
Chapter 15 Forest Fire Management JDI)

h. Forest Fire Ignitions and Arrivals

It is important to distinguish between the occurrence and arrival of forest
fires. The occurrence time is the time a fire is ignited. Another important time
is the time the fire is first reported to the fire management agency or it "arrives"
and demands the organization's attention. Given the hold-over behavior of some
fires that may smolder undetected for several days before they are detected, par-
ticularly those caused by lightning, it is important to distinguish between pre-
dictions concerning fire occurrences and fire arrivals. Not all fires that occur
will eventually arrive since some of them will be extinguished naturally by
weather or a lack of fuel.

c. Probabilistic Models of People-Caused Fire Occurrence

The binomial probability distribution can be used to model the daily occur-
rence of people-caused fires in a specific area. The binomial distribution is ap-
plicable to processes with repeated independent trials (e.g., Blake, 1979) such
as people subjecting forest fuels to firebrands. Suppose that there are n people
in a forested area during a particular day and that each of those people subjects
the forest fuel complex to a single firebrand. Let each firebrand represent an in-
dependent trial which results in success (a fire occurs) or failure (a fire does not
occur). Let 6 denote the probability that a trial results in a fire, and assume that
it is the same for all the people in the forest. If the probability of success is con-
stant and the trials are independent (people do not influence each other), the
probability distribution of the number of fires that will occur is binomial with
parameters n and 6:

b(x; n, 0) = P{x successes in n trials}

n! (16)
x!(n — x)!r ( l - Oy-'' X - 0,1,2, . . . n

In most jurisdictions, it would be very difficult to determine how many

people are present in a designated area each day. It is, therefore, difficult to use
historical data to estimate 6 and to determine n to predict fire occurrence dur-
ing any particular day. Fortunately, it can be shown (e.g., Blake, 1979), that the
Poisson probability distribution is the limiting form of the binomial distribution
as 6 becomes very small, n becomes very large, and nO remains constant. That
is, lim as ^ —> 0, n —> oo, and n6 remains constant,

X^e~^
b(x; n, 6) = —^ (17)

where A = n^ is the expected number of fires per day.

556 David L. Martell

Since there are usually many people in the forest and the probability that any
one of them will start a fire is very small, it is reasonable to use the Poisson dis-
tribution to model daily people-caused forest fire occurrence. Cunningham and
Martell (1973) studied daily people-caused forest fire occurrence in the Sioux
Lookout area of northwestern Ontario and found that it was reasonable to use
the Poisson distribution to model fire occurrence there.
The Poisson distribution is a single parameter distribution and

P(x) = ^^-^— for X = 0, 1, 2 . . . (18)

where P(x) = probability that x fires will occur and A = expected number of
fires per day. If one accepts the validity of the Poisson model, the task is to de-
velop operational procedures for estimating A for each fire management com-
partment each day.

d. A Simple People-Caused Forest Fire Occurrence Model

Daily people-caused forest fire occurrence is influenced by many factors
including the number of people present in the forest, their behavior which is
influenced by the land use activities in which they are engaged and the pre-
vention measures they have encountered, and the condition of the forest fuel
complex.
The major factors which influence the occurrence of people-caused forest
fires are the ease of ignition of the fine fuel and the potential number of sources
of ignition to which that fuel is exposed. The Fine Fuel Moisture Code (FFMC)
is one of the component indices of the Canadian Forest Fire Weather Index
(Van Wagner, 1987). Since the FFMC was designed as a measure of the mois-
ture content of the fine fuels, one would expect it to have an important effect
upon people-caused fire occurrence. Previous research results (e.g., Cunning-
ham and Martell, 1973) indicate that this is the case.
One of the simplest procedures for predicting daily people-caused fire oc-
currence is to use daily records of the observed FFMC and the corresponding
number of fires that occurred to estimate A, the expected number of fires per
day. This can be accomplished by compiling historical fire occurrence and fire
weather data into the format shown in the following numerical example.

e. A Numerical Fire Occurrence Prediction Example

Suppose that one analyzed a set of historical data and obtained the results
shown in Table 3. If the current day's forecast FFMC is 83, then using the for-
Chapter 15 Forest Fire Management 557

TABLE 3 Relationship between Average Daily People-Caused

Fire Occurrence and the FFMC for a Hypothetical Area

Average
number
of fires
Category FFMC Days Fires per day

1 0-74 200 2 0.01

2 75-79 150 9 0.06
3 80-84 120 18 0.15
4 85-89 100 24 0.24
5 90-100 90 32 0.36

mula for the Poisson distribution with an average of 0.15 fires per day, one
would obtain the probabiUstic prediction shown in Table 4.
Some forest fire management agencies use more sophisticated daily fire oc-
currence predictions such as the logistic regression analysis procedure devel-
oped by Martell et al. (1987) which can be used to predict daily people-caused
fire occurrence by subseason and the procedure developed by Martell et al.
(1989) which models seasonality explicitly. Poulin-Costello (1993) used Pois-
son regression techniques to relate the expected number of fires per day to fire
danger rating indices. The development of lightning stroke counters and light-
ning strike location systems has resulted in a substantial improvement in the
quantity and quality of information available for predicting lightning-caused
forest fire occurrence, and analogous lightning-caused fire occurrence predic-
tion models have been developed. For example, Kourtz and Todd (1992) devel-
oped a daily lightning fire occurrence prediction model that uses fuel moisture
and lightning stroke data to predict fire ignitions. The holdover smoldering
process is modeled to predict how many "detectable" fires are burning unde-
tected in an area each day.

TABLE 4 Probabilistic People-Caused Fire Occurrence Prediction for

a Category 3 FFMC Day with an Expected Value of 0.15 Fires per Day

Number of fires Probability

0 .8607
1 .1291
2 .0097
3 or more .0005
558 David L. Martell

4. Fire Detection
The objectives of the forest fire detection system are to find and report fires
while they are small and to provide the initial attack dispatcher with informa-
tion that will enable him or her to prioritize fires as they are reported and to tai-
lor the initial attack response to the potential impact of each fire. The sooner the
fire is reported and the better the information concerning its precise location,
accessibility, its current size, its spread rate, its intensity, the fuel complexes
ahead of the fire, and the values at risk, the more likely that threatening fires
will be contained while they are still small.
The fire detection system's primary effect is on the size or extent component
of the fire regime. However, since fire detection measures increase the likeli-
hood that a fire will be contained soon after it is ignited and while it remains
small, they reduce the likelihood that fire will continue to burn during later pe-
riods when more severe fire weather prevails or through more hazardous fuel
complexes located elsewhere on the landscape. Fire detection activities, there-
fore, indirectly decrease the area burned by intense fire and thereby influence
the intensity aspects of the fire regime as well.
During the years 1 9 7 6 - 1 9 9 9 the detection size of forest fires in Ontario
ranged from less than 0.1 ha to 20,000 ha, but 0.1 ha or less is the smallest de-
tection size that is assigned to a fire on an official fire report form in Ontario.
The average detection size was 6.4 ha, but the distribution was highly skewed
with a median of 0.1 ha and a mode of 0.1 ha. Of the fires, 95% were less than
or equal to 3.4 ha when first detected. The skewness is largely a result of a few
large fires in the extensive protection zone in the far north.
Forest fire management agencies use fixed lookout towers and patrol aircraft
to detect fires, but they also depend upon the public to find and report fires in
and near populated areas. Each agency's strategy is driven by the relative cost
effectiveness of the different modes of detection and the values at risk. Towers
and patrol aircraft comprise what is commonly referred to as the organized de-
tection system, and the public constitutes the unorganized detection system.
Those terms reflect the belief that, although fire managers can influence the be-
havior and effectiveness of the public, they have much more direct control over
the performance of the organized detection system.
Satellites equipped with infrared scanners and other remote sensing devices
can detect forest fires, but the resolution of currently publicly available satellite
technology is such that satellites cannot find fires until they grow to sizes that
are considered too large for effective initial attack. Satellites are, therefore, used
primarily to monitor large, on-going fires in some remote low-priority areas
where they can provide fire size information with some delay at less cost than
conventional fixed-wing aircraft.
Chapter 15 Forest Fire Management JJ^

a. The Unorganized Detection System

In most jurisdictions, the public views fire as a potentially destructive force
and assumes it has a civic duty to report fires that are burning out of control in
forested areas. Given such attitudes, fire management agencies need not devote
scarce resources to searching for fires that will be detected and reported by the
public while they are still small. It makes little sense, for example, to build a fire
lookout tower at the edge of a forested community or toflya detection patrol air-
craft along a highway as the public will find and report most of the fires that oc-
cur in such areas before they are detected by trained detection observers, but at
little or no cost. Fire management agencies are therefore free to let the public de-
tect fires in and near populated areas and to devote their efforts to more remote
areas where fires might burn undetected for long periods of time.
Fire managers can and should devote some resources to the unorganized de-
tection system and enhance its cost-effectiveness by influencing public attitudes
and enhancing the communications infrastructure to facilitate the reporting of
fires, thereby reducing the time interval between detection and formal report-
ing. Fire management agencies, therefore, publicly stress the importance of re-
porting fires and use both print and broadcast media to publicize how to report
forest fires. The unorganized detection system has benefitted from the increas-
ing availability of cellular telephone technology in many rural areas in recent
years. Since the cost-effectiveness of the unorganized detection system has not
been well studied, fire managers must rely upon their intuition and experience
when they decide how to influence the behavior of the people and thus the per-
formance of the unorganized detection system in their area.

h. The Organized Detection System

Most forest fire management agencies use fire towers or lookouts located
atop high hills or mountains, patrol aircraft, or mixed systems with both towers
and patrol aircraft, for fire detection purposes. Towers are fixed and expensive
to construct and operate, but they provide constant surveillance of an area
when the observer is in the tower. Tower systems, therefore, provide excellent
but costly coverage.
Aerial detection systems are very flexible because patrol routes can vary from
day to day or even hour to hour as fire weather conditions vary, and a patrol
aircraft can be diverted from its planned route to enable the onboard observer
to check suspicious smokes reported by the public. Although aircraft are ex-
pensive to charter or own and operate, most North American forest fire man-
agement agencies that rely on aircraft are satisfied with aerial coverage levels
that cost less than the former towers would cost had they remained in service.
However, detection aircraft provide intermittent surveillance, and once they
560 David L. Martell

pass over a specific area, any fires they miss or any new fires that occur before
they pass near that area again can burn freely undetected and have the poten-
tial to remain undetected so long that they cannot be controlled by the initial
attack force. While this problem can, to a large extent, be addressed by using
many aircraft to fly almost continuous patrols over areas where potentially de-
structive fires are likely to occur, detection patrols never really provide the con-
tinuous coverage possible with extensive tower networks.
Kourtz and O'Regan (1968) studied the economics of fire detection systems
and concluded that tower systems are best for high-value areas under intensive
forest management and that aircraft are best for less-valuable extensive forest
management areas with relatively small detection budgets.
Strategic detection planners must decide how many lookout towers are re-
quired and where they will be constructed. Computer-based spatial analysis
techniques (see Mees, 1976, for example) can be used to assess the "seen area"
surrounding each potential tower site and the number and type of fires that have
occurred within the seen areas in the past.

Detection Patrol Route Planning Many forest fire management agencies

use patrol aircraft to search for fires, and each day detection planners must de-
cide the times at which airborne patrols will leave specified airports and the
routes they will follow Since the time and location of fires is uncertain, this con-
stitutes a very difficult stochastic combinatorial planning problem.
Forest fire managers use the term visihility to describe the ability of an ob-
server to see a smoke, the smoke plume that rises above a small fire burning
under a forest canopy. In Ontario's boreal forest region, the visibility is usually
assumed to be roughly 24 km either side of a patrol aircraft in good weather.
Visibility varies from day to day and depends upon many factors including vege-
tation, atmospheric haze conditions, the location of the sun with respect to the
observer and the fire, the altitude of the observer, the characteristics of the fire
(fuel, weather, size, diurnal variation), and the observer.
Aerial detection observers do not detect every smoke they pass, and the fire
detection process can be modeled as a stochastic process for planning purposes.
Detection planners can divide the fire region into many small cells and define
the detection probability as the conditional probability that the observer finds a
fire when he or she looks in a cell, if there is one fire burning undetected in that
cell. The detection probability depends upon the visibility factors described
earlier. The following problem is a very simplified example of a daily detection
patrol route planning problem which illustrates some of the basic principles of
daily detection patrol aircraft management.

A Simple Detection Patrol Routing Problem Suppose that a fire is known

to be burning in one of a number of cells some distance from the initial attack
Chapter 15 Forest Fire Management 561

• Airport
FIGURE 10 Detection patrol routing.

base but the precise cell in which the fire is located is unknown [as depicted in
Figure 10]. The duty officer must know the precise location and behavior of the
fire before dispatching an initial attack force. Should he/she use a patrol aircraft
or rely on the public to provide that information?
The detection planner must decide whether to send a detection aircraft to
look for the fire and, if so, at what time that patrol should be dispatched. To sim-
plify the problem, we will assume that (1) the fire started at 08:00, (2) the fire
spreads in the shape of a circle and the radius increases at a constant rate of
36 m hr"\ and (3) the fire damage is $200/ha based on the size of the fire when
it is detected. Given these parameters, the fire loss depending upon the time the
fire is found is shown in Table 5.
Assume that the detection probability varies throughout the day as shown
in Table 6 and Figure 11. The expected cost if a patrol is dispatched early
enough to look in the cell at 10:00 A.M. is
(1000 + 320)(0.2) + (1000 + 11,720)(1 - 0.2) = 10 440

TABLE 5 Fire Loss Assuming Fire Is Circular

Time found Hours Area Fire cost

(hr) burned (ha) ($)
10:00 2 1.6 320
12:00 4 6.5 1,300
14:00 6 14.7 2,940
16:00 8 26.1 5,220
18:00 10 40.7 8,140
20:00 12 58.6 11,720
562 David L. Martell

TABLE 6 Detection Probability Function

Look Aircraft detection Public detection

time probability probability

10:00 0.2
12:00 0.4
14:00 0.6
16:00 0.8
18:00 0.6
20:00 —

Detection
Probability

Time
FIGURE 11 Detection probability function.

The expected costs associated with detection flights dispatched at later times are
shown in Table 7. The expected cost plus loss is convex, as shown in Figure 12.
The optimum solution is to dispatch a detection patrol to look in the cell at
14:00 hours.
The tactical detection patrol routing problems faced by detection planners
are much more complex than this simple patrol timing problem. One can ap-

TABLE 7 Expected Cost Plus Loss

Time the Expected cost

detection observer Flying cost plus loss
looks in the cell ($) ($)
10:00 1000 10,440
12:00 1000 8,552
14:00 1000 7,452 Optimum
16:00 1000 7,520
18:00 1000 10,572
ic finds fire at20:00 0 11,720 Do not fly
Chapter 15 Forest Fire Management 563

Expected
cost plus loss

Time
FIGURE 12 Expected detection cost plus loss function.

proach such problems by partitioning the fire region into a large number of small
rectangular cells and predicting the expected number of fires or the parameters
of the probability distribution of the number of new fires that occur in each cell
each hour of the day. Forest vegetation or fuel type maps, weather and fire dan-
ger rating indices, and "values at risk" maps can be used to help identify poten-
tially critical cells that "must" be visited. The task is then to design "good" sets
of patrol routes that will route aircraft over or close to the critical cells.
Kourtz developed and field tested many novel approaches to daily detection
routing problems, many of which were reviewed by Martell (1982). His most
recent approach (Kourtz and Mroske, 1991) was to formulate the problem as
what operational researchers refer to as a multiple salesperson-traveling sales-
person problem, and he used a simulated annealing algorithm to solve such
problems on a daily basis.

C. FIRE SUPPRESSION

Free-burning fires grow in size, become more difficult to control, and, if they
are destructive, cause more damage over time. In the simplest case, a fire on flat
terrain covered with a homogeneous fuel complex will burn in the shape of a
circle in the absence of wind. The perimeter and therefore the difficulty of con-
taining the fire will increase as a linear function of time. One can view fire
suppression as a race between a fire that produces burning perimeter and fire
564 David L. Martell

fighters that produce fire hne. The fire is contained when the amount of fire line
constructed by the initial attack crew is greater than or equal to the perimeter
of the fire. Since the perimeter of a free-burning fire increases as a linear func-
tion of time, the greater the head start of the fire and the faster it is spreading,
the longer it takes to contain the fire.
If one assumes fire damage is a linear function of the area burned, then fire
damage increases as a quadratic or nonlinear function of time, as illustrated in
Figure 8. Fire management systems are, therefore, designed to predict when
and where those fires are likely to occur, deploy fire fighters, helicopters, air-
tankers, and other suppression resources close to areas where fires are expected
to occur, find and report fires soon after they are ignited, dispatch initial attack
resources to fires soon after they are reported, carry out suppression action to
contain fires as quickly as possible, and minimize the damage of large poten-
tially destructive escaped fires that result when fires are not controlled by the
initial attack force. In the following sections, we describe each of these fire sup-
pression subsystems, the decision making associated with their management,
and some of the planning models and decision support systems that have been
developed to enhance their performance.
Fire suppression systems act on individual fires, but we are primarily inter-
ested in their impact on landscape fire regimes. Presuppression planning mea-
sures to support decision making concerning fire suppression resource acquisi-
tion, home basing, and daily deployment help fire managers design and operate
their initial attack system to minimize response times and decrease the number
of fires that escape initial attack and grow to large sizes. They influence the Jire
size or extent component of a fire regime directly but, like detection, they also
have an indirect impact on fire intensity.

1. Fire Suppression Resource Acquisition and Strategic Deployment

Strategic fire suppression resource management entails deciding how many per-
manent initial attack bases and other facilities will be established and where they
will be located, the number and type of air tankers and transport aircraft that
will be purchased and kept for long periods of time, the number of fire fighters
that will be hired, and the number and type of aircraft that will be chartered for
a particular fire season. The more resources acquired, the less likely the agency
will have to pay premium rates for aircraft chartered for short periods of time
and the less likely they will have to incur the transportation costs and waiting
times associated with short-term aircraft charters and borrowing fire fighters
and equipment from other agencies. Once a base has been established, it is dif-
ficult to relocate it. Airtankers are very effective but costly to own and operate,
and, since they do not fly long hours each year, they can last for many years.
Well-trained fire fighters are essential skilled personnel that cannot readily be
Chapter 15 Forest Fire Management 565

replaced. The way such strategic decision-making problems are resolved, there-
fore, has significant impacts that can ripple throughout a fire organization for
many years.
It is difficult to determine how much of each type of fire suppression resource
to acquire and where to base those resources so that they are readily available to
be dispatched to fires as they are reported. Fire management agencies often use
large computer-based decision support systems to help resolve such decision-
making problems. Martell (1982) described some of the many strategic plan-
ning models, most of which focused on airtanker operations, that have been de-
veloped. We discuss the use of such models in Section V where we study level
of fire protection planning.
Fire management planning is a hierarchical process, and decisions concern-
ing the mix of resources required depends upon where those resources will be
permanently based, how they will be repositioned or deployed each day as fire
occurrence ebbs and flows across the protected area, how they will be dis-
patched to fires, and how they will be used on each fire. Fire management sys-
tems are so large and complex that managers must decompose them into hier-
archical systems of decisions such that decisions made at any level within a
hierarchy are compatible with decisions being made above and below that level
in the hierarchy, as described in Section 1.
Consider, for example, the decision of where to home-base airtankers. Fire
occurrence processes vary significantly over both time and space, and the de-
mand for fire suppression resources shifts from day to day and place to place.
Fire suppression resources are very mobile, and fire managers move them from
their home bases to areas where they are most needed on a daily basis. In the
province of Ontario, for example, it is not uncommon to require a large num-
ber of fire fighters and airtankers in different areas, and, when such needs arise,
resources are quickly moved or deployed from their home bases or their cur-
rent locations to the areas where they are most needed. Once the agency has de-
cided how many and what type of airtankers to purchase or lease, it must de-
cide where to home-base them. Given their mobility and the fact that they can
be shared with other agencies, fire managers attempt to develop home-basing
strategies that will minimize the time and flying costs incurred while meeting
shifting daily deployment needs.
Greulich (1976) developed an integrated airtanker home-basing/daily de-
ployment model for two bases in California, but his model would not have been
mathematically tractable for the number of airtankers and bases used in Ontario.
MacLellan and Martell (1996) decomposed the problem into separate home-
basing and daily deployment decision-making systems and developed a mathe-
matical programming model which the Ontario Ministry of Natural Resources
used to help decide how it should home base its airtankers. They dealt with
daily airtanker deployment needs by consulting fire managers and asking them
566 David L. Martell

to express their daily airtanker deployment needs in terms of the number of air-
tankers required at each base as a function of the forecasted fire weather and
the number of fires expected to occur in the area surrounding each base. Their
mathematical programming model accounted for historical fire weather and
fire occurrence patterns and identified home-basing strategies (how many air-
tankers should be home-based at each airport) to minimize the cost of satisfying
the daily airtanker deployment rules identified by the fire managers. That home-
basing system is only as good as the daily deployment rules embedded in the
model and, since they and the Ontario fire managers that developed the model
opted to use subjective deployment rules, it begs the question: How should air-
tankers be deployed at each base each day? Daily airtanker deployment poses
many complex challenges due to the need to deal with stochastic fire occur-
rence and behavior processes that vary over both time and space, so it is rea-
sonable to approximate the deployment practices fire managers will use as an
interim measure. In the next section, we describe how daily suppression de-
ployment needs can be formally modeled and assessed.

2. Initial Attack Resource Deployment

Fire occurrence rates vary over both time and space, and each day fire managers
must decide where to deploy their initial attack resources to minimize initial
attack response times. They must, for example, decide how many fire fighters
to place on initial attack standby at each base each day. The more crews they
place on alert, the more costs they will incur, but the less likely they will expe-
rience crew shortages. They can reduce the number of crews on initial attack
standby but, if they do so, they run the risk of experiencing shortages which
will mean delays in acquiring crews from other bases or agencies, extra costs in
transporting them in to the area experiencing the shortage, and an increased
likelihood that fires will escape initial attack. They face similar decision-making
problems associated with the daily deployment of airtankers and transport air-
craft. Martell (1982) described some of the models that have been developed to
enhance daily deployment decision-making.
Initial attack resource deployment analysis is complicated by uncertainty
concerning the timing and location of fires and interactions between fires. For ex-
ample, if the response time to one fire is excessive, that fire will grow as it waits,
and its size at attack will increase. Initial attack resources will be required on
that fire for a longer period of time so that the next fire that arrives may have
to wait even longer. A delayed response to one fire can therefore produce im-
pacts that can ripple over both time and space and affect the ability of the ini-
tial attack system to respond to far-distant fires several days later. That calls for
systems-level approaches to deployment analysis.
Chapter 15 Forest Fire Management 30 /

Many service systems are designed such that, when the number of customers
that require service is greater than the number of servers, the surplus custom-
ers join a queue where they wait for service. Queues are prevalent in our soci-
ety, and we are accustomed, for example, to queueing for supermarket cashiers,
automatic banking machines, and airline ticket agents. Service system man-
agers must balance the cost of service (more servers cost more money) with the
cost (or inconvenience) to customers associated with waiting for service. For
an introduction to the basic principles of queueing theory, see Ross (1985).

a. Initial Attack Queues

There are many queueing systems in forest fire management systems, one
example of which is an airtanker initial attack system in which fires are cus-
tomers and airtankers are servers. The initial attack queue is a list of fires that
require initial attack action. Airtanker systems are complex queueing systems
since fires arrive at rates that vary over the course of the day, fires grow and
their service times can increase as they wait in the queue, and the airtanker
servers must fly out to serve their customers where they are burning.
To model the behavior of a queueing system, we need to model the cus-
tomer arrival process, the queue, the service discipline, and the service process.
The arrival of customers that need service (fires) can be modeled as a Poisson
process.
One must also describe the queue where customers wait for service. The ini-
tial attack queue is a list of fires with an infinite capacity and the queueing dis-
cipline describes how waiting customers are selected for service. It may be a
first-come-first-served (FCFS) discipline, or it may be a priority queue. If it is
a priority queue, the high-priority customers may preempt lower priority cus-
tomers that are in service when they arrive.
The description of the service process includes the number of service chan-
nels and the service time distribution which is typically exponential or Erlang.
Interarrival times and service time distributions are often modeled as exponen-
tial distributions because their Markovian or memoryless property simplifies
the mathematical analysis of queueing systems (see Ross, 1985, p. 190).
Let T be a random variable with an exponential distribution.

jj{t) = lie ^^ t> 0 Probability density function

(19)
= 0 t<0

Fj{t) = 1 — e~^^ t> 0 Cumulative distribution function

(20)
E{T) = 1//X Var(T) = l//x^
568 David L. Martell

Queueing models can provide a rich variety of performance measures includ-

ing the expected time a customer waits in the queue, the probability that a
customer has to wait for service, and the expected number of customers in the
queue.
The simple M/M/s queueing system model is a model of a system in which
customers arrive according to a Poisson process which has parameters that re-
main constant over time, the service time is exponential, and there is one queue
in front of s servers. More complex models can be developed to deal with situ-
ations in which the fire arrival rate varies over the day due to diurnal variation
in fire occurrence and behavior processes and long travel times make it inap-
propriate to use exponential service time distributions.

A Simple Daily Airtanker Deployment Problem Consider a simple case in

which the fire region depicted in Figure 13 is partitioned into two sectors, and
each sector has one airtanker base that houses all the airtankers that respond
to fires within the sector. Assume that airtankers cannot be dispatched to fight
fires outside the sector in which they are based.
We ignore diurnal variation in the fire arrival rates and the fact that finite
travel time is incompatible with the use of an exponential service time distri-
bution and model each sector as an independent M/M/s queueing system. It can
be shown that W^, the expected waiter time in the queue for an M/M/s queue,
is given by the following formula (see Blake, 1979, p. 346):
A = average fire arrival rate (fires/hr)
l//ji = average time to fight (service) a fire (hr)
p = X/jji
PQ = probability the system is empty

Sector 1

Base 1

FIGURE 13 A hypothetical fire region with two airtanker bases.

Chapter 15 Forest Fire Management 569
s-l fe
(21)
;s-l)!(s-p)
W^, the expected waiting time in the queue, is

(22)
i)\{s-py
Suppose that the fire management pohcy for the area calls for a fire waiting
time of 0.10 hr or less. The expression for W^ can be used to determine how
many airtankers (servers) are required in sectors 1 and 2 to achieve that goal.
Response time targets by sector are W^i 0.10 and W^j - 0-10. The resuks
presented in Table 8 indicate we would need 5 + 9 or 14 airtankers.
To demonstrate the importance of resource sharing, we could aggregate the
sectors 1 and 2 into 1 region with AR = A^ + A2. The results for the aggregate
region are shown in Table 9.
We would need only 12 airtankers to achieve the same level of service for
two cooperating bases. It should be noted, however, that we have ignored the
"extra" travel time required to move between sectors, so 12 is actually a lower
bound estimate of the number of airtankers required and the true number may
be greater than 12.
The queueing models we have described are for simple steady state M/M/s
queueing systems with single bases. It is possible to develop queueing models
of more realistic fire management systems. Martell and Tithecott (1991) ap-
plied an M(t)/M/s for airtankers at a single base when the fire arrival rate varies
over the course of the day. They used numerical methods to solve the differen-
tial equations that described the system and produced time-dependent perfor-
mance measures. They developed software that enabled them to field-test the

TABLE 8 Airtanker Response Time by Sector

Fire Fire Expected

arrival service Number waiting time
rate rate of airtankers, in the queue
Sector (A fires/hr) ilJL fires/hr) s {W,hr)

1 2 0.75 1-2 Undefined

3 3.19
4 0.38
5* 0.09
2 4 0.75 1-5 Undefined
6 1.43
7 0.33
8 0.11
9* 0.04
570 David L. Martell

TABLE 9 Airtanker Response by Region

Fire Fire Expected

arrival service Number waiting time
rate rate of airtankers, in the queue
(A fires/hr) (/I fires/hr) s (%hr)

6 0.75 1-8 Undefined

9 0.87
10 0.27
11 0.11
12 0.05

use of the model in the northwestern region of Ontario. Managers indicated

that, since it did not model base interaction, it would not satisfy their needs. Is-
lam (1998) subsequently studied initial attack airtanker systems and developed
several models for spatial airtanker queueing systems with time-dependent ar-
rivals and interaction between bases, one of which is described in Islam and
Martell (1998).

3. Initial Attack Dispatching

As soon as a fire has been reported, the initial attack dispatcher or duty officer
must decide what suppression resources will be dispatched to that fire and
when they will depart for the fire. Fires are placed in the initial attack queue as
they are reported, and some have higher priorities than others depending upon
their location and the threats they pose to public safety, property, and other val-
ues. Initial attack dispatching is complicated by uncertainty concerning what
might happen later in the day. While airtankers are quite mobile and can read-
ily be diverted from one fire to another as the need arises, initial attack crews
are not as flexible. If an initial attack crew is allocated to a fire, it will not be
available to fight higher priority fires that might be reported later that same day
and, since crews are usually tied up on fires for several days, today's initial at-
tack dispatching effects can ripple throughout the organization for the next
several days.

4. Initial Attack

Once the initial attack crew arrives at the fire, the initial attack fire boss must
assess the situation and devise a suppression strategy. The current fire behavior,
surrounding vegetation, values at risk, number and type of resources currently
available, and the forecast weather will influence his or her decision concern-
Chapter 15 Forest Fire Management 571

ing the attack strategy. He or she may, for example, opt to have the initial attack
crew set up a power pump, lay hose up to the fire, and gradually work around
the fire perimeter while extinguishing the flame front. If there is insufficient
water available, the crew may use hand tools such as axes, shovels, and other
specialized line building equipment. Airtankers may drop water or special fire-
spread-retarding chemicals on the fire while the initial attack crew is traveling
to the fire, and helicopters slinging buckets that can hover and drop water pre-
cisely might work with them as they construct fire line and attack hot spots near
the fire perimeter. The decision making associated with initial attack opera-
tions is the domain of skilled and experienced fire fighters, and there have been
few attempts to develop decision-making aids to enhance their effectiveness.
During the years 1976-1999 the fire size at the start of initial attack in On-
tario ranged from less than 0.1 ha to 27,774 ha, but 95% of the fires were at-
tacked at less than 5.5 ha. The average fire size at attack was 24.8 ha, but the dis-
tribution is highly skewed with a median of 0.1 ha and a mode of 0.1 ha. Again,
the skewness is largely a result of a few large fires in the extensive protection
zone in the far north.

5. Large Fire Management

In many jurisdictions, a very small proportion of the fires burn a large portion
of the area that is burned each year. In the province of Ontario, for example,
95% of the fires burned to a final size of 33 ha or less during the years 1976-
1999. The average final fire size was 141.1 ha, but the distribution was highly
skewed with a median of 0.2 ha and a mode of 0.1 ha. Fires that are not con-
trolled by the initial attack force are classed as escaped fires, and they have the
potential to grow to very large sizes and cause considerable damage. However,
the small number of large fires can have very significant impacts, some of which
are beneficial, on natural forest ecosystems.
Large fires are difficult and very costly to contain, and observers sometimes
question what may appear to be futile suppression activities. One reason, of
course, is that most agencies operate under what are essentially fire exclusion
policies in much of their area, and they and the governments that fund them
might be found legally liable if they simply admitted defeat and decided not to
attempt to control large fires. Nevertheless, many fire managers believe that
they can limit the damage caused by large fires in a cost-effective manner. Their
rationale is as follows.
Large fires are driven across the landscape by weather and in many regions
(e.g., the boreal forest region of Canada) suppression action may not only be
ineffective but also very dangerous to fire fighters while high intensity fast-
spreading fires are moving on wide fronts. Large fires can move as fast as
7 km hr"^ on fronts as wide as several kilometers with flame fronts in excess of
572 David L. Martell

50 m in the boreal forest. It makes little or no sense to attempt to contain fires

while they are making such runs. Under such circumstances, fire fighters some-
times carry out what they refer to as value protection measures. For example,
they may use portable power pumps, hose, and irrigation sprinklers to estab-
lish sprinkler lines around isolated cottages and other valuable structures in the
paths of such fires.
Fire fighters also combat running wildfires by burning out from natural or
human-made barriers. For example, they may use helicopters equipped with
helitorches to fly close to a lake or river in the fire's path, ignite fire at the edge
of the lake, and then continue to light fire progressively closer to the advanc-
ing fire. Their objective is to augment an existing barrier, the lake or river, and
thereby create a much wider fuel break in the hope that they can stop the prog-
ress of the fire when it reaches that barrier.
Fire suppression organizations often simply monitor and project the growth
of running fires but, while they do so, they identify strategic areas where they
hope to establish control lines. They move suppression resources up close to
the fire and establish base camps from which they can launch their attack when
weather conditions turn in their favor and the fire intensity subsides. Fires that
are allowed to burn freely in the largely continuous fuel complexes of the boreal
forest have the potential to continue burning until the advent of winter snow-
fall and, in some cases, they have even been known to smolder in the deep duff
under the snow during particularly bad drought years. But such fires usually
move in short fast-paced bursts. As fire weather conditions worsen, the fire be-
comes more active. If the drying continues for a sufficiently long time, the fire
will, usually under the influence of a strong wind, move rapidly for part of a day
or more until the weather changes and it settles down to smolder until burning
conditions worsen yet again. The fire manager's objective is to mobilize his or
her resources so that the fire can be contained during one of those lulls. If and
when such efforts are successful, the damage that might have been incurred dur-
ing subsequent high-burning-hazard periods will be averted.
Large fire suppression teams may have to determine what values to sacrifice
during burning out operations to enhance the likelihood that they can control
the fire at a later date. They carry out what they refer to as escaped fire situa-
tion analyses (EFSA) to evaluate alternative strategies for dealing with large es-
caped fires.

V. LEVEL OF FIRE PROTECTION PLANNING

The least cost plus damage model provides valuable insight into the potential
long-term economic impact of fire management, and supplementary models like
Chapter 15 Forest Fire Management 573

the fire and timber supply model augment that understanding with detailed in-
formation concerning the timber supply implications of fire management from
landscape and regional perspectives. But fire management objectives should be
based on the potential social, economic, and ecological impacts of fire, and we
have, at best, only begun to scratch the surface in terms of our understanding
and ability to quantify such impacts. Fire management policy analysts and oth-
ers will ultimately have to provide land managers and the public with compre-
hensive planning models that can be used to assess the impact of fire and fire
management on social systems and forest ecosystems. It will take considerable
time and effort to develop our understanding sufficiently to assess the social,
economic, and ecological impacts of fire management. But the public safety,
property, and timber supply concerns that lead to the development of variants
of current policies remain, and fire managers must continue to suppress fire, al-
beit with more discretion than was the case in the past, until the policies that
govern their agencies direct them to do otherwise. Fire managers and their many
diverse clients, therefore, need pragmatic interim solutions that can be refined
as time passes and our knowledge and understanding grow.
In 1935, the U.S. Forest Service developed an effective fire exclusion policy
by using what it referred to as the 10:00 A.M. policy—that all fires should be
contained by 10:00 A.M. the day following the day the fire is detected (Pyne,
1982, p. 116). That surrogate initial attack objective guided their prevention,
detection, and suppression activities to minimize area burned for many years.
Most North American forest fire management agencies developed and imple-
mented variants of the 10:00 A.M. rule. Some agencies recognized explicitly that
they need not and cannot achieve such objectives on all fires, particularly those
that occur in remote areas and do not threaten people, property, and timber
supplies. The Ontario Ministry of Natural Resources, for example (see Martell,
1994), partitioned their fire region into intensive, measured, and extensive pro-
tection zones. All wildfires that occur in the intensive protection zone are at-
tacked aggressively until they are controlled. Fires in the extensive protection
zone are not attacked unless they pose significant threats to people or property.
Initial attack action is taken on all fires that occur in the measured protection
zone. Measured protection zone fires that are not controlled by the initial at-
tack force are subjected to an escaped fire situation analysis that may call for
continued aggressive suppression, modified suppression, or continued moni-
toring without suppression. Even though such zoning schemes suited the needs
of the OMNR and other fire management organizations well in the past, the
small number of very large zones are no longer adequate to meet the complex
needs of fire management agencies that will have to assess social, economic,
and ecological impacts of their activities in far more detail than was ever the
case in the past.
574 David L. Martell

Martell and Boychuk (1994) described conceptually how zoning schemes

can be refined to suit the needs of the many diverse clients that want fire man-
agement on tracts of land they deem as being important. One can begin by par-
titioning the fire region into a large number of zones that are reasonably homo-
geneous with respect to forest ecosystems, land use, and values at risk and by
specifying an average annual burn fraction for each zone. Such fractions could
then be used to help quantify the potential impacts of fire on public safety and
property, and they could readily be incorporated into timber supply models to
assess the timber supply implications of the proposed burn fractions by zone.
Ecologists who study fire regimes recognize the average fire return interval and
burn fraction as important components of fire regimes and will no doubt de-
velop models that relate ecological impacts to burn fractions.
The interested stake holders in each zone could then consider the potential
impact of the proposed burn fraction and the ecosystems in those zones and ar-
rive at a decision, preferably a consensus, as to what average annual burn frac-
tion is acceptable in each zone. The next step would be for the fire management
agency to use a strategic level of protection planning model to determine how
they could meet that objective and the cost of doing so.
Many fire management agencies use modern information technology and
OR/MS methodologies to guide their planning. One approach is to decompose
the fire management system into many small relatively simple subsystems and
then develop large comprehensive computer simulation models that predict the
consequences of specified fire management program alternatives and are used
for fire management budget planning purposes. Examples include LEOPARDS
(McAlpine and Hirsch, 1999), National Fire Management Analysis System
(NFMAS), and California Fire Economics Simulator (CFES) (Fried and Gilless,
1988). Figure 14 is a schematic representation of the basic structure of a fire
suppression computer simulation model. Such models are designed to enable
managers to specify what fire suppression resources they would hire and then
simulate how the proposed system would perform were those resources to be
used to fight historical fires or representative sets of hypothetical fires. The key
issue is model validity or the extent to which such models can truly represent
the real world, and increased effort is expected to be devoted to the develop-
ment, testing, and use of such models in the near future.
Martell et al. (1984) developed an initial attack simulation model (1AM)
which the Ontario Ministry of Natural Resources used to help determine the
number and type of air tankers required in Ontario. That model predicts several
measures of system performance including the fraction of fires that escape ini-
tial attack depending upon the number and type of airtankers, fire fighters, and
transport aircraft allocated to fire suppression. Martell et al (1995) later modi-
fied lAM to produce LANIK, a modern desktop computer implementation of
lAM, to facilitate its use for strategic level of protection planning in Ontario.
Chapter 15 Forest Fire Management 575

Initialize system parameters and fire

management strategies and tactics

Advance calendar by 1 fire season

I
Advance calendar by 1 day and generate the weather

Advance clock to the ignition time of the

next fire and generate its attributes

Allocate suppression resources

Dispatch allocated resources to the

fire as they become available

Fight the fire

Generate system
performance and
fire impact
statistics report

FIGURE 14 Structure of a forest fire management system computer simulation model.

McAlpine and Hirsch (1999) later extended LANIK and embedded it in a GIS
to produce a Windows'95 model called LEOPARDS.
LEOPARDS has since been used for a variety of planning purposes including
an assessment of the potential implications of climate change, changes in ini-
tial attack fire crew staffing levels, and the upgrading of the OMNR's airtanker
fleet with modern CL-415s. Many other forest fire management agencies have
developed similar systems (e.g., the U.S. Forest Service's NFMAS, the California
Division of Forestry's CFES, and Chile's KITRAL [Pedernera and Julio, 1999]).
576 David L. Martell

There is no reason to believe that all the stake holders will agree with the fire
management costs that will emerge when they engage in detailed zonal level of
protection planning exercises, so iterative procedures that hopefully converge
to an acceptable agency wide strategy will have to be used.
After the burn fractions have been specified by zone, they collectively can be
used as a strategic objective to be met by the fire organization. Its task then be-
comes one of determining a detailed site-specific plan for minimizing the cost
of achieving that objective. Level of protection planning models could again be
used to help transform those objectives into a refined fire management plan.
Strategic models like LANIK, LEOPARDS, NFMAS, and KITRAL are aggre-
gate models that will need to be enhanced or replaced as time passes. Fire man-
agers can incorporate simple proposed strategies and tactics in such models in
the form of rules that govern the use of fire suppression resources, but it will
be difficult to translate those strategies and tactics into precise operational de-
ployment, initial attack, and large fire suppression in each zone. They will have
to develop simple operational guidelines for prevention, detection, initial attack,
and large fire suppression that vary by fire danger as measured by, for example,
fire danger rating indices. For example, they will have to specify daily detection
size and initial attack response time objectives by zone given the fire danger.

VI. SOME CHALLENGES

The development and implementation of a fire exclusion policy is relatively
simple and straightforward when compared with the challenges that fire man-
agers will have to deal with in the 21st century. It is, of course, no mean feat
to decide what fire-fighting resources are required and where they should be
based, predict daily fire occurrence, deploy fire suppression resources close to
areas where fires are expected to occur, find the fires that do occur while they
are still small, dispatch initial attack resources to fight those fire and contain
them quickly and at small sizes, and deal with any large escaped fires that re-
sult. These are enormous and potentially costly logistic problems, but they are
manageable given the widespread acceptance that fire is destructive and that fire
managers must minimize the cost of achieving specified fire control. But such
challenges quickly fade when compared with the need to understand the social
and ecological impacts of fire and how human intervention will shape forest
landscapes and the animal populations that will struggle to survive there and to
resolve the many conflicts that different constituencies will bring to the table as
society attempts to develop site-specific fire management strategies for every
significant parcel of forest land. We address some of those issues on the follow-
ing pages.
Chapter 15 Forest Fire Management 577

A. RELATING OUTCOMES TO STRATEGIES

Fire management programs produce fire regimes that forest management clients
and others affected can use to assess the impact of fire management on what
they do and the things they value. It is, however, difficult to predict the fire re-
gime consequences of specified fire management strategies, and even more dif-
ficult to determine how best to achieve a specified fire management regime.
Consider, for example, the simplest case of a small forest management region
that functions independently. It would indeed be difficult to predict what regime
would result by using a specified set of fire management resources according to
some specified sets of rules. Researchers can and have, of course, developed
planning models that are designed to help resolve such problems. Examples in-
clude NFMAS, C F E E S , and LEOPARDS, all of which have already been dis-
cussed. But these systems are primarily initial attack planning tools that focus
on initial attack operations, and their ability to model large fires that escape ini-
tial attack is either nonexistent or, at best, very simplistic.
We can expect progress in this area, but the real challenge lies in the fact that
fire management units are not independent units. Consider the case of two in-
terdependent fire management units that lie adjacent to each other and suppose
airtankers are used for initial attack purposes in both units. One form of fire
management might call for specified initial attack time targets that vary by
compartment and fire weather within each unit. If there is a shortage of air-
tankers in one unit, then fires will have to wait longer in the initial attack queue
during which they will grow and demand even longer service times. That will
reduce the likelihood that the busy unit can share its resources with the adja-
cent unit which will increase the likelihood that fires escape initial attack in the
adjacent unit. In turn, that will lead to an increase in the need for suppression
resources in the adjacent unit and produce shortages in both units that may last
for days or even weeks. Thus, a particularly troublesome fire or a shortage of fire
suppression resources in one unit can have significant impacts that percolate
across time and space. Furthermore, fire management agencies can borrow re-
sources from other provinces, states, and countries. The availability of such re-
sources will vary depending upon the weather in those jurisdictions. The first
challenge is that the set of empirical data required to document what strategies
were used in the past on some designated area is enormous. For example, if the
fire regime is thought to be a function of the initial attack response time strat-
egy, one needs to know how easy it was to borrow resources from other agen-
cies to meet initial attack response time targets. Equally challenging is the fact
that when a manager sets initial attack response time targets for one manage-
ment unit, he or she needs to model the ability of that unit to borrow resources
from other widely scattered agencies.
578 David L. Martell

B. DEALING WITH UNCERTAINTY

Fire is a stochastic disturbance agent that introduces enormous uncertainty into

forest land management planning. Forest managers recognized that fire threat-
ened their abihty to exploit forest resources and took measures to limit its ex-
tent and impact. They developed contingency plans that were designed to en-
hance their ability to cope with fire losses, but their understanding of fire and
the lack of mathematical modeling techniques and computing resources made
it impossible for them to incorporate fire loss explicitly in their planning sys-
tems in anything other than a rudimentary fashion. Reed and Errico (1986) de-
veloped a forest-level timber harvest scheduling model that makes it possible to
account for fire loss in large forest management units if fire losses are not severe
and some deterministic average fire loss is assumed to occur each year. Boy-
chuk and Martell (1996) showed how stochastic programming methods could
be used to deal with variability in fire processes, but stochastic modeling meth-
odologies and computer technology limit the ability to do so. There is, of course,
a need to enhance stochastic programming planning methods, but even more
important is the need for forest-based societies to learn to live in harmony with
the variability that is characteristic of natural forests.
Gunn (1996) has suggested that replanning after significant uncertain events
have taken place is a reasonable strategy for coping with uncertain fire losses
and, assuming forests are not "pushed to the limit" with respect to timber pro-
duction, they may be robust enough to "respond." One cannot maintain both
natural ecosystems and stable timber harvest flows simultaneously. Societies
must develop site-specific compromises for such problems, and they have to
learn to cope with the variability that will result. That might, for example, call
for explicit recognition that harvest levels will rise and fall and the aggregation
of relatively small independently owned forest management units into larger
woodsheds in which forest landowners agree to share their resources with those
that suffer loss so that each of them enjoys a measure of insurance concerning
resource flows. We will also have to learn how to deal with the problems that
will arise when we embed islands of flammable "natural" forest landscapes that
are designed to provide habitats for rare and endangered species in larger land-
scapes that are not so well endowed with suitable habitat. To date the forest fire
management research community has focused on timber production under un-
certainty. It is essential that ecologists and land management planners work to-
gether to deal with the much more complex issues of wildlife conservation in
managed forests that are expected to provide both resources and suitable wild-
life habitats.
Chapter 15 Forest Fire Management 579

C. MOVING BEYOND FIRE EXCLUSION

Most modern wildland fire management agencies were establislied and devel-
oped in response to threats to public safety, property, and timber production.
Despite the growing recognition that fire is a natural component of many for-
est ecosystems and that it is neither economically nor ecologically sound to at-
tempt to exclude it completely from forested landscapes, there has not been a
significant shift away from traditional fire exclusion practices in most jurisdic-
tions. Fire has, of course, been "reintroduced" into some areas, primarily parks
and wilderness areas where the desire to maintain or restore natural ecosystem
processes is paramount and the land managers responsible believe that they can
increase the amount of fire on the landscape without undermining public safety
and property significantly.
Choosing not to fight all fires aggressively does, however, pose significant
challenges to the managers who must develop and implement such policies, the
fire fighters who must deliver them, and the public who must live with the con-
sequences. The primary motivation for fire exclusion is the desire to reduce
uncertainty and minimize the risk that significant losses will be incurred in the
near future. While fire exclusion may lead to hazardous fuel buildups and po-
tentially even more destructive fires in the distant future in some biomes, every
wildfire that is not extinguished represents a potential short-term threat to pub-
lic safety, property, and timber. Historical accounts of wildland fire disasters in
North America during the early decades of the 1900s describe how hundreds
of lives were lost as fires swept across the landscape and engulfed entire com-
munities and the rural homesteads between them, almost without warning (see,
for example, Holbrook, 1943; Lambert and Pross, 1967; Pyne, 1982). There is
no simple common explanation for all such tragedies, but, in many cases, small
land clearing and lightning-caused fires had been left to burn largely unat-
tended across the landscape. That posed no significant threat as long as benign
fire weather conditions prevailed, but from time to time nature conspired to
produce several good drying days and one or more days with high tempera-
tures, low relative humidities, and high winds. The heretofore small benign fires
then began to spread, joined up, and raced across the landscape on wide fronts
pushed by strong winds. The surveillance, telecommunications, and transpor-
tation systems were not up to the task, and the resulting damage precipitated
calls for fire exclusion that largely persist to the present.
Modern forest fire management agencies have the capability to monitor on-
going fires, and they can rely on meteorologists to forecast weather so that they
can predict potential fire behavior some days in advance. They can, in principle,
predict when fires are about to make a run, intervene, and extinguish them be-
fore that happens. If and when they are unable to do so, they can use modern
580 David L. Martell

transportation technology to evacuate threatened communities. The irony is

that, since such pohcies would not produce the large burned areas characteris-
tic of natural fire regimes, it is questionable what environmental benefits would
be gained by such practices.
There are no simple answers. Martell (1984) addressed this issue as it per-
tains to the boreal forest region of Canada and advocated the adoption of what
he described as "fire impact management policies" whereby decisions concern-
ing wildfire suppression and the use of prescribed fire are based on sound so-
cial, economic, and ecological principles. He also discussed some of the practi-
cal problems that would complicate the development and implementation of
such policies.
The simple truth is that fire cannot be reintroduced into fire-prone land-
scapes without significant threat to public safety, property, and other values. A
fire manager that cannot control what proves to be a destructive fire will not be
criticized if he or she is judged to have done his or her best to deal with the sit-
uation in a professional manner. However, if he or she decides to let what ulti-
mately proves to be a destructive fire burn with little or no significant suppres-
sion, he or she and the agency that employs him or her will no doubt be liable
for civil litigation. The issue is not unlike the concern Howard et al. (1972) iden-
tified when they realized that decisions to seed hurricanes to diminish their
destructive potential should be tempered by the possibility that a change in the
storm's characteristics might, rightly or wrongly, be attributed to the seeding
operation and could open the government to civil litigation.
Mitchell (1995, p. 4) explains that the precautionary principle that emerged
from the Earth Summit in Rio de Janeiro in 1992 recognizes that even though
"uncertainty creates a serious dilemma for resource and environmental manag-
ers. . . . incomplete understanding should not be used as an excuse for delay-
ing action when environmental degradation appears imminent." Some might
well be tempted to use the precautionary principle to support calls for nonex-
clusion policies, but that must be tempered by concern for public safety and
property damage and the possibility that, although fire is natural, some forest
ecosystems may have been so perturbed by humans that "natural" fire regimes
would simply hasten their demise. Fire management is ecosystem management
over massive spatial and temporal landscapes, and it is fraught with considerable
uncertainty concerning what will happen in the next few days, weeks, decades,
or centuries. Given the importance of fire in natural ecosystem processes, we
cannot afford to simply continue on with the way we in North America have be-
haved for the last century. It should all make for an interesting future.

ACKNOWLEDGMENTS
J. Beverly and A. Tithecott provided helpful comments on earlier versions of this chapter. The
Aviation, Flood and Fire Management Branch of the Ontario Ministry of Natural Resources pro-
Chapter 15 Forest Fire Management 5ol

vided a digital file of their 1976-1999 fire report data that was used to compute the Ontario fire sta-
tistics reported in this chapter.

FURTHER READING
Winston, W. L. (1994). "Operations Research: Applications and Algorithms." Duxbury Press,
Belmont.

REFERENCES
Baumgartner, D. C , and Simard, A.J. (1982). "Wildland Fire Economics: A State of the Art Review
and Bibliography." Gen. Tech. Rep. NC-72. USDA Forest Service, North Central Forest Experi-
ment Station, St. Paul.
Blake, I. F. (1979). "An Introduction to Apphed Probability." John Wiley and Sons, New York.
Boxall, P. C , Watson, D. O., and Englin, J. (1996). Backcountry recreationists' valuation of forest
and park management features in wilderness parks of the western Canadian Shield. Can. J. For.
Res. 26, 982-990 .
Boychuk, D. B., and Martell, D. L. (1996). A multistage stochastic programming model for sus-
tainable forest-level timber supply under risk of fire. For. Sci. 42, 10-26.
Clough, D.J. (1963). "Concepts in Management Science." Prentice-Hall, Inc., Englewood Cliffs.
Cunningham, A. A., and Martell, D. L. (1973). A stochastic model for the occurrence of man-
caused forest fires. Can. J. For. Res. 3, 282-287.
Fried, J. S., and Gilless, J. K. (1988). "The California Fire Economics Simulator Initial Attack Model
(CFES-IAM): MS-DOS Version 1.11 User's Guide." Bulletin 1925. Division of Agriculture and
Natural Resources, University of California, Oakland.
Fried, J. S., and Fried, B. D. (1996). Simulating wildfire containment with realistic tactics. For. Sci.
42, 267-281.
Gass, S. I. (1994). Public sector analysis and operations research/management science. In "Opera-
tions Research and the Public Sector" (S. M. Pollock, M. H. Rothkopf, and A. Barnett, Eds.),
Handbooks in Operations Research and Management Science Vol. 6, pp. 23 - 46. North-Holland,
New York.
Gorte, J. K., and Gorte, R. W (1979). "Application of Economic Techniques to Fire Management—
A Status Review and Evaluation." Gen. Tech. Rep. INT-53. USDA Forest Service, Intermountain
Forest and Range Experiment Station, Ogden.
Greulich, F. E. (1976). "A Model for the Seasonal Assignment of Airtankers to Home Bases Under
Optimal Daily Transfer Rules." PhD dissertation. University of California, Berkeley.
Gunn, E. A. (1996). Hierarchical planning processes in forestry: A stochastic programming-decision
analytic perspective. In "Workshop on Hierarchical Approaches to Forest Management in Pub-
he and Private Organizations" (D. L. Martell, L. S. Davis, and A. Weintraub, Eds.), pp. 85-95.
Inf. Rep. PI-X-124. Canadian Forest Service, Petawawa National Forestry Institute, Chalk River.
Holbrook, S. H. (1943). "Burning an Empire." The Macmillan Company, New York.
Howard, R. A., Matheson, J. E., and North, D. W. (1972). The decision to seed hurricanes. Science
176,1191-1202.
Islam, K. M. S. (1998). "Spatial Dynamic Queueing Models for the Daily Deployment of Airtankers
for Forest Fire Control." PhD dissertation. University of Toronto, Toronto.
Islam, K. M. S., and Martell, D. L. (1998). Performance of initial attack airtanker systems with in-
teracting bases and variable initial attack ranges. Can. J. For. Res. 28, 1448-1455.
Johnson, K. N., and Scheurman, H. L. (1977). Techniques for prescribing optimal timber harvest and
investment under different objectives—Discussion and synthesis. For. Sci. Monogr. 18, 1-31.
582 David L. Martell

Kourtz, P. H., and Mroske, B. (1991). "Routing Forest Fire Detection Aircraft: A Multiple-Salesman,
Travelling Salesman Problem." Canadian Forest Service, Petawawa National Forestry Institute,
Chalk River.
Kourtz, P. H., and O'Regan, W. G. (1968). A cost-effectiveness analysis of simulated forest fire de-
tection systems. Hilgardia 39, 341-366.
Kourtz, P. H., and Todd, B. (1992). "Predicting the Daily Occurrence of Lightning-Caused Forest
Fires." Inf. Rep. PI-X-112. Canadian Forest Service, Petawawa National Forestry Institute, Chalk
River.
Lambert, R. S., and Pross, P. (1967). "Renewing Nature's Wealth." Ontario Department of Lands and
Forests, Toronto.
MacLellan, J. I., and Martell, D. L. (1996). Basing airtankers for forest fire control in Ontario. Oper.
Res. 4^, 677-686.
Martell, D. L. (1982). A review of operational research studies in forest fire management. Can. J.
For. Res. 12, 119-140.
Martell, D. L. (1984). Fire impact management in the boreal forest region of Canada. In "Resources
and Dynamics of the Boreal Zone, Proceedings of a Conference held at Thunder Bay, Ontario,
August 1982" (R. W. Wein, R. R. Riewe, and I. R. Methven, Eds.), pp. 526-533. Association of
Canadian Universities for Northern Studies, Ottawa.
Martell, D. L. (1994). The impact of fire on timber supply in Ontario. For. Chron. 70, 164-173.
Martell, D. L., and Boychuk, D. (1994). "Levels of Fire Protection for Sustainable Forestry in On-
tario: Final Report." Report prepared for the Canada-Ontario Northern Ontario Development
Agreement, Sustainable Forestry Development/Decision Support Project: Developing Analyti-
cal Procedures for Establishing the Level of Protection for Forest Fire Management to Support
Sustainable Forestry in Ontario.
Martell, D. L., and Tithecott, A. (1991). Development of daily airtanker deployment models. In
"Proceedings of the 1991 Symposium on Systems Analysis in Forest Resources" (M. A. Buford,
Ed.), pp. 366-368. Gen. Tech. Rep. SE-74. USDA Forest Service, Southeastern Forest Experi-
ment Station, Asheville.
Martell, D. L., Bevilacqua, E., and Stocks, B. J. (1989). Modeling seasonal variation in daily people-
caused forest fire occurrence. Can. J. For. Res. 19, 1555-1563.
Martell, D. L., Boychuk, D., MacLellan, J. I., Sakowicz, B. M., and Saporta, R. (1995). Decision
analysis of the level of forest fire protection in Ontario. In "Symposium on Systems Analysis in
Forest Resources: Management Systems for a Global Economy with Global Resource Concerns"
(J. Sessions and J. D. Brodie, Eds.), pp. 138-149. Pacific Grove, California.
Martell, D. L., Drysdale, R. J., Doan, G. E., and Boychuk, D. (1984). An evaluation of forest fire ini-
tial attack resources. Interfaces 14(5), 20-32.
Martell, D. L., Gunn, E. A., and Weintraub, A. (1998). Forest management challenges for opera-
tional researchers. Euro. J. Oper. Res. 104, 1-17.
Martell, D. L., Kourtz, P. H., Tithecott, A., and Ward, P. C. (1999). The development and imple-
mentation of forest fire management decision support systems in Ontario, Canada. In "Pro-
ceedings of the symposium on fire economics, planning, and policy: Bottom Lines" (P. N. Omi
and A. Gonzalez-Caban, technical coordinators), pp. 131-142. Gen. Tech. Rep. PSW-GTR-173.
USDA Forest Service, Pacific Southwest Research Station, Albany, Cahfornia.
Martell, D. L., Otukol, S., and Stocks, B. J. (1987). A logistic model for predicting daily people-
caused forest fire occurrence in Ontario. Can. J. For. Res. 17, 394-401.
McAlpine, R. S., and Hirsch, K. G. (1999). An overview of LEOPARDS: The level of protection
analysis system. For. Chron. 75, 615-621.
Mees, R. M. (1976). "Computer Evaluation of Existing and Proposed Fire Lookouts." Gen. Tech.
Rep. PSW-19. USDA Forest Service, Pacific Southwest Forest Experiment Station, Berkeley.
Merrill, D. R, and Alexander, M. E., Eds. (1987). "Glossary of Forest Fire Management Terms."
Chapter 15 Forest Fire Management 583

Fourth Edition. Canadian Committee on Forest Fire Management, National Research Council
of Canada. Ottawa.
Mitchell, B. (1995). Addressing conflict and uncertainty. In "Resource and Environmental Man-
agement in Canada: Addressing Conflict and Uncertainty" (B. Mitchell, Ed.), pp. 1-8. Oxford
University Press, Toronto.
Parks, G. M. (1964). Development and application of a model for suppression of forest fires. Man-
age. Set 10,760-766.
Pedernera. P., and Julio, G. (1999). Improving the economic efficiency of combatting forest fires in
Chile: The KITRAL system. In "Proceedings of the symposium on fire economics, planning, and
policy: Bottom lines" (P. N. Omi and A. Gonzalez-Caban, technical coordinators), pp. 149-155.
Gen. Tech. Rep. PSW-GTR-173. USDA Forest Service, Pacific Southwest Research Station, Al-
bany, California.
Plonski, W. L. (1974). Normal yield tables (metric) for major forest species of Ontario. Division of
Forests, Ontario Ministry of Natural Resources, Toronto.
Pollock, S. M., and Maltz, M. D. (1994). Operations research in the public sector: An introduction
and a brief history. In "Operations Research and the Public Sector" (S. M. Pollock, M. H. Roth-
kopf, and A. Barnett, Eds.), Handbooks in Operations Research and Management Science, Vol-
ume 6, pp. 1-22. North-Holland, New York.
Poulin-Costello, M. (1993). "People-Caused Forest Fire Prediction Using Poisson and Logistic Re-
gression." MSc thesis. University of Victoria, Victoria.
Pyne, S.J. (1982). "Fire in America: A Cultural History of Wildland and Rural Fire." Princeton Uni-
versity Press, Princeton.
Pyne, S. J. (1997). "World Fire: The Culture of Fire on Earth." University of Washington Press,
Seatde.
Reed, W J., and Errico, D. (1986). Optimal harvest scheduling at the forest level in the presence of
the risk of fire. Can. J. For. Res. 16, 266-278.
Ross, S. M. (1985). "Introduction to Probability Models," 3rd ed. Academic Press, Orlando.
Sparhawk, W N. (1925). The use of liability ratings in planning forest fire protection. J. Agric. Res.
30,693-762.
Van Wagner, C. E. (1979). The economic impact of individual fires on the whole forest. For. Chron.
55, 47-50.
Van Wagner, C. E. (1987). "Development and Structure of the Canadian Forest Fire Weather In-
dex System." Forestry Technical Report 35. Canadian Forestry Service, Ottawa.
Whalen, R. J. (1995). "The Ecology of Fire." Cambridge University Press, Cambridge.
This Page Intentionally Left Blank
 SUBJECT INDEX

thermal diffusivity, 495-496

Adsorption, see Sorption thickness, 491,518
Advection, 173 Bayes Information Criterion, 430-431
Albedo, 306 BEHAVE, 158, 183
Apoplast, 85, 101 Biomass, burned, 56
Approach Biosphere-Atmosphere Transfer Scheme,
empirical, 442-443 336-338
process-based, 2, 443 Biot number, 94, 128,510
Area burned, 534, 541, 550 Black body, 445
atmospheric conditions, 358 Boundary layer, 4, 107-108, 266
future warming, 365-369 planetary, 315
Ontario, 545 Brunauer-Emmett-Teller theory, 120
predictors, 354 Burn fraction, 544
Arrhenius Burst, turbulent, 266
equation, 489-491 Byram index, 162
number, 490
Atmosphere
height anomalies, 356
mixed phase region, 376 Cambium, vascular, 486
upper, 354-357 mortality, 486
vertical structure, 357-358 necrosis, 488-489, 497-505
Canadian Fire Weather Index System, 343,
354,385
B Capillary
Backfire, 181 flow model, 124
Balance pressure, 99
cyclostrophic, 264-265 Capillarity, 123-124
energy, 317-318 Carbon, release, 6 3 - 6 5
Bark, 486 Cell wall
density, 496 primary, 85
heat capacity, 495 secondary, 88

585
586 Index

Cells, convective, 295-299 Contagion effect, 424

Centrifugal pump effect, 501 Dirichlet distribution, 424
Change points, 422-424 overdispersed multinomial model, 4 2 4 -
location, 430-435 427
significance test, 426 Convection, 4, 268, 275, 290, 481-482, 487
Char, 450 atmospheric, 3
composition, 452 coefficient, heat transfer, 483, 505, 508,
oxidation, 452-455, 460, 462 510-513
production, 83, 443, 450-452 forced, 482
Charge separation, 376-377 free, 481
Chimney effect, 502-503 Cooling
Chromatography, gas, 21 Newtonian model, 236-237
Circulation, 290-292 Newton's law, 482, 507, 511
atmospheric modeling, 269-272 Coupling, atmosphere-fire, 266-268
Climate, 351 modeling approach, 272-277
Cloud, smoke ingestion, 411 problems, 289
Combustion, 3, 12, 13 Crown death, 230, 507-508
chemistry, 58-59
efficiency, 16, 6 4 - 6 5
flaming, 62, 83, 442, 443-448 E)
glowing, 62, 448-449 Damkohler number, 27-28, 212
governing equations, 13-15 Darcy's law, 99, 124
incomplete Decision support systems, 529
products, 15 Deployment, air tanker, 530-531
flame temperature, 15 Desorption, see Sorption
phases, 6 0 - 6 2 Diffusion, see also Water transport mechanisms
products, 58-59 bound water, 99, 124
residence time, 235-236 surface, 124
smoldering, see Smoldering combustion theory, 93, 115, 128, 129
Concentration, chemical species, 18 transverse, 92
measurement, 21 Diffusivity
Condensation, 132 moisture, 85, 92-93, 129, 135
Condensed phase, 26 thermal, 8,159, 488, 491-497
Conduction, 4, 456-457, 459, 480-481 bark, 493,495
Fourier's law, 480, 494, 511 equation, 492
transient, 487 measurement, 493-494
Conductivity, thermal, 481 Dimension, characteristic, 510
air, 401 Dimensional analysis, 5, 226
bark, 495 Dirichlet distribution, 424
measurement, 494 Distillation, 132
soils, 514 Drag, frictional, 263, 265
Cone, opening, 508-509 Drought Code, 343
Conflagration, 259 Duff, 81,437
Conservation characteristics, 439-441
angular momentum, 263 consumption, 83, 438
energy, 198 empirical approach, 443, 464
equations, 13-15, 98-100, 154, 226, 228, empirical studies, 441-443
245,270, 293, 394, 506 flaming combustion, 444-448
Constant-rate period, 126-127 heat budget model, 444-447
Index 587
depth, 404, 440-441, 447, 465, 467 temperature, 326-327
moisture content, 91-92, 467-468 Thornthwaite, 326
moisture transport, 98 potential, 319, 327-328
specific heat, 446 characteristics, 329-333
types, 391 elevation dependence, 330 -332
Duff Moisture Code, 343, 354 estimation, 317-324
Dynamics seasonal cycle, 329-330
atmospheric, model, 286 topographic effects, 332-333
vorticity, 260-266 rate, 109
understory, 333
Extinction criterion
E Damkohler number, 27-28, 212
Ecology, fire, 2, 5 forced-convection, 201
El Nino, 359-361 Zel'dovich, 27-28
fire occurrence, 360 Extractives, 8 4 - 8 5
seasonal precipitation, 363
Emission
factors, 63-64, 67
trace gases, 69-72 Falling-rate period, 126-127
carbon monoxide, 71-72 FARSITE, 183
nitrogen, 69-70 Fiber saturation point, 89-90
nonmethane hydrocarbons, 72 Fick's law, 99
sulfur, 70-71 Fine Fuel Moisture Code, 343, 387, 556
methane, 72 Fire(s), 12
methyl chloride, 71 arrival, 555
Emissivity, 445 behavior, 271
Energy budget, surface, 337 models, 158, 183-184,286
remote sensing, 339-340 blowup, 258, 276
Entrainment, mass, 244-245 control status, 547
Environment, near-surface, 333-336 crib, wood, 3 8 - 4 0
Epochs, 420 cycle, 427-433
determining, 430-435 danger, 138, 156
Eulerian approach, 185 indices, 366-368, 385, 341-343,
Evaporation 353,385
pan, 325-326 detection, 558
coefficients, 326 unorganized system, 559
rate, 310 organized system, 559-563
potential, 107 patrol route planning, 560
resistance, 88 probabihty, 560 -562
temperature effects, 313 economics, 534
vertical mixing, 313-317 least cost plus damage, 535-538
processes, 314 suppression, 550-553
wind speed, 313 effects
Evapotranspiration, 108, 305 bole, tree, 483-505
actual, 319, 327-328 canopy, 505-514
characteristics, 329 mortality, tree, 515-517
functional dependence, 327-328 roots, 514-515
models exclusion, 579
radiation, 324-325 policy, 573
588 Index

Fire(s) (continued) model, 546-553

extreme events, 172 strategic, 564-566
flanks, 188 surface, 258
frequency, 3, 420, 553, see also Time- temperature profile, 154, 225-226
since-fire transition, surface to crown, 80, 100
grass, 267 visibility, 560
head(ing) 152-158, 185 vortices, 258, 287
holdover, 391 weather rating systems
impacts Canada, 343
assessing, 534-544 United States, 341-343
recreation, 543 whirls, 259
timber supply, 538-544 wildland, 176
large, 258 Firebrand, trajectories, 181, 251-252, 357
life cycle, 547-548 Firefront, 172-176
line, 278-282 elliptic, 184
intensity, 485 locus, 189
parabolic/hyperbolic, 280-281 perimeter, 173-175, 184, 189, 217-218
wind effects, 278 velocity, 186-187
management, see Management, fire propagation, 180-181, 185, 188, 197-
modeling, whole fire, 160, 269 203
resolution, 273 tracking, 189
number, 545 Fire spread, 42
occurrence atmospheric stability effects, 282-285
daily, 554-557 Australia, 155-156
lightning-caused, 557 Canada, 158
people-caused, 555-556 chaparral, 186-192,208
pool, 36, 230 equation, fundamental, 456
smoke, 17 equilibrium, 218
scaling laws, 37 fuel factors, 179, 206, 208, 212, 214-217
prevention, 530, 553-554 fundamental equation, 153
propagation function, 186 models
protection containment, 550
planning, 572-576 McArthur, 155-156
zones, 573-574 Parks, 548
regime, 530, 544 probabilistic, 163, 194
scar, 498-504, 516 Rothermel, 156-158, 202-203
science, 528 wind-aided, 179-194
season, 368 one-dimensional, 192, 193, 201
shape, 160-163, 172-176, 184-186 radiation effects, 272
circular, 276-277, 549 rate, 152-158, 173-174, 178, 185, 196,
ellipse model, 160-161, 175, 184 202,249,258-259,485,504
fire intensity, 162-163 slope effects, 157, 193, 203, 282-285
wedge, 175, 190-191 surface, 180
wind speed, 161-162 three-dimensional, 227
size, detection, 558 two-dimensional, 202
stand-replacement, 198 United States, 156-158, 180-181
structures, 288 wind-aided, see Chapter 6
suppression, 528, 563-572 wind effects, 157, 178, 203, 212,
economics, 550-553 214-217
Index 589
FIRETEC, 286 characteristics, 84
Fire Weather Index, 366 -368 chemical, 8 4 - 8 5
Fisher information matrix, 425, 428 structure, 93-93
Flame, 3, 12, 198 physical, 93-96
characteristics, 483-486 chemistry, 57, 65-66, 8 4 - 8 5
color, 20 consumption, 83
curvature, 185 curing, 114
diffusion, 13, 23-24, 185-186, 398 depth, 231,465
color, 33 drying, 126-129
extinction, 27-28 forest, 80, 81-96
laminar, 29-35 classification, 81-82
methane-air, 30-35 inert content, 209-212
model, 61 loading, 83, 194, 201, 205, 207
scaling analysis, 29 management, 553
turbulent, 30, 36 moisture content, 3, 79, 304, 403-404,
height, 18, 231, 503 486
measurement, 22, 32 burning rates, 82-84
premixed, 23-24 coarse fuels, 305
velocity, burning, 24-25 dead fuel, 115-137,305
thickness, 24-25 diurnal prediction models, 136
residence time, 485, 489, 497-498, equilibrium, 115
500-501 estimation, 137-139
spreading, 41-45 fine fuels, 305
downward, 43 fire sensitivity, 286
horizontal, 43 hysteresis, 115
upward, 44 indicator sticks, 138
structure, 4 5 - 4 7 live fuel, 110-115
temperature, 484 timelag, 130
adiabatic, 15, 27 response time, 130
maxima, 229-243, 517-518 soil moisture effect, 133
measurement, 19, 484-485, 517- weather factors, 130, 304, 353
518 moisture relationships, 96
moisture effects, 83 porosity, 205
profile, 503-504 reduction, 251
vegetation fires, 69 shrubland, 236
velocity, 18 state, 383
measurement, 20 type, 43, 383
Flaming combustion, 62, 83, 442, 443-448 Fuel-bed
Flammability, 517 porous, 45
factors, 111-113 parameters, 157
limits, 26 properties, 193, 205, 233
Floor, forest, 437
Flow, rotational/irrotational, 499-500
Fourier G
law of heat conduction, 100 Gas law, ideal, 246
number, 128 Gasification, 13
Froude number, 29, 259, 267-268 General Circulation Models, 365-366
Fuel(s), see also Smoldering combustion Gibbs free energy, 96-97, 120
arrangement, 205-209 Girdling, 497
590 Index

Gradients experiments, 402-406

moisture/nutrient, 467 fuel depth, 404
vertical, 333-334 heat of, 82, 159
Graupel, 377,411 lightning, see Lightning, ignition
minimum energy, 27
modeling, 393
H piloted, 26, 398-399,483
Hazard, burning, 420, 423 probabihty, 82, 368
confidence interval, 426, 433 process, 391-401
maximum likelihood estimates, 423-424, spontaneous, 26, 398-399
427,431-433 temperature, 26
variance, 425 thunderstorm type, 381-382
Heat time, 82
capacity, 492 tracing, 274-277
measurement, 494-495 unpiloted, 26
flux, 241, 319, 399-401, 481 Infiltration, 123
downward, 199 Infrared
latent, 4, 337, 340 observations, 286-288, 340
vaporization, 305, 446, 514 thermograph, 19
loss, 19 Initial attack, 546, 570-571
sensible, 275, 288, 340 dispatching, 570
sofl,337-338 queues, 567-568
transfer, 3 resource deployment, 566-570
bole, 486-505 Intensity, fire or reaction, 156
equation, one-dimensional, 486 Interception, precipitation, 339
fire to trees, 477-479 Intertropical convergence zone, 412
lightning to forest fuels, 401 Inversion, 357
mechanisms, 479-483 Islands, unburned, 176
model, 7 - 8
rates,4
scales, 273 K
Heating Keetch-Byram Drought Index, 114
sofl, 514-515 Kirchhoff's law, 445
unsteady state, 486
Hessian matrix, 425, 428
HIGRAD, 286 L
Humidity, relative, 311-313 La Niiia, 359-361
vertical profile, 334 Laminar sublayer, 333-334
Humus, 437 Land-surface interactions, 336-339
Huygens' principle, 184 Land Surface Model, 336-338
Hygroscopicity, 84 - 85 Laplace transform, 199-200, 487
Lapse rate, 357
adiabatic, 314
Layer, surface, 315-316
Ice, crystal growth, 377 Leader, stepped, 377
Ignition, 25-26, 184 Leaf-atmosphere models, 107-110
autocatalytic, 25 Leaf, drying, 109-110
canopy, 269 Least cost plus damage, 535-538
defining, 403 limitations, 536-538
Index 591
Lightning, 3, 376 Mass flow, 96-97
channel, 378 Matts, root, 438, 440
collapse, 396 Maximum likelihood estimate, 423-424, 427,
structure, 394-397 431-433
temperature, 396-397 Mechanics, fluid, 227-229
cloud microphysics, 410-411 Meteorology, 3
cloud-to-ground, 377 Meteorological, 333
continuing current, 378 Modeling, grid scales, 271
probabihty distributions, 407-408 Moisture, see also Fuel moisture content. Water
fires, 383, 386, 411-412 characteristic curve, 101
flashover, 389 content, canopy, 110-115
heat, 379 Australian annuals, perennials, 114
ignition, 376 California chaparral, 113
arc duration, 403 Canadian conifer. 111
criteria, 398 grasslands, 114
efficiency, 386-389, 405-406, 413 seasonal variation. 111
probability, 406-407, 408-409 Monin-Obukhov length, 321
requirements, 398-399 Mortality
studies, 381-383 tree, 515-517
intracloud, 377 understory, 438, 439, 441
location, 385-386, 407
maps, global, 411-412
reduction experiments, 384
return stroke, 378
N
National Fire-Danger Rating System, 341-343,
sheet, 379
353,385
sprites and elves, 379
National Lightning Detection Network, 379
strike path, 390
Navier-Stokes equations, 226, 228, 270
stroke multiplicity, 386
Necrosis
tree damage, 390
cambium, vascular, 488-489, 497-505,
Lignin, 439
513-514
Line fire, 196
canopy, 506-514
Litter, 81,437, 440
carbon budget, 516
Local volume averaging, 98
foliage, bud and twig, 513
Lower Atmosphere Severity Index, 357-358
root, 514-515
Lumped-heat-capacity model, 509, 518
tissue and ceU, 489-491
Newton's law, 482, 507, 511
Nocturnal drainage flow, 285
IVl
Normalized Difference Vegetation Index, 138
Management, 534
North Pacific Oscillation, 361, 363
fire, 3, 528, 534
Nusselt number, 401, 511-513
decision-making, 530-531
load, 553-563
models, 574-576
objectives, 527-528, 537-538, 545
planning process, 544
o
Ohm's law, 103-106, 107, 294, 320
production function, 537
Operations research, 529
fuel, 553
Optical Transient Detector, 412
land, planning, 532-533
Osmosis, 86
large fire, 571-572
Osmotic potential, 97, 101
592 Index

Overdispersion parameter, 424-425, 428- cellulose, 450-452

429,431 products, 450
Pearson estimate, 426 temperature, 450
Pyrometer, two-color, 20

Pacific North America Pattern, 357, 362

Parameterization, subgrid-scale, 274
Quasi likelihood function, 424
Peat, characteristics, 449
Peclet number, 29
Penman-Monteith
equation, 108, 319 R
method, 317-324 Radiation, 4, 272, 306-310, 324
application, 323-324 net, 307-310, 324
Percolation thermal, 482-483
approach, 163 Radiometry, microwave, 340
theory, 172, 194 Rainfall, simulated, 91
Permeability Ratio
longitudinal, 90-91 mixing, 311-312
specific, 90-92 packing, 94-95, 156, 444
Phase, mixed, 376 temperature, excess, 487
Pits, 90 Rayleigh scattering, 379
Plants, components, 478-479 Reactions, chemical, 25
Plasmalemma, 86, 101 thermal, 25
Plume, fire, 195-196,226 autocatalytic, 25
axisymmetric, 243 Regeneration, boreal forest, 438, 441
characteristics, 505-506 Regional Climate Model, 367
constant, 232-233 Residence time
cross-wind, 243-252 particle, 82
equation, 227 Resistance
parameters, 233 aerodynamic, 320-322
experimental data, 233-235 excess, 320-321,335
models, 227 heat transfer, 481,510
three-region, 232-235 leaf, 105-106, 108
two-dimensional, 243-247 stomatal, 108, 322-323
modeling approaches, 227-228 water flow, 90
residence time, 513 Reynolds number, 29, 401, 498-499, 504,
studies, 230-231 511
temperature, 227, 229 Richards equation, 106, 338
profile, 230, 232, 505-506 Richardson number, 315, 321
turbulent, 227 Ridge
Prandd number, 511, 512 blocking, 356
Prefire properties, 193 breakdown, 355-356
Presuppression, 536 Rossby waves, 355
Priestly-Taylor equation, 325 Rotation, vertical
Processes, wildfire, 2 - 3 convective cells, 295-297
Project Skyfire, 380, 383-385, 391, 407 development, 292-295
Protoplast, 85, 101 generation, 297-299
Pyrolysis, 59-60, 198, 398, 403, 443, 455 Roughness length, 316-317, 321
Index 593
S hysteresis, 117-118, 123
Scaling isotherms, 116, 120-123
analysis, 29 models, 119-121
laws, 36-38 types, 119-120
Scorch Southern Oscillation, 359
crown, 506 Index, 360-361
height, 227, 243, 506-507 SPAC
Seasonal Severity Rating, 366 concept, 97
Seeding, cloud, 383-385 water transport, 102-110
Shear, vertical, 260 spotting, 181, 251-252, 357
Smoke, 16-18,560 Stability, atmospheric, 314-316, 358
liquid pool fires, 17 Stand age distribution, 539-540
particles, 65-69 Stefan-Boltzmann law, 306, 445, 482
production, 62-72 Suction, matric, 123
minimizing, 72-74 SURFRAD, 308
solid fuels, 18 Survival, seeds, 242
Smoldering combustion, 3, 62, 158, 188, 448-
450
dependent, 448
downward, 458 Teleconnections, 359-365
extinguishment, 459-462, 466 Temperature
forward, 453-455 air, vertical profile, 334
free flow, 454 bark, 5
fuel dew point, 310
depth, 465 fire, see Fire
moisture, 467 ignition, 202
parameters, 465-466 lethal time, 241
models, 456-464 plume, see Plume
vertical, 458-462 potential, 315, 335
horizontal, 462-463 pre-heating zone, 199
peat, moisture content, 405 pseudo-wet-bulb, 125
propagation, 453-455 radiative, surface, 339
velocity, 456-457, 459, 462 time
reverse, 453 curves, 238-240, 250
roots, 479, 515 history, 235-238
spread rate, 158-159 profiles, 249-230
transient, 444 trace, transient, 487-488
transition to flaming, 159, 449 Thermal death, 238
wave, 453-455 number, 242
length, 453 Thermocouples, 5 - 6 , 19, 229, 231
structure, 453-454, 462-463 shielded-aspirated, 484
zones, 452 time history, 179
Soil, organic, 437 Timber harvest, 540
Soil-root models, 106-107 mean value formulation, 541
Soot scheduling models, 542
formation, 15-16 supply models, 542
collection, 16-17 Time-since-fire
Sorption, 93, 115-119, 130-131 distributions, 421-422, 427, 432-434
Anderson-McCarthy equation, 120 graphical analysis, 420-422
594 Index

Time-since-fire (continued) potential, 97

map, 419 leaf, 103
sample, 419 relative water content, 101
sample vs map, 427-430 root uptake, 107, 304
contagion effect, 429-430 transport, 96-97
efficiency, 428 leaf-atmosphere models, 107-
examples, 428-430 110
Tolerance, thermal, 489-491 live fuel, 102-110
Torus, 90 mechanisms, 123-126
Transfer Ohm's law, 103-106, 107, 320
mass, 4 radial, 107
momentum, 4, 314 soil-root models, 106-107
efficiency, 320-321 vapor
Transpiration, 108 atmospheric content, 310-313
cuticular, 87 pressure
rate, 104 deficit, 310
stomatal, 87 profile, 334
Turgor, 86, 97, 101, 103 surface content, 336
Weather, 351
area burned, 352-358
V synoptic surface, 352-354
Vapor pressure upper air, 354-358
deficit, 310 Wilting, 114
profile, 334 point, physiological, 110
Velocity, friction, 316 process, 109-110
Viscosity, kinematic, 512 Wind
Volatilization, 444 profile, 247-249, 316-317, 334
Vortex shear, 357
street, 499 velocity attenuation coefficient, 317
structures, 273 Wind tunnel, facility, 176-179, 204-205
tilting, 266 fuel
Vortices, 498 arrangement, 205-209
fire-scale, 280-282 bed width, 201-205
horizontal roll, 357 inert content, 209-212
microscale, 282 moisture, 212
Vorticity, 290-292 substratum, 212-214
vertical, 293 test results, 214-217

w Zel'dovich extinction criterion, 27-28

Water
capillary, 122
movement, see transport
 Pale blue

Thick ^ 0.25 mm
blue
t ID

CHAPTER 2, FIGURE 3 An ethylene-air microdiffusion flame established on a 0.25-mm-

diameter hypodermic needle (fuel flow rate at the burner port is 2 m/s). From Ban et al. (1994).
J. Heat Transfer 116, 331.

CHAPTER 2, FIGURE 4 A hydrogen-air jet

diffusion flame established on a 1.43-mm-
diameter stainless-steel burner nozzle (hydro-
gen flow rate at the burner port is 500 m/s)
(Takahashi et d., 1996).
CHAPTER 2, FIGURE 9 (A) A diesel oil pool fire using a 15-m square shape open-top container
(the test was conducted by the Building and Fire Research Laboratory at the National Institute of
Standards and Technology, at the U.S. Coast Guard Fire and Safety Test Detachment in Mobile,
AL). (B) An infrared image of the flame in Figure 9A.
 Q40O-599 str^kes^lOO&q.lcrn
• 600-799 slfik«^100»q km
•800-947 gtnkw^lOOwj.km

CHAPTER 11, FIGURE 1 Fire locations

(black points) with lightning density (flashes/
100 km2) as background.

I Douglas fir D Barren/agriculture/riparian

I Spruce-fir, mixed conifir flf Poderosa pine
D Harwood/shrub/tundra • Annual^rennial grass • 0.172-0.267 fires/stnke/IOOsq.km ,
ii Water • Lodgepole pine/pinyon-juniper m 0,258-0.342 fires/strike/100sq,km I 10-0.085 fires/strike/100sq.km
• 0.343-0,429 fires/strike/100sq.km ^ 10,0.086-0.171 fires/strike/iOOsqkm

CHAPTER 11, FIGURE 2 Fire locations CHAPTER 11, FIGURE 3 Fire locations
(black points) with fuel type as background. (black points) with ignition efficiency as back-
ground.
 Annual Flash Density (flashes/krn^}

0.1 02 0.5 1.0 2.0 5.0 10. 20. 50. 100.

CHAPTER 11, FIGURE 18 Worldwide lightning map as observed by the Optical Transient
Detector in 1997. Courtesy of Global Hydrology and Climate Center and NASA/MSFC.

CHAPTER 11, FIGURE 19 World Fires, April 1992-May 1993 (courtesy of Dwyer et ah, 1999).