Adw.Space Res. Vol. 15. No. 11, pp. (11)91-(1l)lOO.

1995
1995 COSPAR
0273- 1177~95NO079-8 Printedin GreatBritain.
0273-l 177/T’S$9.50 + 0.00

APPLICATION OF VEGETATION INDEX

AND BRIGHTNESS TEMPERATURE FOR
DROUGHT DETECTION

F.N.Kogan
National Oceanic and Atmospheric Administration, Washington, DC 20233,
U.S.A.

ABSTRACT

In recent years the National Oceanic and Atmospheric Administration

(NOAA) has designed a new AVHRR-based Vegetation Condition Index
(VCI) that has showed to be useful for drought detection and
tracking. Validations showed that the VCI has excellent ability to
detect drought and to measure time of its onset, intensity,
duration, and impact on vegetation. The VCI provides accurate
drought information not only for well-defined, prolonged,
widespread, and intensive droughts, but also for very localized,
short-term, and non well-defined droughts. In addition to the VCI,
the AVHRR-based observations in thermal bands were used to develop
the Temperature Condition Index (TCI). This index was used to
determine temperature-related vegetation stress and also stress
caused by an excessive wetness. This paper provides principles of
these indices, describes data processing, and gives examples of
VCI/TCI application in different ecological environments of the
United States.

INTRODUCTION

Since 1981, the Advanced Very High Resolution Radiometer (AVHRR)

sensor flown aboard NOAA polar orbiting satellites has been
observing spectral reflectance in five spectral bands: visible
(Chl), near infrared (Ch2), and three thermal (Ch3, Ch4, Ch5) /l/.
Green and healthy vegetation reflects much less solar radiation in
the Chl than in Ch2 bands. More importantly, when vegetation is
under stress, Chl values increase and Ch2 decrease /2/. Vegetation
identification is enhanced if the two channel values are expressed
as a ratio. There are different expression for this ratio /3/, but
the most popular is the Normalized Difference Vegetation Index
(NDVI) .

NDVI = (Ch2 - Chl)/(ChZ + Chl) (1)

NDVI values are much higher for vegetation than for bare soil,
cloud and water. Healthy and dense vegetation has the largest
NDVI. Because of these properties, NDVI has become the primary
tool for description of vegetation phenology, continental land
cover, vegetation classification and dynamics, and also cropping
practice /4-8/. These applications continue up to now /9/,
although the current emphases tends toward NDVI use for rainfall
monitoring, estimating weather impacts, calculating biomass, crop
yield and pasture production /lo-13/.

The NDVI was used for development of the VCI that showed success

(II)91
(11)92 F. N. Kogan

when applied to drought detection and tracking /12, 14/.

Validations showed that the VCI-derived drought-watch tool is quite
promising. However, in some cases, only VCI data were not
sufficient for accurate drought analysis. Therefore, thermal
channels were studied to receive additional information about
drought. As the result, the Temperature Condition Index (TCI) was
developed.

This paper presents the algorithm development for drought

monitoring and validation. The results presented here are the
first attempt to use both NDVI and thermal channels on a large area
with a very diversified ecological resources. The application of
VCI and TCI are illustrated and validated based on two drought
cases in the United States. These droughts were the most
interesting for analysis and validation because they were different
in the size of affected area, duration, intensity, time of their
occurrence, and damage to agriculture.

SATELLITE DATA PROCESSING

The drought-monitoring algorithm was developed using the Global

Vegetation Index (GVI) data set that is NOAA product described in
/I, 5/. The GVI is produced by sampling and mapping the 4 km daily
radiance, measured on board NOAA polar orbiting satellites to a 16
km map. The Chl and Ch2 data were used to calculated the NDVI
based on equation (1). The daily maps of all five channel
radiance, NDVI, satellites and sun angles are cornposited over a 7
day period by saving those values that have the largest difference
between Ch2 and Chl during the 7 days for each map cell. This
procedure has the effect of minimizing cloud contamination in the
weekly composite. The weekly GVI data from April 1985 through the
end of 1993 were used in the development of the drought-watch
system. In this study we used Chl, Ch2, Ch4, Ch5, and NDVI data.

The GVI data were collected from two polar orbiting satellites,
NOAA-9 and NOAA-11. During the nine years, the performance of the
Chl and Ch2 differed between satellites and most importantly,
deteriorated over time for each satellite. The seven-day composite
Chl and Ch2 values were corrected for degradation using post-launch
calibration coefficients /16/. Following /17/, the new calibration
coefficients considerably improved the stability of NDVI over time
for NOAA-9 satellite. Unfortunately, these coefficients degraded
slightly the stability for the NOAA-11 data. The Ch4 and Ch5 were
used to calculate the brightness temperature to which a nonlinear
correction was applied /18/.

NDVI AS AN INDICATOR OF ENVIRONMENTAL RESOURCES

The amount of vegetation on the Earth is mainly controlled by the

environment /19/. Climate, soil, vegetation type, weather, and
relief are main components of the physical environment that
determine the amount and distribution of vegetation on the Earth.
Limited environmental resources yield scarce vegetation (semi-
desert zone), while unlimited resources are conducive to a large
 Drought Detectioo (II)93

amount of vegetation (tropical forest). The NDVI is widely used as

an indicator of the environmental resources /lo, 11, 15/.

TABLE 1 Source of noise in the Global Vegetation Index data

Source Source
________-___________~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~-~~~-~~~-~
Clouds Changing foot print
Viewing Geometry Satellite Orbital Drift
bias towards off nadir view Satellite Change
surface anisotropy Sensor degradation
Atmospheric Attenuation Truncation
Raleigh scattering Daily Sampling
aerosols Random Noise
water vapor represantativeness of sample
Mapping Procedure human errors
off nadir view transmission
g-day cycle adjustment of satellite
parameters
___-____---____--___---~~--~~~--~~~_~~~_~~~_~~~_~~~_-~~_~~__~~___

Each location on the Earth has certain amount of ecosystem

resources that determine ecosystem potential or "carrying capacity"
/20/ of geographic area. This carrying capacity can be expressed
in the amount of vegetation measured directly or in NDVI value
measured indirectly. We used NDVI values to estimate carrying
capacity of ecosystems. The maximum and minimum NDVI value were
used as a criterion for assessment of ecosystem resources. The
absolute maximum of NDVI (the highest value for the period of
observations) approximates the highest ecosystem potential or
carrying capacity and the absolute minimum of NDVI characterizes
the lowest ecosystem potential (carrying capacity) of a geographic
area.

DROUGHT MONITORING ALGORITHM

The algorithm was designed to reduce noise and enhance the weather-
related component in NDVI data. Noise in AVHRR data creates
fundamental constraints to the remote sensing of vegetation. As a
result of noise, the annual curve of weekly NDVI is always erratic
(Fig. 1). The sources of noise in GVI data set are summarized in
Table 1 /21, 22/. The largest noise comes from clouds which
considerably depress NDVI. Other constituents of the atmosphere
also attenuate NDVI. In case of unusual events, such as sharp
increase in aerosol due to volcanic eruption, NDVI reduces
significantly /23/. Changes in viewing geometry can lead to both
an increase and decrease in NDVI depending on location, type of
vegetation and illumination. Satellite orbital drift, sensor
degradation and satellite change create long-term noise in NDVI
data, especially after satellite has been in service for more than
three years.

Some techniques /16, 22, 24/ were designed to reduce noise in NDVI

JASR 15-11-G
(1094 F.N.Kogao

data. However, complete physically-grounded correction for all

disturbances are currently not available. It is unlikely to
develop algorithm for random noise reduction. Unfortunately, ten
to fifteen percent of AVHRR data have this type of noise (see
extreme outlier in Fig. 1). In addition, when clouds are detected,
the NDVI value for such pixel are condemned. Since clouds cover a
large area, the number of such pixels can be very enormous.
Following /22/, the fine cloud screening discarded 60 percent of
pixels in the area with normal cloud climatology. Therefore, noise
was reduced by smoothing the weekly NDVI time series with a
combination of a compound median filter and the least square
technique /25/. An example of this smoothing is shown in Fig. 1.

Fig. 1. Weekly NDVI: smoothed (solid line) and unsmoothed (dotted

line) for one 16 km pixel (lat. 39.14', long. -80.86') in 1987.

After smoothing, difference in NDVI between years become more

apparent. These differences are due to weather variation. In the
example in Fig. 2, the 1988 drought year had the lowest mid-season
NDVI for both locations. The largest mid-season NDVI was in 1990
(North Dakota, Fig. 2b) and 1986 (Illinois Fig. 2a). Figure 2 also
shows that the same NDVI should be interpreted differently in
various ecosystems. The highest NDVI of 0.39 in Fig. 2a would
indicate excellent vegetation conditions in North Dakota ecosystem
and only poor conditions in Illinois. Therefore, the drought-
monitoring algorithm was designed to take into account ecosystem
differences.
-.--_ ___ __.
4
U6.
Illinois
U.
I9l. .,.a. .I North Dakota
I
q 5: : .;%!I,
,,&:,& &, 4.. i;:ipi;:.*y*
jp*..* *. ... ... ...:!;.-*
I.3 _;:! * I ‘:
_.*a y.!
I.¶ . .;:* , ‘-‘::*
l
l
- .;;;;:;$i’
I.. -.
.‘I:
. :”
a..
‘A..r*....>:
::..2- ‘G:
,?.

Fig. 2. Smoothed weekly NDVI during 1985-1990 for one 16 km pixel

in Illinois and North Dakota.
 Drought
Detection (1W
Ecosystems were stratified based on the multi-year maximum and
minimum of NDVI. They were calculated from the 1985-1993 period
for each pixel and week. The mean growing season (May-August)
absolute maximum and minimum of NDVI for each 16 km by 16 km grid
is shown in Fig. 3a. The maximum estimates productivity of
ecosystem in case of optimal weather and the minimum - if weather
is unfavorable. There is a very good pattern agreement between
ecosystem resources of the USA estimated from NDVI data (Fig. 3a)
and vegetation zones in Fig 3b determined from ground data /26/.
A very typical east-west Zoning iS clearly identified in both
figures. Coniferous forest areas of the western and eastern of the
United States identified from NDVI data matches with ground data.
Desert and mountains environment have good correspondence on both
maps. Even small features of the vegetation zones are reflected on
the NDVI maps of the estimated ecological resources.

Fig. 3. Multi-year absolute maximum and minimum of NDVI average

for the period May through August and vegetation zones of US /26/.

The drought monitoring algorithm also considers separation of the

short-term weather-related NDVI fluctuations from the long-term
ecosystem changes. This is a very important procedure because the
weather signal in an NDVI value is weaker than the ecological one
/12, 14/. Therefore, weather-related NDVI fluctuations are not
easily detectable. The weather-related NDVI envelope (Fig. 2) was
linearly scaled from zero, minimum NDVI, to 100, maximum NDVI for
each grid cell and week. The resulting component was named the
Vegetation Condition Index (VCI) and was defined by the following
expression:

where NDVI, NDVI and NDVI,,, are the smoothed weekly NDVI, its
multi-year absol%L maximum, and minimum, respectively.

The VCI approximates the weather component in NDVI value. It

(11)96 F.N.Kogan
changes from 0 to 100, corresponding to the changes in vegetation
conditions from extremely bad to optimal. The range of VCI values
appropriate for drought analysis was determined by correlating corn
yield with VCI /15/. The 20 percent reduction in corn yield during
drought years was associated with 0 to 35 values of VCI. This
range was accepted as VCI-derived drought indicator.

7 .k..I.
1 I.' 8. .01..I. 1
... ..

P Y

L__ week

Fig. 4. Smoothed weekly NDVI for 1988 drought and 1987 non-drought
years and multi-year absolute maximum and minimum of NDVI.

The Fig. 4 shows NDVI for favorable 1987 and unfavorable, 1988 in
relation to the NDVI max/min envelope for one pixel in Illinois.
During the period of the intensive vegetation growth (weeks 13-24),
the 1987 and 1988 NDVI were almost identical (relative to the
envelope), indicating that VCI-estimated conditions in both years
were similar. In the middle of the growing season (week 24-38) the
1988 NDVIs were the lowest (at the multi-year minimum level) and
VCI values were equal to zero, indicating very severe drought. The
1987 NDVI during the same period grew significantly, approaching to
the maximum of NDVI. VCI values were in the range between 50 (week
24) and 90 (week 28) indicating medium to very good vegetation
condition, respectively. During vegetation senescence, VCI was
near zero for the 1987 and slightly higher for the 1988.

The Temperature Condition Index (TCI) algorithm was similar to the

VCI one. The conditions were estimated relative to the max/min's
temperature envelope. However, the formula was modified to reflect
different response of vesetation to temnerature. Onnosite to the
NDVI, high temperature -in the middle- of the se&on indicates
unfavorable or drought conditions while low temperature indicates
mostly favorable conditions. The TCI has the following expression.

TCI=lOO*(T,, - T)/(T,, - Tmin) (3)

where T, T,,, and Tnrinare the smoothed weekly temperature its
multi-year maximum, and minimum, respectively. We tested
vegetation response to the TCI derived from Ch4 and Ch5. Neither
of the channels showed advantages in assessing vegetation
conditions. We selected Ch4-derived temperature since Ch4 is less
responsive to water vapor in the atmosphere.
Drought Detection ou97
WP8 F.N.Kogan
In order to estimate drought conditions, we used additive
combination of VCI and TCI. However, some weight was assigned to
each of these indices. Because the NDVI, that was used for VCI
development, reflects both temperature and precipitation
conditions, the higher weight was assigned to the VCI.

V/TCI = 0.7O*VCI + 0.30*TCI

With the development of the validation data set, the indicated

weights will be reexamined based on correlation analysis.

ASSESSMENT OF DROUGHT CONDITIONS

In recent years, the VCI has been successfully used for monitoring
large- and small-scale droughts in the United States of America,
the Former Soviet Union, Eastern Europe, and China /15, 25, 27/.
One of the major concerns in the application of VCI for drought
assessment is excessive soil wetness and/or long cloudiness (longer
than three weeks). In case of these events, the NDVI is normally
very depressed and the VCI has low values which can be interpreted
erroneously as a drought. In such cases, the TCI is used to
distinguish drought from non-drought events.

Fig. 6. In-situ data for assessment of environmental conditions in

the USA: (al) percent of normal rainfall during April-July 1993
(dark shading 200 to 250 percent of normal: light shading 75
percent of normal): (a2) 4 to 6 deg. F above normal temperature
during June-July 1993; (b) corn yield reduction in 1988.

These conditions occurred in the US during the 1993 growing season

/28/ (Fig. 6a). Excessive rainfall (150 to 250 percent of normal)
has fallen on the upper Midwest and central US during four-month
period. They produced record flooding and extreme soil saturation.
These conditions reduced planting of agricultural crops and damaged
crops already planted. The vegetation was very depressed. The VCI
clearly identified stressed vegetation in Midwest (Fig. 5).
However, it was not drought-related stress since TCI values were
high, 85 to 100, indicating favorable temperature conditions. In
contrast, the southeastern US, was affected by rainfall deficit
and above normal temperature during the growing season (Fig. 6a).
They caused considerable dryness, especially in the North and South
Carolina and adjacent areas of the neighboring states. These areas
coincide with the areas of the V/TCI-derived drought (Fig. 5).
 Drought
Detection (11)99

Figure 5b shows large-scale 1988 drought in the US. This drought

was unique due to its occurrence early in the season, rapid
development, expansion, unusual severity, and damages to crops,
natural resources and to the US economy /27/. VCI data were used
widely to monitor this drought and these results were validated
with in-situ data /15, 25, 27/. As seen in Fig. 5b, the TCI
provided additional information that was used for more accurate
estimate of drought area and intensity with the combined V/TCI
index. The area of corn yield reduction, shown in Fig. 6b /15/,
matches better with this index, then with either VCI or TCI.

CONCLUSIONS

The VCI showed to be a good tool to detect drought and to measure

time of its onset, intensity, duration, dynamics, and impacts on
vegetation. The VCI-derived drought matched very well with those
outlined from in-situ data. The VCI detects both well defined,
prolonged, widespread, intensive droughts and localized, short-
term, and non well-defined droughts. The TCI provides additional
information about vegetation stress. It indicates weather stress
is due to dryness or excessive wetness. The TCI improve VCI-
derived drought assessments and the combination of these two
indices, V/TCI should be employed as a tool to monitor both drought
and excessive wetness.

REFERENCES

1. K.B. Kidwell, Global Vegetation Index User's Guide, U.S.

Department of Commerce, NOAA, Washington, D.C. (1990).
2. T.I.Gray and D.G.McCrary, The environmental vegetation index, a
tool potentially useful for arid land management, AaRISTARS Report
EW-Nl-04076 JSC-17132 (1981).
3. B.L.Deekshatulu and R.K.Gupta, Remote sensing and vegetation, in
Proc. Indian Nat. Science Acad., 60, #l, 299-333 (1994).
4. C.J.Tucker, J.Gatlin, S.R.Schneider, and M.A.Kuchinos,
Monitoring large scale vegetation dynamics in the Nile delta and
river valley from NOAA AVHRR data, in Proc. Conference on Remote
Sensing of Arid and Semi-Arid Lands, Cairo, Egypt, 1982, p. 973.
5. J.P.Tarpley, S.R.Schnieder, and R.L.Money, Global vegetation
indices from NOAA-7 Meteorological satellite, J. Climate and
Applied Meteorology, #23, 491 (1984).
6. S.N.Goward, C.J.Tucker, and D.G.Dye, North American vegetation
patterns observed with the NOAA-7 Advanced Very High Resolution
Radiometer, Vegetatio, 64, 3-14 (1985).
7. C.O.Justice, J.R.G.Townshend, B.N.Holben, and C.J.Tucker,
Analysis of the phenology of global vegetation using meteorological
satellite data, Int. J. Remote Sensing, #6, 1271-1318 (1985).
8. J.P.Malingreau, Global vegetation in dynamics, satellite
observations over Asia, Int. J. Remote Sens., #7, 1121-1146 (1986).
9. V.Gond, P.Loudjani, F.Cabot, and N.Viovy, Classification of West
Africa vegetation based upon temporal evolution of corrected NDVI
data, in Proc. 6th AVHRR data user's meeting, Italy, (1993).
10. A. van Dijk, A crop condition and crop yield estimation method
based on NOAA/AVHRR satellite data, in Ph.D. Dissertation,
(11)loa F. N. Kogan

University of Missouri, Columbia (1986).

ll.C.J.Tucker and P.J.Sellers, Satellite remote sensing of primary
production, Int. J. Remote Sensing, #7, 1395-1416 (1986).
12. F.Kogan, Remote sensing of weather impacts on vegetation in
non-homogeneous areas, Int.J.Remote Sens, 11, #8, 1405-1419 (1990).
13. K.Hall-Konyves, Crop monitoring in Sweden, Int. J. Remote
Sensing, #3, 461-484 (1990).
14. F.Kogan, Vegetation index for area1 analysis of crop
conditions, in Proc. 18th Conference on Agricultural and Forest
Meteorology, AMS, W.Lafayette, 1987, p. 103.
15. F.Kogan, How drought looks from space, Geocarto International,
in press, (1995).
16. C.R.N.Rao, J.Chen, Calibration of the visible and near-infrared
channels of the Advanced Very High Resolution Radiometer (AVHRR)
after launch, In Recent Advances in Sensors, Radiometric
Calibration, and Processing of Remote Sensed Data, Vol. 1938.
Proc. SPIE-The International Society for Optical Engineering,
Orlando, FL, April 14-16, 56-66, (1993).
17. F.Kogan and P.Bu, Improvement in 1985-1993 NDVI time-series for
world desert targets using post-launch calibration, in Proc.
IGARSS-94, (1994).
18. M.P.Weinreb, G.Hamilton, and S.Brown, Nonlinearity correction
in calibration of the Advanced Very High Resolution Radiometer
infrared channels, J. Geophys. Res., vol. 95, #5, 7381-7388 (1990).
19. W.C.Clark, Scales of climate impacts, in The Sensitivity of
Natural Ecosystems and Agriculture to Climate Change (M.L.Parry,
Ed.), International Institute for Applied Systems Analysis.
Laxenburg, Austria (1985).
20. P.Reining, the use of ERTS-1 data in carrying capacity
estimates for sites in Upper Volta and Niger, Paper presented at
the 1974 Annual Meeting of the American Anthropological
Association, Mexico City (1974).
21. S.N.Goward, B.Markham, D.G.Dye, W.Dulaney, J.Yang, Normalized
Difference Vegetation Index measurements from the Advanced Very
high Resolution Radiometer, Remote Sens. Environ., 35, (1991).
22. G.G.Gutman, Vegetation Indices from AVHRR Data: An Update and
Future Prospects, Remote Sens. Environ., 35, 121-136 (1991).
23. F. Kogan, J.Sullivan, R.Carrey, D.Tarpley, Post-Pinatubo
Vegetation index in Central Africa, Geocarto International (1994).
24. T.F.Eck and V.L.Kalb, Cloud screening for Africa using a
geographically and seasonally variable infrared threshold, Int. J.
Rem. Sens., v.12, 1205-1221 (1991).
25. F.Kogan and J.Sullivan, Development of global drought-watch
system using NOAA/AVHRR data, Adv.Space Res., vol. 13, #5, (1993).
26. Hammond, World Atlas, Gemini Edition, Hammond Inc. Maplewood,
New Jersey, 189-192 (1991).
27. G.E.Johnson, V.R. Achutuni, S. Thiruvengadachari, and F. Kogan,
The Role of NOAA Satellite Data in Drought Early Warning and
Monitoring, in book Drought Assessment, Management, and Planning:
Theory and Case Studies, Ed. D.A.Wilhite, Kluwer Academic
Publishers, Boston/Dordrecht/London, 31-49, 1993.
28. NOAA, Special Climate Summary 93/l: Midwestern Floods, Heat and
Drought in the East, National Weather Service Reoort, (1993).