THE SOIL AND WATER ASSESSMENT TOOL:

HISTORICAL DEVELOPMENT, APPLICATIONS,

AND FUTURE RESEARCH DIRECTIONS Invited Review Series

P. W. Gassman, M. R. Reyes, C. H. Green, J. G. Arnold

ABSTRACT. The Soil and Water Assessment Tool (SWAT) model is a continuation of nearly 30 years of modeling efforts
conducted by the USDA Agricultural Research Service (ARS). SWAT has gained international acceptance as a robust
interdisciplinary watershed modeling tool as evidenced by international SWAT conferences, hundreds of SWAT‐related papers
presented at numerous other scientific meetings, and dozens of articles published in peer‐reviewed journals. The model has
also been adopted as part of the U.S. Environmental Protection Agency (USEPA) Better Assessment Science Integrating Point
and Nonpoint Sources (BASINS) software package and is being used by many U.S. federal and state agencies, including the
USDA within the Conservation Effects Assessment Project (CEAP). At present, over 250 peer‐reviewed published articles
have been identified that report SWAT applications, reviews of SWAT components, or other research that includes SWAT. Many
of these peer‐reviewed articles are summarized here according to relevant application categories such as streamflow
calibration and related hydrologic analyses, climate change impacts on hydrology, pollutant load assessments, comparisons
with other models, and sensitivity analyses and calibration techniques. Strengths and weaknesses of the model are presented,
and recommended research needs for SWAT are also provided.
Keywords. Developmental history, Flow analysis, Modeling, SWAT, Water quality.

T
he Soil and Water Assessment Tool (SWAT) model of Hydrological Processes (volume 19, issue 3) and proceed‐
(Arnold et al., 1998; Arnold and Fohrer, 2005) has ings for the second (TWRI, 2003), third (EAWAG, 2005), and
proven to be an effective tool for assessing water re‐ fourth (UNESCO-IHE, 2007) conferences.
source and nonpoint‐source pollution problems for Reviews of SWAT applications and/or components have
a wide range of scales and environmental conditions across been previously reported, sometimes in conjunction with
the globe. In the U.S., SWAT is increasingly being used to comparisons with other models (e.g., Arnold and Fohrer,
support Total Maximum Daily Load (TMDL) analyses (Bo‐ 2005; Borah and Bera, 2003, 2004; Shepherd et al., 1999).
rah et al., 2006), research the effectiveness of conservation However, these previous reviews do not provide a compre‐
practices within the USDA Conservation Effects Assessment hensive overview of the complete body of SWAT applica‐
Program (CEAP, 2007) initiative (Mausbach and Dedrick, tions that have been reported in the peer‐reviewed literature.
2004), perform “macro‐scale assessments” for large regions There is a need to fill this gap by providing a review of the
such as the upper Mississippi River basin and the entire U.S. full range of studies that have been conducted with SWAT and
(e.g., Arnold et al., 1999a; Jha et al., 2006), and a wide range to highlight emerging application trends. Thus, the specific
of other water use and water quality applications. Similar objectives of this study are to: (1) provide an overview of
SWAT application trends have also emerged in Europe and SWAT development history, including the development of
other regions, as shown by the variety of studies presented in GIS interface tools and examples of modified SWAT models;
four previous European international SWAT conferences, (2) summarize research findings or methods for many of the
which are reported for the first conference in a special issue more than 250 peer‐reviewed articles that have been identi‐
fied in the literature, as a function of different application
categories; and (3) describe key strengths and weaknesses of
Submitted for review in November 2006 as manuscript number SW the model and list a summary of future research needs.
6726; approved for publication by the Soil & Water Division of ASABE in
May 2007.
The authors are Philip W. Gassman, ASABE Member Engineer,
Assistant Scientist, Center for Agricultural and Rural Development, SWAT DEVELOPMENTAL HISTORY AND
Department of Economics, Iowa State University, Ames, Iowa; Manuel R.
Reyes, ASABE Member Engineer, Professor, Biological Engineering OVERVIEW
Program, Department of Natural Resources and Environmental Design, The development of SWAT is a continuation of USDA
School of Agriculture and Environmental Sciences, North Carolina A&T Agricultural Research Service (ARS) modeling experience
State University, Greensboro, North Carolina; Colleen H. Green, ASABE that spans a period of roughly 30 years. Early origins of
Member, Soil Scientist, and Jeffrey G. Arnold, Agricultural Engineer,
USDA‐ARS Grassland, Soil and Water Research Laboratory, Temple, SWAT can be traced to previously developed USDA‐ARS
Texas. Corresponding author: Philip W. Gassman, Center for Agricultural models (fig. 1) including the Chemicals, Runoff, and Erosion
and Rural Development, Department of Economics, 560A Heady Hall, from Agricultural Management Systems (CREAMS) model
Iowa State University, Ames, IA 50011‐1070; phone: 515‐294‐6313; fax: (Knisel, 1980), the Groundwater Loading Effects on
515‐294‐6336; e‐mail: [email protected].

Transactions of the ASABE

Vol. 50(4): 1211-1250 2007 American Society of Agricultural and Biological Engineers ISSN 0001-2351 1211
 Figure 1. Schematic of SWAT developmental history, including selected SWAT adaptations.

Agricultural Management Systems (GLEAMS) model valuable simulation model, while allowing simulations of
(Leonard et al., 1987), and the Environmental Impact Policy very extensive areas.
Climate (EPIC) model (Izaurralde et al., 2006), which was SWAT has undergone continued review and expansion of
originally called the Erosion Productivity Impact Calculator capabilities since it was created in the early 1990s. Key en‐
(Williams, 1990). The current SWAT model is a direct de‐ hancements for previous versions of the model (SWAT94.2,
scendant of the Simulator for Water Resources in Rural Ba‐ 96.2, 98.1, 99.2, and 2000) are described by Arnold and Foh‐
sins (SWRRB) model (Arnold and Williams, 1987), which rer (2005) and Neitsch et al. (2005a), including the incorpora‐
was designed to simulate management impacts on water and tion of in‐stream kinetic routines from the QUAL2E model
sediment movement for ungauged rural basins across the (Brown and Barnwell, 1987), as shown in figure 1. Documen‐
U.S. tation for some previous versions of the model is available at
Development of SWRRB began in the early 1980s with the SWAT web site (SWAT, 2007d). Detailed theoretical doc‐
modification of the daily rainfall hydrology model from umentation and a user's manual for the latest version of the
CREAMS. A major enhancement was the expansion of sur‐ model (SWAT2005) are given by Neitsch et al. (2005a,
face runoff and other computations for up to ten subbasins, 2005b). The current version of the model is briefly described
as opposed to a single field, to predict basin water yield. Oth‐ here to provide an overview of the model structure and execu‐
er enhancements included an improved peak runoff rate tion approach.
method, calculation of transmission losses, and the addition
of several new components: groundwater return flow (Arnold SWAT OVERVIEW
and Allen, 1993), reservoir storage, the EPIC crop growth SWAT is a basin‐scale, continuous‐time model that oper‐
submodel, a weather generator, and sediment transport. Fur‐ ates on a daily time step and is designed to predict the impact
ther modifications of SWRRB in the late 1980s included the of management on water, sediment, and agricultural chemi‐
incorporation of the GLEAMS pesticide fate component, op‐ cal yields in ungauged watersheds. The model is physically
tional USDA‐SCS technology for estimating peak runoff based, computationally efficient, and capable of continuous
rates, and newly developed sediment yield equations. These simulation over long time periods. Major model components
modifications extended the model's capability to deal with a include weather, hydrology, soil temperature and properties,
wide variety of watershed water quality management prob‐ plant growth, nutrients, pesticides, bacteria and pathogens,
lems. and land management. In SWAT, a watershed is divided into
Arnold et al. (1995b) developed the Routing Outputs to multiple subwatersheds, which are then further subdivided
Outlet (ROTO) model in the early 1990s in order to support into hydrologic response units (HRUs) that consist of homo‐
an assessment of the downstream impact of water manage‐ geneous land use, management, and soil characteristics. The
ment within Indian reservation lands in Arizona and New HRUs represent percentages of the subwatershed area and are
Mexico that covered several thousand square kilometers, as not identified spatially within a SWAT simulation. Alterna‐
requested by the U.S. Bureau of Indian Affairs. The analysis tively, a watershed can be subdivided into only subwa‐
was performed by linking output from multiple SWRRB runs tersheds that are characterized by dominant land use, soil
and then routing the flows through channels and reservoirs in type, and management.
ROTO via a reach routing approach. This methodology over‐
came the SWRRB limitation of allowing only ten subbasins; Climatic Inputs and HRU Hydrologic Balance
however, the input and output of multiple SWRRB files was Climatic inputs used in SWAT include daily precipitation,
cumbersome and required considerable computer storage. To maximum and minimum temperature, solar radiation data,
overcome the awkwardness of this arrangement, SWRRB relative humidity, and wind speed data, which can be input
and ROTO were merged into the single SWAT model (fig. 1). from measured records and/or generated. Relative humidity
SWAT retained all the features that made SWRRB such a is required if the Penman‐Monteith (Monteith, 1965) or

1212 TRANSACTIONS OF THE ASABE

Priestly‐Taylor (Priestly and Taylor, 1972) evapotranspira‐ form of inorganic fertilizer and/or manure inputs. An alterna‐
tion (ET) routines are used; wind speed is only necessary if tive automatic fertilizer routine can be used to simulate fertil‐
the Penman‐Monteith method is used. Measured or generated izer applications, as a function of nitrogen stress. Biomass
sub‐daily precipitation inputs are required if the Green‐Ampt removal and manure deposition can be simulated for grazing
infiltration method (Green and Ampt, 1911) is selected. The operations. SWAT2005 also features a new continuous ma‐
average air temperature is used to determine if precipitation nure application option to reflect conditions representative of
should be simulated as snowfall. The maximum and mini‐ confined animal feeding operations, which automatically
mum temperature inputs are used in the calculation of daily simulates a specific frequency and quantity of manure to be
soil and water temperatures. Generated weather inputs are applied to a given HRU. The type, rate, timing, application
calculated from tables consisting of 13 monthly climatic efficiency, and percentage application to foliage versus soil
variables, which are derived from long‐term measured can be accounted for simulations of pesticide applications.
weather records. Customized climatic input data options in‐ Selected conservation and water management practices
clude: (1) simulation of up to ten elevation bands to account can also be simulated in SWAT. Conservation practices that
for orographic precipitation and/or for snowmelt calcula‐ can be accounted for include terraces, strip cropping, con‐
tions, (2) adjustments to climate inputs to simulate climate touring, grassed waterways, filter strips, and conservation
change, and (3) forecasting of future weather patterns, which tillage. Simulation of irrigation water on cropland can be
is a new feature in SWAT2005. simulated on the basis of five alternative sources: stream
The overall hydrologic balance is simulated for each reach, reservoir, shallow aquifer, deep aquifer, or a water
HRU, including canopy interception of precipitation, parti‐ body source external to the watershed. The irrigation applica‐
tioning of precipitation, snowmelt water, and irrigation water tions can be simulated for specific dates or with an auto‐
between surface runoff and infiltration, redistribution of wa‐ irrigation routine, which triggers irrigation events according
ter within the soil profile, evapotranspiration, lateral subsur‐ to a water stress threshold. Subsurface tile drainage is simu‐
face flow from the soil profile, and return flow from shallow lated in SWAT2005 with improved routines that are based on
aquifers. Estimation of areal snow coverage, snowpack tem‐ the work performed by Du et al. (2005) and Green et al.
perature, and snowmelt water is based on the approach de‐ (2006); the simulated tile drains can also be linked to new
scribed by Fontaine et al. (2002). Three options exist in routines that simulate the effects of depressional areas (pot‐
SWAT for estimating surface runoff from HRUs, which are holes). Water transfer can also be simulated between differ‐
combinations of daily or sub‐hourly rainfall and the USDA ent water bodies, as well as “consumptive water use” in
Natural Resources Conservation Service (NRCS) curve num‐ which removal of water from a watershed system is assumed.
ber (CN) method (USDA‐NRCS, 2004) or the Green‐Ampt HRU‐level and in‐stream pollutant losses can be esti‐
method. Canopy interception is implicit in the CN method, mated with SWAT for sediment, nitrogen, phosphorus, pesti‐
while explicit canopy interception is simulated for the Green‐ cides, and bacteria. Sediment yield is calculated with the
Ampt method. Modified Universal Soil Loss Equation (MUSLE) developed
A storage routing technique is used to calculate redistribu‐ by Williams and Berndt (1977); USLE estimates are output
tion of water between layers in the soil profile. Bypass flow for comparative purposes only. The transformation and
can be simulated, as described by Arnold et al. (2005), for movement of nitrogen and phosphorus within an HRU are
soils characterized by cracking, such as Vertisols. SWAT2005 simulated in SWAT as a function of nutrient cycles consisting
also provides a new option to simulate perched water tables of several inorganic and organic pools. Losses of both N and
in HRUs that have seasonal high water tables. Three methods P from the soil system in SWAT occur by crop uptake and in
for estimating potential ET are provided: Penman‐Monteith, surface runoff in both the solution phase and on eroded sedi‐
Priestly‐Taylor, and Hargreaves (Hargreaves et al., 1985). ET ment. Simulated losses of N can also occur in percolation be‐
values estimated external to SWAT can also be input for a low the root zone, in lateral subsurface flow including tile
simulation run. The Penman‐Monteith option must be used drains, and by volatilization to the atmosphere. Accounting
for climate change scenarios that account for changing atmo‐ of pesticide fate and transport includes degradation and
spheric CO2 levels. Recharge below the soil profile is parti‐ losses by volatilization, leaching, on eroded sediment, and in
tioned between shallow and deep aquifers. Return flow to the the solution phase of surface runoff and later subsurface flow.
stream system and evapotranspiration from deep‐rooted Bacteria surface runoff losses are simulated in both the solu‐
plants (termed “revap”) can occur from the shallow aquifer. tion and eroded phases with improved routines in
Water that recharges the deep aquifer is assumed lost from the SWAT2005.
system.
Flow and Pollutant Loss Routing, and Auto‐Calibration
Cropping, Management Inputs, and HRU‐Level Pollutant and Uncertainty Analysis
Losses Flows are summed from all HRUs to the subwatershed
Crop yields and/or biomass output can be estimated for a level, and then routed through the stream system using either
wide range of crop rotations, grassland/pasture systems, and the variable‐rate storage method (Williams, 1969) or the
trees with the crop growth submodel. New routines in Muskingum method (Neitsch et al., 2005a), which are both
SWAT2005 allow for simulation of forest growth from seed‐ variations of the kinematic wave approach. Sediment, nutri‐
ling to mature stand. Planting, harvesting, tillage passes, nu‐ ent, pesticide, and bacteria loadings or concentrations from
trient applications, and pesticide applications can be each HRU are also summed at the subwatershed level, and the
simulated for each cropping system with specific dates or resulting losses are routed through channels, ponds, wet‐
with a heat unit scheduling approach. Residue and biological lands, depressional areas, and/or reservoirs to the watershed
mixing are simulated in response to each tillage operation. outlet. Contributions from point sources and urban areas are
Nitrogen and phosphorus applications can be simulated in the also accounted for in the total flows and pollutant losses ex‐

Vol. 50(4): 1211-1250 1213

ported from each subwatershed. Sediment transport is simu‐ tion System (GIS) and other interface tools to support the
lated as a function of peak channel velocity in SWAT2005, input of topographic, land use, soil, and other digital data into
which is a simplified approach relative to the stream power SWAT. The first GIS interface program developed for SWAT
methodology used in previous SWAT versions. Simulation of was SWAT/GRASS, which was built within the GRASS
channel erosion is accounted for with a channel erodibility raster‐based GIS (Srinivasan and Arnold, 1994). Haverkamp
factor. In‐stream transformations and kinetics of algae et al. (2005) have adopted SWAT/GRASS within the Input-
growth, nitrogen and phosphorus cycling, carbonaceous bio‐ OutputSWAT (IOSWAT) software package, which incorpo‐
logical oxygen demand, and dissolved oxygen are performed rates the Topographic Parameterization Tool (TOPAZ) and
on the basis of routines developed for the QUAL2E model. other tools to generate inputs and provide output mapping
Degradation, volatilization, and other in‐stream processes support for both SWAT and SWAT‐G.
are simulated for pesticides, as well as decay of bacteria. The ArcView‐SWAT (AVSWAT) interface tool (Di Luzio
Routing of heavy metals can be simulated; however, no trans‐ et al., 2004a, 2004b) is designed to generate model inputs
formation or decay processes are simulated for these pollu‐ from ArcView 3.x GIS data layers and execute SWAT2000
tants. within the same framework. AVSWAT was incorporated
A final feature in SWAT2005 is a new automated sensitiv‐ within the U.S. Environmental Protection Agency (USEPA)
ity, calibration, and uncertainty analysis component that is Better Assessment Science Integrating point and Nonpoint
based on approaches described by van Griensven and Meix‐ Sources (BASINS) software package versions 3.0 (USEPA,
ner (2006) and van Griensven et al. (2006b). Further discus‐ 2006a), which provides GIS utilities that support automatic
sion of these tools is provided in the Sensitivity, Calibration, data input for SWAT2000 using ArcView (Di Luzio et al.,
and Uncertainty Analyses Section. 2002). The most recent version of the interface is denoted
AVSWAT‐X, which provides additional input generation
SWAT ADAPTATIONS functionality, including soil data input from both the USDA‐
A key trend that is interwoven with the ongoing develop‐ NRCS State Soils Geographic (STATSGO) and Soil Survey
ment of SWAT is the emergence of modified SWAT models Geographic (SSURGO) databases (USDA‐NRCS, 2007a,
that have been adapted to provide improved simulation of 2007b) for applications of SWAT2005 (Di Luzio et al., 2005;
specific processes, which in some cases have been focused on SWAT, 2007b). Automatic sensitivity, calibration, and uncer‐
specific regions. Notable examples (fig. 1) include SWAT‐G, tainty analysis can also be initiated with AVSWAT‐X for
Extended SWAT (ESWAT), and the Soil and Water Integrated SWAT2005. The Automated Geospatial Watershed Assess‐
Model (SWIM). The initial SWAT‐G model was developed ment (AGWA) interface tool (Miller et al., 2007) is an alter‐
by modifying the SWAT99.2 percolation, hydraulic conduc‐ native ArcView‐based interface tool that supports data input
tivity, and interflow functions to provide improved flow pre‐ generation for both SWAT2000 and the KINEROS2 model,
dictions for typical conditions in low mountain ranges in including options for soil inputs from the SSURGO, STATS‐
Germany (Lenhart et al., 2002). Further SWAT‐G enhance‐ GO, or United Nations Food and Agriculture Organization
ments include an improved method of estimating erosion loss (FAO) global soil maps. Both AGWA and AVSWAT have
(Lenhart et al., 2005) and a more detailed accounting of CO2 been incorporated as interface approaches for generating
effects on leaf area index and stomatal conductance (Eck‐ SWAT2000 inputs within BASINS version 3.1 (Wells, 2006).
hardt and Ulbrich, 2003). The ESWAT model (van Griensven A SWAT interface compatible with ArcGIS version 9.1
and Bauwens, 2003, 2005) features several modifications rel‐ (ArcSWAT) has recently been developed that uses a geodata‐
ative to the original SWAT model including: (1) sub‐hourly base approach and a programming structure consistent with
precipitation inputs and infiltration, runoff, and erosion loss Component Object Model (COM) protocol (Olivera et al.,
estimates based on a user‐defined fraction of an hour; (2) a 2006; SWAT, 2007a). An ArcGIS 9.x version of AGWA
river routing module that is updated on an hourly time step (AGWA2) is also being developed and is expected to be re‐
and is interfaced with a water quality component that features leased near mid‐2007 (USDA‐ARS, 2007).
in‐stream kinetics based partially on functions used in A variety of other tools have been developed to support
QUAL2E as well as additional enhancements; and (3) multi‐ executions of SWAT simulations, including: (1) the interac‐
objective (multi‐site and/or multi‐variable) calibration and tive SWAT (i_SWAT) software (CARD, 2007), which sup‐
autocalibration modules (similar components are now incor‐ ports SWAT simulations using a Windows interface with an
porated in SWAT2005). The SWIM model is based primarily Access database; (2) the Conservation Reserve Program
on hydrologic components from SWAT and nutrient cycling (CRP) Decision Support System (CRP‐DSS) developed by
components from the MATSALU model (Krysanova et al., Rao et al. (2006); (3) the AUTORUN system used by Kannan
1998, 2005) and is designed to simulate “mesoscale” (100 to et al. (2007b), which facilitates repeated SWAT simulations
100,000 km2) watersheds. Recent improvements to SWIM with variations in selected parameters; and (4) a generic in‐
include incorporation of a groundwater dynamics submodel terface (iSWAT) program (Abbaspour et al., 2007), which au‐
(Hatterman et al., 2004), enhanced capability to simulate for‐ tomates parameter selection and aggregation for iterative
est systems (Wattenbach et al., 2005), and development of SWAT calibration simulations.
routines to more realistically simulate wetlands and riparian
zones (Hatterman et al., 2006). SWAT APPLICATIONS
Applications of SWAT have expanded worldwide over the
GEOGRAPHIC INFORMATION SYSTEM INTERFACES AND past decade. Many of the applications have been driven by
OTHER TOOLS the needs of various government agencies, particularly in the
A second trend that has paralleled the historical develop‐ U.S. and the European Union, that require direct assessments
ment of SWAT is the creation of various Geographic Informa‐ of anthropogenic, climate change, and other influences on a

1214 TRANSACTIONS OF THE ASABE

Figure 2. Distribution of the 2,149 8‐digit watersheds within the 18 Major Water Resource Regions (MWRRs) that comprise the conterminous U.S.

wide range of water resources or exploratory assessments of will be used to help determine the pollutant sources and po‐
model capabilities for potential future applications. tential solutions for many of these forthcoming TMDLs. Ex‐
One of the first major applications performed with SWAT tensive discussion of applying SWAT and other models for
was within the Hydrologic Unit Model of the U.S. (HUMUS) TMDLs is presented in Borah et al. (2006), Benham et al.
modeling system (Arnold et al., 1999a), which was imple‐ (2006), and Shirmohammadi et al. (2006).
mented to support USDA analyses of the U.S. Resources SWAT has also been used extensively in Europe, including
Conservation Act Assessment of 1997 for the conterminous projects supported by various European Commission (EC)
U.S. The system was used to simulate the hydrologic and/or agencies. Several models including SWAT were used to
pollutant loss impacts of agricultural and municipal water quantify the impacts of climate change for five different wa‐
use, tillage and cropping system trends, and other scenarios tersheds in Europe within the Climate Hydrochemistry and
within each of the 2,149 U.S. Geological Survey (USGS) Economics of Surface‐water Systems (CHESS) project,
8‐digit Hydrologic Cataloging Unit (HCU) watersheds which was sponsored by the EC Environment and Climate
(Seaber et al., 1987), referred to hereafter as “8‐digit wa‐ Research Programme (CHESS, 2001). A suite of nine models
tersheds”. Figure 2 shows the distribution of the 8‐digit wa‐ including SWAT were tested in 17 different European wa‐
tersheds within the 18 Major Water Resource Regions tersheds as part of the EUROHARP project, which was spon‐
(MWRRs) that comprise the conterminous U.S. sored by the EC Energy, Environment and Sustainable
SWAT is also being used to support the USDA Conserva‐ Development (EESD) Programme (EUROHARP, 2006). The
tion Effects Assessment Project, which is designed to quanti‐ goal of the research was to assess the ability of the models to
fy the environmental benefits of conservation practices at estimate nonpoint‐source nitrogen and phosphorus losses to
both the national and watershed scales (Mausbach and De‐ both freshwater streams and coastal waters. The EESD‐
drick, 2004). SWAT is being applied at the national level sponsored TempQsim project focused on testing the ability of
within a modified HUMUS framework to assess the benefits SWAT and five other models to simulate intermittent stream
of different conservation practices at that scale. The model is conditions that exist in southern Europe (TempQsim, 2006).
also being used to evaluate conservation practices for wa‐ Volk et al. (2007) and van Griensven et al. (2006a) further de‐
tersheds of varying sizes that are representative of different scribe SWAT application approaches within in the context of
regional conditions and mixes of conservation practices. the European Union (EU) Water Framework Directive.
SWAT is increasingly being used to perform TMDL analy‐ The following application discussion focuses on the wide
ses, which must be performed for impaired waters by the dif‐ range of specific SWAT applications that have been reported
ferent states as mandated by the 1972 U.S. Clean Water Act in the literature. Some descriptions of modified SWAT model
(USEPA, 2006b). Roughly 37% of the nearly 39,000 current‐ applications are interspersed within the descriptions of stud‐
ly listed impaired waterways still require TMDLs (USEPA, ies that used the standard SWAT model.
2007); SWAT, BASINS, and a variety of other modeling tools

Vol. 50(4): 1211-1250 1215

 Table 1. Overview of major application categories in the literature. However, Moriasi et al. (2007) proposed that
of SWAT studies reported in the literature.[a] NSE values should exceed 0.5 in order for model results to be
Hydrologic judged as satisfactory for hydrologic and pollutant loss evalua‐
and Pollutant
Hydrologic Pollutant Loss tions performed on a monthly time step (and that appropriate re‐
Primary Application Category Only Loss Only laxing and tightening of the standard be performed for daily and
Calibration and/or sensitivity analysis 15 20 2 annual time step evaluations, respectively). Assuming this crite‐
Climate change impacts 22 8 -- rion for both the NSE and r2 values at all time steps, the majority
GIS interface descriptions 3 3 2 of statistics listed in table2 would be judged as adequately repli‐
Hydrologic assessments 42 -- -- cating observed streamflows and other hydrologic indicators.
Variation in configuration or data input 21 15 -- However, it is clear that poor results resulted for parts or all of
effects some studies. The poorest results generally occurred for daily
Comparisons with other models or 5 7 1 predictions, although this was not universal (e.g., Grizzetti et al.,
techniques 2005). Some of the weaker results can be attributed in part to
Interfaces with other models 13 15 6 inadequate representation of rainfall inputs, due to either a lack
Pollutant assessments -- 57 6 of adequate rain gauges in the simulated watershed or subwa‐
[a] Includes studies describing applications of ESWAT, SWAT-G, SWIM, tershed configurations that were too coarse to capture the spatial
and other modified SWAT models. detail of rainfall inputs (e.g., Cao et al., 2006; Conan et al.,
2003b; Bouraoui et al., 2002; Bouraoui et al., 2005). Other fac‐
SPECIFIC SWAT APPLICATIONS tors that may adversely affect SWAT hydrologic predictions in‐
SWAT applications reported in the literature can be cate‐ clude a lack of model calibration (Bosch et al., 2004),
gorized in several ways. For this study, most of the peer‐ inaccuracies in measured streamflow data (Harmel et al., 2006),
reviewed articles could be grouped into the nine and relatively short calibration and validation periods (Muleta
subcategories listed in table 1, and then further broadly de‐ and Nicklow, 2005b).
fined as hydrologic only, hydrologic and pollutant loss, or Example Calibration/Validation Studies
pollutant loss only. Reviews are not provided for all of the ar‐ The SWAT hydrologic subcomponents have been refined
ticles included in the table 1 summary; a complete list of the and validated at a variety of scales (table 2). For example, Ar‐
SWAT peer‐reviewed articles is provided at the SWAT web nold and Allen (1996) used measured data from three Illinois
site (SWAT, 2007c), which is updated on an ongoing basis. watersheds, ranging in size from 122 to 246 km2, to success‐
fully validate surface runoff, groundwater flow, groundwater
HYDROLOGIC ASSESSMENTS ET, ET in the soil profile, groundwater recharge, and ground‐
Simulation of the hydrologic balance is foundational for water height parameters. Santhi et al. (2001a, 2006) per‐
all SWAT watershed applications and is usually described in formed extensive streamflow validations for two Texas
some form regardless of the focus of the analysis. The major‐ watersheds that cover over 4,000 km2. Arnold et al. (1999b)
ity of SWAT applications also report some type of graphical evaluated streamflow and sediment yield data in the Texas
and/or statistical hydrologic calibration, especially for Gulf basin with drainage areas ranging from 2,253 to
streamflow, and many of the studies also report validation re‐ 304,260km 2. Streamflow data from approximately 1,000
sults. A wide range of statistics has been used to evaluate stream monitoring gauges from 1960 to 1989 were used to
SWAT hydrologic predictions. By far the most widely used calibrate and validate the model. Predicted average monthly
statistics reported for hydrologic calibration and validation streamflows for three major river basins (20,593 to
are the regression correlation coefficient (R2) and the Nash‐ 108,788km 2) were 5% higher than measured flows, with
Sutcliffe model efficiency (NSE) coefficient (Nash and Sut‐ standard deviations between measured and predicted within
cliffe, 1970). The R2 value measures how well the simulated 2%. Annual runoff and ET were validated across the entire
versus observed regression line approaches an ideal match continental U.S. as part of the Hydrologic Unit Model for the
and ranges from 0 to 1, with a value of 0 indicating no correla‐ U.S. (HUMUS) modeling system. Rosenthal et al. (1995)
tion and a value of 1 representing that the predicted disper‐ linked GIS to SWAT and simulated ten years of monthly
sion equals the measured dispersion (Krause et al., 2005). streamflow without calibration. SWAT underestimated the
The regression slope and intercept also equal 1 and 0, respec‐ extreme events but produced overall accurate streamflows
tively, for a perfect fit; the slope and intercept are often not (table 2). Bingner (1996) simulated runoff for ten years for a
reported. The NSE ranges from −∞ to 1 and measures how watershed in northern Mississippi. The SWAT model pro‐
well the simulated versus observed data match the 1:1 line duced reasonable results in the simulation of runoff on a daily
(regression line with slope equal to 1). An NSE value of 1 and annual basis from multiple subbasins (table 2), with the
again reflects a perfect fit between the simulated and mea‐ exception of a wooded subbasin. Rosenthal and Hoffman
sured data. A value of 0 or less than 0 indicates that the mean (1999) successfully used SWAT and a spatial database to sim‐
of the observed data is a better predictor than the model out‐ ulate flows, sediment, and nutrient loadings on a 9,000 km2
put. See Krause et al. (2005) for further discussion regarding watershed in central Texas to locate potential water quality
the R2, NSE, and other efficiency criteria measures. monitoring sites. SWAT was also successfully validated for
An extensive list of R2 and NSE statistics is presented in streamflow (table 2) for the Mill Creek watershed in Texas for
table 2 for 115 SWAT hydrologic calibration and/or validation 1965‐1968 and 1968‐1975 (Srinivasan et al., 1998). Monthly
results reported in the literature. These statistics provides valu‐ streamflow rates were well predicted, but the model overesti‐
able insight regarding the hydrologic performance of the model mated streamflows in a few years during the spring/summer
across a wide spectrum of conditions. To date, no absolute crite‐ months. The overestimation may be accounted for by vari‐
ria for judging model performance have been firmly established able rainfall during those months.

1216 TRANSACTIONS OF THE ASABE

Vol. 50(4): 1211-1250 1217
1218 TRANSACTIONS OF THE ASABE
Vol. 50(4): 1211-1250 1219
1220 TRANSACTIONS OF THE ASABE
Vol. 50(4): 1211-1250 1221
1222 TRANSACTIONS OF THE ASABE
Vol. 50(4): 1211-1250 1223
 Van Liew and Garbrecht (2003) evaluated SWAT's ability subwatersheds. Other studies that report the use of multiple
to predict streamflow under varying climatic conditions for gauges to perform hydrologic calibration and validation with
three nested subwatersheds in the 610 km2 Little Washita SWAT include Cao et al. (2006), White and Chaubey (2005),
River experimental watershed in southwestern Oklahoma. Vazquez‐Amábile and Engel (2005), and Santhi et al.
They found that SWAT could adequately simulate runoff for (2001a).
dry, average, and wet climatic conditions in one subwa‐
Applications Accounting for Base Flow and/or for
tershed, following calibration for relatively wet years in two
of the subwatersheds. Govender and Everson (2005) report Karst‐Influenced Systems
Arnold et al. (1995a) and Arnold and Allen (1999) de‐
relatively strong streamflow simulation results (table 2) for
scribe a digital filter technique that can be used for determin‐
a small (0.68 km2) research watershed in South Africa. How‐
ing separation of base and groundwater flow from overall
ever, they also found that SWAT performed better in drier
streamflow, which has been used to estimate base flow and/or
years than in a wet year, and that the model was unable to ade‐
groundwater flow in several SWAT studies (e.g., Arnold et
quately simulate the growth of Mexican Weeping Pine due to
al., 2000; Santhi et al., 2001a; Hao et al., 2004; Cheng et al.,
inaccurate accounting of observed increased ET rates in ma‐
2006; Kalin and Hantush, 2006; Jha et al., 2007). Arnold et
ture plantations.
al. (2000) found that SWAT groundwater recharge and dis‐
Qi and Grunwald (2005) point out that, in most studies,
charge (base flow) estimates for specific 8‐digit watersheds
SWAT has usually been calibrated and validated at the drain‐
compared well with filtered estimates for the 491,700 km2
age outlet of a watershed. In their study, they calibrated and
upper Mississippi River basin. Jha et al. (2007) report accu‐
validated SWAT for four subwatersheds and at the drainage
rate estimates of streamflow (table 2) for the 9,400 km2 Rac‐
outlet (table 2). They found that spatially distributed calibra‐
coon River watershed in west central Iowa, and that their
tion and validation accounted for hydrologic patterns in the

1224 TRANSACTIONS OF THE ASABE

predicted base flow was similar to both the filtered estimate and a temporal resolution of one week. The simulated soil
and a previous base flow estimate. Kalin and Hantush (2006) moisture was evaluated on the basis of vegetation response,
report accurate surface runoff and streamflow results for the by using 16 years of normalized difference vegetation index
120 km2 Pocono Creek watershed in eastern Pennsylvania (NDVI) data derived from NOAA‐AVHRR satellite data.
(table 2); their base flow estimates were weaker, but they The predicted soil moistures were well correlated with agri‐
state those estimates were not a performance criteria. Base culture and pasture NDVI values. Narasimhan and Sriniva‐
flow and other flow components estimated with SWAT by san (2005) describe further applications of a soil moisture
Srivastava et al. (2006) for the 47.6 km2 West Branch Bran‐ deficit index and an evapotranspiration deficit index.
dywine Creek watershed in southwest Pennsylvania were Arnold et al. (2005) validated a crack flow model for
found to be generally poor (table 2). Peterson and Hamlett SWAT, which simulates soil moisture conditions with depth
(1998) also found that SWAT was not able to simulate base to account for flow conditions in dry weather. Simulated
flows for the 39.4 km2 Ariel Creek watershed in northeast crack volumes were in agreement with seasonal trends, and
Pennsylvania, due to the presence of soil fragipans. Chu and the predicted daily surface runoff levels also were consistent
Shirmohammadi (2004) found that SWAT was unable to sim‐ with measured runoff data (table 2). Sun and Cornish (2005)
ulate an extremely wet year for a 3.46 km2 watershed in simulated 30 years of bore data for a 437 km2 watershed.
Maryland. After removing the wet year, the surface runoff, They used SWAT to estimate recharge in the headwaters of
base flow, and streamflow results were within acceptable ac‐ the Liverpool Plains in New South Wales, Australia. These
curacy on a monthly basis. Subsurface flow results also im‐ authors determined that SWAT could estimate recharge and
proved when the base flow was corrected. incorporate land use and land management at the watershed
Spruill et al. (2000) calibrated and validated SWAT with scale. A code modification was performed by Vazquez‐
one year of data each for a small experimental watershed in Amábile and Engel (2005) that allowed reporting of soil
Kentucky. The 1995 and 1996 daily NSE values reflected moisture for each soil layer. The soil moisture values were
poor peak flow values and recession rates, but the monthly then converted into groundwater table levels based on the ap‐
flows were more accurate (table 2). Their analysis confirmed proach used in DRAINMOD (Skaggs, 1982). It was con‐
the results of a dye trace study in a central Kentucky karst wa‐ cluded that predictions of groundwater table levels would be
tershed, indicating that a much larger area contributed to useful to include in SWAT.
streamflow than was described by topographic boundaries. Modifications were performed by Du et al. (2006) to
Coffey et al. (2004) report similar statistical results for the SWAT2000 to improve the original SWAT tile drainage func‐
same Kentucky watershed (table 2). Benham et al. (2006) re‐ tion. The modified model was referred to as SWAT‐M and re‐
port that SWAT streamflow results (table 2) did not meet cal‐ sulted in clearly improved tile drainage and streamflow
ibration criteria for the karst‐influenced 367 km2 Shoal Creek predictions for the relatively flat and intensively cropped
watershed in southwest Missouri, but that visual inspection 51.3 km2 Walnut Creek watershed in central Iowa (table 2).
of the simulated and observed hydrographs indicated that the Green et al. (2006) report a further application of the revised
system was satisfactorily modeled. They suggest that SWAT tile drainage routine using SWAT2005 for a large tile‐drained
was not able to capture the conditions of a very dry year in watershed in north central Iowa, which resulted in a greatly
combination with flows sustained by the karst features. improved estimate of the overall water balance for the wa‐
Afinowicz et al. (2005) modified SWAT in order to more tershed (table 2). This study also presented the importance of
realistically simulate rapid subsurface water movement ensuring that representative runoff events are present in both
through karst terrain in the 360 km2 Guadalupe River wa‐ the calibration and validation in order to improve the model's
tershed in southwest Texas. They report that simulated base effectiveness.
flows matched measured streamflows after the modification,
Snowmelt‐Related Applications
and that the predicted daily and monthly and daily results
(table 2) fell within the range of published model efficiencies Fontaine et al. (2002) modified the original SWAT snow
accumulation and snowmelt routines by incorporating im‐
for similar systems. Eckhardt et al. (2002) also found that
proved accounting of snowpack temperature and accumula‐
their modifications for SWAT‐G resulted in greatly improved
simulation of subsurface interflow in German low mountain tion, snowmelt, and areal snow coverage, and an option to
input precipitation and temperature as a function of elevation
conditions (table 2).
bands. These enhancements resulted in greatly improved
Soil Water, Recharge, Tile Flow, and Related Studies streamflow estimates for the mountainous 5,000 km2 upper
Mapfumo et al. (2004) tested the model's ability to simu‐ Wind River basin in Wyoming (table 2). Abbaspour et al.
late soil water patterns in small watersheds under three graz‐ (2007) calibrated several snow‐related parameters and used
ing intensities in Alberta, Canada. They observed that SWAT four elevation bands in their SWAT simulation of the
had a tendency to overpredict soil water in dry soil conditions 1,700km 2 Thur watershed in Switzerland that is character‐
and to underpredict in wet soil conditions. Overall, the model ized by a pre‐alpine/alpine climate. They report excellent
was adequate in simulating soil water patterns for all three SWAT discharge estimates.
watersheds with a daily time step. SWAT was used by Delib‐ Other studies have reported mixed SWAT snowmelt simu‐
erty and Legates (2003) to document 30‐year (1962‐1991) lation results, including three that reported poor results for
long‐term average soil moisture conditions and variability, watersheds (0.395 to 47.6 km2) in eastern Pennsylvania. Pet‐
and topsoil variability, for Oklahoma. The model was judged erson and Hamlett (1998) found that SWAT was unable to ac‐
to be able to accurately estimate the relative magnitude and count for unusually large snowmelt events, and Srinivasan et
variability of soil moisture in the study region. Soil moisture al. (2005) found that SWAT underpredicted winter stream‐
was simulated with SWAT by Narasimhan et al. (2005) for six flows; both studies used SWAT versions that predated the
large river basins in Texas at a spatial resolution of 16 km2 modifications performed by Fontaine et al. (2002). Srivasta‐

Vol. 50(4): 1211-1250 1225

va et al. (2006) also found that SWAT did not adequately pre‐ in order to continuously function over the entire study period.
dict winter flows. Qi and Grunwald found that SWAT did not Conan et al. (2003b) found that SWAT adequately simulated
predict winter season precipitation‐runoff events well for the conversion of wetlands to dry land for the upper Guadiana
3,240 km2 Sandusky River watershed. Chanasyk et al. (2003) River basin in Spain but was unable to represent all of the dis‐
found that SWAT was not able to replicate snowmelt‐ charge details impacted by land use alterations. Wu and John‐
dominated runoff (table 2) for three small grassland wa‐ ston (2007) accounted for wetlands and lakes in their SWAT
tersheds in Alberta that were managed with different grazing simulation of a Michigan watershed, which covered over
intensities. Wang and Melesse (2005) report that SWAT accu‐ 23% of the watershed. The impact of flood‐retarding struc‐
rately simulated the monthly and annual (and seasonal) dis‐ tures on streamflow for dry, average, and wet climatic condi‐
charges for the Wild Rice River watershed in Minnesota, in tions in Oklahoma was investigated with SWAT by Van Liew
addition to the spring daily streamflows, which were predom‐ et al. (2003b). The flood‐retarding structures were found to
inantly from melted snow. Accurate snowmelt‐dominated reduce average annual streamflow by about 3% and to effec‐
streamflow predictions were also found by Wang and Me‐ tively reduce annual daily peak runoff events. Reductions of
lesse (2006) for the Elm River in North Dakota. Wu and John‐ low streamflows were also predicted, especially during dry
ston (2007) found that the snow melt parameters used in conditions. Mishra et al. (2007) report that SWAT accurately
SWAT are altered by drought conditions and that streamflow accounted for the impact of three checkdams on both daily
predictions for the 901 km2 South Branch Ontonagon River and monthly streamflows for the 17 km2 Banha watershed in
in Michigan improved when calibration was based on a northeast India (table 2). Hotchkiss et al. (2000) modified
drought period (versus average climatic conditions), which SWAT based on U.S. Army Corp of Engineers reservoir rules
more accurately reflected the drought conditions that charac‐ for major Missouri River reservoirs, which resulted in greatly
terized the validation period. Statistical results for all these improved simulation of reservoir dynamics over a 25‐year
studies are listed in table 2. period. Kang et al. (2006) incorporated a modified impound‐
Benaman et al. (2005) found that SWAT2000 reasonably ment routine into SWAT, which allowed more accurate simu‐
replicated streamflows for the 1,200 km2 Cannonsville Res‐ lation of the impacts of rice paddy fields within a South
ervoir watershed in New York (table 2), but that the model un‐ Korean watershed (table 2).
derestimated snowmelt‐driven winter and spring stream-
Green‐Ampt Applications
flows. Improved simulation of cumulative winter stream‐
Very few SWAT applications in the literature report the use
flows and spring base flows were obtained by Tolston and
of the Green‐Ampt infiltration option. Di Luzio and Arnold
Shoemaker (2007) for the same watershed (table 2) by modi‐
(2004) report sub‐hourly results for two different calibration
fying SWAT2000 so that lateral subsurface flow could occur
in frozen soils. Francos et al. (2001) also modified SWAT to methods using the Green‐Ampt method (table 2). King et al.
(1999) found that the Green‐Ampt option did not provide any
obtain improved streamflow results for the Kerava River wa‐
significant advantage as compared to the curve number ap‐
tershed in Finland (table 2) by using a different snowmelt
submodel that was based on degree‐days and that could ac‐ proach for uncalibrated SWAT simulations for the 21.3 km2
Goodwin Creek watershed in Mississippi (table 2). Kannan
count for variations in land use by subwatershed. Incorporat‐
et al. (2007b) report that SWAT streamflow results were more
ing modifications such as those described in these two studies
may improve the accuracy of snowmelt‐related processes in accurate using the curve number approach as compared to the
Green‐Ampt method for a small watershed in the U.K.
future SWAT versions.
(table2). However, they point out that several assumptions
Irrigation and Brush Removal Scenarios were not optimal for the Green‐Ampt approach.
Gosain et al. (2005) assessed SWAT's ability to simulate
return flow after the introduction of canal irrigation in a basin POLLUTANT LOSS STUDIES
in Andra Pradesh, India. SWAT provided the assistance water Nearly 50% of the reviewed SWAT studies (table 1) report
managers needed in planning and managing their water re‐ simulation results of one or more pollutant loss indicator.
sources under various scenarios. Santhi et al. (2005) describe Many of these studies describe some form of verifying pollu‐
a new canal irrigation routine that was used in SWAT. Cumu‐ tant prediction accuracy, although the extent of such report‐
lative irrigation withdrawal was estimated for each district ing is less than what has been published for hydrologic
for each of three different conservation scenarios (relative to assessments. Table 3 lists R2 and NSE statistics for 37 SWAT
a reference scenario). The percentage of water that was saved pollutant loss studies, which again are used here as key indi‐
was also calculated. SWAT was used by Afinowicz et al. cators of model performance. The majority of the R2 and NSE
(2005) to evaluate the influence of woody plants on water values reported in table 3 exceed 0.5, indicating that the mod‐
budgets of semi‐arid rangeland in southwest Texas. Baseline el was able to replicate a wide range of observed in‐stream
brush cover and four brush removal scenarios were evaluat‐ pollutant levels. However, poor results were again reported
ed. Removal of heavy brush resulted in the greatest changes for some studies, especially for daily comparisons. Similar to
in ET (approx. 32 mm year-1 over the entire basin), surface the points raised for the hydrologic results, some of the weak‐
runoff, base flow, and deep recharge. Lemberg et al. (2002) er results were due in part to inadequate characterization of
also describe brush removal scenarios. input data (Bouraoui et al., 2002), uncalibrated simulations
of pollutant movement (Bärlund et al., 2007), and uncertain‐
Applications Incorporating Wetlands, Reservoirs, and
ties in observed pollutant levels (Harmel et al., 2006).
Other Impoundments
Arnold et al. (2001) simulated a wetland with SWAT that Sediment Studies
was proposed to be sited next to Walker Creek in the Fort Several studies showed the robustness of SWAT in predict‐
Worth, Texas, area. They found that the wetland needed to be ing sediment loads at different watershed scales. Saleh et al.
above 85% capacity for 60% of a 14‐year simulation period,

1226 TRANSACTIONS OF THE ASABE

 Table 3. Summary of reported SWAT environmental indicator calibration and validation
coefficient of determination (R2) and Nash‐Sutcliffe model efficiency (NSE) statistics.
Drainage Time Period Calibration Validation
Area (C = calib., Daily Monthly Annual Daily Monthly Annual
Reference Watershed (km2)[a] Indicator[b] V = valid.) R2 NSE R2 NSE R2 NSE R2 NSE R2 NSE R2 NSE
Arabi et al. Dreisbach and 6.2 Suspended C: 1974-1975 0.97 0.92 0.86 0.75
(2006b) [c] Smith Fry and solids V: 1976 to and and and and
(Indiana) 7.3 May 1977 0.94 0.86 0.85 0.68
Total P 0.93 0.78 0.90 0.79
and and and and
0.64 0.51 0.73 0.37
Total N 0.76 0.54 0.75 0.85
and and and and
0.61 0.50 0.52 0.72
Bärlund et al. Lake Pyhäjärvi -- Sediment 1990-1994 0.01
(2007) [d],[e] (Finland)
Behera and Kapgari (India) 9.73 Sediment C: 2002 0.93 0.84 0.89 0.86
Panda V: 2003
(2006) (rainy season)
Nitrate 0.93 0.92 0.87 0.83
Total P 0.92 0.83 0.94 0.89
Bouraoui et al. Ouse River 3,500 Nitrate 1986-1990 0.64
(2002) (Yorkshire, U.K.) Ortho P 0.02
Bouraoui et al. Vantaanjoki 295 Susp. solids 1982-1984 0.49
(2004) (Finland); Total N 0.61
subwatershed
Total P 0.74
Entire 1,682 Nitrate 1974-1998 0.34
watershed Total P 0.62
Bracmort et al. Dreisbach and 6.2 Mineral P C: 1974-1975 0.92 0.84 0.86 0.74
(2006) [c] Smith Fry and V: 1976 to and and and and
(Indiana) 7.3 May 1977 0.90 0.78 0.73 0.51
Cerucci and Townbrook 36.8 Sediment Oct. 1999- 0.70
Conrad (New York) Sept. 2000
(2003) [f]
Dissolved P 0.91
Particulate P 0.40
Chaplot et al. Walnut Creek 51.3 Nitrate 1991-1998 0.56
(2004)
Cheng et al. Heihe River 7,241 Sediment C: 1992-1997 0.70 0.74 0.78 0.76
(2006) (China) V: 1998-1999
Ammonia C: 1992-1997 0.75 0.76 0.74 0.72
V: 1998-1999
Chu et al. Warner Creek 3.46 Sediment Varying 0.10 0.05 0.19 0.11 0.91 0.90
(2004) [g] periods
Nitrate 0.27 0.16 0.38 0.36 0.96 0.90
Ammonium 0.38 -0.05 0.80 0.19
Total 0.40 0.15 0.66 -0.56
Kjeldahl N
Soluble P 0.39 -0.08 0.65 0.64 0.87 0.80
Total P 0.38 0.08 0.83 0.19
Cotter et al. Moores Creek 18.9 Sediment 1997-1998 0.48
(2003) (Arkansas) Nitrate 0.44
Total P 0.66
Di Luzio et al. Upper North 932.5 Sediment Jan. 1993 to 0.78
(2002) Bosque River July 1998
(Texas)
Organic N 0.60
Nitrate 0.60
Organic P 0.70
Ortho P 0.58

Vol. 50(4): 1211-1250 1227

 Table 3 (cont'd). Summary of reported SWAT environmental indicator calibration and validation
coefficient of determination (R2) and Nash‐Sutcliffe model efficiency (NSE) statistics.
Calibration Validation
Drainage Time Period
Daily Monthly Annual Daily Monthly Annual
Area (C = calib.,
Reference Watershed (km2)[a] Indicator[b] V = valid.) R2 NSE R2 NSE R2 NSE R2 NSE R2 NSE R2 NSE
Du et al. Walnut Creek (Iowa); 51.3 Nitrate C: 1992-1995 -0.37 -0.21 -0.14 -0.21
(2006) [d],[h],[i] subwatershed (stream V: 1996-2001 and and and and
(site 310) and flow) (SWAT2000) -0.41 -0.26 -0.18 -0.22
watershed outlet
Subwatershed -- Nitrate (SWAT2000) -0.60 -0.08 -0.16 -0.31
(site 210) (tile flow)
Subwatershed 51.3 Nitrate (SWAT-M)[j] 0.61 0.91 0.41 0.80
(site 310) and (stream and and and and
watershed outlet flow) 0.53 0.85 0.26 0.67
Subwatershed -- Nitrate (SWAT-M) 0.25 0.73 0.42 0.71
(site 210) (tile flow)
Subwatershed 51.3 Atrazine (SWAT2000) -0.05 -0.01 -0.02 -0.04
(site 310) and (stream and and and and
watershed outlet flow) -0.12 -0.02 -0.39 0.06
Subwatershed -- Atrazine (SWAT2000) -0.47 -0.04 -0.46 -0.06
(site 210) (tile flow)
Subwatershed 51.3 Atrazine (SWAT-M) 0.21 0.50 0.12 0.53
(site 310) and (stream and and and and
watershed outlet flow) 0.47 0.73 -0.41 0.58
Subwatershed -- Atrazine (SWAT-M) 0.51 0.92 0.09 0.31
(site 210) (tile flow)
Gikas et al. Vistonis Lagoon 1,349 Sediment C: May 1998 0.40 0.34
(2005) [d],[k] (Greece); to June 1999 to to
nine gauges V: Nov. 1999 0.98 0.98
to Jan. 2000
Nitrate 0.51 0.57
to to
0.87 0.89
Total P 0.50 0.43
to to
0.82 0.97
Grizzetti et al. Parts of four 1,380 Nitrate 1995-1999 0.24 0.32 0.004 -0.66 0.68
(2005) [d] watersheds (U.K.); to and and and
C: one gauge, 8,900 nitrite 0.28 0.38
V: two gauges,
annual: 50 gauges
Grizzetti et al. Vantaanjoki 295 Total N Varying 0.59 0.43 0.10
(2003) (Finland); to periods and and
three gauges 1,682 0.51 0.30
Total P 0.74 0.54 0.63
and and
0.44 0.64
Grunwald Sandusky (Ohio); 90.3 Suspended C: 1998-1999 -5.1 -1.0
and Qi three gauges to sediment V: 2000-2001 to to
(2006) 3,240 0.2 0.02
Total P -0.89 0.08
to to
0.07 0.45
Nitrite -4.6 -0.16
to to
0.19 0.48
Nitrate -0.12 -0.1
to to
0.29 0.57
Ammonia -0.44 -0.44
to to
-0.24 -0.21

1228 TRANSACTIONS OF THE ASABE

 Table 3 (cont'd). Summary of reported SWAT environmental indicator calibration and validation
coefficient of determination (R2) and Nash‐Sutcliffe model efficiency (NSE) statistics.
Drainage Time Period Calibration Validation
Area (C = calib., Daily Monthly Annual Daily Monthly Annual
Reference Watershed (km2)[a] Indicator[b] V = valid.) R2 NSE R2 NSE R2 NSE R2 NSE R2 NSE R2 NSE
Hanratty and Cottonwood 3,400 Suspended 1967-1991 0.59
Stefan (Minnesota) sediment
(1998)
Nitrate 0.68
and nitrite
Total P 0.54
Organic N 0.57
and
ammonia
Hao et al. Lushi (China) 4,623 Sediment C: 1992-1997 0.72 0.72 0.98 0.94
(2004) V: 1998-1999
Jha et al. Raccoon River 8,930 Sediment C: 1981-1992 0.55 0.53 0.97 0.93 0.80 0.78 0.89 0.79
(2007) [l] (Iowa) V: 1993-2003
Nitrate 0.76 0.73 0.83 0.78 0.79 0.78 0.91 0.84
Kang et al. Baran 29.8 Suspended C: 1996-1997 0.77 0.70 0.89 0.89
(2006) [k] (South Korea) solids V: 1999-2000
Total N 0.84 0.73 0.85 0.65
Total P 0.81 0.42 0.85 0.19
Kaur et al. Nagwan (India) 9.58 Sediment C: 1984 0.54 -0.67 0.65 0.70
(2004) and 1992
V: 1981-1983,
1985-1989,
and 1991
Kirsch et al. Rock River 190 Sediment 1991-1995 0.82 0.75
(2002) (Wisconsin);
Windsor gauge Total P 0.95 0.07
Mishra et al. Banha (India) 17 Sediment C: 1996 0.82 0.82 0.99 0.98 0.77 0.58 0.89 0.63
(2007) V: 1997-2001
Muleta and Big Creek 86.5 Sediment 1999-2001 0.42
Nicklow (Illinois)
(2005a)
Muleta and Big Creek 23.9 Sediment C: June 1999 0.46 -0.005
Nicklow (Illinois); and to Aug. 2001
(2005b) separate gauges 86.5 V: Apr. 2000
for C and V to Aug. 2001
Nasr et al. Clarianna, Dripsey, 15 Total P Varying 0.44
(2007) [c] and Oona Water to periods to
(Ireland) 96 0.59
Plus et al. Thau Lagoon 280 Nitrate 1993-1999 0.44
(2006) [d],[m] (France); and
two gauges 0.27
Ammonia 0.31
and
0.15
Organic N 0.66
and
0.20
Saleh et al. Upper North 932.5 Sediment Oct. 1993 to 0.81 0.94
(2000) [n] Bosque River Aug. 1995
(Texas);
Nitrate 0.27 0.65
C: one gauge,
V: 11 gauges Organic N 0.78 0.82
Total N 0.86 0.97
Ortho P 0.94 0.92
Particulate 0.54 0.89
P
Total P 0.83 0.93

Vol. 50(4): 1211-1250 1229

 Table 3 (cont'd). Summary of reported SWAT environmental indicator calibration and validation
coefficient of determination (R2) and Nash‐Sutcliffe model efficiency (NSE) statistics.
Drainage Time Period Calibration Validation
Area (C = calib., Daily Monthly Annual Daily Monthly Annual
Reference Watershed (km2)[a] Indicator[b] V = valid.) R2 NSE R2 NSE R2 NSE R2 NSE R2 NSE R2 NSE
Saleh and Du Upper North 932.5 Total C: Jan. 1994 -2.5 0.83 -3.5 0.59
(2004) Bosque River suspended to June 1995
(Texas) solids V: July 1995
to July 1999
Nitrate 0.04 0.29 0.50 0.50
and nitrite
Organic N -0.07 0.87 0.69 0.77
Total N 0.01 0.81 0.68 0.75
Ortho P 0.08 0.76 0.45 0.40
Particulate -0.74 0.59 0.59 0.73
P
Total P -0.08 0.77 0.63 0.71
Santhi et al. Bosque River 4,277 Sediment C: 1993-1997 0.81 0.80 0.98 0.70
(2001a) [d],[o] (Texas); V: 1998 and and and and
two gauges 0.87 0.69 0.95 0.23
Mineral N 0.64 0.59 0.89 0.75
and and and and
0.72 -0.08 0.72 0.64
Organic N 0.61 0.58 0.92 0.73
and and and and
0.60 0.57 0.71 0.43
Mineral P 0.60 0.59 0.83 0.53
and and and and
0.66 0.53 0.93 0.81
Organic P 0.71 0.70 0.95 0.72
and and and and
0.61 0.59 0.80 0.39
Stewart et al. Upper North 932.5 Sediment C: 1994-1999 0.94 0.80 0.82 0.63
(2006) Bosque River V: 2001-2002
(Texas)
Mineral N 0.80 0.60 0.57 -0.04
Organic N 0.87 0.71 0.89 0.73
Mineral P 0.88 0.75 0.82 0.37
Organic P 0.85 0.69 0.89 0.58
Tolson and Cannonsville 37 Total Varying 0.70 0.67 0.42 0.33 0.72 0.52
Shoemaker (New York) to suspended periods (0.47) (0.24) and and and and
(2007) [d],[j],[p] 913[q] solids 0.83 0.83 0.83 0.76
Total 0.79 0.78 0.62 0.61 0.93 0.89
dissolved (0.84) (0.84) and and and and
P 0.71 -5.3 0.89 -6.5
Particulate 0.67 0.61 0.37 0.32 0.63 0.48
P (0.50) (0.26) and and and and
0.85 0.85 0.88 0.79
Total P 0.73 0.78 0.43 0.40 0.75 0.63
(0.58) (0.37) and and and and
0.87 0.78 0.92 0.92
Tripathi et al. Nagwan (India) 92.5 Sediment June-Oct. 1997 0.89 0.89 0.89 0.79
(2003)
Nitrate 0.89
Organic N 0.82
Soluble P 0.82
Organic P 0.86
Vazquez- St. Joseph River 628.2 Atrazine 1996-1999 0.14 0.42
Amabile et al. (Indiana, Michigan, to
(2006) [i] and Ohio); 1620
ten sampling sites
Main outlet at 2,620 Atrazine 2000-2004 0.27 -0.31 0.59 0.28
Fort Wayne, Indiana

1230 TRANSACTIONS OF THE ASABE

 Table 3 (cont'd). Summary of reported SWAT environmental indicator calibration and validation
coefficient of determination (R2) and Nash‐Sutcliffe model efficiency (NSE) statistics.
Drainage Time Period Calibration Validation
Area (C = calib., Daily Monthly Annual Daily Monthly Annual
Reference Watershed (km2)[a] Indicator[b] V = valid.) R2 NSE R2 NSE R2 NSE R2 NSE R2 NSE R2 NSE
Veith et al. Watershed FD-36 0.395 Sediment 1997-2000 0.04 -0.75
(2005) (Pennsylvania)
White and Beaver Reservoir 362 Sediment C: 2000 0.45 0.23 0.69 0.32
Chaubey (Arkansas); to or 2001 to to to to
(2005) [r],[s] three gauges 1,020 V: 2001 0.85 0.76 0.82 0.85
or 2002
Nitrate 0.01 -2.36 0.59 0.13
and to to and and
nitrite 0.84 0.29 0.71 0.49
Total P 0.50 0.40 0.58 -0.29
to to and and
0.82 0.67 0.76 0.67
[a] Based on drainage areas to the gauge(s)/sampling site(s) rather than total watershed area where reported (see footnote [d] for further information).
[b] The reported indicators are listed here as reported in each respective study; the standard SWAT variables for relevant in-stream constituents are: sediment,
organic nitrogen (N), organic phosphorus (P), nitrate (NO3-N), ammonium (NH4-N), nitrite (NO2-N), and mineral P (Neitsch et al., 2005b).
[c] Arabi et al. (2006b) and Bracmort et al. (2006) reported the same set of r2 and NSE statistics for sediment and total P; the calibration time periods were
reported by Arabi et al. (2006b), and the validation time periods were inferred from graphical results reported by Bracmort et al. (2006).
[d] Explicit or estimated drainage areas were not reported for some or all of the gauge sites; the total watershed area is listed for those studies that reported it.
[e] The exact time scale of comparison was not explicitly stated and thus was inferred from other information provided.
[f] The statistics reported for sediment and organic P excluded the months of February and March 2000; large underestimations of both constituents occurred
in those two months.
[g] The nutrient statistics were based on adjusted flows that accounted for subsurface flows that originated from outside the watershed as reported by Chu and
Shirmohammadi (2004); the annual sediment, nitrate, and soluble P statistics were based on the combined calibration and validation periods.
[h] The daily and monthly statistics were based only on the days that sampling occurred.
[i] Other statistics were reported for different time periods, conditions, gauge combinations, and/or variations in selected in input data.
[j] A modified SWAT model was used.
[k] The exact time scale of comparison was not explicitly stated and thus was inferred from other information provided.
[l] A similar set of Raccoon River watershed statistics were reported for slightly different time periods by Secchi et al. (2007).
[m] Specific calibration and/or validation time periods were reported, but the statistics were based on the overall simulated time period (calibration plus
validation time periods).
[n] The APEX model (Williams and Izaurralde, 2006) was interfaced with SWAT for this study. The calibration statistics were based on a comparison between
simulated and measured flows at the watershed outlet, while the validation statistics were based on a comparison between simulated and measured flows
averaged across 11 different gauges.
[o] The calibration and validation statistics were also reported by Santhi et al. (2001b).
[p] The calibration statistics in parentheses include January 1996; an unusually large runoff and erosion event occurred during that month.
[q] As reported by Benamen et al. (2005).
[r] These statistics were computed on the basis of comparisons between simulated and measured data within specific years, rather than across multiple years.
[s] The statistics for the War Eagle Creek subwatershed gauge were also reported by Migliaccio et al. (2007).

(2000) conducted a comprehensive SWAT evaluation for the for both the effects of snow cover and snow runoff depth (the
932.5 km2 upper North Bosque River watershed in north cen‐ latter is not accounted for in the standard SWAT model) to
tral Texas, and found that predicted monthly sediment losses overcome snowmelt‐induced prediction problems identified
matched measured data well but that SWAT daily output was by Benaman et al. (2005) for the Cannonsville Reservoir wa‐
poor (table 3). Srinivasan et al (1998) concluded that SWAT tershed in New York. They also reported improved sediment
sediment accumulation predictions were satisfactory for the loss predictions (table 3). Jha et al. (2007) found that the sedi‐
279 km2 Mill Creek watershed, again located in north central ment loads predicted by SWAT were consistent with sedi‐
Texas. Santhi et al. (2001a) found that SWAT‐simulated sedi‐ ment loads measured for the Raccoon River watershed in
ment loads matched measured sediment loads well (table 3) Iowa (table 3). Arabi et al. (2006b) report satisfactory SWAT
for two Bosque River (4,277 km2) subwatersheds, except in sediment simulation results for two small watersheds in Indi‐
March. Arnold et al. (1999b) used SWAT to simulate average ana (table 3). White and Chaubey (2005) report that SWAT
annual sediment loads for five major Texas river basins sediment predictions for the Beaver Reservoir watershed in
(20,593 to 569,000 km2) and concluded that the SWAT‐ northeast Arkansas (table 3) were satisfactory. Sediment re‐
predicted sediment yields compared reasonably well with es‐ sults are also reported by Cotter et al. (2003) for another Ar‐
timated sediment yields obtained from rating curves. kansas watershed (table 3). Hanratty and Stefan (1998)
Besides Texas, the SWAT sediment yield component has calibrated SWAT using water quality and quantity data mea‐
also been tested in several Midwest and northeast U.S. states. sured in the Cottonwood River in Minnesota (table3). In
Chu et al. (2004) evaluated SWAT sediment prediction for the Wisconsin, Kirsch et al. (2002) calibrated SWAT annual pre‐
Warner Creek watershed located in the Piedmont physio‐ dictions for two subwatersheds located in the Rock River ba‐
graphic region of Maryland. Evaluation results indicated sin (table 3), which lies within the glaciated portion of south
strong agreement between yearly measured and SWAT‐ central and eastern Wisconsin. Muleta and Nicklow (2005a)
simulated sediment load, but simulation of monthly sediment calibrated daily SWAT sediment yield with observed sedi‐
loading was poor (table 3). Tolston and Shoemaker (2007) ment yield data from the Big Creek watershed in southern Il‐
modified the SWAT2000 sediment yield equation to account linois and concluded that sediment fit seems reasonable

Vol. 50(4): 1211-1250 1231

(table 3). However, validation was not conducted due to lack approach used in SWAT2000. However, Jha et al. (2007) re‐
of data. port accurate nitrate loss predictions (table 3) for the Raccoon
SWAT sediment simulations have also been evaluated in River watershed in Iowa using SWAT2000. In Arkansas, Cot‐
Asia, Europe, and North Africa. Behera and Panda (2006) ter et al. (2003) calibrated SWAT with measured nitrate data
concluded that SWAT simulated sediment yield satisfactorily for the Moores Creek watershed and reported an NSE of 0.44.
throughout the entire rainy season based on comparisons with They state that SWAT's response was similar to that of other
daily observed data (table 3) for an agricultural watershed lo‐ published reports.
cated in eastern India. Kaur et al. (2004) concluded that Bracmort et al. (2006) and Arabi et al. (2006b) found that
SWAT predicted annual sediment yields reasonably well for SWAT could account for the effects of best management
a test watershed (table 3) in Damodar‐Barakar, India, the sec‐ practices (BMPs) on phosphorus and nitrogen losses for two
ond most seriously eroded area in the world. Tripathi et al. small watersheds in Indiana, with monthly validation NSE
(2003) found that SWAT sediment predictions agreed closely statistics ranging from 0.37 to 0.79 (table 3). SWAT tended
with observed daily sediment yield for the same watershed to underpredict both mineral and total phosphorus yields for
(table 3). Mishra et al. (2007) found that SWAT accurately the months with high measured phosphorus losses, but over‐
replicated the effects of three checkdams on sediment trans‐ predicted the phosphorus yields for months with low mea-
port (table 3) within the Banha watershed in northeast India. sured losses. Cerucci and Conrad (2003) calibrated SWAT
Hao et al. (2004) state that SWAT was the first physically soluble phosphorus predictions using measured data ob‐
based watershed model validated in China's Yellow River ba‐ tained for the Townbrook watershed in New York. They re‐
sin. They found that the predicted sediment loading accurate‐ ported monthly NSE values of 0.91 and 0.40, if the measured
ly matched loads measured for the 4,623 km2 Lushi data from February and March were excluded. Kirsch et al.
subwatershed (table 3). Cheng et al. (2006) successfully (2002) reported that SWAT phosphorus loads were consider‐
tested SWAT (table 3) using sediment data collected from the ably higher than corresponding measured loads for the Rock
7,241 km2 Heihe River, another tributary of the Yellow River. River watershed Wisconsin. Veith et al. (2005) found that
In Finland, Bärlund et al. (2007) report poor results for uncal‐ SWAT‐predicted losses were similar in magnitude to mea‐
ibrated simulations performed within the Lake Pyhäjärvi wa‐ sured watershed exports of dissolved and total phosphorus
tershed (table 3). Gikas et al. (2005) conducted an extensive during a 7‐month sampling period from a Pennsylvania wa‐
evaluation of SWAT for the Vistonis Lagoon watershed, a tershed.
mountainous agricultural watershed in northern Greece, and SWAT nutrient predictions have also been evaluated in
concluded that agreement between observed and SWAT‐ several other countries. In India, SWAT N and P predictions
predicted sediment loads were acceptable (table 3). Bouraoui were tested using measured data within the Midnapore (Beh‐
et al. (2005) evaluated SWAT for the Medjerda River basin era and Panda, 2006) and Hazaribagh (Tripathi et al., 2003)
in northern Tunisia and reported that the predicted concentra‐ districts of eastern India (table 3). Both studies concluded
tions of suspended sediments were within an order of magni‐ that the SWAT model could be successfully used to satisfac‐
tude of corresponding measured values. torily simulate nutrient losses. SWAT‐predicted ammonia
was close to the observed value (table 3) for the Heihe River
Nitrogen and Phosphorus Studies
study in China (Cheng et al., 2006). Three studies conducted
Several published studies from the U.S. showed the ro‐ in Finland for the Vantaanjoki River (Grizzetti et al. 2003;
bustness of SWAT in predicting nutrient losses. Saleh et al. Bouraoui et al. 2004) and Kerava River (Francos et al., 2001)
(2000), Saleh and Du (2004), Santhi et al. (2001a), Stewart
watersheds reported that SWAT N and P simulations were
et al. (2006), and Di Luzio et al. (2002) evaluated SWAT by generally satisfactory. Plus et al. (2006) evaluated SWAT
comparing SWAT nitrogen prediction with measured nitro‐ from data on two rivers in the Thau Lagoon watershed, which
gen losses in the upper North Bosque River or Bosque River
drains part of the French Mediterranean coast. The best cor‐
watersheds in Texas. They all concluded that SWAT reason‐ relations were found for nitrate loads, and the worst for am‐
ably predicted nitrogen loss, with most of the average month‐ monia loads (table 3). Gikas et al. (2005) evaluated SWAT
ly validation NSE values greater than or equal to 0.60
using nine gauges within the Vistonis Lagoon watershed in
(table3). Phosphorus losses were also satisfactorily simu‐ Greece and found that the monthly validation statistics gener‐
lated with SWAT in these four studies, with validation NSE ally indicated good model performance for nitrate and total
values ranging from 0.39 to 0.93 (table 3). Chu et al. (2004)
P (table 3). SWAT nitrate and total phosphorus predictions
applied SWAT to the Warner Creek watershed in Maryland were found to be excellent and good, respectively, by Abbas‐
and reported satisfactory annual but poor monthly nitrogen pour et al. (2007) for the 1700 km2 Thur River basin in Swit‐
and phosphorus predictions (table 3). Hanratty and Stefan
zerland. Bouraoui et al. (2005) applied SWAT to a part of the
(1998) calibrated SWAT nitrogen predictions using measured Medjerda River basin, the largest surface water reservoir in
data collected for the Cottonwood River, Minnesota, and Tunisia, and reported that SWAT was able to predict the range
concluded that if properly calibrated, SWAT is an appropriate
of nitrate concentrations in surface water, but lack of data pre‐
model to use for simulating the effect of climate change on vented in‐depth evaluation.
water quality; they also reported satisfactory SWAT phospho‐
rus results (table 3). Pesticide and Surfactant Studies
In Iowa, Chaplot et al. (2004) calibrated SWAT using nine Simulations of isoaxflutole (and its metabolite RPA
years of data for the Walnut Creek watershed and concluded 202248) were performed by Ramanarayanan et al. (2005)
that SWAT gave accurate predictions of nitrate load (table 3). with SWAT for four watersheds in Iowa, Nebraska, and Mis‐
Du et al. (2006) showed that the modified tile drainage func‐ souri that ranged in size from 0.49 to 1,434.6 km2. Satisfacto‐
tions in SWAT‐M resulted in far superior nitrate loss predic‐ ry validation results were obtained based on comparisons
tions for Walnut Creek (table 3), as compared to the previous with measured data. Long‐term simulations indicated that

1232 TRANSACTIONS OF THE ASABE

accumulation would not be a problem for either compound CLIMATE CHANGE IMPACT STUDIES
in semistatic water bodies. Kannan et al. (2006) report that Climate change impacts can be simulated directly in
SWAT accurately simulated movement of four pesticides for SWAT by accounting for: (1) the effects of increased atmo‐
the Colworth watershed in the U.K. The results of different spheric CO2 concentrations on plant development and tran‐
application timing and split application scenarios are also de‐ spiration, and (2) changes in climatic inputs. Several SWAT
scribed. Two scenarios of surfactant movement are described studies provide useful insights regarding the effects of arbi‐
by Kannan et al. (2007a) for the same watershed. Prediction trary CO2 fertilization changes and/or other climatic input
of atrazine greatly improved using SWAT‐M as reported by shifts on plant growth, streamflow, and other responses, in‐
Du et al. (2006) for the Walnut Creek watershed in Iowa cluding Stonefelt et al. (2000), Fontaine et al. (2001), and Jha
(table 3), which is a heavily tile‐drained watershed. Vazquez‐ et al. (2006). The SWAT results reported below focus on ap‐
Amabile et al. (2006) found that SWAT was very sensitive to proaches that relied on downscaling of climate change pro‐
the estimated timing of atrazine applications in the 2,800 km2 jections generated by general circulation models (GCMs) or
St. Joseph River watershed in northeast Indiana. The pre‐ GCMs coupled with regional climate models (RCMs).
dicted atrazine mass at the watershed outlet was in close
SWAT Studies Reporting Climate Change Impacts on
agreement with measured loads for the period of September
Hydrology
through April during 2000‐2003. Graphical and statistical
Muttiah and Wurbs (2002) used SWAT to simulate the im‐
analyses indicated that the model replicated atrazine move‐
pacts of historical climate trends versus a 2040‐2059 climate
ment trends well, but the NSE statistics (e.g., table 3) were
change projection for the 7,300 km2 San Jacinto River basin
generally weak.
in Texas. They report that the climate change scenario re‐
Scenarios of BMP and Land Use Impacts on Pollutant sulted in a higher mean streamflow due to greater flooding
Losses and other high flow increases, but that normal and low
Simulation of hypothetical scenarios in SWAT has proven streamflows decreased. Gosain et al. (2006) simulated the
to be an effective method of evaluating alternative land use, impacts of a 2041‐2060 climate change scenario on the
BMP, and other factors on pollutant losses. SWAT studies in streamflows of 12 major river basins in India, ranging in size
India include identification of critical or priority areas for soil from 1,668 to 87,180 km2. Surface runoff was found to gener‐
and water management in a watershed (Kaur et al., 2004; Tri‐ ally decrease, and the severity of both floods and droughts in‐
pathi et al., 2003). Santhi et al. (2006) report the impacts of creased, in response to the climate change projection.
manure and nutrient related BMPs, forage harvest manage‐ Rosenberg et al. (2003) simulated the effect of down‐
ment, and other BMPs on water quality in the West Fork wa‐ scaled HadCM2 GCM (Johns et al., 1997) climate projec‐
tershed in Texas. The effects of BMPs related to dairy manure tions on the hydrology of the 18 MWRRs (fig. 2) with SWAT
management and municipal wastewater treatment plant ef‐ within the HUMUS framework. Water yields were predicted
fluent were evaluated by Santhi et al. (2001b) with SWAT for to change from -11% to 153% and from 28% to 342% across
the Bosque River watershed in Texas. Stewart et al. (2006) the MWRRs in 2030 and 2095, respectively, relative to base‐
describe modifications of SWAT for incorporation of a turf‐ line conditions. Thomson et al. (2003) used the same
grass harvest routine, in order to simulate manure and soil HadCM2‐HUMUS (SWAT) approach and found that three El
Pexport that occurs during harvest of turfgrass sod within the Niño/Southern Oscillation (ENSO) scenarios resulted in
upper North Bosque River watershed in north central Texas. MWRR water yield impacts ranging from -210% to 77% rel‐
Kirsch et al. (2002) describe SWAT results showing that im‐ ative to baseline levels, depending on seasonal and dominant
proved tillage practices could result in reduced sediment weather patterns. An analysis of the impacts of 12 climate
yields of almost 20% in the Rock River in Wisconsin. Chaplot change scenarios on the water resources of the 18 MWRRs
et al. (2004) found that adoption of no tillage, changes in ni‐ was performed by Thomson et al. (2005) using the HUMUS
trogen application rates, and land use changes could greatly approach, as part of a broader study that comprised the entire
impact nitrogen losses in the Walnut Creek watershed in cen‐ issue of volume 69 (number 1) of Climatic Change. Water
tral Iowa. Analysis of BMPs by Vaché et al. (2002) for the yield shifts exceeding ±50% were predicted for portions of
Walnut Creek and Buck Creek watersheds in Iowa indicated Midwest and Southwest U.S., relative to present water yield
that large sediment reductions could be obtained, depending levels. Rosenberg et al. (1999) found that driving SWAT with
on BMP choice. Bracmort et al. (2006) present the results of a different set of 12 climate projections generally resulted in
three 25‐year SWAT scenario simulations for two small wa‐ Ogallala Aquifer recharge decreases (of up to 77%) within
tersheds in Indiana in which the impacts of no BMPs, BMPs the Missouri and Arkansas‐White‐Red MWRRs (fig. 2).
in good condition, and BMPs in varying condition are re‐ Stone et al. (2001) predicted climate change impacts on
ported for streamflow, sediment, and total P. Nelson et al. Missouri River basin (fig. 2) water yields by inputting down‐
(2005) report that large nutrient and sediment loss reductions scaled climate projections into SWAT, which were generated
occurred in response to simulated shifts of cropland into by nesting the RegCM RCM (Giorgi et al., 1998) within the
switchgrass production within the 3,000 km2 Delaware River CISRO GCM (Watterson et al., 1997) into the previously de‐
basin in northeast Kansas. Benham et al. (2006) describe a scribed version of SWAT that was modified by Hotchkiss et
TMDL SWAT application for a watershed in southwest Mis‐ al. (2000). A structure similar to the HUMUS approach was
souri. Frequency curves comparing simulated and measured used, in which 310 8‐digit watersheds were used to define the
bacteria concentrations were used to calibrate SWAT. The subwatersheds. Water yields declined at the basin outlet by
model was then used to simulate the contributions of different 10% to 20% during the spring and summer months, but in‐
bacteria sources to the stream system, and to assess the im‐ creased during the rest of the year. Further research revealed
pact of different BMPs that could potentially be used to miti‐ that significant shifts in Missouri River basin water yield im‐
gate bacteria losses in the watershed. pacts were found when SWAT was driven by downscaled

Vol. 50(4): 1211-1250 1233

CISRO GCM projections only versus the nested RegCM‐ narios downscaled from three GCMs for the 2,796 km2 Pinios
CISRO GCM approach (Stone et al., 2003). watershed in Greece. Bouraoui et al. (2002) reported that six
Jha et al. (2004b), Takle et al. (2005), and Jha et al. (2006) different climate change scenarios resulted in increased total
all report performing GCM‐driven studies for the nitrogen and phosphorus loads of 6% to 27% and 5% to 34%,
447,500km 2 upper Mississippi River basin (fig. 2), with an respectively, for the 3,500 km2 Ouse River watershed located
assumed outlet at Grafton, Illinois, using a framework con‐ in the Yorkshire region of the U.K. Bouraoui et al. (2004) fur‐
sisting of 119 8‐digit subwatersheds and land use, soil, and to‐ ther found for the Vantaanjoki River watershed, which covers
pography data that was obtained from BASINS. Jha et al. 1,682 km2 in southern Finland, that snow cover decreased,
(2004b) found that streamflows in the upper Mississippi Riv‐ winter runoff increased, and slight increases in annual nutri‐
er basin increased by 50% for the period 2040‐2049, when ent losses occurred in response to a 34‐year scenario repre‐
climate projections generated by a nested RegCM2‐HadCM2 sentative of observed climatic changes in the region.
approach were used to drive SWAT. Jha et al. (2006) report Boorman (2003) evaluated the impacts of climate change for
that annual average shifts in upper Mississippi River basin five different watersheds located in Italy, France, Finland,
streamflows, relative to the baseline, ranged from -6% to and the UK., including the three watersheds analyzed in the
38% for five 2061‐2090 GCM projections and increased by Varanou et al. (2002), Bouraoui et al. (2002), and Bouraoui
51% for a RegCM‐CISRO projection reported by Giorgi et al. et al. (2004) studies.
(1998). An analysis of driving SWAT with precipitation out‐
put generated with nine GCM models indicated that GCM SENSITIVITY, CALIBRATION, AND UNCERTAINTY ANALYSES
multi‐model results may be used to depict 20th century annu‐ Sensitivity, calibration, and uncertainty analyses are vital
al streamflows in the upper Mississippi River basin, and that and interwoven aspects of applying SWAT and other models.
the interface between the single high‐resolution GCM used Numerous sensitivity analyses have been reported in the
in the study and SWAT resulted in the best replication of ob‐ SWAT literature, which provide valuable insights regarding
served streamflows (Takle et al., 2005). which input parameters have the greatest impact on SWAT
Krysanova et al. (2005) report the impacts of 12 different output. As previously discussed, the vast majority of SWAT
climate scenarios on the hydrologic balance and crop yields applications report some type of calibration effort. SWAT in‐
of a 30,000 km2 watershed in the state of Brandenburg in Ger‐ put parameters are physically based and are allowed to vary
many using the SWIM model. Further uncertainty analysis of within a realistic uncertainty range during calibration. Sensi‐
climate change was performed by Krysanova et al. (2007) for tivity analysis and calibration techniques are generally re‐
the 100,000 km2 Elbe River basin in eastern Germany, based ferred to as either manual or automated, and can be evaluated
on an interface between a downscaled GCM scenario and with a wide range of graphical and/or statistical procedures.
SWIM. Eckhardt and Ulbrich (2003) found that the spring Uncertainty is defined by Shirmohammadi et al. (2006) as
snowmelt peak would decline, winter flooding would likely “the estimated amount by which an observed or calculated
increase, and groundwater recharge and streamflow would value may depart from the true value.” They discuss sources
decrease by as much as 50% in response to two climate of uncertainty in depth and list model algorithms, model cal‐
change scenarios simulated in SWAT‐G. Their approach fea‐ ibration and validation data, input variability, and scale as
tured variable stomatal conductance and leaf area responses key sources of uncertainty. Several automated uncertainty
by incorporating different stomatal conductance decline fac‐ analyses approaches have been developed, which incorpo‐
tors and leaf area index (LAI) values as a function of five rate various sensitivity and/or calibration techniques, which
main vegetation types; these refinements have not been are briefly reviewed here along with specific sensitivity anal‐
adopted in the standard SWAT model. ysis and calibration studies.
SWAT Studies Reporting Climate Change Impacts on Sensitivity Analyses
Pollutant Loss Spruill et al. (2000) performed a manual sensitivity/cal‐
Several studies report climate change impacts on both ibration analysis of 15 SWAT input parameters for a 5.5 km2
hydrology and pollutant losses using SWAT, including four watershed with karst characteristics in Kentucky, which
that were partially or completely supported by the EU showed that saturated hydraulic conductivity, alpha base
CHESS project (Varanou et al., 2002; Bouraoui et al., 2002; flow factor, drainage area, channel length, and channel width
Boorman, 2003; Bouraoui et al., 2004). Nearing et al. (2005) were the most sensitive parameters that affected streamflow.
compared runoff and erosion estimates from SWAT versus Arnold et al. (2000) show surface runoff, base flow, recharge,
six other models, in response to six climate change scenarios and soil ET sensitivity curves in response to manual varia‐
that were simulated for the 150 km2 Lucky Hills watershed tions in the curve number, soil available water capacity, and
in southeastern Arizona. The responses of all seven models soil evaporation coefficient (ESCO) input parameters for
were similar across the six scenarios for both watersheds, and three different 8‐digit watersheds within their upper Missis‐
it was concluded that climate change could potentially result sippi River basin SWAT study. Lenhart et al. (2002) report on
in significant soil erosion increases if necessary conservation the effects of two different sensitivity analysis schemes using
efforts are not implemented. Hanratty and Stefan (1998) SWAT‐G for an artificial watershed, in which an alternative
found that streamflows and P, organic N, nitrate, and sedi‐ approach of varying 44 parameter values within a fixed per‐
ment yields generally decreased for the 3,400 km2 Cotton‐ centage of the valid parameter range was compared with the
wood River watershed in southwest Minnesota in response to more usual method of varying each initial parameter by the
a downscaled 2×CO2 GCM climate change scenario. Vara‐ same fixed percentage. Both approaches resulted in similar
nou et al. (2002) also found that average streamflows, sedi‐ rankings of parameter sensitivity and thus could be consid‐
ment yields, organic N losses, and nitrate losses decreased in ered equivalent.
most months in response to nine different climate change sce‐

1234 TRANSACTIONS OF THE ASABE

 A two‐step sensitivity analysis approach is described by Automated techniques involve the use of Monte Carlo or
Francos et al. (2003), which consists of: (1) a “Morris” other parameter estimation schemes that determine automat‐
screening procedure that is based on the one factor at a time ically what the best choice of values are for a suite of parame‐
(OAT) design, and (2) the use of a Fourier amplitude sensitiv‐ ters, usually on the basis of a large set of simulations, for a
ity test (FAST) method. The screening procedure is used to calibration process. Govender and Everson (2005) used the
determine the qualitative ranking of an entire input parameter automatic Parameter Estimation (PEST) program (Doherty,
set for different model outputs at low computational cost, 2004) and identified soil moisture variables, initial ground‐
while the FAST method provides an assessment of the most water variables, and runoff curve numbers to be some of the
relevant input parameters for a specific set of model output. sensitive parameters in SWAT applications for two small
The approach is demonstrated with SWAT for the 3,500 km2 South African watersheds. They also report that manual cal‐
Ouse watershed in the U.K. using 82 input and 22 output pa‐ ibration resulted in more accurate predictions than the PEST
rameters. Holvoet et al. (2005) present the use of a Latin hy‐ approach (table 2). Wang and Melesse (2005) also used PEST
percube (LH) OAT sampling method, in which initial LH to perform an automatic SWAT calibration of three
samples serve as the points for the OAT design. The method snowmelt‐related and eight hydrologic‐related parameters
was used for determining which of 27 SWAT hydrologic‐ for the 4,335 km2 Wild Rice River watershed in northwest
related input parameters were the most sensitive regarding Minnesota, which included daily and monthly statistical
streamflow and atrazine outputs for 32 km2 Nil watershed in evaluation (table 2).
central Belgium. The LH‐OAT method was also used by van Applications of an automatic shuffled complex evolution
Griensven et al. (2006b) for an assessment of the sensitivity (SCE) optimization scheme are described by van Griensven
of 41 input parameters on SWAT flow, sediment, total N, and and Bauwens (2003, 2005) for ESWAT simulations, primari‐
total P estimates for both the UNBRW and the 3,240 km2 San‐ ly for the Dender River in Belgium. Calibration parameters
dusky River watershed in Ohio. The results show that some and ranges along with measured daily flow and pollutant data
parameters, such as the curve number (CN2), were important are input for each application. The automated calibration
in both watersheds, but that there were distinct differences in scheme executes up to several thousand model runs to find
the influences of other parameters between the two wa‐ the optimum input data set. Similar automatic calibration
tersheds. The LH‐OAT method has been incorporated as part studies were performed with a SCE algorithm and SWAT‐G
of the automatic sensitivity/calibration package included in by Eckhardt and Arnold (2001) and Eckhardt et al. (2005) for
SWAT2005. watersheds in Germany. Di Luzio and Arnold (2004) de‐
scribed the background, formulation and results (table 2) of
Calibration Approaches
an hourly SCE input‐output calibration approach used for a
The manual calibration approach requires the user to SWAT application in Oklahoma. Van Liew et al. (2005) de‐
compare measured and simulated values, and then to use ex‐
scribe an initial test of the SCE automatic approach that has
pert judgment to determine which variables to adjust, how
been incorporated into SWAT2005, for streamflow predic‐
much to adjust them, and ultimately assess when reasonable tions for the Little River watershed in Georgia and the Little
results have been obtained. Coffey et al. (2004) present near‐
Washita River watershed in Oklahoma. Van Liew et al.
ly 20 different statistical tests that can be used for evaluating
(2007) further evaluated the SCE algorithm for five wa‐
SWAT streamflow output during a manual calibration pro‐ tersheds with widely varying climatic characteristics
cess. They recommended using the NSE and R2 coefficients
(table2), including the same two in Georgia and Oklahoma
for analyzing monthly output and median objective func‐
and three others located in Arizona, Idaho, and Pennsylvania.
tions, sign test, autocorrelation, and cross‐correlation for as‐
sessing daily output, based on comparisons of SWAT Uncertainty Analyses
streamflow results with measured streamflows (table 2) for Shirmohammadi et al. (2006) state that Monte Carlo simu‐
the same watershed studied by Spruill et al. (2000). Cao et al. lation and first‐order error or approximation (FOE or FOA)
(2006) present a flowchart of their manual calibration ap‐ analyses are the two most common approaches for perform‐
proach that was used to calibrate SWAT based on five hydro‐ ing uncertainty analyses, and that other methods have been
logic outputs and multiple gauge sites within the 2075 km2 used, including the mean value first‐order reliability method,
Motueka River basin on the South Island of New Zealand. LH simulation with constrained Monte Carlo simulations,
The calibration and validation results were stronger for the and generalized likelihood uncertainty estimation (GLUE).
overall basin as compared to results obtained for six subwa‐ They present three case studies of uncertainty analyses using
tersheds (table 2). Santhi et al. (2001a) successfully cali‐ SWAT, which were based on the Monte Carlo, LH‐Monte
brated and validated SWAT for streamflow and pollutant loss Carlo, and GLUE approaches, respectively, within the con‐
simulations (tables 2 and 3) for the 4,277 km2 Bosque River text of TMDL assessments. They report that uncertainty is a
in Texas. They present a general procedure, including a flow‐ major issue for TMDL assessments, and that it should be tak‐
chart, for manual calibration that identifies sensitive input en into account during both the TMDL assessment and imple‐
parameters (15 were used), realistic uncertainty ranges, and mentation phases. They also make recommendations to
reasonable regression results (i.e., satisfactory r2 and NSE improve the quantification of uncertainty in the TMDL pro‐
values). A combined sensitivity and calibration approach is cess.
described by White and Chaubey (2005) for SWAT stream‐ Benaman and Shoemaker (2004) developed a six‐step meth‐
flow and pollutant loss estimates (tables 2 and 3) for the od that includes using Monte Carlo runs and an interval‐spaced
3,100km 2 Bear Reservoir watershed, and three subwa‐ sensitivity approach to reduce uncertain parameter ranges. After
tersheds, in northwest Arkansas. They also review calibra‐ parameter range reduction, their method reduced the model out‐
tion approaches, including calibrated input parameters, for put range by an order of magnitude, resulting in reduced uncer‐
previous SWAT studies. tainty and the amount of calibration required for SWAT.

Vol. 50(4): 1211-1250 1235

However, significant uncertainty remained with the SWAT sedi‐ (2007) found that SWAT streamflow predictions were gener‐
ment routine. Lin and Radcliffe (2006) performed an initial two‐ ally insensitive to variations in HRU and/or subwatershed de‐
stage automatic calibration streamflow prediction process with lineations for watersheds ranging in size from 21.3 to
SWAT for the 1,580 km2 Etowah River watershed in Georgia in 17,941km 2. Tripathi et al. (2006) and Muleta et al. (2007)
which an SCE algorithm was used for automatic calibration of further discuss HRU and subwatershed delineation impacts
lumped SWAT input parameters, followed by calibration of het‐ on other hydrologic components. Haverkamp et al. (2002) re‐
erogeneous inputs with a variant of the Marquardt‐Levenberg port that streamflow accuracy was much greater when using
method in which “regularization” was used to prevent parame‐ multiple HRUs to characterize each subwatershed, as op‐
ters taking on unrealistic values. They then performed a nonlin‐ posed to using just a single dominant soil type and land use
ear calibration and uncertainty analysis using PEST, in which within a subwatershed, for two watersheds in Germany and
confidence intervals were generated for annual and 7‐day one in Texas. However, the gap in accuracy between the two
streamflow estimates. Their resulting calibrated statistics are approaches decreased with increasing numbers of subwa‐
shown in table 2. Muleta and Nicklow (2005b) describe a study tersheds.
for the Big Creek watershed that involved three phases: (1) pa‐ Bingner et al. (1997) report that the number of simulated
rameter sensitivity analysis for 35 input parameters, in which subwatersheds affected predicted sediment yield and suggest
LH samples were used to reduce the number of Monte Carlo that sensitivity analyses should be performed to determine
simulations needed to conduct the analysis; (2) automatic cal‐ the appropriate level of subwatersheds. Jha et al. (2004a)
ibration using a genetic algorithm, which systematically deter‐ found that SWAT sediment and nitrate predictions were sen‐
mined the best set of input parameters using a sum of the square sitive to variations in both HRUs and subwatersheds, but
of differences criterion; and (3) a Monte Carlo‐based GLUE ap‐ mineral P estimates were not. The effects of BMPS on SWAT
proach for the uncertainty analysis, in which LH sampling is sediment, total P, and total N estimates was also found by
again used to generate input samples and reduce the computa‐ Arabi et al. (2006b) to be very sensitive to watershed subdivi‐
tion requirements. Uncertainty bounds corresponding to the sion level. Jha et al. (2004a) suggest setting subwatershed
95% confidence limit are reported for both streamflow and sedi‐ areas ranging from 2% to 5% of the overall watershed area,
ment loss, as well as final calibrated statistics (tables 2 and 3). depending on the output indicator of interest, to ensure accu‐
Arabi et al. (2007b) used a three‐step procedure that included racy of estimates. Arabi et al. (2006b) found that an average
OAT and interval‐spaced sensitivity analyses, and a GLUE subwatershed equal to about 4% of the overall watershed area
analysis to assess uncertainty of SWAT water quality predictions was required to accurately account for the impacts of BMPs
of BMP placement in the Dreisbach and Smith Fry watersheds in the model.
in Indiana. Their results point to the need for site‐specific cal‐ FitzHugh and Mackay (2000, 2001) and Chen and Mackay
ibration of some SWAT inputs, and that BMP effectiveness (2004) found that sediment losses predicted with SWAT did
could be evaluated with enough confidence to justify using the not vary at the outlet of the 47.3 km2 Pheasant Branch wa‐
model for TMDL and similar assessments. tershed in south central Wisconsin as a function of increasing
Additional uncertainty analysis insights are provided by numbers of HRUs and subwatersheds due to the transport‐
Vanderberghe et al. (2007) for an ESWAT‐based study and by limited nature of the watershed. However, sediment genera‐
Huisman et al. (2004) and Eckhardt et al. (2003), who as‐ tion at the HRU level dropped 44% from the coarsest to the
sessed the uncertainty of soil and/or land use parameter varia‐ finest resolutions (FitzHugh and Mackay, 2000), and sedi‐
tions on SWAT‐G output using Monte Carlo‐based ment yields varied at the watershed outlet for hypothetical
approaches. Van Greinsven and Meixner (2006) describe sev‐ source‐limited versus transport‐limited scenarios (FitzHugh
eral uncertainty analysis tools that have been incorporated and Mackay, 2001) in response to eight different HRU/sub‐
into SWAT2005, including a modified SCE algorithm called watershed combinations used in both studies. Chen and
“parameter solutions” (ParaSol), the Sources of Uncertainty Mackay (2004) further found that SWAT's structure in‐
Global Assessment using Split Samples (SUNGLASSES), fluences sediment predictions in tandem with spatial data ag‐
and the Confidence Analysis of Physical Inputs (CANOPI), gregation effects. They suggest that errors in MUSLE
which evaluates uncertainty associated with climatic data sediment estimates can be avoided by using only subwa‐
and other inputs. tersheds, instead of using HRUs, within subwatersheds.
In contrast, Muleta et al. (2007) found that sediment gen‐
EFFECTS OF HRU AND SUBWATERSHED DELINEATION AND erated at the HRU level and exported from the outlet of the
OTHER INPUTS ON SWAT OUTPUT 133 km2 Big Creek watershed in Illinois decreased with in‐
Several studies have been performed that analyzed im‐ creasing spatial coarseness, and that sediment yield varied
pacts on SWAT output as a function of: (1) variation in HRU significantly at the watershed outlet across a range of HRU
and/or subwatershed delineations, (2) different resolutions in and subwatershed delineations, even when the channel prop‐
topographic, soil, and/or land use data, (3) effects of spatial erties remained virtually constant.
and temporal transfers of inputs, (4) actual and/or hypotheti‐
DEM, Soil, and Land Use Resolution Effects
cal shifts in land use, and (5) variations in precipitation inputs Bosch et al. (2004) found that SWAT streamflow estimates
or ET estimates. These studies serve as further SWAT sensi‐
for a 22.1 km2 subwatershed of the Little River watershed in
tivity analyses and provide insight into how the model re‐
Georgia were more accurate using high‐resolution topo‐
sponds to variations in key inputs. graphic, land use, and soil data versus low‐resolution data ob‐
HRU and Subwatershed Delineation Effects tained from BASINS. Cotter et al. (2003) report that DEM
Bingner et al. (1997), Manguerra and Engel (1998), Fitz‐ resolution was the most critical input for a SWAT simulation
Hugh and Mackay (2000), Jha et al. (2004a), Chen and of the 18.9 km2 Moores Creek watershed in Arkansas, and
Mackay (2004), Tripathi et al. (2006), and Muleta et al. provide minimum DEM, land use, and soil resolution recom‐

1236 TRANSACTIONS OF THE ASABE

mendations to obtain accurate flow, sediment, nitrate, and to‐ hydrologic impacts with SWAT‐G (e.g., Fohrer et al., 2002,
tal P estimates. Di Luzio et al. (2005) also found that DEM 2005) and SWIM (Krysanova et al., 2005) and on nutrient and
resolution was the most critical for SWAT simulations of the sediment loss predictions with SWAT‐G (Lenhart et al.,
21.3 km2 Goodwin Creek watershed in Mississippi; land use 2003).
resolution effects were also significant, but the resolution of
Climate Data Effects
soil inputs was not. Chaplot (2005) found that SWAT surface
Chaplot et al. (2005) analyzed the effects of rain gauge
runoff estimates were sensitive to DEM mesh size, and that
nitrate and sediment predictions were sensitive to both the distribution on SWAT output by simulating the impacts of cli‐
matic inputs for a range of 1 to 15 rain gauges in both the Wal‐
choice of DEM and soil map resolution, for the Walnut Creek
nut Creek watershed in central Iowa and the upper North
watershed in central Iowa. The most accurate results did not
Bosque River watershed in Texas. Sediment predictions im‐
occur for the finest DEM mesh sizes, contrary to expecta‐
proved significantly when the densest rain gauge networks
tions. Di Luzio et al. (2004b) and Wang and Melesse (2006)
were used; only slight improvements occurred for the corre‐
present additional results describing the impacts of STATS‐
sponding surface runoff and nitrogen predictions. However,
GO versus SSURGO soil data inputs on SWAT output.
Hernandez et al. (2000) found that increasing the number of
Effects of Different Spatial and Temporal Transfers of simulated rain gauges from 1 to 10 resulted in clear estimated
Inputs streamflow improvements (table 2). Moon et al. (2004) found
Heuvelmans et al. (2004a) evaluated the effects of trans‐ that SWAT's streamflow estimates improved when Next‐
ferring seven calibrated SWAT hydrologic input parameters, Generation Weather Radar (NEXRAD) precipitation input
which were selected on the basis of a sensitivity analysis, in was used instead of rain gauge inputs (table 2). Kalin and
both time and space for three watersheds ranging in size from Hantush (2006) report that NEXRAD and rain gauge inputs
51 to 204 km2 in northern Belgium. Spatial transfers resulted resulted in similar streamflow estimates at the outlet of the
in the greatest loss of streamflow efficiency, especially be‐ Pocono Creek watershed in Pennsylvania (table 2), and that
tween watersheds. Heuvelmans et al. (2004b) further evalu‐ NEXRAD data appear to be a promising source of alternative
ated the effect of four parameterization schemes on SWAT precipitation data. A weather generator developed by Schuol
streamflow predictions, for the same set of seven hydrologic and Abbaspour (2007) that uses climatic data available at
inputs, for 25 watersheds that covered 2.2 to 210 km2 within 0.5° intervals was found to result in better streamflow esti‐
the 20,000 km2 Scheldt River basin in northern Belgium. The mates than rain gauge data for a region covering about 4 mil‐
highest model efficiencies were achieved when optimal pa‐ lion km2 in western Africa that includes the Niger, Volta, and
rameters for each individual watershed were used; optimal Senegal river basins. Sensitivity of precipitation inputs on
parameters selected on the basis of regional zones with simi‐ SWAT hydrologic output are reported for comparisons of dif‐
lar characteristics proved superior to parameters that were ferent weather generators by Harmel et al. (2000) and Watson
averaged across all 25 watersheds. et al. (2005). The effects of different ET options available in
SWAT on streamflow estimates are further described by
Historical and Hypothetical Land Use Effects Wang et al. (2006) and Kannan et al. (2007b).
Miller et al. (2002) describe simulated streamflow im‐
pacts with SWAT in response to historical land use shifts in
COMPARISONS OF SWAT WITH OTHER MODELS
the 3,150 km2 San Pedro watershed in southern Arizona and
the Cannonsville watershed in south central New York. Borah and Bera (2003, 2004) compared SWAT with sever‐
al other watershed‐scale models. In the 2003 study, they re‐
Streamflows were predicted to increase in the San Pedro wa‐
port that the Dynamic Watershed Simulation Model
tershed due to increased urban and agricultural land use,
while a shift from agricultural to forest land use was predicted (DWSM) (Borah et al., 2004), Hydrologic Simulation Pro‐
gram - Fortran (HSPF) model (Bicknell et al., 1997), SWAT,
to result in a 4% streamflow decrease in the Cannonsville wa‐
and other models have hydrology, sediment, and chemical
tershed. Hernandez et al. (2000) further found that SWAT
routines applicable to watershed‐scale catchments and con‐
could accurately predict the relative impacts of hypothetical
cluded that SWAT is a promising model for continuous simu‐
land use change in an 8.2 km2 experimental subwatershed
lations in predominantly agricultural watersheds. In the 2004
within the San Pedro watershed. Heuvelmans et al. (2005) re‐
study, they found that SWAT and HSPF could predict yearly
port that SWAT produced reasonable streamflow and erosion
flow volumes and pollutant losses, were adequate for month‐
estimates for hypothetical land use shifts, which were per‐
ly predictions except for months having extreme storm
formed as part of a life cycle assessment (LCA) of CO2 emis‐
events and hydrologic conditions, and were poor in simulat‐
sion reduction scenarios for the 29.2 km2 Meerdaal
ing daily extreme flow events. In contrast, DWSM reason‐
watershed and the 12.1 km2 Latem watersheds in northern
ably predicted distributed flow hydrographs and
Belgium. However, they state that an expansion of the SWAT
concentration or discharge graphs of sediment and chemicals
vegetation parameter dataset is needed in order to fully sup‐
at small time intervals. Shepherd et al. (1999) evaluated
port LCA analyses. Increased streamflow was predicted with
14models and found SWAT to be the most suitable for esti‐
SWAT for the 59.8 km2 Aar watershed in the German state of
mating phosphorus loss from a lowland watershed in the U.K.
Hessen, in response to a grassland incentive scenario in
Van Liew et al. (2003a) compared the streamflow predic‐
which the grassland area increased from 20% to 41% while
tions of SWAT and HSPF on eight nested agricultural wa‐
the extent forest coverage decreased by about 70% (Weber et
tersheds within the Little Washita River basin in south-
al., 2001). The impacts of hypothetical forest and other land
western Oklahoma. They concluded that SWAT was more
use changes on total runoff using SWAT are presented by
consistent than HSPF in estimating streamflow for different
Lorz et al. (2007) in the context of comparisons with three
climatic conditions and may thus be better suited for investi‐
other models. The impacts of other hypothetical land use
gating the long‐term impacts of climate variability on surface
studies for various German watersheds have been reported on

Vol. 50(4): 1211-1250 1237

water resources. Saleh and Du (2004) found that the average results were reported, with respective monthly NSE values
daily flow, sediment loads, and nutrient loads simulated by for streamflow and nitrate of 0.88 and 0.87.
SWAT were closer than HSPF to measured values collected Menking et al. (2003) interfaced SWAT with both MOD‐
at five sites during both the calibration and verification peri‐ FLOW and the MODFLOW LAK2 lake modeling package
ods for the upper North Bosque River watershed in Texas. to assess how current climate conditions would impact water
Singh et al. (2005) found that SWAT flow predictions were levels in ancient Lake Estancia (central New Mexico), which
slightly better than corresponding HSPF estimates for the existed during the late Pleistocene era. The results indicated
5,568 km2 Iroquois River watershed in eastern Illinois and that current net inflow from the 5,000 km2 drainage basin
western Indiana, primarily due to better simulation of low would have to increase by about a factor of 15 to maintain
flows by SWAT. Nasr et al. (2007) found that HSPF predicted typical Late Pleistocene lake levels. Additional analyses of
mean daily discharge most accurately, while SWAT simu‐ Lake Estancia were performed by Menking et al. (2004) for
lated daily total phosphorus loads the best, in a comparison the Last Glacial Maximum period. SWAT was interfaced
of three models for three Irish watersheds that ranged in size with a 3‐D lagoon model by Plus et al. (2006) to determine
from 15 to 96 km2. El‐Nasr et al. (2005) found that both nitrogen loads from a 280 km2 drainage area into the Thau
SWAT and the MIKE‐SHE model (Refsgaard and Storm, Lagoon, which lies along the south coast of France. The main
1995) simulated the hydrology of Belgium's Jeker River ba‐ annual nitrogen load was estimated with SWAT to be 117 t
sin in an acceptable way. However, MIKE‐SHE predicted the year -1; chlorophyll a concentrations, phytoplankton produc‐
overall variation of river flow slightly better. tion, and related analyses were performed with the lagoon
Srinivasan et al. (2005) found that SWAT estimated flow model. Galbiati et al. (2006) interfaced SWAT with
more accurately than the Soil Moisture Distribution and QUAL2E, MODFLOW, and another model to create the Inte‐
Routing (SMDR) model (Cornell, 2003) for 39.5 ha FD‐36 grated Surface and Subsurface model (ISSm). They found
experimental watershed in east central Pennsylvania, and that the system accurately predicted water and nutrient inter‐
that SWAT was also more accurate on a seasonal basis. SWAT actions between the stream system and aquifer, groundwater
estimates were also found to be similar to measured dissolved dynamics, and surface water and nutrient fluxes at the wa‐
and total P for the same watershed, and 73% of the 22 fields tershed outlet for the 20 km2 Bonello coastal watershed in
in the watershed were categorized similarly on the basis of northern Italy.
the SWAT analysis as compared to the Pennsylvania P index
SWAT with Environmental Models or Genetic Algorithms
(Veith et al., 2005). Grizzetti et al. (2005) reported that both
for BMP Analyses
SWAT and a statistical approach based on the SPARROW
Renschler and Lee (2005) linked SWAT with the Water
model (Smith et al., 1997) resulted in similar total oxidized
nitrogen loads for two monitoring sites within the 1,380 km2 Erosion Prediction Project (WEPP) model (Ascough et al.,
1997) to evaluate both short‐ and long‐term assessments, for
Great Ouse watershed in the U.K. They also state that the sta‐
pre‐ and post‐implementation, of grassed waterways and
tistical reliability of the two approaches was similar, and that
the statistical model should be viewed primarily as a screen‐ field borders for three experimental watersheds ranging in
size from 0.66 to 5.11 ha. SWAT was linked directly to the
ing tool while SWAT is more useful for scenarios. Srivastava
Geospatial Interface for WEPP (GeoWEPP), which facili‐
et al. (2006) found that an artificial neural network (ANN)
model was more accurate than SWAT for streamflow simula‐ tated injection of WEPP output as point sources into SWAT.
The long‐term assessment results were similar to SWAT‐only
tions of a small watershed in southeast Pennsylvania.
evaluations, but the short‐term results were not. Cerucci and
Conrad (2003) determined the optimal riparian buffer config‐
INTERFACES OF SWAT WITH OTHER MODELS urations for 31 subwatersheds in the 37 km2 Town Brook wa‐
Innovative applications have been performed by interfac‐
tershed in south central New York, by using a binary
ing SWAT with other environmental and/or economic mod‐
optimization approach and interfacing SWAT with the
els. These interfaces have expanded the range of scenarios Riparian Ecosystem Model (REMM) (Lowrance et al.,
that can be analyzed and allowed for more in‐depth assess‐
2000). They determined the marginal utility of buffer widths
ments of questions that cannot be considered with SWAT by
and the most affordable parcels in which to establish riparian
itself, such as groundwater withdrawal impacts or the costs buffers. Pohlert et al. (2006) describe SWAT‐N, which was
incurred from different choices of management practices.
created by extending the original SWAT2000 nitrogen
SWAT with MODFLOW and/or Surface Water Models cycling routine primarily with algorithms from the
Sophocleus et al. (1999) describe an interface between Denitrification‐Decomposition (DNDC) model (Li et al.,
SWAT and the MODFLOW groundwater model (McDonald 1992). They state that SWAT‐N was able to replicate nitrogen
and Harbaugh, 1988) called SWATMOD, which they used to cycling and loss processes more accurately than SWAT.
evaluate water rights and withdrawal rate management sce‐ Muleta and Nicklow (2005a) interfaced SWAT with a ge‐
narios on stream and aquifer responses for the Rattlesnake netic algorithm and a multiobjective evolutionary algorithm
Creek watershed in south central Kansas. The system was to perform both single and multiobjective evaluations for the
used by Sophocleus and Perkins (2000) to investigate irriga‐ 130 km2 Big Creek watershed in southern Illinois. They
tion effects on streamflow and groundwater levels in the low‐ found that conversion of 10% of the HRUs into conservation
er Republican River watershed in north central Kansas and on programs (cropping system/tillage practice BMPs), within a
streamflow and groundwater declines within the Rattlesnake maximum of 50 genetic algorithm generations, would result
Creek watershed. Perkins and Sophocleous (1999) describe in reduced sediment yield of 19%. Gitau et al. (2004) inter‐
drought impact analyses with the same system. SWAT was faced baseline P estimates from SWAT with a genetic algo‐
coupled with MODFLOW to study for the 12 km2 Coët‐Dan rithm and a BMP tool containing site‐specific BMP
watershed in Brittany, France (Conan et al., 2003a). Accurate effectiveness estimates to determine the optimal on‐farm

1238 TRANSACTIONS OF THE ASABE

placement of BMPs so that P losses and costs were both mini‐ different policies were demonstrated by Attwood et al.
mized. The two most efficient scenarios met the target of re‐ (2000) by showing economic and environmental impacts at
ducing dissolved P loss by at least 60%, with corresponding the U.S. national scale and for Texas by linking SWAT with
farm‐level cost increases of $1,430 and $1,683, respectively, an agricultural sector model. Volk et al. (2007) and Turpin et
relative to the baseline. SWAT was interfaced with an eco‐ al. (2005) describe respective modeling systems that include
nomic model, a BMP tool, and a genetic algorithm by Arabi interfaces between SWAT, an economic model, and other
et al. (2006a) to determine optimal placement for the Dreis‐ models and data to simulate different watershed scales and
bach and Smith Fry watersheds in Indiana. The optimization conditions in European watersheds.
approach was found to be three times more cost‐effective as
SWAT with Ecological and Other Models
compared to environmental targeting strategies.
Weber et al. (2001) interfaced SWAT with the ecological
SWAT with Economic and/or Environmental Models model ELLA and the Proland economic model to investigate
A farm economic model was interfaced with the Agricul‐ the streamflow and habitat impacts of a “grassland incentive
tural Policy Extender (APEX) model (Williams and Izaur‐ scenario” that resulted in grassland area increasing from 21%
ralde, 2006) and SWAT to simulated the economic and to 40%, and forest area declining by almost 70%, within the
environmental impacts of manure management scenarios 59.8 km2 Aar watershed in Germany. SWAT‐predicted
and other BMPs for the 932.5 km2 upper North Bosque River streamflow increased while Skylark bird habitat decreased in
and 1,279 km2 Lake Fork Reservoir watersheds in Texas and response to the scenario. Fohrer et al. (2002) used SWAT‐G,
the 162.2 km2 upper Maquoketa River watershed in Iowa the YELL ecological model, and the Proland to assess the ef‐
(Gassman et al., 2002). The economic and environmental im‐ fects of land use changes and associated hydrologic impacts
pacts of several manure application rate scenarios are de‐ on habitat suitability for the Yellowhammer bird species. The
scribed for each watershed, as well as for manure haul‐off, authors report effects of four average field size scenarios (0.5,
intensive rotational grazing, and reduced fertilizer scenarios 0.75, 1.0, and 2.0 ha) on land use, bird nest distribution and
that were simulated for the upper North Bosque River wa‐ habitat, labor and agricultural value, and hydrological re‐
tershed, Lake Fork Reservoir watershed, and upper Maquo‐ sponse. SWAT is also being used to simulate crop growth,
keta River watershed, respectively. Osei et al. (2003) report hydrologic balance, soil erosion, and other environmental re‐
additional stocking density scenario results for pasture‐based sponses by Christiansen and Altaweel (2006) within the EN‐
dairy productions in the Lake Fork Reservoir watershed. KIMDU modeling framework (named after the ancient
They concluded that appropriate pasture nutrient manage‐ Sumerian god of agriculture and irrigation), which is being
ment, including stocking density adjustments and more effi‐ used to study the natural and societal aspects of Bronze Age
cient application of commercial fertilizer, could lead to Mesopotamian cultures.
significant reductions in nutrient losses in the Lake Fork Res‐
ervoir watershed. Gassman et al. (2006) further assessed the
impacts of seven individual BMPs and four BMP combina‐ SWAT STRENGTHS, WEAKNESSES, AND
tions for upper Maquoketa River watershed. Terraces were
predicted to be very effective in reducing sediment and or‐ RESEARCH NEEDS
ganic nutrient losses but were also the most expensive prac‐ The worldwide application of SWAT reveals that it is a
tice, while no‐till or contouring in combination with reduced versatile model that can be used to integrate multiple envi‐
fertilizer rates were predicted to result in reductions of all ronmental processes, which support more effective wa‐
pollutant indictors and also positive net returns. tershed management and the development of better‐informed
Lemberg et al. (2002) evaluated the economic impacts of policy decisions. The model will continue to evolve as users
brush control in the Frio River basin in south central Texas determine needed improvements that: (1) will enable more
using SWAT, the Phytomass Growth Simulator (PHY‐ accurate simulation of currently supported processes, (2) in‐
GROW) model (Rowan, 1995), and two economic models. corporate advancements in scientific knowledge, or (3) pro‐
It was determined that subsidies on brush control would not vide new functionality that will expand the SWAT simulation
be worthwhile. Economic evaluations of riparian buffer domain. This process is aided by the open‐source status of the
benefits in regards to reducing atrazine concentration and SWAT code and ongoing encouragement of collaborating
other factors were performed by Qiu and Prato (1998) using scientists to pursue needed model development, as demon‐
SWAT, a budget generator, and an economic model for the strated by a forthcoming set of papers in Hydrological
77.4 km2 Goodwater Creek watershed in north central Mis‐ Sciences Journal describing various SWAT research needs
souri (riparian buffers were not directly simulated). The im‐ that were identified at the 2006 Model Developer's Work‐
plementation of riparian buffers was found to result in shop held in Potsdam, Germany. The model has also been in‐
substantial net economic return and savings in government cluded in the Collaborative Software Development
costs, due to reduced CRP rental payments. Qiu (2005) used Laboratory that facilitates development by multiple scien‐
a similar approach for the same watershed to evaluate the tists (CoLab, 2006).
economic and environmental impacts of five different alter‐ The foundational strength of SWAT is the combination of
native scenarios. SWAT was interfaced with a data envelope upland and channel processes that are incorporated into one
analysis linear programming model by Whittaker et al. simulation package. However, every one of these processes
(2003) to determine which of two policies would be most ef‐ is a simplification of reality and thus subject to the need for
fective in reducing N losses to streams in the 259,000 km2 improvement. To some degree, the strengths that facilitate
Columbia Plateau region in the northwest U.S. The analysis widespread use of SWAT also represent weaknesses that need
indicated that a 300% tax on N fertilizer would be more effi‐ further refinement, such as simplified representations of
cient than a mandated 25% reduction in N use. Evaluation of HRUs. There are also problems in depicting some processes

Vol. 50(4): 1211-1250 1239

accurately due to a lack of sufficient monitoring data, inade‐ HYDROLOGIC RESPONSE UNITS (HRUS)
quate data needed to characterize input parameters, or insuf‐ The incorporation of nonspatial HRUs in SWAT has sup‐
ficient scientific understanding. The strengths and ported adaptation of the model to virtually any watershed,
weaknesses of five components are discussed here in more ranging in size from field plots to entire river basins. The fact
detail, including possible courses of action for improving that the HRUs are not landscape dependent has kept the mod‐
current routines in the model. The discussion is framed to el simple while allowing soil and land use heterogeneity to
some degree from the perspective of emerging applications, be accounted for within each subwatershed. At the same
e.g., bacteria die‐off and transport. Additional research needs time, the nonspatial aspect of the HRUs is a key weakness of
are also briefly listed for other components, again in the con‐ the model. This approach ignores flow and pollutant routing
text of emerging application trends where applicable. within a subwatershed, thus treating the impact of pollutant
losses identically from all landscape positions within a sub‐
HYDROLOGIC INTERFACE watershed. Thus, potential pollutant attenuation between the
The use of the NRCS curve number method in SWAT has source area and a stream is also ignored, as discussed by Bry‐
provided a relatively easy way of adapting the model to a ant et al. (2006) for phosphorus movement. Explicit spatial
wide variety of hydrologic conditions. The technique has representation of riparian buffer zones, wetlands, and other
proved successful for many applications, as evidenced by the BMPs is also not possible with the current SWAT HRU ap‐
results reported in this study. However, the embrace of the proach, as well as the ability to account for targeted place‐
method in SWAT and similar models has proved controver‐ ment of grassland or other land use within a given
sial due to the empirical nature of the approach, lack of com‐ subwatershed. Incorporation of greater spatial detail into
plete historical documentation, poor results obtained for SWAT is being explored with the initial focus on developing
some conditions, inadequate representation of “critical routing capabilities between distinct spatially defined land‐
source areas” that generate pollutant loss (which can occur scapes (Volk et al., 2005), which could be further subdivided
even after satisfactory hydrologic calibration of the model), into HRUs.
and other factors (e.g., Ponce and Hawkins, 1996; Agnew et
al., 2006; Bryant et al., 2006; Garen and Moore, 2005). SIMULATION OF BMPS
The Green‐Ampt method provides an alternative option in A key strength of SWAT is a flexible framework that al‐
SWAT, which was found by Rawls and Brakenseik (1986) to lows the simulation of a wide variety of conservation practic‐
be more accurate than the curve number method and also to es and other BMPs, such as fertilizer and manure application
account for the effects of management practices on soil prop‐ rate and timing, cover crops (perennial grasses), filter strips,
erties in a more rational manner. However, the previously dis‐ conservation tillage, irrigation management, flood‐preven-
cussed King et al. (1999) and Kannan et al. (2007b) SWAT tion structures, grassed waterways, and wetlands. The major‐
applications did not find any advantage to using the Green‐ ity of conservation practices can be simulated in SWAT with
Ampt approach, as compared to the curve number method. straightforward parameter changes. Arabi et al. (2007a) have
These results lend support to the viewpoint expressed by proposed standardized approaches for simulating specific
Ponce and Hawkins (1996) that alternative point infiltration conservation practices in the model, including adjustment of
techniques, including the Green‐Ampt method, have not the parameters listed in table 4. Filter strips and field borders
shown a clear superiority to the curve number method. can be simulated at the HRU level, based on empirical func‐
Improved SWAT hydrologic predictions could potentially tions that account for filter strip trapping effects of bacteria
be obtained through modifications in the curve number meth‐ or sediment, nutrients, and pesticides (which are invoked
odology and/or incorporation of more complex routines. Bo‐ when the filter strip width parameter is set input to the mod‐
rah et al. (2007) propose inserting a combined curve number‐ el). However, assessments of targeted filter strip placements
kinematic wave methodology used in DWSM into SWAT, within a watershed are limited, due to the lack of HRU spatial
which was found to result in improved simulation of daily definition in SWAT. There are also further limitations in sim‐
runoff volumes for the 8,400 km2 Little Wabash River wa‐ ulating grassed waterways, due to the fact that channel rout‐
tershed in Illinois. Bryant et al. (2006) propose modifications ing is not simulated at the HRU level. Arabi et al. (2007a)
of the curve number initial abstraction term, as a function of proposed simulating grassed waterways by modifying sub‐
soil physical characteristics and management practices, that watershed channel parameters, as shown in table 4. However,
could result in more accurate simulation of extreme (low and this approach is usually only viable for relatively small wa‐
high) runoff events. Model and/or data input modifications tersheds, such as the example they present in their study.
would be needed to address phenomena such as variable Wetlands can be simulated in SWAT on the basis of one
source area (VSA) saturated excess runoff, which dominants wetland per subwatershed, which is assumed to capture dis‐
runoff in some regions including the northeast U.S., where charge and pollutant loads from a user‐specified percentage
downslope VSA saturated discharge often occurs due to sub‐ of the overall subwatershed. The ability to site wetlands with
surface interflow over relatively impermeable material (Ag‐ more spatial accuracy within a subwatershed would clearly
new et al., 2006; Walter et al., 2000). Steenhuis (2007) has provide improvements over the current SWAT wetland simu‐
developed a method of reclassifying soil types and associated lation approach, although this can potentially be overcome
curve numbers that provides a more accurate accounting of for some applications by subdividing a watershed into small‐
VSA‐driven runoff and pollutant loss for a small watershed er subwatersheds.
in New York. The modified SWAT model described by The lack of spatial detail in SWAT also hinders simulation
Watson et al. (2005), which accounts for VSA-dominated of riparian buffer zones and other conservation buffers,
hydrology in southwest Victoria, Australia, by incorporating which again need to be spatially defined at the landscape or
a saturated excess runoff routine in SWAT, may also provide HRU level in order to correctly account for upslope pollutant
useful insights. source areas and the pollutant mitigation impacts of the buff‐

1240 TRANSACTIONS OF THE ASABE

ers. The riparian and wetland processes recently incorporated pension as function of sediment particles rather than just dis‐
into the SWIM model (Hatterman et al., 2006) may prove charge.
useful for improving current approaches used in SWAT.
IN‐STREAM KINETIC FUNCTIONS
BACTERIA LIFE CYCLE AND TRANSPORT The ability to simulate in‐stream water quality dynamics
Benham et al. (2006) state that SWAT is one of two prima‐ is a definite strength of SWAT. However, Horn et al. (2004)
ry models used for watershed‐scale bacteria fate and trans‐ point out that very few SWAT‐related studies discuss whether
port assessments in the U.S. The strengths of the SWAT the QUAL2E‐based in‐stream kinetic functions were used or
bacteria component include: (1) simultaneous assessment of not. Santhi et al. (2001a) opted to not use the in‐stream func‐
fecal coliform (as an indicator pathogen) and a more persis‐ tions for their SWAT analysis of the Bosque River in central
tent second pathogen that possesses different growth/die‐off Texas because the functions do not account for periphyton
characteristics, (2) different rate constants that can be set for (attached algae), which dominates phosphorus‐limited sys‐
soluble versus sediment‐bound bacteria, and (3) the ability to tems including the Bosque River. This is a common limita‐
account for multiple point and/or nonpoint bacteria sources tion of most water quality models with in‐stream
such as land‐applied livestock and poultry manure, wildlife components, which focus instead on just suspended algae.
contributions, and human sources such as septic tanks. Jamie‐ Migliaccio et al. (2007) performed parallel SWAT analyses
son et al. (2004) further point out that SWAT is the only model of total P and nitrate (including nitrite) movement for the
that currently simulates partitioning of bacteria between ad‐ 60km 2 War Eagle Creek watershed in northwest Arkansas
sorbed and non‐adsorbed fractions; however, they also state by: (1) loosely coupling SWAT with QUAL2E (with the
that reliable partitioning data is currently not available. SWAT in‐stream component turned off), and (2) executing
Bacteria die‐off is simulated in SWAT on the basis of a first‐ SWAT by itself with and without the in‐stream functions acti‐
order kinetic function (Neitsch et al., 2005a), as a function of vated. They found no statistical difference in the results gen‐
time and temperature. However, Benham et al. (2006), Ja‐ erated between the SWAT‐QUAL2E interface approach
mieson et al. (2004), and Pachepsky et al. (2006) all cite sev‐ versus the stand‐alone SWAT approach, or between the two
eral studies that show that other factors such as moisture stand‐alone SWAT simulations. They concluded that further
content, pH, nutrients, and soil type can influence die‐off testing and refinement of the SWAT in‐stream algorithms are
rates. Leaching of bacteria is also simulated in SWAT, al‐ warranted, which is similar to the views expressed by Horn
though all leached bacteria are ultimately assumed to die off. et al. (2004). Further investigation is also needed to deter‐
This conflicts with some actual observations in which patho‐ mine if the QUAL2E modifications made in ESWAT should
gen movement has been observed in subsurface flow (Pa‐ be ported to SWAT, which are described by Van Griensven
chepsky et al., 2006; Benham et al., 2006), which is and Bauwens (2003, 2005).
especially prevalent in tile‐drained areas (Jamieson et al.,
2004). Benham et al. (2006), Jamieson et al. (2004), and Pa‐ ADDITIONAL RESEARCH NEEDS
chepsky et al. (2006) list a number of research needs and S Development of concentrated animal feeding opera‐
modeling improvements needed to perform more accurate tion and related manure application routines, that sup‐
bacteria transport simulations with SWAT and other models port simulation of surface and integrated manure
including: (1) more accurate characterization of bacteria application techniques and their influence on nutrient
sources, (2) development of bacteria life cycle equations that fractionation, distribution in runoff and soil, and sedi‐
account for different phases of die‐off and the influence of ment loads. Current development is focused on a ma‐
multiple factors on bacteria die‐off rates, (3) accounting of nure cover layer.
subsurface flow bacteria movement including transport via S All aspects of stream routing need further testing and
tile drains, and (4) depiction of bacteria deposition and resus‐ refinement, including the QUAL2E routines as dis‐
cussed above.

Table 4. Proposed key parameters to adjust for accounting of different conservation practice effects in SWAT (source: Arabi et al., 2007a).
Channel Manning SCS
Channel Channel Manning Channel Filter Hillside N for Runoff USLE USLE
Channel Channel Erodibility Cover Roughness Slope Strip Slope Overland Curve C P
Conservation Practice Depth Width Factor Factor Coeff. Segment Width[a] Length Flow Number Factor Factor
Contouring X X
Field border X
Filter strips X
Grade stabilization structures X X
Grassed waterways X X X X
Lined waterways X X X X
Parallel terraces X X X
Residue management[b] X X X
Stream channel stabilization X X X X
Strip cropping X X X X
[a] Setting a filter strip width triggers one of two filter strip trapping efficiency functions (one for bacteria and the other for sediment, pesticides, and nutrients)
that account for the effect of filter strip removal of pollutants.
[b] Soil incorporation of residue by tillage implements is also a key aspect of simulated residue management in SWAT.

Vol. 50(4): 1211-1250 1241

 S Improved stream channel degradation and sediment de‐ robust tool that can be used to simulate a variety of watershed
position routines are needed to better describe sediment problems. The process of configuring SWAT for a given wa‐
transport, and to account for nutrient loads associated tershed has also been greatly facilitated by the development
with sediment movement, as discussed by Jha et al. of GIS‐based interfaces, which provide a straightforward
(2004a). Channel sediment routing could be improved by means of translating digital land use, topographic, and soil
accounting for sediment size effects, with separate algo‐ data into model inputs. It can be expected that additional sup‐
rithms for the wash and bed loads. Improved flood plain port tools will be created in the future to facilitate various ap‐
deposition algorithms are needed, and a stream bank ero‐ plications of SWAT. The ability of SWAT to replicate
sion routine should be incorporated. hydrologic and/or pollutant loads at a variety of spatial scales
S SWAT currently assumes that soil carbon contents are on an annual or monthly basis has been confirmed in numer‐
static. This approach will be replaced by an updated ous studies. However, the model performance has been inad‐
carbon cycling submodel that provides more realistic equate in some studies, especially when comparisons of
accounting of carbon cycling processes. predicted output were made with time series of measured dai‐
S Improvements to the nitrogen cycling routines should ly flow and/or pollutant loss data. These weaker results un‐
be investigated based on the suggestions given by Bo‐ derscore the need for continued testing of the model,
rah et al. (2006). Other aspects of the nitrogen cycling including more thorough uncertainty analyses, and ongoing
process should also be reviewed and updated if needed, improvement of model routines. Some users have addressed
including current assumptions of plant nitrogen uptake. weaknesses in SWAT by component modifications, which
Soil phosphorus cycling improvements have been initi‐ support more accurate simulation of specific processes or re‐
ated and will continue. The ability to simulate leaching gions, or by interfacing SWAT with other models. Both of
of soil phosphorus through the soil profile, and in later‐ these trends are expected to continue. The SWAT model will
al, groundwater, and tile flows, has recently been incor‐ continue to evolve in response to the needs of the ever‐
porated into the model. increasing worldwide user community and to provide im‐
S Expansion of the plant parameter database is needed, proved simulation accuracy of key processes. A major
as pointed out by Heuvelmans et al. (2005), to support challenge of the ongoing evolution of the model will be meet‐
a greater range of vegetation scenarios that can be sim‐ ing the desire for additional spatial complexity while main‐
ulated in the model. In general, more extensive testing taining ease of model use. This goal will be kept in focus as
of the crop growth component is needed, including re‐ the model continues to develop in the future.
visions to the crop parameters where needed.
S Modifications have been initiated by McKeown et al. ACKNOWLEDGEMENTS
(2005) in a version of the model called SWAT2000‐C Partial support for this study was provided by the U.S. EPA
to more accurately simulate the hydrologic balance and Office of Policy, Economics, and Innovation and Office of
other aspects of Canadian boreal forest systems includ‐ Wastewater Management under cooperating agreement
ing: (1) incorporation of a surface litter layer into the number CR 820374‐02‐7 and the Cooperative State Re‐
soil profile, (2) accounting of water storage and release search, Education, and Extension Service of the USDA, Proj‐
by wetlands, and (3) improved simulation of spring ect No. NCX‐186‐5‐04‐130‐1, in the Agricultural Research
thaw generated runoff. These improvements will ulti‐ Program, North Carolina Agricultural and Technical State
mately be grafted into SWAT2005. University. The opinions expressed in this document remain
S Advancements have been made in simulating subsur‐ the sole responsibility of the authors and do not necessarily
face tile flows and nitrate losses (Du et al., 2005, 2006). express the position of the U.S. EPA or the USDA.
Current research is focused on incorporating a second
option, based on the DRAINMOD (Skaggs, 1982) ap‐
proach, that includes the effects of tile drain spacing
and shallow water table depth. Future research should
REFERENCES
Abbaspour, K. C., J. Yang, I. Maximov, R. Siber, K. Bogner, J.
also be focused on controlled drainage BMPs.
Mieleitner, J. Zobrist, and R. Srinivasan. 2007. Modelling
S Routines for automated sensitivity, calibration, and in‐ hydrology and water quality in the pre‐alpine/alpine Thur
put uncertainty analysis have been added to SWAT watershed using SWAT. J. Hydrol. 333(2‐4): 413‐430.
(vanGriensven and Bauwens, 2003). These routines Afinowicz, J. D., C. L. Munster, and B. P. Wilcox. 2005. Modeling
are currently being tested on several watersheds, in‐ effects of brush management on the rangeland water budget:
cluding accounting of uncertainty encountered in mea‐ Edwards Plateau, Texas. J. American Water Resour. Assoc.
sured water quality data, as discussed by Harmel et al. 41(1): 181‐193.
(2006). Agnew, L. J., S. Lyon, P. Gérard‐Marchant, V. B. Collins, A. J.
S The effects of atmospheric CO2 on plant growth need Lembo, T. S. Steenhuis, and M. T. Walter. 2006. Identifying
to be revised to account for varying stomatal conduc‐ hydrologically sensitive areas: Bridging the gap between science
and application. J. Environ. Mgmt. 78(1): 63‐76.
tance and leaf area responses as a function of plant spe‐
Arabi, M., R. S. Govindaraju, and M. M. Hantush. 2006a.
cies, similar to the procedure developed for SWAT‐G Cost‐effective allocation of watershed management practices
by Eckhardt et al. (2003). using a genetic algorithm. Water Resour. Res. 42.W10429,
doi:10,1029/2006WR004931.
Arabi, M., J. Frankenberger, B. Engel, and J. Arnold. 2007a.
Representation of agricultural management practices with
CONCLUSIONS SWAT. Hydrol. Process. (submitted).
The wide range of SWAT applications that have been de‐ Arabi, M., R. S. Govindaraju, and M. M. Hantush. 2007b. A
scribed here underscores that the model is a very flexible and probabilistic approach for analysis of uncertainty in evaluation

1242 TRANSACTIONS OF THE ASABE

 of watershed management practices. J. Hydrol. 333(2‐4): Benham, B. L., C. Baffaut, R. W. Zeckoski, K. R. Mankin, Y. A.
459‐471. Pachepsky, A. M. Sadeghi, K. M. Brannan, M. L. Soupir, and
Arabi, M., R. S. Govindaraju, M. M. Hantush, and B. A. Engel. M. J. Habersack. 2006. Modeling bacteria fate and transport in
2006b. Role of watershed subdivision on modeling the watershed models to support TMDLs. Trans. ASABE 49(4):
effectiveness of best management practices with SWAT. J. 987‐1002.
American Water Resour. Assoc. 42(2): 513‐528. Bicknell, B. R., J. C. Imhoff, A. S. Donigian, and R. C. Johanson.
Arnold, J. G., and J. R. Williams. 1987. Validation of SWRRB: 1997. Hydrological simulation program - FORTRAN (HSPF):
Simulator for water resources in rural basins. J. Water Resour. User's manual for release 11. EPA‐600/R‐97/080. Athens, Ga.:
Plan. Manage. ASCE 113(2): 243‐256. U.S. Environmental Protection Agency.
Arnold, J. G., and P. M. Allen. 1993. A comprehensive Bingner, R. L. 1996. Runoff simulated from Goodwin Creek
surface‐ground water flow model. J. Hydrol. 142(1‐4): 47‐69. watershed using SWAT. Trans. ASAE 39(1): 85‐90.
Arnold, J. G., and P. M. Allen. 1996. Estimating hydrologic budgets Bingner, R. L., J. Garbrecht, J. G. Arnold, and R. Srinivasan. 1997.
for three Illinois watersheds. J. Hydrol. 176(1‐4): 57‐77. Effect of watershed subdivision on simulated runoff and fine
Arnold, J. G., and P. M. Allen. 1999. Automated methods for sediment yield. Trans. ASAE 40(5): 1329‐1335.
estimating baseflow and groundwater recharge from streamflow Boorman, D. B. 2003. Climate, Hydrochemistry, and Economics of
records. J. American Water Resour. Assoc. 35(2): 411‐424. Surface‐water Systems (CHESS): Adding a European dimension
Arnold, J. G., and N. Fohrer. 2005. SWAT2000: Current capabilities to the catchment modelling experience developed under LOIS.
and research opportunities in applied watershed modeling. Sci. Total Environ. 314‐316: 411‐437.
Hydrol. Process. 19(3): 563‐572. Borah, D. K., and M. Bera. 2003. Watershed‐scale hydrologic and
Arnold, J. G., P. M. Allen, R. S. Muttiah, and G. Bernhardt. 1995a. nonpoint‐source pollution models: Review of mathematical
Automated base flow separation and recession analysis bases. Trans. ASAE 46(6): 1553‐1566.
techniques. Groundwater 33(6): 1010‐1018. Borah, D. K., and M. Bera. 2004. Watershed‐scale hydrologic and
Arnold, J. G., J. R. Williams, and D. R. Maidment. 1995b. nonpoint‐source pollution models: Review of applications.
Continuous‐time water and sediment‐routing model for large Trans. ASAE 47(3): 789‐803.
basins. J. Hydrol. Eng. ASCE 121(2): 171‐183. Borah, D. K., M. Bera, M. and R. Xia. 2004. Storm event flow and
Arnold, J. G., R. Srinivasan, R. S. Muttiah, and J. R. Williams. sediment simulations in agricultural watersheds using DWSM.
1998. Large‐area hydrologic modeling and assessment: Part I. Trans. ASAE 47(5): 1539‐1559.
Model development. J. American Water Resour. Assoc. 34(1): Borah, D. K., G. Yagow, A. Saleh, P. L. Barnes, W. Rosenthal, E. C.
73‐89. Krug, and L. M. Hauck. 2006. Sediment and nutrient modeling
Arnold, J. G., R. Srinivasan, R. S. Muttiah, P. M. Allen, and C. for TMDL development and implementation. Trans. ASABE
Walker. 1999a. Continental‐scale simulation of the hydrologic 49(4): 967‐986.
balance. J. American Water Resour. Assoc. 35(5): 1037‐1052. Borah, D. K., J. G. Arnold, M. Bera, E. C. Krug, and X. Z. Liang.
Arnold, J. G., R. Srinivasan, T. S. Ramanarayanan, and M. Di 2007. Storm event and continuous hydrologic modeling for
Luzio. 1999b. Water resources of the Texas gulf basin. Water comprehensive and efficient watershed simulations. J. Hydrol.
Sci. Tech. 39(3): 121‐133. Eng. (in press).
Arnold, J. G., R. S. Muttiah, R. Srinivasan, and P. M. Allen. 2000. Bosch, D. D., J. M. Sheridan, H. L. Batten, and J. G. Arnold. 2004.
Regional estimation of base flow and groundwater recharge in Evaluation of the SWAT model on a coastal plain agricultural
the upper Mississippi basin. J. Hydrol. 227(1‐4): 21‐40. watershed. Trans. ASAE 47(5): 1493‐1506.
Arnold, J. G., P. M. Allen, and D. Morgan. 2001. Hydrologic model Bouraoui, F., L. Galbiati, and G. Bidoglio. 2002. Climate change
for design of constructed wetlands. Wetlands 21(2): 167‐178. impacts on nutrient loads in the Yorkshire Ouse catchment (UK).
Arnold, J. G., K. N. Potter, K. W. King, and P. M. Allen. 2005. Hydrol. Earth System Sci. 6(2): 197‐209.
Estimation of soil cracking and the effect on surface runoff in a Bouraoui, F., B. Grizzetti, K. Granlund, S. Rekolainen, and G.
Texas Blackland Prairie watershed. Hydrol. Process. 19(3): Bidoglio. 2004. Impact of climate change on the water cycled
589‐603. and nutrient losses in a Finnish catchment. Clim. Change
Ascough II, J. C., C. Baffaut, M. A. Nearing, and B. Y. Liu. 1997. 66(1‐2): 109‐126.
The WEPP watershed model: I. Hydrology and erosion. Trans. Bouraoui, F., S. Benabdallah, A. Jrad, and G. Bidoglio. 2005.
ASAE 40(4): 921‐933. Application of the SWAT model on the Medjerda River basin
Attwood, J. D., B. McCarl, C. C. Chen, B. R. Eddleman, B. Nayda, (Tunisia). Phys. Chem. Earth 30(8‐10): 497‐507.
and R. Srinivasan. 2000. Assessing regional impacts of change: Bracmort, K. S., M. Arabi, J. R. Frankenberger, B. A. Engel, and J.
Linking economic and environmental models. Agric. Syst. 63(3): G. Arnold. 2006. Modeling long‐term water quality impact of
147‐159. structural BMPs. Trans. ASABE 49(2): 367‐374.
Bärlund, I., T. Kirkkala, O. Malve, and J. Kämäri. 2007. Assessing Brown, L. C., and T. O. Barnwell, Jr. 1987. The enhanced water
the SWAT model performance in the evaluation of management quality models QUAL2E and QUAL2E‐UNCAS:
actions for the implementation of the Water Framework Documentation and user manual. EPA document
Directive in a Finnish catchment. Environ. Model. Soft. 22(5): EPA/600/3‐87/007. Athens, Ga.: USEPA.
719‐724. Bryant, R. B., W. J. Gburek, T. L. Veith, and W. D. Hively. 2006.
Behera, S., and R. K. Panda. 2006. Evaluation of management Perspectives on the potential for hydropedology to improve
alternatives for an agricultural watershed in a sub‐humid watershed modeling of phosphorus loss. Geoderma 131(3‐4):
subtropical region using a physical process model. Agric. 299‐307.
Ecosys. Environ. 113(1‐4): 62‐72. CARD. 2007. CARD interactive software programs. Ames, Iowa:
Benaman, J., and C. A. Shoemaker. 2004. Methodology for Iowa State University, Center for Agricultural and Rural
analyzing ranges of uncertain model parameters and their impact Development. Available at: www.card.iastate.edu/environment/
on total maximum daily load processes. J. Environ. Eng. 130(6): interactive_programs.aspx. Accessed 12 February 2007.
648‐656. Cao, W., W. B. Bowden, T. Davie, and A. Fenemor. 2006.
Benaman, J., C. A. Shoemaker, and D. A. Haith. 2005. Calibration Multi‐variable and multi‐site calibration and validation of SWAT
and validation of Soil and Water Assessment Tool on an in a large mountainous catchment with high spatial variability.
agricultural watershed in upstate New York. J. Hydrol. Eng. Hydrol. Proc. 20(5): 1057‐1073.
10(5): 363‐374.

Vol. 50(4): 1211-1250 1243

Cerucci, M., and J. M. Conrad. 2003. The use of binary Di Luzio, M., and J. G. Arnold. 2004. Formulation of a hybrid
optimization and hydrologic models to form riparian buffers. J. calibration approach for a physically based distributed model
American Water Resour. Assoc. 39(5): 1167‐1180. with NEXRAD data input. J. Hydrol. 298(1‐4): 136‐154.
Chanasyk, D. S., E. Mapfumo, and W. Willms. 2003. Quantification Di Luzio, M., R. Srinivasan, and J. G. Arnold. 2002. Integration of
and simulation of surface runoff from fescue grassland watershed tools and SWAT model into BASINS. J. American
watersheds. Agric. Water Mgmt. 59(2): 137‐153. Water Resour. Assoc. 38(4): 1127‐1141.
Chaplot, V. 2005. Impact of DEM mesh size and soil map scale on Di Luzio, M., R. Srinivasan, and J. G. Arnold. 2004a. A
SWAT runoff, sediment, and NO3-N loads predictions. J. GIS‐coupled hydrological model system for the watershed
Hydrol. 312(1‐4): 207‐222. assessment of agricultural nonpoint and point sources of
Chaplot, V., A. Saleh, D. B. Jaynes, and J. Arnold. 2004. Predicting pollution. Trans. GIS 8(1): 113‐136.
water, sediment, and NO3-N loads under scenarios of land‐use Di Luzio, M., J. G. Arnold, and R. Srinivasan 2004b. Integration of
and management practices in a flat watershed Water Air Soil SSURGO maps and soil parameters within a geographic
Pollut. 154(1‐4): 271‐293. information system and nonpoint‐source pollution model
Chaplot, V., A. Saleh, and D. B. Jaynes. 2005. Effect of the system. J. Soil Water Cons. 59(4): 123‐133.
accuracy of spatial rainfall information on the modeling of Di Luzio, M., J. G. Arnold, and R. Srinivasan. 2005. Effect of GIS
water, sediment, and NO3-N loads at the watershed level. J. data quality on small watershed streamflow and sediment
Hydrol. 312(1‐4): 223‐234. simulations. Hydrol. Process. 19(3): 629‐650.
CEAP. 2007. Conservation Effects Assessment Project. Doherty, J. 2004. PEST: Model‐Independent Parameter Estimation
Washington, D.C.: USDA Natural Resources Conservation User Manual. 5th ed. Brisbane, Australia: Watermark Numerical
Service. Available at: www.nrcs.usda.gov/technical/NRI/ceap/. Computing. Available at: www.simulistics.com/documents/
Accessed 14 February 2007. pestman.pdf. Accessed 18 February 2007.
Chen, E., and D. S. Mackay. 2004. Effects of distribution‐based Du, B., J. G. Arnold, A. Saleh, and D. B. Jaynes. 2005.
parameter aggregation on a spatially distributed agricultural Development and application of SWAT to landscapes with tiles
nonpoint‐source pollution model. J. Hydrol. 295(1‐4): 211‐224. and potholes. Trans. ASAE 48(3): 1121‐1133.
Cheng, H., W. Ouyang, F. Hao, X. Ren, and S. Yang. 2006. The Du, B., A. Saleh, D. B. Jaynes, and J. G. Arnold. 2006. Evaluation
nonpoint‐source pollution in livestock‐breeding areas of the of SWAT in simulating nitrate nitrogen and atrazine fates in a
Heihe River basin in Yellow River. Stoch. Environ. Res. Risk watershed with tiles and potholes. Trans. ASABE 49(4):
Assess. doi:10.1007/s00477‐006‐0057‐2. 949‐959.
CHESS. 2001. Climate, hydrochemistry, and economics of EAWAG. 2005. Proc. 3rd International SWAT Conference. Zurich,
surface‐water systems. Available at: www.nwl.ac.uk/ih/www/ Switzerland: Swiss Federal Institute for Environmental Science
research/images/chessreport.pdf. Accessed 25 August 2006. and Technology. Available at: www.brc.tamus.edu/swat/
Christiansen, J. H., and M. Altaweel. 2006. Simulation of natural 3rdswatconf/SWAT%20Book%203rd%20Conference.pdf.
and social process interactions: An example from Bronze Age Accessed 14 February 2007.
Mesopotamia. Soc. Sci. Comp. Rev. 24(2): 209‐226. Eckhardt, K., and J. G. Arnold. 2001. Automatic calibration of a
Chu, T. W., and A. Shirmohammadi. 2004. Evaluation of the SWAT distributed catchment model. J. Hydrol. 251(1‐2): 103‐109.
model's hydrology component in the Piedmont physiographic Eckhardt, K., and U. Ulbrich. 2003. Potential impacts of climate
region of Maryland. Trans. ASAE 47(4): 1057‐1073. change on groundwater recharge and streamflow in a central
Chu, T. W., A. Shirmohammadi, H. Montas, and A. Sadeghi. 2004. European low mountain range. J. Hydrol. 284(1‐4): 244‐252.
Evaluation of the SWAT model's sediment and nutrient Eckhardt, K., S. Haverkamp, N. Fohrer, and H.‐G. Frede. 2002.
components in the Piedmont physiographic region of Maryland. SWAT‐G, a version of SWAT99.2 modified for application to
Trans. ASAE 47(5): 1523‐1538. low mountain range catchments. Phys. Chem. Earth 27(9‐10):
Coffey, M. E., S. R. Workman, J. L. Taraba, and A. W. Fogle. 2004. 641‐644.
Statistical procedures for evaluating daily and monthly Eckhardt, K., L. Breuer, and H.‐G. Frede. 2003. Parameter
hydrologic model predictions. Trans. ASAE 47(1): 59‐68. uncertainty and the significance of simulated land use change
CoLab. 2006. CoLab: Project Integration - Change Control - Life effects. J. Hydrol. 273(1‐4): 164‐176.
Cycle Management. Washington, D.C.: USDACollaborative Eckhardt, K., N. Fohrer, and H.‐G. Frede. 2005. Automatic model
Software Development Laboratory. Collaborative Software calibration. Hydrol. Process. 19(3): 651‐658.
Development Laboratory. Available at: colab.sc.egov.usda.gov/ El‐Nasr, A. J. G. Arnold, J. Feyen, and J. Berlamont. 2005.
cb/ sharedProjectsBrowser.do. Accessed 30 October 2006. Modelling the hydrology of a catchment using a distributed and
Conan, C., F. Bouraoui, N. Turpin, G. de Marsily, and G. Bidoglio. a semi‐distributed model. Hydrol. Process. 19(3): 573‐587.
2003a. Modeling flow and nitrate fate at catchment scale in EUROHARP. 2006. Towards European harmonised procedures for
Brittany (France). J. Environ. Qual. 32(6): 2026‐2032. quantification of nutrient losses from diffuse sources. Available
Conan, C., G. de Marsily, F. Bouraoui, and G. Bidoglio. 2003b. A at: euroharp.org/pd/pd/index.htm#5. Accessed 25 August 2006.
long‐term hydrological modelling of the upper Guadiana river FitzHugh, T. W., and D. S. Mackay. 2000. Impacts of input
basin (Spain). Phys. Chem. Earth 28(4‐5): 193‐200. parameter spatial aggregation on an agricultural nonpoint‐source
Cornell. 2003. SMDR: The soil moisture distribution and routing pollution model. J. Hydrol. 236(1‐2): 35‐53.
model. Documentation version 2.0. Ithaca, N.Y.: Cornell FitzHugh, T. W., and D. S. Mackay. 2001. Impact of subwatershed
University Department of Biological and Environmental partitioning on modeled source‐ and transport‐limited sediment
Engineering, Soil and Water Laboratory. Available at: yields in an agricultural nonpoint‐source pollution model. J.
soilandwater.bee.cornell.edu/Research/smdr/downloads/SMDR‐ Soil Water Cons. 56(2): 137‐143.
manual‐v200301.pdf. Accessed 11 February 2007. Fohrer, N., D. Möller, and N. Steiner. 2002. An interdisciplinary
Cotter, A. S., I. Chaubey, T. A. Costello, T. S. Soerens, and M. A. modelling approach to evaluate the effects of land use change.
Nelson. 2003. Water quality model output uncertainty as Phys. Chem. Earth 27(9‐10): 655‐662.
affected by spatial resolution of input data. J. American Water Fohrer, N., S. Haverkamp, and H.‐G. Frede. 2005. Assessment of
Res. Assoc. 39(4): 977‐986. the effects of land use patterns on hydrologic landscape
Deliberty, T. L., and D. R. Legates. 2003. Interannual and seasonal functions: Development of sustainable land use concepts for low
variability of modelled soil moisture in Oklahoma. Intl. J. mountain range areas. Hydrol. Process. 19(3): 659‐672.
Climatol. 23(9): 1057‐1086.

1244 TRANSACTIONS OF THE ASABE

Fontaine, T. A., J. F. Klassen, T. S. Cruickshank, and R. H. Grunwald, S., and C. Qi. 2006. GIS‐based water quality modeling
Hotchkiss. 2001. Hydrological response to climate change in the in the Sandusky watershed, Ohio, USA. J. American Water
Black Hills of South Dakota, USA. Hydrol. Sci. J. 46(1): 27‐40. Resour. Assoc. 42(4): 957‐973.
Fontaine, T. A., T. S. Cruickshank, J. G. Arnold, and R. H. Hao, F. H., X. S. Zhang, and Z. F. Yang. 2004. A distributed
Hotchkiss. 2002. Development of a snowfall‐snowmelt routine nonpoint‐source pollution model: Calibration and validation in
for mountainous terrain for the Soil and Water Assessment Tool the Yellow River basin. J. Environ. Sci. 16(4): 646‐650.
(SWAT). J. Hydrol. 262(1‐4): 209‐223. Hanratty, M. P., and H. G. Stefan. 1998. Simulating climate change
Francos, A., G. Bidoglio, L. Galbiati, F. Bouraoui, F. J. Elorza, S. effects in a Minnesota agricultural watershed. J. Environ. Qual.
Rekolainen, K. Manni, and K. Granlund. 2001. Hydrological 27(6): 1524‐1532.
and water quality modelling in a medium‐sized coastal basin. Hargreaves, G. L., G. H. Hargreaves, and J. P. Riley. 1985.
Phys. Chem. Earth (B) 26(1): 47‐52. Agricultural benefits for Senegal River basin. J. Irrig. Drain.
Francos, A., F. J. Elorza, F. Bouraoui, G. Bidoglio, and L. Galbiati. Eng. 108(3): 225‐230.
2003. Sensitivity analysis of distributed environmental Harmel, R. D., C. W. Richardson, and K. W. King. 2000.
simulation models: Understanding the model behaviour in Hydrologic response of a small watershed model to generated
hydrological studies at the catchment scale. Real. Eng. Syst. precipitation. Trans. ASAE 43(6): 1483‐1488.
Safe. 79(2): 205‐218. Harmel, R. D., R. J. Cooper, R. M. Slade, R. L. Haney, and J. G.
Galbiati, L., F. Bouraoui, F. J. Elorza, and G. Bidoglio. 2006. Arnold. 2006. Cumulative uncertainty in measured streamflow
Modeling diffuse pollution loading into a Mediterranean lagoon: and water quality data for small watersheds. Trans. ASABE
Development and application of an integrated surface‐subsurface 49(3): 689‐701.
model tool. Ecol. Model. 193(1‐2): 4‐18. Hatterman, F., V. Krysanova, F. Wechsung, and M. Wattenbach.
Garen, D. C., and D. S. Moore. 2005. Curve number hydrology in 2004. Integrating groundwater dynamics in regional
water quality modeling: Uses, abuses, and future directions. J. hydrological modelling. Environ. Model. Soft. 19(11):
American Water Resour. Assoc. 41(2): 377‐388. 1039‐1051.
Gassman, P. W., E. Osei, A. Saleh, and L. M. Hauck. 2002. Hatterman, F. F., V. Krysanova, A. Habeck, and A. Bronstert. 2006.
Application of an environmental and economic modeling system Integrating wetlands and riparian zones in river basin modeling.
for watershed assessments. J. American Water Resour. Assoc. Ecol. Model. 199(4): 379‐392.
38(2): 423‐438. Haverkamp, S., R. Srinivasan, H.‐G. Frede, and C. Santhi. 2002.
Gassman, P. W., E. Osei, A. Saleh, J. Rodecap, S. Norvell, and J. Subwatershed spatial analysis tool: Discretization of a
Williams. 2006. Alternative practices for sediment and nutrient distributed hydrologic model by statistical criteria. J. American
loss control on livestock farms in northeast Iowa. Agric. Ecosys. Water Resour. Assoc. 38(6): 1723‐1733.
Environ. 117(2‐3): 135‐144. Haverkamp, S., N. Fohrer, and H.‐G. Frede. 2005. Assessment of
Geza, M., and J. E. McCray. 2007. Effects of soil data resolution on the effect of land use patterns on hydrologic landscape functions:
SWAT model stream flow and water quality predictions. J. A comprehensive GIS‐based tool to minimize model uncertainty
Environ. Mgmt. (in press). resulting from spatial aggregation. Hydrol. Process. 19(3):
Gikas, G. D., T. Yiannakopoulou, and V. A. Tsihrintzis. 2005. 715‐727.
Modeling of nonpoint‐source pollution in a Mediterranean Hernandez, M., S. C. Miller, D. C. Goodrich, B. F. Goff, W. G.
drainage basin. Environ. Model. Assess. 11(3): 219‐233 Kepner, C. M. Edmonds, and K. B. Jones. 2000. Modeling
Gitau, M. W., T. L. Veith, and W. J. Gburek. 2004. Farm‐level runoff response to land cover and rainfall spatial variability in
optimization of BMP placement for cost‐effective pollution semi‐arid watersheds. Environ. Monitoring Assess. 64(1):
reduction. Trans. ASAE 47(6): 1923‐1931. 285‐298.
Giorgi, F., L. O. Mearns, C. Shields, and L. McDaniel. 1998. Heuvelmans, G., B. Muys, and J. Feyen. 2004a. Analysis of the
Regional nested model simulations of present day and 2×CO2 spatial variation in the parameters of the SWAT model with
climate over the central plains of the U.S. Clim. Change 40(3‐4): application in Flanders, northern Belgium. Hydrol. Earth Syst.
457‐493. Sci. 8(5): 931‐939.
Gosain, A. K., S. Rao, R. Srinivasan, and N. Gopal Reddy. 2005. Heuvelmans, G., B. Muys, and J. Feyen. 2004b. Evaluation of
Return‐flow assessment for irrigation command in the Palleru hydrological model parameter transferability for simulating the
River basin using SWAT model. Hydrol. Process. 19(3): impact of land use on catchment hydrology. Phys. Chem. Earth
673‐682. 29(11‐12): 739‐747.
Gosain, A. K., S. Rao, and D. Basuray. 2006. Climate change Heuvelmans, G., J. F. Garcio‐Qujano, B. Muys, J. Feyen, and P.
impact assessment on hydrology of Indian river basins. Current Coppin. 2005. Modelling the water balance with SWAT as part
Sci. 90(3): 346‐353. of the land use impact evaluation in a life cycle study of CO2
Govender, M., and C. S. Everson. 2005. Modelling streamflow emission reduction scenarios. Hydrol. Process. 19(3): 729‐748.
from two small South African experimental catchments using Heuvelmans, G., B. Muys, and J. Feyen. 2006. Regionalisation of
the SWAT model. Hydrol. Process. 19(3): 683‐692. the parameters of a hydrological model: Comparison of linear
Green, W. H., and G. A. Ampt. 1911. Studies on soil physics: 1. regression models with artificial neural nets. J. Hydrol. 319(1‐4):
The flow of air and water through soils. J. Agric. Sci. 4: 11‐24. 245‐265.
Green, C. H., M. D. Tomer, M. Di Luzio, and J. G. Arnold. 2006. Holvoet, K., A. van Griensven, P. Seuntjens, and P. A.
Hydrologic evaluation of the Soil and Water Assessment Tool Vanrolleghem. 2005. Sensitivity analysis for hydrology and
for a large tile‐drained watershed in Iowa. Trans. ASABE 49(2): pesticide supply towards the river in SWAT. Phys. Chem. Earth
413‐422. 30(8‐10): 518‐526.
Grizzetti, B., F. Bouraoui, K. Granlund, S. Rekolainen, and G. Horn, A. L., F. J. Rueda, G. Hörmann, and N. Fohrer. 2004.
Bidoglio. 2003. Modelling diffuse emission and retention of Implementing river water quality modelling issues in mesoscale
nutrients in the Vantaanjoki watershed (Finland) using the watersheds for water policy demands: An overview on current
SWAT model. Ecol. Model. 169(1): 25‐38. concepts, deficits, and future tasks. Phys. Chem. Earth
Grizzetti, B., F. Bouraoui, and G. De Marsily. 2005. Modelling 29(11‐12): 725‐737.
nitrogen pressure in river basins: A comparison between a Hotchkiss, R. H., S. F. Jorgensen, M. C. Stone, and T. A. Fontaine.
statistical approach and the physically‐based SWAT model. 2000. Regulated river modeling for climate change impact
Physics and Chemistry of the Earth 30(8‐10): 508‐517. assessment: The Missouri River. J. American Water Res. Assoc.
36(2): 375‐386.

Vol. 50(4): 1211-1250 1245

Huisman, J. A., L. Breuer, and H. G. Frede. 2004. Sensitivity of hydrological/water quality model for mesoscale watersheds.
simulated hydrological fluxes towards changes in soil properties Ecol. Model. 106(2‐3): 261‐289.
in response to land use change. Phys. Chem. Earth 29(11‐12): Krysanova, V., F. Hatterman, and F. Wechsung. 2005. Development
749‐758. of the ecohydrological model SWIM for regional impact studies
Izaurralde, R. C., J. R. Williams, W. B. McGill, N. J. Rosenberg, and vulnerability assessment. Hydrol. Process. 19(3): 763‐783.
and M. C. Quiroga Jakas. 2006. Simulating soil C dynamics Krysanova, V., F. Hatterman, and F. Wechsung. 2007. Implications
with EPIC: Model description and testing against long‐term of complexity and uncertainty for integrated modelling and
data. Ecol. Model. 192(3‐4): 362‐384. impact assessment in river basins. Environ. Model. Soft. 22(5):
Jamieson, R., R. Gordon, D. Joy, and H. Lee. 2004. Assessing 701‐709.
microbial pollution of rural surface waters: A review of current Lemberg, B., J. W. Mjelde, J. R. Conner, R. C. Griffin, W. D.
watershed‐scale modeling approaches. Agric. Water Mgmt. Rosenthal, and J. W. Stuth. 2002. An interdisciplinary approach
70(1): 1‐17. to valuing water from brush control. J. American Water Resour.
Jha, M., P. W. Gassman, S. Secchi, R. Gu, and J. Arnold. 2004a. Assoc. 38(2): 409‐422.
Effect of watershed subdivision on SWAT flow, sediment, and Lenhart, T., K. Eckhardt, N. Fohrer, and H.‐G. Frede. 2002.
nutrient predictions. J. American Water Resour. Assoc. 40(3): Comparison of two different approaches of sensitivity analysis.
811‐825. Phys. Chem. Earth 27(9‐10): 645‐654.
Jha, M., Z. Pan, E. S. Takle, and R. Gu. 2004b. Impacts of climate Lenhart, T., N. Fohrer, and H.‐G. Frede. 2003. Effects of land use
change on streamflow in the upper Mississippi River basin: A changes on the nutrient balance in mesoscale catchments. Phys.
regional climate model perspective. J. Geophys. Res. 109: Chem. Earth 28(33‐36): 1301‐1309.
D09105, doi:10.1029/2003JD003686. Lenhart, T., A. Van Rompaey, A. Steegen, N. Fohrer, H.‐G. Frede,
Jha, M., J. G. Arnold, P. W. Gassman, F. Giorgi, and R. Gu. 2006. and G. Govers. 2005. Considering spatial distribution and
Climate change sensitivity assessment on upper Mississippi river deposition of sediment in lumped and semi‐distributed models.
basin steamflows using SWAT. J. American Water Resour. Hydrol. Process. 19(3): 785‐794.
Assoc. 42(4): 997‐1015. Leonard, R. A., W. G. Knisel, and D. A. Still. 1987. GLEAMS:
Jha, M., P. W. Gassman, and J. G. Arnold. 2007. Water quality Groundwater loading effects of agricultural management
modeling for the Raccoon River watershed using SWAT2000. systems. Trans. ASAE 30(5): 1403‐1418.
Trans. ASABE 50(2): 479‐493. Li, C., J. Aber, F. Stange, K. Butterbach‐Bahl, and H. Papen. 1992.
Johns, T. C., R. E. Carnell, J. F. Crossley, J. M. Gregory, J. F. B. A model of nitrous oxide evolution driven from soil driven by
Mitchell, C. A. Senior, S. F. B. Tett, and R. A. Wood. 1997. The rainfall events: 1. Model structure and sensitivity. J. Geophys.
second Hadley Centre coupled ocean‐atmosphere GCM: Mode Res. 97(D9): 9759‐9776.
description, spinup, and validation. Clim. Dynam. 13(2): Limaye, A. S., T. M. Boyington, J. F. Cruise, A. Bulus, and E.
103‐134. Brown. 2001 Macroscale hydrologic modeling for regional
Kalin, L., and M. H. Hantush. 2006. Hydrologic modeling of an climate assessment studies in the southeastern United States. J.
eastern Pennsylvania watershed with NEXRAD and rain gauge American Water Resour. Assoc. 37(3): 709‐722.
data. J. Hydrol. Eng. 11(6): 555‐569. Lin, Z., and D. E. Radcliffe. 2006. Automatic calibration and
Kang, M. S., S. W. Park, J. J. Lee, and K. H. Yoo. 2006. Applying predictive uncertainty analysis of a semidistributed watershed
SWAT for TMDL programs to a small watershed containing rice model. Vadose Zone J. 5(1): 248‐260.
paddy fields. Agric. Water Mgmt. 79(1): 72‐92. Lorz, C., M. Volk, and G. Schmidt. 2007. Considering spatial
Kannan, N., S. M. White, F. Worrall, and M. J. Whelan. 2006. distribution and functionality of forests in a modeling
Pesticide modeling for a small catchment using SWAT‐2000. J. framework for river basin management. For. Ecol. Mgmt.
Environ. Sci. Health, Part B 41(7): 1049‐1070. 248(1‐2): 17‐25.
Kannan, N., S. M. White, and M. J. Whelan. 2007a. Predicting Lowrance, R., L. S. Altier, R. G. Williams, S. P. Inamdar, J. M.
diffuse‐source transfers of surfactants to surface waters using Sheridan, D. D. Bosch, R. K. Hubbard, and D. L. Thomas. 2000.
SWAT. Chemosphere 66(7): 1336‐1345 REMM: The riparian ecosystem management model. J. Soil
Kannan, N., S. M. White, F. Worrall, and M. J. Whelan. 2007b. Water Cons. 55(1): 27‐34.
Sensitivity analysis and identification of the best Manguerra, H. B., and B. A. Engel. 1998. Hydrologic
evapotranspiration and runoff options for hydrological modeling parameterization of watersheds for runoff prediction using
in SWAT‐2000. J. Hydrol. 332(3‐4): 456‐466. SWAT. J. American Water Res. Assoc. 34(5): 1149‐1162.
Kaur, R., O. Singh, R. Srinivasan, S. N. Das, and K. Mishra. 2004. Mapfumo, E., D. S. Chanasyk, and W. D. Willms. 2004. Simulating
Comparison of a subjective and a physical approach for daily soil water under foothills fescue grazing with the Soil and
identification of priority areas for soil and water management in Water Assessment Tool model (Alberta, Canada). Hydrol.
a watershed: A case study of Nagwan watershed in Hazaribagh Process. 18(3): 2787‐2800.
District of Jharkhand, India. Environ. Model. Assess. 9(2): Mausbach, M. J., and A. R. Dedrick. 2004. The length we go:
115‐127. Measuring environmental benefits of conservation practices. J.
King, K. W., J. G. Arnold, and R. L. Bingner. 1999. Comparison of Soil Water Cons. 59(5): 96A‐103A.
Green‐Ampt and curve number methods on Goodwin Creek McDonald, M. G., and A. W. Harbaugh. 1988. A modular
watershed using SWAT. Trans. ASAE 42(4): 919‐925. three‐dimensional finite‐differences ground‐water flow model.
Kirsch, K., A. Kirsch, and J. G. Arnold. 2002. Predicting sediment In Techniques of Water‐Resources Investigations. Reston, Va.:
and phosphorus loads in the Rock River basin using SWAT. U.S. Geological Survey.
Trans. ASAE 45(6): 1757‐1769. McKeown, R., G. Putz, J. Arnold, and M. Di Luzio. 2005.
Knisel, W. G. 1980. CREAMS, a field‐scale model for chemicals, Modifications of the Soil and Water Assessment Tool (SWAT‐C)
runoff, and erosion from agricultural management systems. for streamflow modeling in a small, forested watershed on the
USDA Conservation Research Report No. 26. Washington, Canadian boreal plain. In Proc. 3rd International SWAT Conf.,
D.C.: USDA. 189‐199. R. Srinivasan, J. Jacobs, D. Day, and K. Abbaspour,
Krause, P., D. P. Boyle, and F. Bäse. 2005. Comparison of different eds. Zurich, Switzerland: Swiss Federal Institute for
efficiency criteria for hydrological model assessment. Adv. Environmental Science and Technology (EAWAG). Available at:
Geosci. 5: 89‐97. ww.brc.tamus.edu/swat/3rdswatconf/. Accessed 30 October
Krysanova, V., D.‐I.Müller‐Wohlfeil, and A. Becker. 1998. 2006.
Development and test of a spatially distributed

1246 TRANSACTIONS OF THE ASABE

Menking, K. M., K. H. Syed, R. Y. Anderson, N. G. Shafike, and J. andWater Research Laboratory. Available at: www.brc.tamus.
G. Arnold. 2003. Model estimates of runoff in the closed, edu/swat/doc.html. Accessed 1 November 2006.
semiarid Estancia basin, central New Mexico, USA. Hydrol. Sci. Neitsch, S. L., J. G. Arnold, J. R. Kiniry, R. Srinivasan, and J. R.
J. 48(6): 953‐970. Williams. 2005b. Soil and Water Assessment Tool Input/Output
Menking, K. M., R. Y, Anderson, N. G. Shafike, K. H. Syed, and B. File Documentation, Version 2005. Temple, Tex.: USDA‐ARS
D. Allen. 2004. Wetter or colder during the last glacial Grassland, Soil andWater Research Laboratory. Available at:
maximum? Revisiting the pluvial lake question in southwestern www.brc.tamus.edu/swat/doc.html. Accessed 1 November 2006.
North America. Quart. Res. 62(3): 280‐288. Nelson, R. G., J. C. Ascough II, and M. R. Langemeier. 2005.
Migliaccio, K. W., I. Chaubey, and B. E. Haggard. 2007. Evaluation Environmental and economic analysis of switchgrass production
of landscape and instream modeling to predict watershed for water quality improvement in northeast Kansas. J. Environ.
nutrient yields. Environ. Model. Soft. 22(7): 987‐999. Mgmt. 79(4): 336‐347.
Miller, S. N., W. G. Kepner, M. H. Mehaffey, M. Hernandez, R. C. Olivera, F., M. Valenzuela, R. Srinivasan, J. Choi, H. Cho, S. Koka,
Miller, D. C. Goodrich, K. K Devonald, D. T. Heggem, and W. and A. Agrawal. 2006. ArcGIS‐SWAT: A geodata model and
P. Miller. 2002. Integrating landscape assessment and hydrologic GIS interface for SWAT. J. American Water Resour. Assoc.
modeling for land cover change analysis. J. American Water 42(2): 295‐309.
Res. Assoc. 38(4): 915‐929. Osei, E., P. W. Gassman, L. M. Hauck, S. Neitsch, R. D. Jones, J.
Miller, S. N., D. J. Semmens, D. C. Goodrich, M. Hernandez, R. C. Mcnitt, and H. Jones. 2003. Using nutrient management to
Miller, W. G. Kepner, and D. P. Guertin. 2007. The automated control nutrient losses from dairy pastures. J. Range Mgmt.
geospatial watershed assessment tool. Environ. Model. Soft. 56(3): 218‐226.
22(3): 365‐377. Pachepsky, Y. A., A. M. Sadeghi, S. A. Bradford, D. R. Shelton, A.
Mishra, A., J. Froebrich, and P. W. Gassman. 2007. Evaluation of K. Gruber, and T. Dao. 2006. Transport and fate of
the SWAT model for assessing sediment control structures in a manure‐borne pathogens: Modeling perspective. Agric. Water
small watershed in India. Trans. ASABE 50(2): 469‐478. Mgmt. 86(1‐2): 81‐92.
Monteith, J. L. 1965. Evaporation and the environment. In The Perkins, S. P., and M. Sophocleous. 1999. Development of a
State and Movement of Water in Living Organisms, Proc. 19th comprehensive watershed model applied to study stream yield
Symp. Swansea, U.K.: Society of Experimental Biology, under drought conditions. Groundwater 37(3): 418‐426.
Cambridge University Press. Peterson, J. R., and J. M. Hamlet. 1998. Hydrologic calibration of
Moon, J., R. Srinivasan, and J. H. Jacobs. 2004. Stream flow the SWAT model in a watershed containing fragipan soils. J.
estimation using spatially distributed rainfall in the Trinity River American Water Resour. Assoc. 34(3): 531‐544.
basin, Texas. Trans. ASAE 47(5): 1445‐1451. Pohlert, T., J. A. Huisman, L. Breuer, and H.‐G. Freude. 2007.
Moriasi, D. N., J. G. Arnold, M. W. Van Liew, R. L. Binger, R. D. Integration of a detailed biogeochemical model into SWAT for
Harmel, and T. Veith. 2007. Model evaluation guidelines for improved nitrogen predictions: Model development, sensitivity,
systematic quantification of accuracy in watershed simulations. and GLUE analysis. Ecol. Model. 203(3‐4): 215‐228.
Trans. ASABE 50(3): 885‐900. Ponce, V. M., and R. H. Hawkins. 1996. Runoff curve number: Has
Muleta, M. K., and J. W. Nicklow. 2005a. Decision support for it reached maturity? J. Hydrol. Eng. 1(1): 11‐19.
watershed management using evolutionary algorithms. J. Water Plus, M., I. La Jeunesse, F. Bouraoui, J.‐M. Zaldívar, A. Chapelle,
Resour. Plan. Mgmt. 131(1): 35‐44. and P. Lazure. 2006. Modelling water discharges and nitrogen
Muleta, M. K., and J. W. Nicklow. 2005b. Sensitivity and inputs into a Mediterranean lagoon: Impact on the primary
uncertainty analysis coupled with automatic calibration for a production. Ecol. Model. 193(1‐2): 69‐89.
distributed watershed model. J. Hydrol. 306(1‐4): 127‐145. Priestly, C. H. B., and R. J. Taylor. 1972. On the assessment of
Muleta, M. K., J. W. Nicklow, and E. G. Bekele. 2007. Sensitivity surface heat flux and evaporation using large‐scale parameters.
of a distributed watershed simulation model to spatial scale. J. Monthly Weather Rev. 100(2): 81‐92.
Hydrol. Eng. 12(2): 163‐172. Qi, C., and S. Grunwald. 2005. GIS‐based hydrologic modeling in
Muttiah, R. S., and R. A. Wurbs. 2002. Modeling the impacts of the Sandusky watershed using SWAT. Trans. ASABE 48(1):
climate change on water supply reliabilities. Water Intl., Intl. 169‐180.
Water Resources Assoc. 27(3): 407‐419. Qiu, Z. 2005. Using multi‐criteria decision models to assess the
Narasimhan, B., and R. Srinivasan. 2005. Development and economic and environmental impacts of farming decisions in an
evaluation of soil moisture deficit index (SMDI) and agricultural watershed. Rev. Agric. Econ. 27(2): 229‐244.
evapotranspiration deficit index (ETDI) for agricultural drought Qiu, Z., and T. Prato. 1998. Economic evaluation of riparian buffers
monitoring. Agric. For. Meteor. 133(1‐4): 69‐88. in an agricultural watershed. J. American Water Resour. Assoc.
Narasimhan, B., R. Srinivasan, J. G. Arnold, and M. Di Luzio. 34(4): 877‐890.
2005. Estimation of long‐term soil moisture using a distributed Rao, M., G. Fan, J. Thomas, G. Cherian, V. Chudiwale, and M.
parameter hydrologic model and verification using remotely Awawdeh. 2006. A web‐based GIS decision support system for
sensed data. Trans. ASABE 48(3): 1101‐1113. managing and planning USDA's Conservation Reserve Program
Nash, J. E., and J. V. Sutcliffe. 1970. River flow forecasting through (CRP). Environ. Model. Soft. 22(9): 1270‐1280.
conceptual models: Part I. A discussion of principles. J. Hydrol. Ramanarayanan, T., B. Narasimhan, and R. Srinivasan. 2005.
10(3): 282‐290. Characterization of fate and transport of isoxaflutole, a
Nasr, A., M. Bruen, P. Jordan, R. Moles, G. Kiely, and P. Byrne. soil‐applied corn herbicide, in surface water using a watershed
2007. A comparison of SWAT, HSPF, and SHETRAN/GOPC model. J. Agric. Food Chem. 53(22): 8848‐8858.
for modeling phosphorus export from three catchments in Rawls, W. J., and D. L. Brakensiek. 1986. Comparison between
Ireland. Water Res. 41(5): 1065‐1073. Green‐Ampt and curve number runoff predictions. Trans. ASAE
Nearing, M. A., V. Jetten, C. Baffaut, O. Cerdan, A. Couturier, M. 29(6): 1597‐1599.
Hernandez, Y. Le Bissonnais, M. H. Nichols, J. P. Nunes, C. S. Refsgaard, J. C., and B. Storm. 1995. MIKE‐SHE. In Computer
Renschler, V. Souchère, and K. van Ost. 2005. Modeling Models in Watershed Hydrology, 809‐846. V. J. Singh, ed.
response of soil erosion and runoff to changes in precipitation Highland Ranch, Colo.: Water Resources Publications.
and cover. Catena 61(2‐3): 131‐154. Renschler, C. S., and T. Lee. 2005. Spatially distributed assessment
Neitsch, S. L., J. G. Arnold, J. R. Kiniry, and J. R. Williams. 2005a. of short‐ and long‐term impacts of multiple best management
Soil and Water Assessment Tool Theoretical Documentation, practices in agricultural watersheds. J. Soil Water Cons. 60(6):
Version 2005. Temple, Tex.: USDA‐ARS Grassland, Soil 446‐455.

Vol. 50(4): 1211-1250 1247

Rosenberg, N. J., D. L. Epstein, D. Wang, L. Vail, R. Srinivasan, Wolfe, J. Frankenberger, C. Graff, and T. M. Sohrabi. 2006.
and J. G. Arnold. 1999. Possible impacts of global warming on Uncertainty in TMDL models. Trans. ASABE 49(4): 1033‐1049.
the hydrology of the Ogallala aquifer region. Clim. Change Singh, J., H. V. Knapp, J. G. Arnold, and M. Demissie. 2005.
42(4): 677‐692. Hydrological modeling of the Iroquois River watershed using
Rosenberg, N. J., R. A. Brown, R. C. Izaurralde, and A. M. HSPF and SWAT. J. American Water Resour. Assoc. 41(2):
Thomson. 2003. Integrated assessment of Hadley Centre 343‐360.
(HadCM2) climate change projections in agricultural Skaggs, R. W. 1982. Field evaluation of a water management
productivity and irrigation water supply in the conterminous simulation model. Trans. ASAE 25(3): 666‐674.
United States: I. Climate change scenarios and impacts on Smith, R. A., G. E. Schwarz, and R. A. Alexander. 1997. Regional
irrigation water supply simulated with the HUMUS model. interpretation of water‐quality monitoring data. Water Resour.
Agric. For. Meteor. 117(1‐2): 73‐96. Res. 33(12): 2781‐2798.
Rosenthal, W. D., and D. W. Hoffman. 1999. Hydrologic Sophocleous, M., and S. P. Perkins 2000. Methodology and
modeling/GIS as an aid in locating monitoring sites. Trans. application of combined watershed and ground‐water models in
ASAE 42(6): 1591‐1598. Kansas. J. Hydrol. 236 (3‐4): 185‐201.
Rosenthal, W. D., R. Srinivasan, and J. G. Arnold. 1995. Alternative Sophocleous, M. A., J. K. Koelliker, R. S. Govindaraju, T. Birdie,
river management using a linked GIS‐hydrology model. Trans. S. R. Ramireddygari, and S. P. Perkins. 1999. Integrated
ASAE 38(3): 783‐790. numerical modeling for basin‐wide water management: The case
Rowan, R. C. 1995. PHYGROW model documentation, version of the Rattlesnake Creek basin in south‐central Kansas. J.
2.0. College Station, Tex.: Texas A&M University, Department Hydrol. 214(1‐4): 179‐196.
of Rangeland Ecology and Management, Ranching Systems Spruill, C. A., S. R. Workman, and J. L. Taraba. 2000. Simulation
Group. of daily and monthly stream discharge from small watersheds
Saleh, A., and B. Du. 2004. Evaluation of SWAT and HSPF within using the SWAT model. Trans. ASAE 43(6): 1431‐1439.
BASINS program for the upper North Bosque River watershed Srinivasan, R., and J. G. Arnold. 1994. Integration of a basin‐scale
in central Texas. Trans. ASAE 47(4): 1039‐1049. water quality model with GIS. Water Resour. Bull. (30)3:
Saleh, A., J. G. Arnold, P. W. Gassman, L. W. Hauck, W. D. 453‐462.
Rosenthal, J. R. Williams, and A. M. S. McFarland. 2000. Srinivasan, R., T. S. Ramanarayanan, J. G. Arnold, and S. T.
Application of SWAT for the upper North Bosque River Bednarz. 1998. Large‐area hydrologic modeling and assessment:
watershed. Trans. ASAE 43(5): 1077‐1087. Part II. Model application. J. American Water Resour. Assoc.
Salvetti, R., A. Azzellino, and R. Vismara. 2006. Diffuse source 34(1): 91‐101.
apportionment of the Po River eutrophying load to the Adriatic Srinivasan, M. S., P. Gerald‐Marchant, T. L. Veith, W. J. Gburek,
Sea: Assessment of Lombardy contribution to Po River nutrient and T. S. Steenhuis. 2005. Watershed‐scale modeling of critical
load apportionment by means of an integrated modelling source areas of runoff generation and phosphorus transport. J.
approach. Chemosphere 65(11): 2168‐2177. American Water Resour. Assoc. 41(2): 361‐375.
Santhi, C., J. G. Arnold, J. R. Williams, W. A. Dugas, R. Srinivasan, Srivastava, P., J. N. McNair, and T. E. Johnson. 2006. Comparison
and L. M. Hauck. 2001a. Validation of the SWAT model on a of process‐based and artificial neural network approaches for
large river basin with point and nonpoint sources. J. American streamflow modeling in an agricultural watershed. J. American
Water Resour. Assoc. 37(5): 1169‐1188. Water Resour. Assoc. 42(2): 545‐563.
Santhi, C., J. G. Arnold, J. R. Williams, L. M. Hauck, and W. A. Steenhuis, T. S. 2007. Personal communication. Ithaca, N.Y.:
Dugas. 2001b. Application of a watershed model to evaluate Cornell University, Department of Biological and Agricultural
management effects on point and nonpoint source pollution. Engineering.
Trans. ASAE 44(6): 1559‐1570. Stewart, G. R., C. L. Munster, D. M. Vietor, J. G. Arnold, A. M. S.
Santhi, C., R. S. Muttiah, J. G. Arnold, and R. Srinivasan. 2005. A McFarland, R. White, and T. Provin. 2006. Simulating water
GIS‐based regional planning tool for irrigation demand quality improvements in the upper North Bosque River
assessment and savings using SWAT. Trans. ASABE 48(1): watershed due to phosphorus export through turfgrass sod.
137‐147. Trans. ASABE 49(2): 357‐366.
Santhi, C., R. Srinivasan, J. G. Arnold, and J. R. Williams. 2006. A Stone, M. C., R. H. Hotchkiss, C. M. Hubbard, T. A. Fontaine, L.
modeling approach to evaluate the impacts of water quality O. Mearns, and J. G. Arnold. 2001. Impacts of climate change
management plans implemented in a watershed in Texas. on Missouri river basin water yield. J. American Water Resour.
Environ. Model. Soft. 21(8): 1141‐1157. Assoc. 37(5): 1119‐1130.
Schomberg, J. D., G. Host, L. B. Johnson, and C. Richards. 2005. Stone, M. C., R. C. Hotchkiss, and L. O. Mearnes. 2003. Water
Evaluating the influence of landform, surficial geology, and land yield responses to high and low spatial resolution climate change
use on streams using hydrologic simulation modeling. Aqua. scenarios in the Missouri River basin. Geophys. Res. Letters
Sci. 67(4): 528‐540. 30(4): 35.1‐35.4.
Schuol, J., and K. C. Abbaspour. 2007. Using monthly weather Stonefelt, M. D., T. A. Fontaine, and R. H. Hotchkiss. 2000.
statistics to generate daily data in a SWAT model application to Impacts of climate change on water yield in the upper Wind
west Africa. Ecol. Model. 201(3‐4): 301‐311. River basin. J. American Water Resour. Assoc. 36(2): 321‐336.
Seaber, P. R., F. P. Kapinos, and G. L. Knapp. 1987. Hydrologic Sun, H., and P. S. Cornish. 2005. Estimating shallow groundwater
units maps. USGS Water‐Supply Paper No. 2294. Reston, Va.: recharge in the headwaters of the Liverpool Plains using SWAT.
U.S. Geological Survey. Hydrol. Process. 19(3): 795‐807.
Secchi, S., P. W. Gassman, M. Jha, L. Kurkalova, H. H. Feng, T. SWAT. 2007a. Soil and Water Assessment Tool: ArcSWAT. College
Campbell, and C. Kling. 2007. The cost of cleaner water: Station, Tex.: Texas A&M University. Available at: www.brc.
Assessing agricultural pollution reduction at the watershed scale. tamus.edu/swat/arcswat.html. Accessed 20 February 2007.
J. Soil Water Cons. 62(1): 10‐21. SWAT. 2007b. Soil and Water Assessment Tool: AVSWAT. College
Shepherd, B., D. Harper, and A. Millington. 1999. Modelling Station, Tex.: Texas A&M University. Available at: www.brc.
catchment‐scale nutrient transport to watercourses in the U.K. tamus.edu/swat/avswat.html. Accessed 13 February 2007.
Hydrobiologia 395‐396: 227‐237. SWAT. 2007c. Soil and Water Assessment Tool: Peer‐reviewed
Shirmohammadi, A., I. Chaubey, R. D. Harmel, D. D. Bosch, R. literature. College Station, Tex.: Texas A&M University.
Muñoz‐Carpena, C. Dharmasri, A. Sexton, M. Arabi, M. L. Available at: www.brc.tamus.edu/swat/pubs_peerreview.html.
Accessed 17 February 2007.

1248 TRANSACTIONS OF THE ASABE

SWAT. 2007d. Soil and Water Assessment Tool: SWAT model. Protection Agency. Available at: www.epa.gov/waterscience/
College Station, Texas: Tex. A&M University. Available at: BASINS/. Accessed 23 August 2006.
www.brc.tamus.edu/swat/soft_model.html. Accessed 21 USEPA. 2006b. Overview of current total maximum daily load -
February 2007. TMDL - Program and regulations. Washington, D.C.: U.S.
Takle, E. S., M. Jha, and C. J. Anderson. 2005. Hydrological cycle Environmental Protection Agency. Available at: www.epa.gov/
in the upper Mississippi River basin: 20th century simulations owow/tmdl/overviewfs.html. Accessed 25 August 2006.
by multiple GCMs. Geophys. Res. Letters 32(18): USEPA. 2007. Total maximum daily loads: National section 303(d)
L18407.1‐L18407.5. list fact sheet. Washington, D.C.: U.S. Environmental Protection
TempQsim. 2006. Evaluation and improvement of water quality Agency. Available at: oaspub.epa.gov/waters/
models for application to temporary waters in southern national_rept.control. Accessed 22 March 2007.
European catchments (TempQsim). Available at: Vaché, K. B., J. M. Eilers, and M. V. Santelman. 2002. Water
www.tempqsim.net/. Accessed 25 August 2006. quality modeling of alternative agricultural scenarios in the U.S.
Thomson, A. M., R. A. Brown, N. J. Rosenberg, R. C. Izaurralde, Corn Belt. J. American Water Resour. Assoc. 38(2): 773‐787.
D. M. Legler, and R. Srinivasan. 2003. Simulated impacts of El Vandenberghe, V., W. Bauwens, and P. A. Vanrolleghem. 2007.
Nino/southern oscillation on United States water resources. J. Evaluation of uncertainty propagation into river water quality
American Water Resour. Assoc. 39(1): 137‐148. predictions to guide future monitoring campaigns. Environ.
Thomson, A. M., R. A. Brown, N. J. Rosenberg, R. Srinivasan, and Model. Soft. 22(5): 725‐732.
R. C. Izaurralde. 2005. Climate change impacts for the van Griensven, A., and W. Bauwens. 2003. Multiobjective
conterminous USA: An integrated assessment: Part 4. Water autocalibration for semidistributed water quality models. Water
resources. Clim. Change 69(1): 67‐88. Resour. Res. 39(12): SWC 9.1‐ SWC 9.9.
Tolson, B. A., and C. A. Shoemaker. 2007. Cannonsville reservoir van Griensven, A., and W. Bauwens. 2005. Application and
watershed SWAT2000 model development, calibration, and evaluation of ESWAT on the Dender basin and Wister Lake
validation. J. Hydrol. 337(1‐2): 68‐86. basin. Hydrol. Process. 19(3): 827‐838.
Tripathi, M. P., R. K. Panda, and N. S. Raghuwanshi. 2003. van Griensven A., and T. Meixner. 2006. Methods to quantify and
Identification and prioritisation of critical sub‐watersheds for soil identify the sources of uncertainty for river basin water quality
conservation management using the SWAT model. Biosys. Eng. models. Water Sci. Tech. 53(1): 51‐59.
85(3): 365‐379. van Griensven, A., L. Breuer, M. Di Luzio, V. Vandenberghe, P.
Tripathi, M. P., N. S. Raghuwanshi, and G. P. Rao. 2006. Effect of Goethals, T. Meixner, J. Arnold, and R. Srinivasan. 2006a.
watershed subdivision on simulation of water balance Environmental and ecological hydroinformatics to support the
components. Hydrol. Process. 20(5): 1137‐1156. implementation of the European Water Framework Directive for
Turpin, N., P. Bontems, G. Rotillon, I. Bärlund, M. Kaljonen, S. river basin management. J. Hydroinformatics 8(4): 239‐252.
Tattari, F. Feichtinger, P. Strauss, R. Haverkamp, M. Garnier, A. van Griensven, A., T. Meixner, S. Grunwald, T. Bishop, M. Diluzio,
Lo Porto, G. Benigni, A. Leone, M. Nicoletta Ripa, O. M. Eklo, and R. Srinivasan. 2006b. A global sensitivity analysis tool for
E. Romstad, T. Bioteau, F. Birgand, P. Bordenave, R. Laplana, J. the parameters of multi‐variable catchment models. J. Hydrol.
M. Lescot, L. Piet, and F. Zahm. 2005. AgriBMPWater: Systems 324(1‐4): 10‐23.
approach to environmentally acceptable farming. Environ. Van Liew, M. W., and J. Garbrecht. 2003. Hydrologic simulation of
Model. Soft. 20(2): 187‐196. the Little Washita River experimental watershed using SWAT. J.
TWRI. 2003. SWAT2003: Proc. 2nd Intl. SWAT Conference. TWRI American Water Resour. Assoc. 39(2): 413‐426.
Technical Report No. 266. College Station, Tex.: Texas Water Van Liew, M. W., J. G. Arnold, and J. D. Garbrecht. 2003a.
Resources Institute, Texas A&M University. Available at: Hydrologic simulation on agricultural watersheds: choosing
www.brc.tamus.edu/swat/pubs_2ndconf.html. Accessed 4 between two models. Trans. ASAE 46(6): 1539‐1551.
February 2007. Van Liew, M. W., J. D. Garbrecht, and J. G. Arnold. 2003b.
UNESCO-IHE. 2007. 4TH International SWAT conference: Book Simulation of the impacts of flood retarding structures on
of abstracts. Delft, Netherlands: United Nations Educational, streamflow for a watershed in southwestern Oklahoma under
Scientific and Cultural Organization, Institute for Water dry, average, and wet climatic conditions. J. Soil Water Cons.
Education. Available at: www.brc.tamus.edu/swat/4thswatconf/ 58(6): 340‐348.
docs/BOOK%20OF%20ABSTRACTS%20final.pdf. Accessed Van Liew, M. W., J. G. Arnold, and D. D. Bosch. 2005. Problems
5 August 2007. and potential of autocalibrating a hydrologic model. Trans.
USDA‐ARS. 2007. The Automated Geospatial Watershed ASABE 48(3): 1025‐1040.
Assessment tool (AGWA). Tucson, Ariz.: USDA Agricultural Van Liew, M. W., T. L. Veith, D. D. Bosch, and J. G. Arnold. 2007.
Research Service. Available at: www.tucson.ars.ag.gov/agwa/. Suitability of SWAT for the Conservation Effects Assessment
Accessed 23 March 2007. Project: A comparison on USDA‐ARS watersheds. J. Hydrol.
USDA‐NRCS. 2004. Part 630: Hydrology. Chapter 10: Estimation Eng. 12(2): 173‐189.
of direct runoff from storm rainfall: Hydraulics and hydrology: Varanou, E, E. Gkouvatsou, E. Baltas, and M. Mimikou. 2002.
Technical references. In NRCS National Engineering Handbook. Quantity and quality integrated catchment modelling under
Washington, D.C.: USDA National Resources Conservation climatic change with use of Soil and Water Assessment Tool
Service. Available at: www.wcc.nrcs.usda.gov/hydro/ model. J. Hydrol. Eng. 7(3): 228‐244.
hydro‐techref‐neh‐630.html. Accessed 14 February 2007. Vazquez‐Amabile, G. G., and B. A. Engel. 2005. Use of SWAT to
USDA‐NRCS. 2007a. Soil Survey Geographic (SSURGO) compute groundwater table depth and streamflow in the
database. Washington, D.C.: USDA National Resources Muscatatuck River watershed. Trans. ASABE 48(3): 991‐1003.
Conservation Service. Available at: www.ncgc.nrcs.usda.gov/ Vazquez‐Amabile, G. G., B. A. Engel, and D. C. Flanagan. 2006.
products/datasets/ssurgo/. Accessed 23 march 2007. Modeling and risk analysis of nonpoint‐source pollution caused
USDA‐NRCS. 2007b. U.S. general soil map (STATSGO). by atrazine using SWAT. Trans. ASABE 49(3): 667‐678.
Washington, D.C.: USDA National Resources Conservation Veith, T. L., A. N. Sharpley, J. L. Weld, and W. J. Gburek. 2005.
Service. Available at: www.ncgc.nrcs.usda.gov/products/ Comparison of measured and simulated phosphorus losses with
datasets/statsgo/. Accessed 23 march 2007. indexed site vulnerability. Trans. ASAE 48(2): 557‐565.
USEPA. 2006a. Better Assessment Science Integrating Point and Volk, M., P. M. Allen, J. G. Arnold, and P. Y. Chen. 2005. Towards
Nonpoint Sources. Washington, D.C.: U.S. Environmental a process‐oriented HRU‐concept in SWAT: Catchment‐related
control on baseflow and storage of landscape units in medium to

Vol. 50(4): 1211-1250 1249

 large river basins. In Proc. 3rd Intl. SWAT Conf., 159‐168. R. Watterson, J. G., S. P. O'Farrell, and M. R. Dix. 1997. Energy and
Srinivasan, J. Jacobs, D. Day, and K. Abbaspour, eds. Zurich, water transport in climates simulated by a general circulation
Switzerland: Swiss Federal Institute for Environmental Science model that includes dynamic sea ice. J. Geophys. Res. 11(D10):
and Technology (EAWAG). Available at: www.brc.tamus.edu/ 11027‐11037.
swat/3rdswatconf/. Accessed 30 October 2006. Weber, A., N. Fohrer, and D. Moller. 2001. Long‐term land use
Volk, M., J. Hirschfeld, G. Schmidt, C. Bohn, A. Dehnhardt, S. changes in a mesocale watershed due to socio‐economic factors:
Liersch, and L. Lymburner. 2007. A SDSS‐based Effects on landscape structures and functions. Ecol. Model.
ecological‐economic modeling approach for integrated river 140(1‐2): 125‐140.
basin management on different scale levels: The project Wells, D. 2006. Personal communication. Washington, D.C.: U.S.
FLUMAGIS. Water Resour. Mgmt. (in press). Environmental Protection Agency.
von Stackelberg, N. O., G. M. Chescheir, R. W. Skaggs, and D. K. White, K. L., and I. Chaubey. 2005. Sensitivity analysis, calibration,
Amatya. 2007. Simulation of the hydrologic effects of and validations for a multisite and multivariable SWAT model. J.
afforestation in the Tacuarembó River basin, Uruguay. Trans. American Water Resour. Assoc. 41(5): 1077‐1089.
ASABE 50(2): 455‐468. Whittaker, G., R. Fare, R. Srinivasan, and D. W. Scott. 2003. Spatial
Walter, M. T., M. F. Walter, E. S. Brooks, T. S. Steenhuis, J. Boll, evaluation of alternative nonpoint nutrient regulatory
and K. Weiler. 2000. Hydrologically sensitive areas: Variable instruments. Water Resour. Res. 39(4): WES 1.1 - WES 1.9.
source area hydrology implications for water quality risk Williams, J. R. 1969. Flood routing with variable travel time or
assessment. J. Soil Water Cons. 55(3): 277‐284. variable storage coefficients. Trans. ASAE 12(1): 100‐103.
Wang, X., and A. M. Melesse. 2005. Evaluation of the SWAT Williams, J. R. 1990. The erosion productivity impact calculator
model's snowmelt hydrology in a northwestern Minnesota (EPIC) model: A case history. Phil. Trans. R. Soc. London
watershed. Trans. ASABE 48(4): 1359‐1376. 329(1255): 421‐428.
Wang, X., and A. M. Melesse. 2006. Effects of STATSGO and Williams, J. R., and H. D. Berndt. 1977. Sediment yield prediction
SSURGO as inputs on SWAT model's snowmelt simulation. J. based on watershed hydrology. Trans. ASAE 20(6): 1100‐4.
American Water Resour. Assoc. 42(5): 1217‐1236. Williams, J. R., and R. C. Izaurralde. 2006. The APEX model. In
Wang, X., A. M. Melesse, and W. Yang. 2006. Influences of Watershed Models, 437‐482. V. P. Singh and D. K. Frevert, eds.
potential evapotranspiration estimation methods on SWAT's Boca Raton, Fla.: CRC Press.
hydrologic simulation in a northwestern Minnesota watershed. Wu, K., and Y. J. Xu. 2006. Evaluation of the applicability of the
Trans. ASABE 49(6): 1755‐1771. SWAT model for coastal watersheds in southeastern Louisiana.
Watson, B. M., R. Srikanthan, S. Selvalingam, and M. Ghafouri. J. American Water Resour. Assoc. 42(5): 1247‐1260.
2005. Evaluation of three daily rainfall generation models for Wu, K., and C. Johnston. 2007. Hydrologic response to climatic
SWAT. Trans. ASABE 48(5): 1697‐1711. variability in a Great Lakes watershed: A case study with the
Wattenbach, M., F. Hatterman, R. Weng, F. Wechsung, V. SWAT model. J. Hydrol. 337(1‐2): 187‐199.
Krysanova, and F. Badeck. 2005. A simplified approach to Zhang, X., R. Srinivasan, and F. Hao. 2007. Predicting hydrologic
implement forest eco‐hydrological properties in regional response to climate change in the Luohe River basin using the
hydrological modelling. Ecol. Model. 187(1): 49‐50. SWAT model. Trans. ASABE 50(3): 901‐910.

1250 TRANSACTIONS OF THE ASABE