HYDROLOGICAL PROCESSES

Hydrol. Process. 23, 748– 763 (2009)

Published online 15 December 2008 in Wiley InterScience
(www.interscience.wiley.com) DOI: 10.1002/hyp.7174

Comparison of AnnAGNPS and SWAT model simulation

results in USDA-CEAP agricultural watersheds
in south-central Kansas
Prem B. Parajuli,1 * Nathan O. Nelson,1 Lyle D. Frees2 and Kyle R. Mankin3
1 Department of Agronomy, Kansas State University, Manhattan, KS 66506
2 USDA/NRCS, Salina, KS 67401
3 Department of Biological and Agricultural Engineering, Kansas State University, Manhattan, KS 66506

Abstract:
This study was conducted under the USDA-Conservation Effects Assessment Project (CEAP) in the Cheney Lake watershed
in south-central Kansas. The Cheney Lake watershed has been identified as ‘impaired waters’ under Section 303(d) of the
Federal Clean Water Act for sediments and total phosphorus. The USDA-CEAP seeks to quantify environmental benefits of
conservation programmes on water quality by monitoring and modelling. Two of the most widely used USDA watershed-scale
models are Annualized AGricultural Non-Point Source (AnnAGNPS) and Soil and Water Assessment Tool (SWAT). The
objectives of this study were to compare hydrology, sediment, and total phosphorus simulation results from AnnAGNPS and
SWAT in separate calibration and validation watersheds. Models were calibrated in Red Rock Creek watershed and validated
in Goose Creek watershed, both sub-watersheds of the Cheney Lake watershed. Forty-five months (January 1997 to September
2000) of monthly measured flow and water quality data were used to evaluate the two models. Both models generally provided
from fair to very good correlation and model efficiency for simulating surface runoff and sediment yield during calibration
and validation (correlation coefficient; R2 , from 0Ð50 to 0Ð89, Nash Sutcliffe efficiency index, E, from 0Ð47 to 0Ð73, root
mean square error, RMSE, from 0Ð25 to 0Ð45 m3 s
1 for flow, from 158 to 312 Mg for sediment yield). Total phosphorus
predictions from calibration and validation of SWAT indicated good correlation and model efficiency (R2 from 0Ð60 to 0Ð70,
E from 0Ð63 to 0Ð68) while total phosphorus predictions from validation of AnnAGNPS were from unsatisfactory to very
good (R2 from 0Ð60 to 0Ð77, E from
2Ð38 to 0Ð32). The root mean square error–observations standard deviation ratio (RSR)
was estimated as excellent (from 0Ð08 to 0Ð25) for the all model simulated parameters during the calibration and validation
study. The percentage bias (PBIAS) of the model simulated parameters varied from unsatisfactory to excellent (from 128 to
3). This study determined SWAT to be the most appropriate model for this watershed based on calibration and validation
results. Copyright  2008 John Wiley & Sons, Ltd.

KEY WORDS flow; sediment; total phosphorus; calibration; validation; R2 ; E; RMSE; RSR; PBIAS

Received 31 March 2008; Accepted 17 September 2008

INTRODUCTION Current water quality assessment techniques generally

The majority of the Kansas population (>70%) uses include two methods: (a) water quality field monitoring
surface water for drinking water and other daily uses and (b) computer/mathematical modelling. Field moni-
(Kansas Water Office [KWO], 2004); but 39% of assessed toring is the most appropriate and reliable method to
river and stream miles and 76% of assessed lake support water quality assessment. However, it is laborious
acreage in Kansas are impaired for one or more des- and expensive. Computer/mathematical models provide
ignated uses (KDHE, 2006). The Cheney Lake water- an alternative to monitoring that can save time, reduce
shed has been identified as ‘impaired waters’ under costs, and minimize the need for testing management
Section 303(d) of the Federal Clean Water Act for alternatives (Shirmohammadi et al., 2006).
sediments and total phosphorus. The TMDLs for the Currently, numerous watershed models with various
Cheney Lake have been established by Kansas Depart- capabilities and degrees of complexity are available. Sev-
ment of Health and Environment (KDHE) for phos- eral of these models are used to estimate runoff, sediment
phorus and sediments, which recommended a reduc- yield, and phosphorus loads, and many are applicable
tion of 45% for both pollutants (CLWMP, 2008). To to water quality goal development and implementation
improve water quality through conservation practices, (Borah et al., 2006). These models are being explored
it is essential to identify sources of contaminants and as tools for developing management strategies to reduce
quantify their contributions to water quality impairment. effects of non-point source pollution on water quality.
Choosing the appropriate model for a given applica-
tion and watershed is critical. Proper model calibration
* Correspondence to: Prem B. Parajuli, 2011D Throckmorton Plant Sci-
ence Center, Kansas State University, Manhattan, KS 66506. E-mail: and validation is necessary to ascertain accuracy and
[email protected]. reliability of results (Das et al., 2007a). Models are first

Copyright  2008 John Wiley & Sons, Ltd.

 ANNAGNPS AND SWAT MODEL SIMULATION RESULTS IN AGRICULTURAL WATERSHEDS 749

calibrated against historically available data collected (1991–1995) and validation (1996–2000) for the two
from the study watershed. The calibrated model can then models. In their study, the AnnAGNPS and SWAT model
be validated by comparing model predictions with mea- demonstrated good to very good (Parajuli, 2007) model
sured data in other watersheds in the same geographic efficiency for flow when compared with observed data
region. Calibrating models in one watershed and vali- during the calibration (E D 0Ð79 for AnnAGNPS and E D
dating them in another watershed enables evaluation of 0Ð70 for SWAT) and validation (E D 0Ð69 for AnnAG-
model capability in different conditions and management NPS and E D 0Ð57 for SWAT) period. For sediment yield
practices (Parajuli et al., 2007, 2008). they used monthly estimated sediment yield data to com-
Because of their proven applicability at the watershed pare with models simulated results for the calibration and
scale in Kansas and other places; popularity and capa- validation period. AnnAGNPS and SWAT showed fair to
bility to predict flow, sediment, and nutrients, the Annu- good (Parajuli, 2007) model efficiency during calibration
alized AGricultural Non-Point Source (AnnAGNPS) and (E D 0Ð53 for AnnAGNPS and E D 0Ð41 for SWAT) and
the Soil and Water Assessment Tool (SWAT) models were validation (E D 0Ð35 for AnnAGNPS and E D 0Ð24 for
chosen for this study. Refereed AnnAGNPS applications SWAT) periods. They did not report model performance
are predominantly for sites in the USA (Yuan et al., using other statistical parameters such as R2 , root mean
2001, 2002; Polyakov et al., 2007); however, applications square errors (RMSE), etc.
in other countries also have been reported, e.g. Aus- The USDA-Conservation Effects Assessment Project
tralia (Baginska et al., 2003), China (Hong et al., 2005), (CEAP) within the USA funded more than 13 individ-
Canada (Das et al., 2006), and Belgium (Licciardello ual CEAP projects using models to evaluate best man-
et al., 2007). AnnAGNPS has been calibrated, validated, agement practices (BMPs). Most of the CEAP projects
and applied for runoff and sediment yield losses from in the USA use the SWAT model for their watershed
watersheds in different geographic locations, conditions, water quality studies. The Cheney Lake watershed (CLW)
and management practices (Gebremeskel et al., 2005; has records of using a single event based Agricultural
Das et al., 2007a; Sadeghi et al., 2007, Licciardello et al., Non Point Source (AGNPS) model (Bhuyan et al., 2003)
2007). and AnnAGNPS model (Lyle F., personal communica-
Refereed applications of the SWAT model are numer- tion, 2007) in some of the sub-watersheds, including Red
ous, and SWAT has been implemented internationally, Rock Creek. However, the SWAT model, which is one
e.g. Greece (Gikas et al., 2006), Ireland (Nasr et al., of the currently available and popularly applied water-
2007), and Switzerland (Abbaspour et al., 2007). SWAT shed water quality modelling tools, has not been applied
has been calibrated, validated, and applied for runoff, to evaluate model performances in any of the CLW sub-
sediment yield and total phosphorus losses from water- watersheds. It is important to test the AnnAGNPS and
sheds in different geographic locations, conditions, and SWAT models to compare their performances to select
management practices (Saleh et al., 2000; Spruill et al., an appropriate model for the conservation effects evalua-
2000; Santhi et al., 2001; Kirsch et al., 2002; Van Liew tion of the CLW because choosing an appropriate model
et al., 2003; White et al., 2004; Qi and Grunwald, 2005; might affect BMPs selection decision making.
White and Chaubey, 2005; Wang et al., 2006; Jha et al., The objectives of this research were to (i) compare
2007; Gassman et al., 2007; Parajuli et al., 2007). AnnAGNPS and SWAT model simulation results for sur-
Limited research has been performed comparing face flow, sediment yield, and total phosphorus, and (ii)
AnnAGNPS and SWAT models for predicting monthly determine the most appropriate model for this watershed
flow, sediment yields, and total phosphorus. Sadeghi based on calibration and validation results. This study
et al. (2007), applied AnnAGNPS and SWAT models in is unique in that it compares model performance in two
a coastal plain watershed, Choptank River, Maryland, separate, but similar, watersheds.
which is also a USDA-CEAP watershed. Portions of the
watershed were impaired due to high levels of nutri-
ents and sediment. They used five years (1991–1995) of METHODS AND MATERIALS
detailed observed flow and water quality data to provide
baseline calibration and validation for the two models. In Study area
their study, the calibrated AnnAGNPS and SWAT models Red Rock Creek watershed and Goose Creek water-
performed fair to good (Parajuli, 2007) in terms of model shed, two sub-watersheds within the Cheney Lake water-
correlation coefficient (R2 ) and Nash Sutcliffe efficiency shed located on the North Fork of the Ninnescah River
index (E) when compared with observed data (R2 D 0Ð51, (HUC 11 030 014) in south-central Kansas, were selected
E D 0Ð49 for AnnAGNPS and R2 D 0Ð50, E D 0Ð34 for for this project due to their similar spatial and land
SWAT). They did not report model performance during use characteristics and history of water quality data.
model validation. The sediment yield and total phospho- Cheney Lake, the primary receiving water in the water-
rus losses from the watershed were not evaluated. shed and primary drinking water source for the City of
Das et al. (2007a) compared performances of AnnAG- Wichita, KS, has been identified as ‘impaired waters’
NPS and SWAT models in a watershed in Ontario, under Section 303(d) of the Federal Clean Water Act due
Canada. They used ten years (1991–2000) of monthly to high levels of sediment and total phosphorus trans-
observed flow data to provide baseline calibration port to the lake. The Kansas Department of Health and

Copyright  2008 John Wiley & Sons, Ltd. Hydrol. Process. 23, 748– 763 (2009)
DOI: 10.1002/hyp
750 P. B. PARAJULI ET AL.

Environment (KDHE) has set TMDLs for eutrophication AnnAGNPS Model. The AnnAGNPS model was
and silt for Cheney Lake (KDA, 2004; KDHE, 2004). designed by the USDA Agriculture Research Service
In recent years, a significant number of state and fed- (USDA-ARS) and the USDA Natural Resources Con-
eral incentive programmes have been implemented for servation Service (USDA-NRCS) to evaluate NPS pol-
water quality improvement in the Cheney Lake water- lution from agriculture-dominated watersheds. It is a
shed. This watershed is part of the USDA Conservation batch-process, continuous simulation, daily time step,
Effects Assessment Project (CEAP), which seeks to quan- pollutant-loading model developed to simulate long-
tify environmental benefits of conservation programmes. term runoff, sediment, and chemical transport from
Models were calibrated in the Red Rock Creek water- agricultural watersheds (Cronshey and Theurer, 1998;
shed and validated in the Goose Creek watershed. Bingner and Theurer, 2003). It is a direct replace-
Red Rock Creek watershed (Figure 1), located in Reno ment for the single event model, Agricultural Non-
county, is of area 136 km2 with an average elevation Point Source (AGNPS) (Young et al., 1989), and retains
of 475 m. Land uses in the study area include grass- many features of AGNPS (Yuan et al., 2001). Unlike
land (32%), cropland (63%), woodland (4%), and 1% AGNPS, AnnAGNPS divides the watershed into drainage
others (water, urban). Fine-loamy textured soils are pre- areas with homogenous land use, soils, etc. and inte-
dominant in this watershed. Goose Creek watershed grates these areas by simulated rivers and streams that
(Figure 1), located in Kingman and Reno counties, is of route runoff and pollutants from each area downstream.
area 136 km2 with an average elevation of 505 m. Study The model uses and combines many modules of other
area land uses include grassland (29%), cropland (64%), widely used models, such as Revised Universal Soil
woodland (6%), and 1% others (water, urban). Fine- Loss Equation (RUSLE) (Renard et al., 1997), Chemi-
loamy textured soils are predominant in this watershed. cals, Runoff, and Erosion from Agricultural Management
Systems (CREAMS) (Knisel, 1980), Erosion Productiv-
Model description ity Impact Calculator (EPIC) (Sharpley and Williams,
Both AnnAGNPS and SWAT are daily time step, 1990), and Groundwater Loading Effects on Agricul-
watershed scale, pollutant-loading models developed to tural Management Systems (GLEAMS) (Leonard et al.,
simulate long-term runoff, sediment, nutrients, and pes- 1987).
ticide transport from agricultural watersheds (Table I) AnnAGNPS allows users to select cell-based spatial
These models differ in structure. While the SWAT model representation, which is characterized by similar land and
uses hydrologic response units (HRUs), AnnAGNPS des- soil properties. Soil moisture conditions are calculated
ignates cells of various sizes; pollutants are routed from with the Soil Conservation Service curve numbers (SCS-
these cells into the associated reaches, and the model CN) method, which serves as the basis for determining
either deposits pollutants within the stream channel sys- surface and subsurface runoff quantities. AnnAGNPS
tem or transports them out of the watershed (Geter and uses the RUSLE to calculate sediment delivered to a field
Theurer, 1998). edge as a result of runoff from any type of precipitation.

Figure 1. Location map of the calibrated (Red Rock Creek) and validated (Goose Creek) watersheds

Copyright  2008 John Wiley & Sons, Ltd. Hydrol. Process. 23, 748–763 (2009)
DOI: 10.1002/hyp
 Table I. Comparison of the hydrologic, sediment yield, chemical and BMP modeling of AnnAGNPS and SWAT models

Description AnnAGNPS SWAT References

Copyright  2008 John Wiley & Sons, Ltd.

Model components and Hydrology, sediment yield, nutrients, Hydrology, sediment yield, nutrients, pesticides, Borah and Bera, 2003
capabilities: pesticides snowmelt, rainfall, irrigation, snow melt, channel and reservoir routing, crop
and capabilities to generate cells and growth, soil temperature, pathogen transport,
stream networks using TOPAZ stream processes with ArcView GIS plateform
Scale: Long-term, daily, or sub-daily steps Long-term, daily, or sub-daily steps
Watershed Homogeneous landareas (cells), reaches, and Sub-basins grouped based on climate, hydrologic
representation: impoundments response units, ponds, ground water, and main
channel
Input weather data: Precipitation, maximum and minimum Precipitation, maximum and minimum temperature, Migliaccio and Srivastava, 2007
temperature, dew point temperature, sky relative humidity, wind speed, and solar radiation
cover, and wind speed
Weather simulator: GEM and complete climate (Bingner et al., WXGEN (Sharpley and Williams, 1990) Migliaccio and Srivastava, 2007
2007)
PET method: Penman (Jenson et al., 1990) Penman-Monteith, Priestly-Taylor, and Hargreaves Migliaccio and Srivastava, 2007
(Monteith, 1965; Allen, 1986; Allen et al., 1989;
Priestley and Taylor, 1972; Hargreaves et al.,
1985)
Infiltration/surface Modified SCS CN2 Modified SCS CN2 with daily time step or King et al., 1999
runoff algorithms: Green-Ampt Mein-Larson infiltration equation
(Mein and Larson, 1973)
Peak runoff rate: SCS TR-55 method (Bingner et al., 2007) Modified rational and SCS TR-55 method s Borah and Bera, 2003
(Neitsch et al., 2005)
Groundwater and Subsurface flow or Interflow - Yes Subsurface flow or Interflow - Yes Migliaccio and Srivastava, 2007
stream flow
simulation:
ANNAGNPS AND SWAT MODEL SIMULATION RESULTS IN AGRICULTURAL WATERSHEDS

Subsurface drainage or tile drainage - Yes Subsurface drainage or tile drainage - Yes
Shallow aquifer or other water storage zone - Shallow aquifer or other water storage zone - Yes
No
Deep aquifer - No Deep aquifer - Yes
Subsurface flow: Lateral subsurface flow using Darcey’s Lateral subsurface flow using kinematic storage Borah and Bera, 2003
equation or tile drain flow using model (Sloan et al., 1983), and groundwater
Hooghoudt’s equation and parallel drain groundwater flow using empirical relations
approximation

DOI: 10.1002/hyp
Hydrol. Process. 23, 748– 763 (2009)
751
 Water routing/channel: Manning’s equation and channel shape Manning’s equation and variable storage routing or Migliaccio and Srivastava, 2007
752
streams/reaches: relationships Muskingum river routing method
Channel for runoff: Trapezoid channel cross-section, rectangular Trapezoidal channel cross-section and trapezoidal Migliaccio and Srivastava, 2007
main channel, rectangular out-of-bank flood plain
(floodplain) section
Sediment Yield: Uses RUSLE to generate sheet and rill Sediment yield based on Modified Universal Soil Borah and Bera, 2003
Overland erosion, HUSLE (Theurer and Clarke, Loss Equation (MUSLE) expressed in terms of
1991) for delivery ratio, and sediment runoff volume, peak flow, and USLE factors
deposition based on particle size (Neitsch et al., 2005)
distribution (Young et al., 1987) and
particle fall velocity, ephemeral gully
erosion model (Gordon et al., 2007)
Channel Modified Einstein equation for sediment Bagnold’s stream power concept for bed Borah and Bera, 2003
transport and Bagnold equation to degradation and sediment transport, adjusted with

Copyright  2008 John Wiley & Sons, Ltd.

determine transport capacity of flow. USLE soil erodibility and cover factors, and
deposition based on particle (Bagnold, 1977)
Bagnold, 1966; Theurer and Cronshey, 1998)
Chemical Simulation: Soil moisture, nutrients, and pesticides in Runoff P based on partitioning factor daily organic Borah and Bera, 2003
each cell are tracked using NRCS soil N and sediment adsorbed functions, crop N and
database and crop information and reach P use from supply and demand, and pesticides
routing includes fate and transport of based on plant leaf-area index, application
nitrogen, phosphorus, and individual efficiency, wash off fraction, organic C
pesticides, and organic carbon (Binger adsorption, and half-life decay (Neitsch et al.,
et al., 2007) 2005)
Best Management Agricultural managements: impact of Agricultural managements: impact of watershed Borah and Bera, 2003
Practices (BMPs) watershed management practices on runoff management practices on runoff and sediment
Evaluation: and sediment BMPs include: vegetative losses, nutrients, pesticides, pathogen transport
filter strip, terracing, contouring, strip BMPs include: grass waterway, grade
P. B. PARAJULI ET AL.

cropping, tillage operation, cover crops, stabilization structure, vegetative filter strip,
crop rotations, tile drainage, irrigation, rice terracing, contouring, strip cropping, tillage
fish farming (Yuan et al., 2001; Yuan operations, cover crops, crop rotations, grazing
et al., 2006; Yuan et al., 2007) operation, pesticide application, irrigation,
nutrient management (Saleh et al., 2000; Gitau
et al., 2003; Bosch et al., 2005; Chu et al., 2005;
Bockhold et al., 2006; Bracmort et al., 2006;
Gitau et al., 2007; Parajuli, 2007)

DOI: 10.1002/hyp
Hydrol. Process. 23, 748–763 (2009)
 ANNAGNPS AND SWAT MODEL SIMULATION RESULTS IN AGRICULTURAL WATERSHEDS 753

The Hydro-geomorphic Universal Soil Loss Equation Model input

(HUSLE) is used to estimate the total sediment yield Both AnnAGNPS and SWAT use various sets of
leaving each field and entering the stream reach while geospatially referenced data to create layers of infor-
accounting for deposition. Output is expressed on an mation to satisfy the necessary input parameters. United
event basis for selected stream reaches and as source State Geological Survey (USGS, 1999) 7Ð5-minute digital
accounting (contribution to outlet) from land or reach elevation data (DEM) was used to delineate watershed
components over the simulation period. boundaries and topography. The Soil Survey Geographic
Database (SSURGO) was used to create a soil database
(USDA, 2005). Land use and land management were
SWAT model. The SWAT model is a product of estimated by analysing Landsat 5 satellite imagery using
USDAnARS. SWAT is a physically based, semi- stacked images from May and August of 1997 for major
distributed parameter, watershed-scale model that oper- crop types and unsupervised classification techniques
ates on a continuous daily time step. It simulates within ArcView Image Analysis with ground truth verifi-
hydrological processes, sediment yield, nutrient loss, and cation using Farm Service Agency records. Image Anal-
pesticide losses into surface/groundwater and the effects ysis in ArcView also is capable of performing the Nor-
of agricultural management practices on water in large malized Difference Tillage Index (NDTI) (band 5-band
ungauged watersheds (Arnold et al., 1998). SWAT incor- 7)/band 5Cband 7) function using Landsat 5 mid-infrared
porates the effects of weather, surface runoff, evapo- bands 5 and 7. Once the NDTI function was completed,
transpiration, crop growth, irrigation, groundwater flow, results were separated into three crop residue covers:
nutrient loading, pesticide loading, and water routing, high, medium and low. Using this information paired
as well as the long-term effects of varying agricul- with local knowledge, land use and land management
tural management practices (Neitsch et al., 2002, 2005). were classified into 24 classes with major land uses
In the hydrologic component, runoff is estimated sepa- including wheat, soybean, grain sorghum, corn, CRP,
rately for each sub-watershed of the total watershed area forestland, pastureland, rangeland, urbanland, and water
and routed to obtain the total runoff for the watershed. (Lyle Frees, 1997, unpublished data). AnnAGNPS input
Runoff volume is estimated from daily rainfall using accepts five types of land use identifiers (cropland, pas-
modified SCS-CN and Green–Ampt methods. Sediment ture, forest, rangeland and urban), and only the pre-
yield is estimated from the Modified Universal Soil Loss dominant land use and management are used to repre-
Equation (MUSLE). SWAT has been applied extensively sent each AnnAGNPS cell. Hence, the land uses in the
for streamflow, sediment yield, and nutrient modelling sub-watershed were reclassified according to the model
(Gosain et al., 2005; Vache et al., 2002; Varanou et al., requirement. Each land use type was included under a
2002). The model requires input of DEM, land use, and land use identifier during input data preparation.
soils, as well as time series of climate data such as daily For the AnnAGNPS model, a threshold critical source
precipitation and temperature. area (CSA) of 100 ha and a minimum source channel
In the SWAT model, the watershed is partitioned into length (MSCL) of 130 m were used to generate spatially
small sub-basins that are further subdivided into HRUs variable stream network parameters. AnnAGNPS divided
based on unique land cover, soil, and topographic con- the calibrated watershed into 169 cells. A MSCL of
ditions. Such divisions are necessary to consider accu- 130 m is a default value given in the model, but the
rately possible effects of spatial variations in parameters CSA value was changed to 100 ha, which represents less
on hydrological processes, sediment, and nutrient sim- than 1% of the calibration watershed area (13 600 ha
ulations. The hydrology component of the model cal- or 136 km2 ) as similar to the SWAT model threshold
culates a soil water balance at each time step based area. The dominant soil and land use for each cell were
on daily amounts of precipitation, runoff, evapotranspi- determined from the soil and land use shape files over the
ration, percolation, and baseflow. Sediment yield from delineated sub-watershed. For the SWAT model, default
each sub-basin or HRU is computed with the MUSLE. threshold critical source area was used, which divided
The MUSLE approach of estimating sediment yield the calibration watershed into 12 sub-basins. The SWAT
makes the sediment computation a non-linear function model further considered spatial variability of parameters
of the HRU area. Simulations are performed at the HRU within each sub-basin by means of HRUs.
level and summarized in each sub-watershed. The sim-
ulated variables (water, sediment, nutrients, and other Model calibration
pollutants) are routed through the stream network to Flow, and sediment yield-related model calibration
the watershed outlet. SWAT evolved from the Simu- parameters were selected (Table II) based on previous
lator for Water Resources in Rural Basins (SWRRB) research (Das et al., 2007a; Parajuli et al., 2007; Sadeghi
and Routing Outputs to Outlet (ROTO) models. Other et al., 2007; Wang et al., 2007). Flow calibration was
models also influenced SWAT development including performed by adjusting the CN parameter. CN is a
CREAMS (Knisel, 1980), GLEAMS (Leonard et al., soil moisture balance parameter that allows the model
1987), and EPIC (Williams et al., 1984; Neitsch et al., to modify the soil moisture condition of the soil to
2002). estimate surface runoff. As the value of CN is reduced,

Copyright  2008 John Wiley & Sons, Ltd. Hydrol. Process. 23, 748– 763 (2009)
DOI: 10.1002/hyp
754 P. B. PARAJULI ET AL.

Table II. Model parameter test and adjustment during calibration

Parameters Default value Test range value Final value

Flow:
Curve Number (CN) in 73–83 73–83 77–79
AnnAGNPS and SWAT Cropland 83 74–83 78
Grassland 79 75–82 79
Woodland 73 73–80 77
Sediment:
USLE cover and management factor (C) in SWAT 0Ð03–0Ð20 0Ð03–0Ð50 0Ð03–0Ð20
Winter wheat 0Ð03–0Ð30 0Ð03
Grain Sorghum 0Ð20–0Ð50 0Ð2
Soybean 0Ð20–0Ð50 0Ð2
Corn 0Ð20–0Ð50 0Ð2
Manning’ss n EGs1 in AnnAGNPS 0Ð04 0Ð04–1Ð00 0Ð04

1 Ephemeral Gully.

the model allows less water to runoff from the surface. SWAT allows users to input initial SCS-CN for mois-
Other parameters than CN factor may also effect flow ture condition II in the management data. AnnAGNPS
prediction but their effect is considered to be small; many allows users to input all CNs based on the soil hydrologic
studies ranked CN as the most sensitive parameter (Arabi group (A, B, C, and D). Initial AnnAGNPS model predic-
et al., 2007; Das et al., 2007a; Feyereisen et al., 2007; tions were made with the default RCN factor (retention
Parajuli et al., 2007; Sadeghi et al., 2007; Wang et al., calibration factor) of 1Ð00, after which the model was
2007). Red Rock Creek and Goose Creek contribute low run automatically nine times using different RCN factors
flow in various seasons, although they are considered (1Ð000–2Ð789).
perennial streams. Baseflow separation analysis (Kyoung The USLE C, default was given as 0Ð03 for winter
et al., 2005; Eckhardt, 2005) in the Red Rock Creek and wheat and 0Ð20 for grain sorghum, soybean, and corn
Goose Creek watersheds indicated about 1Ð2% (weighted crops. During calibration, the C factor was tested in the
average) base flow of the total direct flow. Other studies range 0Ð03 to 0Ð30 for winter wheat and 0Ð20 to 0Ð50 for
confirmed similar results (Li et al., 2006) for these grain sorghum, soybean, and corn crops. Values above
watersheds. the default values were tested because the models were
Models were calibrated for sediment yield by adjusting underpredicting sediment yield. However, increasing the
the Universal Soil Loss Equation crop cover management C factors did not increase model efficiency for sediment
factor (USLE C), one of the most widely used sediment yield prediction.
calibration factors (Parajuli, 2007). The USLE C factor is Therefore, the default C factor given in the model
defined as the ratio of soil loss from land cropped under resulted in the best model efficiency, in this study. The
specified conditions to the corresponding loss from clean- default USLE support practice factor (P) given in the
tilled, continuous fallow land (Wischmeier and Smith, models (1Ð00) was tested in the range from 0Ð20 to 1Ð00
1978). during model calibration. Models always underpredicted
Monthly flow, sediment yield, and total phosphorus sediment yields using both lower and upper P factors. The
data measured from the USGS gauge station for each P factor was fixed at 0Ð50, which generally represents
watershed were used to calibrate and validate the model. the current conditions of the watersheds. Terraces are
Monthly measured data from January 1997 to Septem- one of the conservation practices implemented in the
ber 2000 were used for model calibration and validation. watersheds. P is defined as the ratio of soil loss with
Model predictions were evaluated statistically with the a specific support practice to the corresponding loss with
coefficient of determination (R2 ) and Nash–Sutcliffe Effi- up-and-down slope culture. Support practices include
ciency Index (E) between measured values and model- contour tillage, strip cropping on the contour, and terrace
predicted values after each model run with changed systems. Stabilized waterways for the disposal of excess
parameters. Model input parameters were continuously rainfall are a necessary part of each of these practices
modified during the calibration phase until simulated flow (Wischmeier and Smith, 1978). No calibration parameters
were used to calibrate total phosphorus prediction. After
and sediment yield gave results R2 ½ 0Ð5 and E ½ 0Ð5
model calibration, input parameters were not changed
(Ramanarayanan et al., 1997; Parajuli et al., 2006). Flow
during the model validation process.
calibration was determined first using all default param-
eters. The CN parameters were continuously modified
within the specified range of values during the calibra- Management scenarios
tion phase to find the local optimum value. The CN range
of 77–79 (77 for woodland, 78 for cropland, and 79 for Land in the CRP covers about 16% of the calibra-
grassland) resulted in the best calibration for both models. tion watershed area. The CRP land (high, high terrace,

Copyright  2008 John Wiley & Sons, Ltd. Hydrol. Process. 23, 748–763 (2009)
DOI: 10.1002/hyp
 ANNAGNPS AND SWAT MODEL SIMULATION RESULTS IN AGRICULTURAL WATERSHEDS 755

medium, medium terrace, low, low terrace) was simu- weather station to its sub-watershed whereas AnnAG-
lated with five typical types of grass management: lit- NPS uses a thiessen polygon average of the weather
tle bluestem, big bluestem, Indiangrass, side oats, and data. Daily precipitation data for the calibration water-
switchgrass, which represent the field conditions. These shed was obtained from the Hutchinson south weather
five types of grasses have about equal cover in the station, which is located nearly in the middle of the water-
watershed. Grassland covers about 20% of the watershed shed. The SWAT model fills in missing weather data with
area and typically includes rangeland big bluestem (high, data from the Wichita airport weather station (Sedgwick
medium, low). The CRP grasses are generally not fer- County), which is located about 55 km southeast of the
tilized (Lisa French, Cheney Lake Watershed Inc., 2007, calibration watershed.
personal communication). Daily precipitation data for the validation watershed
A majority (¾64%) of the land use area in both water- were from USGS gauge 3 and Turon weather stations.
sheds is cropland. Grain sorghum and soybean are major The USGS gauge 3 weather station is located at the mouth
warm-season crops, and winter wheat is a primary cool- of the Goose Creek watershed and Turon is located about
season crop grown in a four-year rotation (Lisa French, 10 km from the watershed. To fill in missing data for the
Cheney Lake Watershed Inc., 2007, personal communi- validation watershed, the SWAT model used data from the
cation). Typical planting and harvesting dates are 25 May Pratt weather station (Pratt County), located about 31 km
and 20 October for warm-season crops and 20 October south-west of the watershed. The 1997 to 2000 average
and 29 June for cool-season crops. Crop residue is left on annual rainfall data measured for Hutchinson south was
the ground between the crop periods. Sorghum, soybean, 813 mm (Figure 2a) and for USGS gauge 3 and Turon
and wheat are cultivated primarily with conventional and was about 662 mm (Figure 2b). USGS gauge 3 weather
conservation tillage systems. Primary herbicides used for station had no rainfall data measured from October to
warm-season crops are Bicep II Magnum for sorghum December, 2000.
and Roundup for soybean; Finesse was used for win-
ter wheat. These methods apply to both calibrated and Statistical analysis
validated watersheds. Woodlands cover about 4% of the SWAT and AnnAGNPS model responses for flow were
calibration watershed and about 6% of the validation evaluated based on measured flow data from January
watershed land-use area. Model default parameters were 1997 to December 2000. This study used 45 months
used for woodland areas assuming mixed forest trees in of rainfall–runoff measured/observed events for both
the watersheds. watersheds. Statistical parameters used to evaluate the
relationship between measured and predicted flow, sedi-
Weather and hydrologic data ment yield, and total phosphorus include correlation coef-
ficient (R2 ), Nash–Sutcliffe efficiency index (E), root
Weather data, such as daily precipitation and daily mean square error (RMSE), RMSE–observations stan-
ambient temperatures, were extracted from the National dard deviation ratio (RSR), and percentage bias (PBIAS).
Climatic Data Center (NCDC). The SWAT model requires The R2 value indicates the consistency with which mea-
daily precipitation, daily maximum and minimum tem- sured versus predicted values follow a best fit line. If the
peratures, daily solar radiation, daily wind speed, and R2 values are less than or very close to zero, the model
daily relative humidity. The AnnAGNPS model requires prediction is considered unacceptable or poor. If the val-
daily precipitation, daily maximum and minimum tem- ues are one, then the model prediction is perfect (Santhi
peratures, daily dew point temperatures, daily sky cover, et al., 2001). However, R2 only describes how much of
and daily wind speed data. Daily precipitation data were the observed dispersion is explained by the prediction,
used from only one weather station (Hutchinson South) therefore R2 is not used alone.
for the calibration watershed and from two weather sta- E indicates the consistency with which measured
tions (USGS gauge 3 and Turon) for the validation values match predicted values, or the fit of the data to a
watershed. SWAT uses weather data from the nearest linear 1 : 1 measured versus predicted best-fit line (Nash

(a) 120 (b) 120

90
Rainfall (mm)

90
Rainfall (mm)

60 60

30 30

0 0
1/1/97
5/1/97
9/1/97
1/1/98
5/1/98
9/1/98
1/1/99
5/1/99
9/1/99
1/1/00
5/1/00
9/1/00

1/1/97
5/1/97
9/1/97
1/1/98
5/1/98
9/1/98
1/1/99
5/1/99
9/1/99
1/1/00
5/1/00
9/1/00

Daily rainfall (month/day/year) Daily rainfall (month/day/year)

Figure 2. Daily rainfall data for (a) Red Rock Creek and (b) Goose Creek watersheds

Copyright  2008 John Wiley & Sons, Ltd. Hydrol. Process. 23, 748– 763 (2009)
DOI: 10.1002/hyp
756 P. B. PARAJULI ET AL.

and Sutcliffe, 1970) and estimated using Equation (1). E where RMSE D root mean squared error, Qpi D predicted
ranges from minus infinity (poor model) to 1Ð0 (perfect value for event i, Qoi D observed value for event i, and
model). For example, if the square of the differences n D number of events.
between the model predictions and the observations is RSR is estimated as the ratio of the RMSE and
as large as the variability in the observed data, then standard deviation of measured data using Equation (3).
E D 0Ð0; if it exceeds it, then E < 0Ð0 (i.e. the observed RSR varies from the optimal value of 0, which indicates
mean is better than the predictor). Thus, a value of zero RMSE or residual variation and therefore perfect
zero for E indicates that the observed mean, O, is as model simulation, to a large positive value. The lower
good as a predictor as the model, while negative values RSR, the lower the RMSE, and the better the model
indicate that the observed mean is a better predictor simulation performance (Moriasi et al., 2007).
than the model. E has been widely used to evaluate RMSE
the performance of hydrologic models (Wilcox et al., RSR D
3
1990). A limitation of E is the fact that the differences STDEVobs
between the observed and predicted values are calculated where RSR D RMSE–observations standard deviation
as squared values. As a result larger values in a time ratio, RMSE D root mean squared error, and STDEVobs
series are strongly overestimated whereas lower values D standard deviation of measured data.
are neglected (Legates and McCabe, 1999). PBIAS measures the average tendency of the model-
predicted values to be larger or smaller than their


n

Oi
Pi 2 corresponding measured values. The optimal value of
iD1
PBIAS is 0Ð0, with low-magnitude values indicating
ED1

1 accurate model simulation. Positive values indicate model


n
2 underestimation bias, and negative values indicate model

Oi
O
iD1 overestimation bias (Gupta et al., 1999). PBIAS is the
where O D observed value and P D predicted value. The deviation of data values being evaluated, expressed as a
over-bar denotes the mean (observed or predicted) for the percentage, which is calculated from
 n 
entire time period of the evaluation.
 
obs sim
RMSE summarizes the average error between observed  Yi
Yi ð
100 
 iD1 
and predicted variates using the same units as those PBIAS D 


4
variates, and is estimated using Equation (2). The lower 
n

obs

Yi 
RMSE the better the model performance, and a value of
iD1
zero represents perfect simulation of the observed data
(Chu and Shirmohammadi, 2004). RMSE summarizes the where PBIAS D percentage bias, Yobs i D observed value
mean difference in the units of observed and predicted for event i, and Ysim
i D predicted value for event i.
values. The use of absolute error measures using RMSE In this study, statistics using R2 , E, RMSE, RSE, and
provides an evaluation of the error in the units of the PBIAS values were considered. Model correlations and
variable. The RMSE indicates the bias (deviation of the efficiencies, as modified by Parajuli (2007) from Moriasi
actual slope from the 1 : 1 line) compared with the random et al. (2007), were classified as in Table III.
variation that may occur (Willmott, 1984). RMSE is
biased when large outliers are present.
 RESULTS AND DISCUSSION
 n

 2
 Qpi
Qoi Flow

 iD1 Calibrated models for the Red Rock Creek water-
RMSE D
2
n shed predicted mean monthly flow of the watershed

Table III. Classification of model efficiencies for the different pollutant parameters

Class R2 , E RSR PBIAS PBIAS PBIAS

Flow, sediment, TP Flow, sediment, TP Flow Sediment TP

Excellent <0Ð90 0Ð00–0Ð25 < š 10 < š 15 < š 25

Very good 0Ð75–0Ð89 0Ð26–0Ð50 š11 š 15 š16 š 30 š26 š 40
Good 0Ð50–0Ð74 0Ð51–0Ð60 š16 š 25 š31 š 50 š41 š 60
Fair 0Ð25–0Ð49 0Ð61–0Ð70 š26 š 30 š51 š 60 š61 š 70
Poor 0Ð00–0Ð24 0Ð71–0Ð89 š31 š 35 š61 š 70 š71 š 80
Unsatisfactory <0Ð00 >0Ð90 ½ š 36 ½ š 71 ½ š 81

R2 D Coefficient of determination.
E D Nash sutcliffe efficiency index.
TP D Total phosphorus.
RSR D Root mean square error - observations standard deviation ratio.
PBIAS D Percent bias.

Copyright  2008 John Wiley & Sons, Ltd. Hydrol. Process. 23, 748–763 (2009)
DOI: 10.1002/hyp
 ANNAGNPS AND SWAT MODEL SIMULATION RESULTS IN AGRICULTURAL WATERSHEDS 757

with very good correlation and good agreement (R2 D both models (R2 D 0Ð50, E D 0Ð47, RMSE D 0Ð26 for
0Ð80, E D 0Ð69, RMSE D 0Ð38 for AnnAGNPS; R2 D AnnAGNPS; R2 D 0Ð62, E D 0Ð48, RMSE D 0Ð25 for
0Ð81, E D 0Ð56, RMSE D 0Ð45 for SWAT) between mean SWAT) (Table IV). The AnnAGNPS model predicted
monthly measured and mean monthly predicted flow val- monthly flow (m3 s
1 ) estimated about 4% higher RMSE
ues (Table IV, Figure 3). The SWAT model predicted than SWAT in the validation watershed (Table IV). The
monthly flow (m3 s
1 ) estimated about 18% higher estimated RSR values for both AnnAGNPS and SWAT
RMSE than AnnAGNPS in the calibration watershed. model simulations during calibration and validation were
The calibrated models, when applied to the Goose excellent (0Ð09–0Ð12). The calculated PBIAS values
Creek watershed for validation, predicted mean monthly for both models were unsatisfactory (PBIAS from 49
flow with good correlation and fair agreement for to
95), which means both models were biased to

Table IV. Estimated statistical parameters of model performance for calibration and validation watersheds

Class Calibration Validation

Slope Intercept R2 E RMSE RSR PBIAS Slope Intercept R2 NSE RMSE RSR PBIAS

AnnAGNPS:
Flow 0Ð54
0Ð01 0Ð80 0Ð69 0Ð38 0Ð09 49Ð83 0Ð65 0Ð17 0Ð50 0Ð47 0Ð26 0Ð12
40Ð35
Sediment 0Ð44
3Ð64 0Ð83 0Ð60 230Ð00 0Ð10 57Ð70 0Ð70 43Ð67 0Ð62 0Ð64 312Ð00 0Ð19
67Ð14
TP 0Ð90 385Ð08 0Ð60 0Ð32 704Ð00 0Ð13
128Ð48 2Ð38
55Ð50 0Ð77
2Ð38 476Ð00 0Ð25
117Ð33
SWAT:
Flow 1Ð08 0Ð25 0Ð81 0Ð56 0Ð45 0Ð10
95Ð06 0Ð86 0Ð13 0Ð62 0Ð48 0Ð25 0Ð11
46Ð62
Sediment 0Ð56
1Ð41 0Ð89 0Ð73 186Ð00 0Ð08 45Ð02 0Ð40 46Ð58 0Ð72 0Ð61 158Ð00 0Ð10 9Ð65
TP 0Ð51 126Ð88 0Ð70 0Ð68 487Ð00 0Ð09 3Ð13 0Ð64 33Ð34 0Ð60 0Ð63 178Ð00 0Ð09 13Ð44

R2 D Coefficient of determination.
E D Nash sutcliffe efficiency index.
RSR D Root mean square error - observations standard deviation ratio.
TP D Total phosphorus.
PBIAS D Percent bias.

(a) 5.0
Measured/predicted flow (m3 s-1)

SWAT AnnAGNPS Measured

4.0

3.0

2.0

1.0

0.0
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45
Time period Jan 1997 - Sept. 2000 months (1 - 45)
(b) 3.0
Measured/predicted flow (m3 s-1)

SWAT AnnAGNPS Measured

2.0

1.0

0.0
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45
Time period Jan 1997 - Sept. 2000 months (1 - 45)
Figure 3. Time series measured and models predicted monthly average flow from January 1997 to September 2000 for (a) Red Rock Creek and
(b) Goose Creek watersheds

Copyright  2008 John Wiley & Sons, Ltd. Hydrol. Process. 23, 748– 763 (2009)
DOI: 10.1002/hyp
758 P. B. PARAJULI ET AL.

underpredict or overpredict flow for different months found R2 D 0Ð68 and E D 0Ð84 for mean monthly flow
of the simulation period. AnnAGNPS flow prediction prediction. Spruill et al. (2000) applied the SWAT model
was determined to be less biased than SWAT during in a small, central Kentucky watershed. The SWAT model
model calibration (PBIAS: 50 versus
95) and validation prediction had E values for monthly flows between 0Ð58
(PBIAS:
40 versus
46) study. and 0Ð89. They did not use other statistical parameters as
Although, AnnAGNPS and SWAT models predicted used in this study, such as RMSE, RSR, and PBIAS.
surface runoff equally well in general for both calibration
and validation watersheds, AnnAGNPS slightly under-
Sediment yield
predicted monthly surface flow compared with SWAT
during model calibration (slope: 0Ð54 versus 1Ð08) and Calibrated AnnAGNPS and SWAT models for Red
validation (slope: 0Ð65 versus 0Ð86). Model-predicted val- Rock Creek watershed predicted monthly sediment yield
ues of estimated slopes during model calibration and with very good correlation and good agreement (R2 D
validation indicated no-significant (P values <0Ð001 at 0Ð83, E D 0Ð60, RMSE D 230 for AnnAGNPS; R2 D
˛0Ð05 ) difference. 0Ð89, E D 0Ð73, RMSE D 186 for SWAT) with mea-
Bhuyan et al. (2003) applied a single event-based sured mean monthly sediment yield data (Table IV,
AGNPS model in the Red Rock Creek watershed. The Figure 4). The AnnAGNPS model-predicted monthly
single event-based AGNPS model systematically overes- sediment yield (Mg) estimated about 23% higher RMSE
timated surface runoff, and the model had to be adjusted than SWAT in the calibration watershed.
for antecedent moisture condition (AMC) based on sur- Both models showed decreased, but still good, cor-
face runoff prediction. Compared with predictions by relation and agreement (R2 D 0Ð62, E D 0Ð64, RMSE D
Bhuyan et al. (2003), model predictions from the contin- 312 for AnnAGNPS; R2 D 0Ð72, E D 0Ð61, RMSE D 158
uous version of AnnAGNPS used in this study had better for SWAT) between mean monthly measured and mean
correlation with observed values. Van Liew et al. (2003) monthly predicted sediment yield values during vali-
applied the SWAT model in the Delaware Creek water- dation in the Goose Creek watershed (Table IV). The
shed in Oklahoma, which has land-use conditions similar AnnAGNPS model predicted monthly sediment yield
to Red Rock Creek and Goose Creek watersheds, and (Mg) estimated about 97% higher RMSE than SWAT

(a) 2000
SWAT AnnAGNPS Measured
Measured/predicted sediment

1500
yield (Mg)

1000

500

0
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45
Time period Jan 1997 - Sept. 2000months (1 - 45)
(b) 2400
SWAT AnnAGNPS Measured
Measured/predicted sediment

1800
yield (Mg)

1200

600

0
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45
Time period Jan 1997 - Sept. 2000 months (1 - 45)
Figure 4. Time series measured and models predicted monthly average sediment yield from January 1997 to September 2000 for (a) Red Rock Creek
and (b) Goose Creek watersheds

Copyright  2008 John Wiley & Sons, Ltd. Hydrol. Process. 23, 748–763 (2009)
DOI: 10.1002/hyp
 ANNAGNPS AND SWAT MODEL SIMULATION RESULTS IN AGRICULTURAL WATERSHEDS 759

in the validation watershed (Table IV). The RSR val- Total phosphorus
ues were estimated excellently when both AnnAGNPS
and SWAT model simulation results for the calibration The AnnAGNPS and SWAT models showed good
and validation watersheds were analysed. The estimated correlation and fair to good agreement (R2 D 0Ð60,
PBIAS values using AnnAGNPS model simulation E D 0Ð32, RMSE D 704 for AnnAGNPS; R2 D 0Ð70,
ranged from fair to poor (PBIAS from 57 to
67) E D 0Ð68, RMSE D 487 for SWAT) between predicted
whereas SWAT performance ranged from excellent to and measured mean monthly total phosphorus during
good (PBIAS from 9 to 45) during model calibration and model calibration (Table IV, Figure 5). The AnnAGNPS
validation. model-predicted monthly total phosphorus (Kg) estimated
Although AnnAGNPS and SWAT models predicted about 44% higher RMSE than SWAT in the calibra-
sediment yield equally well in general, AnnAGNPS tion watershed. When both models were applied to the
slightly underpredicted sediment yield compared with Goose Creek watershed for validation, the AnnAGNPS
SWAT during calibration (slope: 0Ð44 versus 0Ð56), and model predicted total phosphorus with good correlation
SWAT slightly underpredicted sediment yield compared but unsatisfactory agreement (R2 D 0Ð77, E D
2Ð38,
with AnnAGNPS during model validation (slope: 0Ð40 RMSE D 476) for mean monthly total phosphorus. The
versus 0Ð70). However, model-predicted values for esti- SWAT model consistently predicted total phosphorus
mated slopes during model calibration and validation with good correlation and agreement (R2 D 0Ð60, E D
indicated no-significant (P values <0Ð001 at ˛0Ð05 ) dif- 0Ð63, RMSE D 178) with measured mean monthly total
ference. phosphorus values (Table IV). The AnnAGNPS model
Yuan et al. (2001) evaluated AnnAGNPS in the Mis- predicted monthly total phosphorus (kg) estimated about
sissippi Delta MSEA watershed. They used three years 167% higher RMSE than SWAT in the validation water-
of measured data to compare model-predicted sediment shed (Table IV). The estimated RSR values using both
yield from the watershed. The model-predicted monthly AnnAGNPS and SWAT models were excellent (RSR
sediment yield showed good correlation (R2 D 0Ð50) from 0Ð09 to 0Ð25). The calculated PBIAS values for the
with measured data. Das et al. (2007a), applied the AnnAGNPS model simulation results for both calibration
AnnAGNPS model in the Grand River Basin in Canada. and validation study were unsatisfactory with overpredic-
Their results demonstrated good agreement (E D 0Ð53) tion bias (PBIAS from
117 to
128) whereas the SWAT
between mean monthly observed and mean monthly model demonstrated excellent performance with PBIAS
model-predicted sediment yield values during the model values from 3 to 13.
calibration period. During the model validation period, Several previous studies determined that the AnnAG-
model efficiencies decreased (E D 0Ð35). NPS model overpredicted total phosphorus loss (Baginska
Santhi et al. (2001) calibrated and validated the SWAT et al., 2003; Yuan et al., 2005; Das et al., 2007b). Yuan
model in the Bosque River watershed in Texas. The et al. (2005), evaluated the AnnAGNPS model in the
calibrated SWAT model showed E values ranging from Deep Hollow watershed of the Mississippi Delta Man-
0Ð69 to 0Ð80 for monthly sediment yield. However, the agement Systems Evaluation Area project. Sensitivity
validated model had decreased E values (0Ð23 to 0Ð70) analyses of the phosphorus component in the model were
for monthly sediment prediction compared with measured evaluated. AnnAGNPS overpredicted dissolved phospho-
data. Kirsch et al. (2002) calibrated the SWAT model rus loss (121%), still the model correlation of the simu-
in the Rock River Basin watershed in Wisconsin. The lated monthly total phosphorus was very good compared
calibrated SWAT model for Yahara and Mendota sub- with the observed total phosphorus (R2 of 0Ð81), model
watersheds in the Rock River Basin had E values of 0Ð75 agreement was not reported. Das et al. (2007b) evalu-
for annual sediment prediction compared with measured ated the nutrient component of the AnnAGNPS model in
sediment data. They did not have enough sediment data a watershed in Ontario, Canada. The model overpredicted
for model validation. Jha et al. (2007), applied the SWAT total phosphorus, which was reported to have very good
model in the Raccoon River watershed in Iowa. The correlation and fair agreement (R2 D 0Ð82, E D 0Ð23).
SWAT model predicted sediment yield with good to None of these studies validated phosphorus prediction of
very good correlation and agreement (R2 D 0Ð55, E D the AnnAGNPS model in either another watershed or for
0Ð53) during the model calibration period based on the a different simulation period. Yuan et al. (2005) reported
performance ratings of Parajuli (2007). SWAT predictions that all the forms and processes involved in the phospho-
improved during the model validation period (R2 D 0Ð80, rous cycle are not completely and scientifically simulated
E D 0Ð78). All of these studies calibrated and validated by the AnnAGNPS model. Although the model calibra-
models in the same watershed using different periods tion might improve some statistics of model output for
of measured data using only two statistical parameters total phosphorus, further research and in-depth investiga-
(R2 and E). In the present study, which used separate tion on fate and transport of the phosphorus component of
calibration and validation watersheds but the same time the AnnAGNPS model is necessary (Yuan et al., 2005;
period, model performance in the calibration watershed Das et al., 2007b). Our study found overprediction of
was similar to or better than in the validation watershed total phosphorus, which was attributed mainly to soluble
and used five statistical parameters. phosphorus loss.

Copyright  2008 John Wiley & Sons, Ltd. Hydrol. Process. 23, 748– 763 (2009)
DOI: 10.1002/hyp
760 P. B. PARAJULI ET AL.

(a) 5000
SWAT AnnAGNPS Measured

Measured/predicted TP (Kg)
4000

3000

2000

1000

0
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45
Time period Jan 1997- Sept. 2000 months (1 - 45)
(b) 4400
SWAT AnnAGNPS Measured
Measured/predicted TP (Kg)

3300

2200

1100

0
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45
Time period Jan 1997 - Sept. 2000 months (1 - 45)
Figure 5. Time series measured and models predicted monthly average total phosphorus (TP) from January 1997 to September 2000 for (a) Red
Rock Creek and (b) Goose Creek watersheds

Santhi et al. (2001) calibrated and validated the SWAT organic phosphorus and insoluble inorganic phosphorus
model in the Bosque River watershed in Texas. The cal- are associated with the clay fraction of the soil and that
ibrated SWAT model showed good agreement with E soil erosion is the mechanism transporting them from the
values ranging from 0Ð53–0Ð70, for monthly mean total soil profile to the water bodies. However, studies show
phosphorus compared with mean monthly measured data. that insoluble inorganic phosphorus, such as particulate
The validated model had fair to good agreement, with phosphorus, could be lost through biological processes
E values ranging from 0Ð39–0Ð72, for mean monthly (Stuck et al., 2001).
total phosphorus prediction compared with mean monthly
measured data. Several other studies successfully cali-
brated and validated the SWAT model for monthly total CONCLUSIONS
phosphorus prediction (Saleh and Du, 2004; White and The objective of this research was to compare AnnAG-
Chaubey, 2005; Arabi et al., 2006; Bracmort et al., 2006; NPS and SWAT model simulation results for surface flow,
Cheng et al., 2006; Tolson and Shoemaker, 2007). More sediment yield, and total phosphorus using 45 months
applications of SWAT are described in Gassman et al. (January 1997 to September 2000) of measured data. The
(2007). uniqueness of this study lies in its comparison of model
Although no calibration parameters were used for performance in two separate yet similar watersheds. It is
phosphorus calibration, the SWAT model uses the important to choose appropriate model to prioritize criti-
QUAL2E stream flow process for model simulation, cal areas in the watershed.
which reduces the amount of phosphorus leaving the According to the classifications of Parajuli (2007), this
watershed outlet or reach (Brown and Barnwell, 1987; study concluded that both AnnAGNPS and SWAT mod-
Neitsch et al., 2005). The SWAT model QUAL2E in- els performed with fair to very good correlation (R2
stream process with default parameters demonstrated fair from 0Ð50 to 0Ð89) and fair to good agreement (E from
to good model correlations and efficiencies in various out- 0Ð47 to 0Ð73) for surface flow and sediment yield when
lets of the Upper Wakarusa watershed in Kansas (Parajuli, comparing model predictions with measured data during
2007). The AnnAGNPS model had no stream process calibration and validation. SWAT also performed consis-
routines in its Input Editor. AnnAGNPS assumes that tently well in terms of correlation (R2 from 0Ð66 to 0Ð77)

Copyright  2008 John Wiley & Sons, Ltd. Hydrol. Process. 23, 748–763 (2009)
DOI: 10.1002/hyp
 ANNAGNPS AND SWAT MODEL SIMULATION RESULTS IN AGRICULTURAL WATERSHEDS 761

and agreement (E from 0Ð63 to 0Ð68) for total phosphorus Borah DK, Bera M. 2003. Watershed-scale hydrologic and nonpoint-
source pollution models: Review of mathematical bases. Transactions
calibration and validation. Total phosphorus predictions of the ASAE 46(6): 1553– 1566.
during validation of AnnAGNPS were unsatisfactory to Borah DK, Yagow G, Saleh A, Barnes PL, Rosenthal W, Krug EC,
good (R2 from 0Ð60 to 0Ð77, E from
2Ð38 to 0Ð32) with Hauck LM. 2006. Sediment and nutrient modelling for TMDL
development and implementation. Transactions of the ASAE 49(4):
RMSE about 167% higher than that for SWAT for mean 967– 986.
monthly total phosphorus prediction. AnnAGNPS over- Bosch DD, Strickland T, Sullivan DG, Wauchope D, Lowrance R,
predicted total phosphorus (PBIAS from
117 to
128), Potter T. 2005. SWAT Application for Conservation Effects
Assessment in the Southeastern Coastal Plain. ASAE Meeting paper
which was attributed mainly to soluble phosphorus loss. number: 052 040. ASAE, St. Joseph, MI.
Because the KDHE has set TMDLs for eutrophication Bracmort KS, Arabi M, Frankenberger JR, Engel BA, Arnold JG. 2006.
and silt for Cheney Lake, SWAT’s ability to predict phos- Modelling long-term water quality impact of structural BMPs.
Transactions of the ASAE 49(2): 367–374.
phorus in conjunction with BMPs is essential. Brown LC, Barnwell TO Jr. 1987. The enhanced water quality models
QUAL2E and QUAL2E-UNCAS documentation and user manual. EPA
document EPA/600/2-87/007. USEPA, Athens, GA.
Cheney Lake Watershed Management Plan (CLWMP). 2003– 2007, 2008.
ACKNOWLEDGEMENTS Cheney Lake Watershed Inc. 18 east 7th street—South Hutchinson, KS,
67 505.
This material is based upon work supported by the Coop- Cheng H, Ouyang W, Hao F, Ren X, Yang S. 2006. The nonpoint-source
erative State Research, Education and Extension Services, pollution in livestock-breeding areas of the Heihe River basin in Yellow
River. Stochastic Environmental Research and Risk Assessment 21(3):
US Dept. of Agriculture, under agreement no. 2006- 213– 221. DOI:10Ð1007/s00 477-006-0057-2.
51130-03703. We acknowledge the contributions of Ms Chu TW, Shirmohammadi A. 2004. Evaluation of the SWAT model’s
Lisa French, project coordinator, and Mr Howard Miller, hydrology component in the piedmont physiographic region of
Maryland. Transaction of the ASAE 47(4): 1057– 1073.
public relations coordinator, at Cheney Lake Watershed Chu TW, Shirmohammadi A, Abbot L, Sadeghi A. 2005. Watershed
Inc., Hutchinson, KS; Mr Daniel S. Moore, hydraulic level BMP Evaluation with SWAT model. ASAE Meeting Paper
engineer, at USDA-NRCS, Portland, OR; Mr Vance number: 052 098. ASAE, St. Joseph, MI.
Cronshey RG, Theurer FG. 1998. AnnAGNPS-non point pollutant
Justice, Jr, hydrologist, and Dr Ron L. Bingner, agricul- loading model. In Proceedings First Federal Interagency Hydrologic
tural engineer, at USDA-ARS, National Sedimentation Modelling Conference. Las Vegas, NV.
Laboratory, Oxford, MS; and Mr Larry M. Pope, project Das S, Rudra RP, Gharabaghi B, Goel PK, Singh A, Ahmed I. 2007a.
Comparing the Performance of SWAT and AnnAGNPS Model in a
chief, at USGS, KS. Contribution no. 08-249-J from the Watershed in Ontario. Watershed Management to Meet Water Quality
Kansas Agricultural Experiment Station. Standards and TMDLS (Total Maximum Daily Load) Proceedings.
ASABE Publication number: 701P0207. ASABE, St. Joseph, MI.
Das S, Rudra RP, Gharabaghi B, Singh A, Ahmed SI, Goel PK. 2007b.
Evaluation of nutrient component of AnnAGNPS model in a watershed
REFERENCES in Ontario. Watershed Management to Meet Water Quality Standards
and TMDLS (Total Maximum Daily Load) Proceedings. ASABE
Abbaspour KC, Yang J, Maximov I, Siber R, Bogner K, Mieleitner J, Publication number: 701P0207. ASABE, St. Joseph, MI.
Zobrist J, Srinivasan R. 2007. Modelling hydrology and water quality Das S, Rudra RP, Goel PK, Gharabaghi B, Gupta N. 2006. Evaluation
in the pre-alpine/alpine Thur watershed using SWAT. Journal of of AnnAGNPS in cold and temperate regions. Water Science and
Hydrology 333(2– 4): 413–430. Technology 53(2): 263– 270.
Allen RG. 1986. A penman for all seasons. Journal of Irrigation and Eckhardt K. 2005. How to construct recursive digital filters for baseflow
Drainage Engineering ASCE 112(4): 348–368. separation. Hydrological Processes 19(2): 507– 515.
Allen RG, Jensen ME, Wright JL, Burman RD. 1989. Operational Feyereisen GW, Strickland TC, Bosch DD, Sullivan DG. 2007. Evalua-
estimates of evapotranspiration. Agronomy Journal 81: 650– 662. tion of SWAT manual calibration and input parameter sensitivity in the
Arabi M, Govindaraju RS, Hantush MM. 2006. Cost-effective allocation Little River watershed. Transactions of the ASABE 50(3): 843– 855.
of watershed management practices using a genetic algorithm. Water Gassman PW, Reyes MR, Green CH, Arnold JG. 2007. The Soil
Resource Research 42: W10 429. DOI:10,1029/2006WR004 931. and Water Assessment Tool: historical development, applications,
Arabi M, Frakenberger JR, Engel BA, Arnold JG. 2007. Representation and future research directions. Transactions of the ASABE 50(4):
of agricultural conservation practices with SWAT. Hydrological 1211– 1250.
Processes 22(16): 3042– 3055. DOI:10Ð1002/hyp.6890. Gebremeskel S, Rudra RP, Gharabaghi B, Das S, Singh A, Bai H,
Arnold JG, Srinivasan R, Muttiah RS, Williams JR. 1998. Large area Jiang G. 2005. Assessing the performance of various hydrological
hydrologic modelling and assessment part I: model development. models in the Canadian Great lakes basin. Watershed Management to
Journal of American Water Resources Association 34(1): 73–89. Meet Water Quality Standards and Emerging TMDL (Total Maximum
Baginska B, Milne-Home W, Cornish PS. 2003. Modelling nutrient Daily Load) Proceedings. ASAE Publication number: 701P0105.
transport in Currency Creek, NSW with AnnAGNPS and PEST. ASABE, St. Joseph, MI.
Environmental Modelling & Software 18: 801–808. Geter WF, Theurer FD. 1998. AnnAGNPS-RUSLE sheet and rill
Bagnold RA. 1966. An approach to the sediment transport problem from erosion. In Proceedings of 1st Federal Interagency Hydrologic
general physics. Prof. Paper 422-J. US Geological Survey, Reston, VA. Modelling Conference. Interagency Advisory Committee on Water
Bagnold RA. 1977. Bedload transport in natural rivers. Water Resources Data, Subcommittee on Hydrology. Washington, DC.
Research 13: 303– 312. Gikas GD, Yiannakopoulou T, Tsihrintzis VA. 2006. Modelling of
Bhuyan SJ, Mankin KR, Koelliker JK. 2003. Watershed-scale AMC non-point source pollution in a Mediterranean drainage basin.
selection on for hydrologic modelling. Transactions of the ASAE 46(2): Environmental Modelling & Assessment 11(3): 219–233.
303– 310. Gitau MW, Chaubey I, Nelson MA, Pennigton JH. 2007. Analysis of
Bingner RL, Theurer FD. 2003. AnnAGNPS technical processes BMP and land use change effects in a Nrthwest Arkansas agricultural
documentation, Version 3Ð2. USDA-ARS, National Sedimentation watershed. ASABE Meeting Paper number: 072 244. ASABE, St.
Laboratory: Oxford, MS. Joseph, MI.
Bingner RL, Theurer FD, Yongping Y. 2007. AnnAGNPS technical pro- Gitau MW, Veith TL, Gburek WJ. 2003. Optimizing Best Management
cesses documentation, Version 4Ð0 . USDA-ARS, National Sedimenta- Practice Selection to Increase Cost-effectiveness. ASAE Meeting Paper
tion Laboratory: Oxford, MS. number: 032 110. ASAE, St. Joseph, MI.
Bockhold AR, Thompson AL, Baffaut C, Sadler EJ. 2006. Evaluating Gordon LM, Bennett SJ, Bingner RL, Theurer FD, Alonso CV. 2007.
BMPs in a Claypan watershed. ASABE Meeting Paper number: Simulating ephemeral gully erosion in AnnAGNPS. Transactions of
062 114. ASABE, St. Joseph, MI. the ASABE 50(3): 857– 866.

Copyright  2008 John Wiley & Sons, Ltd. Hydrol. Process. 23, 748– 763 (2009)
DOI: 10.1002/hyp
762 P. B. PARAJULI ET AL.

Gosain AK, Rao S, Srinivasan R, Reddy NG. 2005. Return-flow Neitsch SL, Arnold JG, Kiniry JR, Williams JR. 2005. Soil and Water
assessment for irrigation command in the Palleru river basin using Assessment Tool (SWAT), Theoretical Documentation. Blackland
SWAT model. Hydrological Processes 19: 673–682. Research Center, Grassland, Soil and Water Research Laboratory,
Gupta HV, Sorooshian S, Yapo PO. 1999. Status of automatic calibration Agricultural Research Service: Temple, TX.
for hydrologic models: comparison with multi-level expert calibration. Parajuli PB. 2007. SWAT bacteria sub-model evaluation and application.
Journal of Hydrologic Engineering 4(2): 135–143. PhD dissertation. Department of Biological and Agricultural
Hargreaves GH, Samani ZA. 1985. Reference crop evapotranspiration Engineering, Kansas State University, Manhattan, KS.
from temperature. Applied Engineering in Agriculture 1: 96–99. Parajuli PB, Mankin KR, Barnes PL. 2006. Calibration and Validation
Hong HS, Huang JL, Zhang LP, Du PF. 2005. Modelling pollutant of SWAT/Microbial sub-model 2005 for Fecal Coliform Bacteria
loads and management alternatives in Jiulong River watershed with Prediction on a Grazed Watershed. ASABE Paper No. 062 196. St.
AnnAGNPS. Huan Jing Ke Xue 26(4): 63–69. Joseph, MI.
Jensen ME, Burman RD, Allen RG (eds). 1990. Evapotranspiration Parajuli PB, Mankin KR, Barnes PL. 2007. New Methods in Modelling
and irrigation water requirements. ASCE Manuals and Reports on Source-Specific Bacteria at Watershed Scale Using SWAT. Watershed
Engineering Practice No. 70, ASCE: NY. Management to meet Water Quality Standards and TMDLs (Total
Jha MK, Gassman PW, Arnold JG. 2007. Water quality modelling for Maximum Daily Load) Proceedings. ASABE Publication No.
the Raccoon river watershed using SWAT. Transactions of the ASABE 701P0207. ASABE: St. Joseph, MI.
50(2): 479– 493. Parajuli PB, Mankin KR, Barnes PL. 2008. Source specific fecal bacteria
Kansas Department of Agriculture. 2004a. Kansas Agricultural Statistics. modelling using SWAT model. Bioresource Technology 100(2):
Kansas Farm Facts 2004. Available at http://www.nass.usda.gov/ks/ 953–963. DOI:10Ð1016/j.biortech.2008Ð06Ð045.
ffacts/2004/pdf/ffpdf.html as of 30 August 2007. Polyakov V, Fares A, Kubo D, Jacobi J, Smith C. 2007. Evaluation of a
Kansas Department of Health and Environment. 2004b. Bureau of non-point source pollution model, AnnAGNPS, in a tropical watershed.
water. Kansas Surface Water Quality Standards and Supporting Docu- Environmental Modelling & Software 22(11): 1617– 1627.
ments. Available at http://www.kdhe.state.ks.us/water/download/kwqs Priestley CHB, Taylor RJ. 1972. On the assessment of surface heat flux
plus supporting.pdf as of 21 December 2007. and evaporation using large-scale parameters. Monthly Weather Review
Kansas Department of Health and Environment. 2006. Bureau of 100: 81–92.
Environmental Field Services. Kansas Water Quality Assessment Qi C, Grunwald S. 2005. GIS-Based Hydrologic Modelling in the
(305(b) Report). Topeka, KS. Sandusky Watershed Using SWAT. Transactions of the ASAE 48(1):
Kansas Water Office. 2004. Reports and Publications. Fact Sheets. Kansas 160–180.
Water Office Fact Sheets. Available at http://www.kwo.org/as of 29 Ramanarayanan TS, Williams JR, Dugas WA, Hauck LM, McFarland
August 2007. AMS. 1997. Using APEX to identify alternative practices for animal
King KW, Arnold JG, Bingner RL. 1999. Comparison of Green-Ampt waste management. ASAE Paper No. 972 209. ASAE, St. Joseph, MI.
and curve number methods on Goodwin Creek watershed using SWAT. Renard KG, Foster GR, Weesies GA, McCool DK, Yoder DC, coordina-
Transactions of the ASAE 42(4): 919–925. tors. 1997. Predicting Soil Erosion by Water: A Guide to Conservation
Kirsch K, Kirsch A, Arnold JG. 2002. Predicting sediment and Planning with the Revised Universal Soil Loss Equation (RUSLE).
phosphorus loads in the Rock River basin using SWAT. Transactions
U.S. Department of Agriculture, Agriculture Handbook No. 703.
of the ASABE 45(6): 1757– 1769.
Sadeghi A, Kwang Y, Graff C, McCarty G, McConnell L,
Knisel WG (Ed.). 1980. A field scale model for chemical, runoff, and
Shirmohammadi A, Hively D, Sefton KA. 2007. Assessing the Per-
erosion from agricultural management systems. Conservation Service
formance of SWAT and AnnAGNPS Models in a Coastal Plain Water-
Report 26, US Department of Agriculture: Washington DC.
shed, Choptank River, Maryland, U.S.A. ASABE Meeting, Paper No.
Kyoung JL, Engel BA, Tang Z, Choi J, Kim KS, Muthukrishnan S,
072 032. ASABE, St. Joseph, MI.
Tripathy D. 2005. Automated Web GIS based Hydrograph Analysis
Saleh A, Du B. 2004. Evaluation of SWAT and HSPF within BASINS
Tool, WHAT. Journal of American Water Resources Association 41(6):
program for the upper North Bosque River watershed in central Texas.
1407– 1416.
Transactions of the ASAE 47(4): 1039– 1049.
Legates DR, McCabe Jr. GJ. 1999. Evaluating the use of “goodness-of-
fit” measures in hydrologic and hydroclimatic model validation. Water Saleh A, Arnold JG, Gassman PW, Hauck LM, Rosenthal WD,
Resources Research 35(1): 233–241. Williams JR, MacFarland AMS. 2000. Application of SWAT for the
Leonard RA, Knisel WG, Still DA. 1987. GLEAMS: Groundwater Upper North Bosque River watershed. Transactions of the ASAE 43(5):
Loading Effects of Agricultural Management Systems. Transactions 1077– 1087.
of the ASAE 30(5): 1403– 1418. Santhi C, Arnold JG, Williams JR, Hauck LM, Dugas WA. 2001.
Li X, Frees L, Moore DS, Wang S. 2006. Calibrating the AnnAGNPS Application of a watershed model to evaluate management effects on
model in the Red Rock Creek watershed. Unpublished report. point and non-point source pollution. Transactions of the ASAE 44(6):
Department of Geography, University of Kansas, Lawrence, KS. 1559– 1570.
Licciardello F, Zema DA, Zimbone SM, Bingner RL. 2007. Runoff Sharpley AN, Williams JR, eds. 1990. EPIC-Erosion Productivity Impact
and soil erosion evaluation by the AnnAGNPS model in a Calculator, 1. model documentation. U. S. Department of Agriculture,
small Mediterranean watershed. Transactions of the ASABE 50(5): Agricultural Research Service, Technical Bulletin, 1768.
1585– 1593. Shirmohammadi A, Chaubey I, Harmel RD, Bosch DD,
Mein RG, Larson CL. 1973. Modelling infiltration during a steady rain. Muñoz-Carpena R, Dharmasri C, Sexton A, Arabi M, Wolfe ML,
Water Resources Research 9(2): 384– 394. Frankenberger J, Graff C, Sohrabi TM. 2006. Uncertainty in TMDL
Migliaccio KW, Srivastava P. 2007. Hydrologic components of Models. Transactions of the ASAE 49(4): 1033– 1049.
watershed scale models. Transactions of the ASABE 50(5): 1695– 1703. Sloan PG, Moore ID, Coltharp GB, Eigel JD. 1983. Modelling surface
Monteith JL. 1965. Evapotranspiration and the environment. The state and subsurface stormflow on steeply sloping forested watersheds.
and movement of water in living organisma, XIXth symposium. Society Water Resources Institute Report No. 142. University of Kentucky,
for Experimental Biology, Swansea, Cambridge University Press. Lexington, KY.
Moriasi DN, Arnold JG, Van Liew MW, Bingner RL, Harmel RD, Spruill CA, Workman SR, Taraba JL. 2000. Simulation of daily and
Veith TL. 2007. Model evaluation guidelines for systematic monthly stream discharge from small watersheds using the SWAT
quantification of accuracy in watershed simulations. Transactions of model. Transactions of the ASAE 43(6): 1431– 1439.
the ASABE 50(3): 885– 900. Stuck JD, Izuno FT, Campbell KL, Bottcher AB, Rice RW. 2001. Farm-
Nash JE, Sutcliffe JV. 1970. River flow forecasting through conceptual level studies of particulate phosphorus transport in the Everglades
models: Part I. A discussion of principles. Journal of Hydrology 10(3): Agricultural Area. Transactions of the ASAE 44(5): 1105– 1116.
282– 290. Theurer FD, Clarke CD. 1991. Wash load component for sediment yield
Nasr A, Bruen M, Jordan P, Moles R, Kiely G, Byrne P. 2007. A modelling. Proceedings of the Fifth Federal Interagency Sedimentation
comparison of SWAT, HSPF, and SHETRAN/GOPC for modelling Conference, Las Vegas, NV.
phosphorus export from three catchments in Ireland. Water Research Theurer FD, Cronshey RG. 1998. AnnAGNPS-reach routing processes.
41(5): 1065– 1073. Proceedings First Federal Interagency Hydrologic Modelling Confer-
Neitsch SL, Arnold JG, Kiniry JR, Williams JR, King KW. 2002. Soil ence, Las Vegas, NV.
and Water Assessment Tool Theoretical Documentation (Version 2000). Tolson BA, Shoemaker CA. 2007. Cannonsville reservoir watershed
Grassland, soil and Water Research Laboratory, Agricultural Research SWAT2000 model development, calibration, and validation. Journal
Service, Temple, TX. of Hydrology 337(1– 2): 68–86.

Copyright  2008 John Wiley & Sons, Ltd. Hydrol. Process. 23, 748–763 (2009)
DOI: 10.1002/hyp
 ANNAGNPS AND SWAT MODEL SIMULATION RESULTS IN AGRICULTURAL WATERSHEDS 763

US Department of Agriculture (USDA), Natural Resources Conservation Williams JR, Jones CA, Dyke PT. 1984. A modelling approach to
Service. 2005. Soil Data Mart. Available at http://soildatamart.nrcs. determining the relationship between erosion and soil productivity.
usda.gov/Default.aspx as of 29 october 2007. Transactions of the ASAE 27(1): 129–144.
US Geological Survey (USGS). 1999. National Elevation Dataset. Willmot CJ. 1984. On the evaluation of model performance in physical
Available at http://gisdasc.kgs.ukans.edu/as of 29 October, 2007. geography. In Spatial Statistics and Models. Gaile GL (ed). D. Reidel;
Vache KB, Eilers JM, Santelmann MV. 2002. Water quality modelling 443– 460.
of alternative agricultural scenarios in the U.S. Corn Belt. Journal of Wischmeier WH, Smith DD. 1978. Predicting rainfall erosion losses: a
American Water Resources Association 38(3): 773–787. guide to conservation planning. Agriculture Handbook 282. United
Van Liew MW, Arnold JG, Garbrecht JD. 2003. Hydrologic Simulation States Department of Agriculture (USDA), Agricultural Research
on agricultural watersheds choosing between two models. Transactions Service (ARS). US Govt Printing Office: Washington, DC.
of the ASABE 46(6): 1539– 1551. Young RA, Onstead CA, Bosch DD, Anderson WP. 1987. AGNPS,
Varanou E, Gkouvatsou E, Baltas E, Mimikou M. 2002. Quantity and Agricultural Non-Point-Source Pollution Model. A Watershed Analysis
quality integrated catchment modelling under climate change with user Tool. USDA Conservation Research Report 35, Washington DC.
of soil and water assessment tool model. ASCE Journal of Hydrological Young RA, Onstead CA, Bosch DD, Anderson WP. 1989. AGNPS: a
Engineering 7(3): 228–244. nonpoint source pollution model for evaluating agricultural watersheds.
Wang X, Melesse AM, Yang W. 2006. Influences of potential Journal of Soil Water Conservation 44(2): 168–173.
evapotranspiration estimation methods on SWAT’s hydrologic Yuan Y, Bingner RL, Rebich RA. 2001. Evaluation of AnnAGNPS on
simulation in a northwestern Minnesota watershed. Transactions of the Mississippi Delta MSEA watersheds. Transactions of the ASAE 44(5):
ASABE 49(6): 1755– 1771. 1183– 1190.
Wang S, Kang S, Zhang L, Li F. 2007. Modelling hydrological response Yuan Y, Bingner RL, Theurer FD. 2006. Subsurface Flow Component
to different land-use and climate change scenarios in the Zamu River for AnnAGNPS. Transactions of the ASABE 22(2): 231–241.
basin of northwest China. Hydrological Processes 22(14): 2502– 2510. Yuan Y, Bingner RL, Theurer FD, Kolian S. 2007. Water quality
DOI: 10Ð1002/hyp.6846. simulation of Rice/Crawfish field ponds within Annualized AGNPS.
White KL, Chaubey I. 2005. Sensitivity analysis, calibration, and Transactions of the ASABE 23(5): 585–595.
validations for a multisite and multivariable SWAT model. Journal Yuan Y, Bingner RL, Theurer FD, Moore PA, Rebich RA. 2005.
of American Water Resources Association 41(5): 1077– 1089. Phosphorus component in AnnAGNPS. ASAE Meeting Paper number:
White KL, Chaubey I, Haggard BE, Matlock MD. 2004. Comparison 052 169. ASAE, St. Joseph, MI.
of two methods for modelling monthly TP yield from a watershed. Yuan Y, Dabney S, Bingner RL. 2002. Cost/benefit analysis of
ASABE Meeting Paper No. 042 162. ASABE: Joseph, MI. agricultural BMPs for sediment reduction in the Mississippi Delta.
Wilcox BP, Rawls WJ, Brakensiek DL, Wight JR. 1990. Predicting Journal of Soil and Water Conservation 57(5): 259– 267.
runoff from rangeland catchments: A comparison of two models. Water
Resources Research 26: 2401– 2410.

Copyright  2008 John Wiley & Sons, Ltd. Hydrol. Process. 23, 748– 763 (2009)
DOI: 10.1002/hyp