Soil & Water Assessment Tool

http://swatmodel.tamu.edu/

Introductory Manual

R. Srinivasan
[email protected]
Soil & Water Assessment Tool
http://swatmodel.tamu.edu/

Beginner SWAT
Training Manual

R. Srinivasan
[email protected]
 Table of Contents

Section 1 ............................. 1
SWAT Overview

Section 2 ............................. 21
ArcSWAT
ArcGIS Interface for SWAT

Section 3 ............................. 93
SWAT Calibration Techniques

Section 4 ............................. 107

SWAT Publications
 Section 1

SWAT Overview

1
2
 Soil and Water
Assessment Tool
http://swatmodel.tamu.edu

R. Srinivasan
[email protected]
Texas A&M University

Model Philosophy

 Readily available input –

Physically based
 Comprehensive – Process Interactions
 Simulate Management

Model Philosophy

“Everything should be made as

simple as possible. I have no interest
in the laws of physics if they can’t be
made simple”

Albert Einstein

3
 ARS Modeling History
Time Line

CREAMS
USLE (CLEAN WATER ACT) EPIC SWRRB SWAT

1960’s 1970’s 1980’s 1990’s

GLEAMS WEPP ANN AGNPS
AGNPS

Environmental Models
Maintained at Temple, TX

 EPIC—Field Scale
 ALMANAC—Field Scale
 APEX—Farm Scale
 SWAT—Watershed Scale

Global Applications
Hydrologic Unit Modeling for Global Eosystem/Environment

HUMER
HUMUS
HUNCH

2001 Giessen, Germany

HUMID
2003 Bari, Italy
2005 Zurich, Switzerland
HUMAF
2006 Potsdam, Germany
2007 Delft, Netherlands
2008 Beijing, China
2009 Chiang Mai, Thailand
2009 Boulder, CO, USA

International SWAT Conferences

2010 New Delhi, India
2011 Toledo, Spain

4
 General Description
 Continuous Time
Daily Time Step
One Day Hundreds of Years
 Distributed Parameter
Unlimited Number of Subwatersheds
 Comprehensive – Process Interactions
 Simulate Management

Example Configuration

 Cells/Subwatersheds
 Hydrologic Response Units
 Output from other Models
 Point Sources - Treatment Plants

SWAT Watershed System

Upland Processes

Channel/Flood Plain
Processes

5
 Subbasins and Streams

Subbasins and Streams

Subbasin 18

Subbasins and Streams

HRU’s
28% Range-Sandy
51% Pasture – Silt
16% Forest – Sandy
4% - Agriculture - Silt

6
 Subbasins and Streams

Upland Processes
 Weather
 Hydrology
 Sedimentation
 Plant Growth
 Nutrient Cycling
 Pesticide Dynamics
 Management
 Bacteria

Climate
• Weather
– Precipitation
– Air Temperature and Solar Radiation
– Wind Speed
– Relative Humidity
• Snow
– Snow Cover
– Snow Melt
– Elevation Bands
• Soil Temperature

7
 Hydrology
• Canopy Storage
• Infiltration
• Redistribution
• Evapotranspiration
• Lateral Subsurface Flow
• Surface Runoff
– Surface runoff volume
– Peak runoff
• Ponds
• Tributary Channels
– Transmission Losses

Hydrologic Balance

Precipitation

Root Zone

Vadose
(unsaturated)
Zone

Shallow
(unconfined)
Aquifer

Confining Layer

Deep (confined)
Aquifer

Hydrologic Balance

Root Zone Surface Runoff

Infiltration
Vadose
(unsaturated)
Zone

Shallow
(unconfined)
Aquifer

Confining Layer

Deep (confined)
Aquifer

8
 Hydrologic Balance
Evaporation and
Transpiration

Root Zone
Infiltration/plant uptake/
Soil moisture redistribution
Vadose
(unsaturated)
Lateral Flow
Zone

Shallow Percolation to
(unconfined)
Aquifer
shallow aquifer

Confining Layer

Deep (confined)
Aquifer

Tile Flow
if

Hydrologic Balance
Root Zone

Tile Flow
Vadose
(unsaturated)
Zone

Shallow
(unconfined)
Aquifer

Confining Layer

Deep (confined)
Aquifer

Hydrologic Balance

Root Zone

Vadose
(unsaturated)
Zone

Shallow Revap from Percolation to

(unconfined) shallow aquifer shallow aquifer
Aquifer Return Flow
Confining Layer

Deep (confined)
Aquifer Recharge to
deep aquifer

9
 Hydrologic Balance

Root Zone

Vadose
(unsaturated)
Zone

Shallow
(unconfined)
Aquifer

Confining Layer

Deep (confined) Flow out of watershed Recharge to

Aquifer
deep aquifer

Hydrologic Balance
Evaporation and
Transpiration
Precipitation

Root Zone Surface Runoff

Infiltration/plant uptake/ Soil
moisture redistribution
Vadose Lateral Flow
(unsaturated)
Zone

Shallow Revap from Percolation to

(unconfined) shallow aquifer shallow aquifer
Aquifer Return Flow
Confining Layer

Deep (confined) Flow out of watershed

Aquifer Recharge to
deep aquifer

Hydrologic Balance
Evaporation and
Transpiration
Precipitation

Root Zone Surface Runoff

Infiltration/plant uptake
Vadose Lateral Flow
(unsaturated)
Zone

Shallow Revap from Percolation to

(unconfined) shallow aquifer shallow aquifer
Aquifer Return Flow
Confining Layer

Deep (confined) Flow out of watershed

Aquifer Recharge to
deep aquifer

10
 Plant Growth

12

0 12
0 2 4 6 8 10
Month

Plant Growth
Yield Prediction Optimum Growth
Harvest Index – Water Stress Radiation Interception LAI
Residue – Cover and Radiation Use Efficiency
Nutrients
Constraints
Water, Temperature, Nitrogen,
Phosphorus

Residue – Cover and Nutrients

Water, Nitrogen
and Root Growth
Phosphorus Uptake

Climate Change
Radiation Use Efficiency
Adjusted for CO2

ET – Penman-Monteith
Canopy Resistance
Adjusted for CO2
Impact on Leaf
Conductance

CGM Estimates of Precip, Temperature,

Humidity, Solar Radiation,
Wind Speed

11
 Nitrogen Cycle
Atmospheric N fixation
(lightning arc discharge)

Symbiotic
fertilizer fixation fertilizer
manures, wastes
and sludge

Soil Organic
Matter
NO -
NO33- NH4+

NO2-

Nitrogen Cycle

immobilization Soil Organic mineralization

Matter
NO -
NO33- immobilization NH4+

NO2- ammonium fixation

clay

nitrification

Nitrogen Cycle
Harvest

NH3
N2
runoff
N2O

ammonia
volatilization
denitrification
Soil Organic
Matter
NO -
NO33- NH4+

NO2-
anaerobic clay
conditions
leaching

12
 Nitrogen Cycle
Atmospheric N fixation
Harvest
(lightning arc discharge)
NH3
Symbiotic
N2 fertilizer fixation fertilizer runoff
N2O
manures, wastes
and sludge

ammonia
volatilization
denitrification
immobilization Soil Organic mineralization
Matter
NO -
NO33- immobilization NH4+

NO2- ammonium fixation

anaerobic clay
conditions nitrification
leaching

Phosphorus Cycle

fertilizer Harvest

runoff manures, wastes, manures, wastes

and sludge and sludge

H2PO4- mineralization Soil Organic

HPO4-2 Matter
immobilization

Adsorbed and fixed

Inorganic
Fe, Al, Ca, and clay

Pesticide Dynamics
Foliar Application

Degradation Surface Application

Runoff
Washoff

Infiltration Degradation

Leaching

13
 Management
 Crop Rotations
 Removal of Biomass as Harvest/
Conversion of Biomass to Residue
 Tillage / Biomixing of Soil
 Fertilizer Applications
 Grazing
 Pesticide Applications

Management

 Irrigation
 Subsurface (Tile) Drainage
 Water Impoundment (e.g. Rice)

Management

 Urban Areas
Pervious/Impervious Areas
Street Sweeping
Lawn Chemicals
 Edge of Field Buffers

14
 Channel Processes

Channel Processes
 Flood Routing
Variable Storage
Muskingum
 Transmission Losses, Evaporation

 Sediment Routing
Degradation and deposition
computed simultaneously

Channel Processes

 Nutrients
modified QUAL2E/WASP
 Pesticide
Toxic balance developed at
University of Colorado

15
 In-stream Nutrient Processes

Atmospheric Aeration

Well-mixed Water Layer

Org N Dissolved Org P

Oxygen Transport out
Transport in NH4
Dissolved P
NO2
Carbonaceous Chlorophyll a
NO3 BOD Algae

Sediment Layer Benthic Benthic

Demand for Sink/Source
Oxygen for Nutrients

In-stream Pesticide Processes

Volatilization

Well-mixed Water Layer

Pesticide Transport out

Transport in
in Water

Degradation
Settling Resuspension

Well-mixed Pesticide in
Sediment Layer Sediment
Degradation Burial

Impoundments
 Water Balance
Inflow
Evaporation
Seepage
Withdrawals
Outflow
Spillway Control
Target Volume Approach
Missouri River Reservoir Operation

16
 Impoundments
 Nutrient Balance
Well-mixed System
Nitrogen & Phosphorus Loss Rates
2 Settling Periods per Year
 Pesticide Balance
Well-mixed System
Toxic balance developed at
University of Colorado

User Options
• PET:
Penman-Monteith, Priestly-Taylor, or
Hargreaves
• Runoff:
Curve Number or Green & Ampt
• Channel Flow:
Variable Storage Coefficient or Muskingham-
Cunge
• Channel Water Quality:
QUAL2E On-Off Switch

More User Options

• ARC GIS 9.3 or 10
• Map Windows (Public Domain GIS)
• SWAT-CUP (Calibration and Uncertainty Program
• VIZSWAT (Output Vizualization)
• Manuals in English, Spanish, Chinese, Korean
• SWAT 2003, 2005, 2009, (2012)

17
 SWAT Strengths
Upland Processes
 Comprehensive Hydrologic Balance
 Physically-Based Inputs
 Plant Growth – Rotations, Crop Yields
 Nutrient Cycling in Soil
 Land Management - BMP
Tillage, Irrigation, Fertilizer, Pesticides,
Grazing, Rotations, Subsurface Drainage,
Urban-Lawn Chemicals, Street Sweeping

SWAT Strengths
Channel Processes
 Flexible Watershed Configuration
 Water Transfer—Irrigation Diversions
 Sediment Deposition/Scour
 Nutrient/Pesticide Transport
 Pond, Wetland and Reservoir Impacts

Collaborators
USDA
Texas A&M
Agricultural Research
University
Service

USDA EPA
Natural Resources Office of Science
Conservation Service and Technology

NOAA Universities

18
 Conclusions
 A product of over 45 years of USDA/Texas A&M
model development

 Widely used for water quality, water supply, and

climate change, carbon sequestration, and
agricultural production assessments
worldwide

•Over 1000’s SWAT users and 30 active developers

worldwide

•Annual meetings alternate in Asia and Europe

The End
http://swatmodel.tamu.edu

19
20
 Section 2

Arc SWAT
ArcGIS Interface for SWAT
Introduction
Objectives
Watershed Delineation
Hydrologic Response Unit Definition
Write Input Tables for SWAT
Edit SWAT Input
SWAT Simulation Setup
Appendix: Installing ArcSWAT

21
22
 ArcSWAT
ArcGIS Interface for Soil and Water Assessment Tool (SWAT)

http://www.brc.tamus.edu/swat

R. Srinivasan
[email protected]
Blackland Research and Extension Center and Spatial Sciences Laboratory
Texas Agricultural Experiment Station
Texas A&M University

R. Srinivasan, [email protected] 23
 Table of Contents

Introduction ................................................................................................................................................ 1

Objectives .................................................................................................................................................. 1

Watershed Delineation ............................................................................................................................... 5

Hydrologic Response Unit Definition........................................................................................................ 29

Write Input Tables for SWAT ................................................................................................................... 31

Edit SWAT Input....................................................................................................................................... 35

SWAT Simulation Setup........................................................................................................................... 56

Appendix: Installing ArcSWAT ................................................................................................................. 58

R. Srinivasan, [email protected] -1- 24

 Introduction
The Soil and Water Assessment Tool (SWAT) is a physically-based continuous-event hydrologic model developed to predict the
impact of land management practices on water, sediment, and agricultural chemical yields in large, complex watersheds with varying
soils, land use, and management conditions over long periods of time. For simulation, a watershed is subdivided into a number of
homogenous subbasins (hydrologic response units or HRUs) having unique soil and land use properties. The input information for
each subbasin is grouped into categories of weather; unique areas of land cover, soil, and management within the subbasin;
ponds/reservoirs; groundwater; and the main channel or reach, draining the subbasin. The loading and movement of runoff,
sediment, nutrient and pesticide loadings to the main channel in each subbasin is simulated considering the effect of several physical
processes that influence the hydrology. For a detailed description of the capabilities of the SWAT, refer to Soil and Water
Assessment Tool User’s Manual, Version 2000 (Neitsch et al., 2002), published by the Agricultural Research Service and the Texas
Agricultural Experiment Station, Temple, Texas. The manual can also be downloaded from the SWAT Web site
(www.brc.tamus.edu/swat/swatdoc.html#new).

Objectives

The objectives of this exercise are to (1) setup a SWAT project and (2) familiarize with the capabilities of SWAT.

R. Srinivasan, [email protected] 1 25
 Create a Project

ArcSWAT extension of ArcGIS 10 creates an ArcMap project file that contains links to your retrieved data and incorporates all
customized GIS functions into your ArcMap project file. The project file contains a customized ArcMap Graphical User Interface (GUI)
including menus, buttons, and tools. The major steps on how to create a SWAT project under then ArcMap environment are
introduced below:

Step 1. Start ArcMap. Under the Customize menu of ArcMap View,

click the Extensions button. You will see an extension entitled
“SWAT Project Manager” and “SWAT Watershed delineator”
under the extensions list (Figure 1). Turn on these two
extensions.

Step 2. Go to the Customize menu of ArcMap, hover the mouse over

the Toolbar button, a list of tools will appear. Click the
ArcSWAT tool, the main interface of ArcSWAT will open

Figure 1 Extensions of ArcMap

R. Srinivasan, [email protected] Texas

26 A & M University 2
 Figure 2 Startup of ArcSWAT Interface

Figure 3 Create a new SWAT project and Setup

working directory and Geodatabases

R. Srinivasan, [email protected] Texas

27 A & M University 3
 Step 3. Create a new SWAT project and Setup working directory and Geodatabases
 Click the New SWAT Project button under then SWAT Project Setup menu. In the prompted dialog entitled “Project
Set Up” (Figure 3), Select a Project Directory for current project. The names of SWAT Project Geodatabase and
Raster Storage Geodatabase are initialized automatically. These two Geodatabases will be created under the
Project Directory.
 The SWAT Parameter Geodatabase stores the parameters that are needed for SWAT model run. The directory of
this Geodatabase is provided by the interface, users should not change this value.
 Click OK after you setup a project. Then a new SWAT project will be created under the Project Directory (Figure 4).
The SWAT project includes two folders, two geodatabases, and one <Project Directory>.mxd file. The < Project
Directory >.mxd file is the file that is currently in use by the user.

Figure 4 The structure of SWAT project directory

R. Srinivasan, [email protected] Texas

28 A & M University 4
 Watershed Delineation
After setup of a new SWAT project, activate the Automatic Watershed Delineation button under the Watershed Delineator menu
by clicking this button. The Watershed Delineation tool will appear (Figure 2). The tool‘s functions are divided into five sections,
namely: DEM setup, Stream Definition, Outlet and Inlet Definition, Watershed Outlet(s) Selection and Definition. This tool is used to
create wateshed delineations using a combination of DEM, digitized network (RF1 or RF3, NHD or User defined), and other user
inputs. The detailed procedures on how to use the Watershed Delineation tool are introduced in the following sections:

Figure 2 Watershed Delineation main dialog

R. Srinivasan, [email protected] 5 29
Step 1: Add DEM Grid

1. Click the icon to add the DEM grid to be setup for modeling with
SWAT. A dialog box (Figure 3) with options Select from Watershed
View and Load DEM grid from disk is opened. Choose Load DEM
grid from disk and select the dem grid located in your disk (the
directory is “C:\AVSWATX\AvSwatDB\Example1” for this example) and
click ADD. The selected dem will be resaved as <Project
Directory>\RasterStore.mdb\SourceDem, which is loaded into the
ArcMap. Figure 3 Projection properties of DEM
Note: the selected dem file must have an effective projection. If
not, an error dialog will prompt.
2. After loading dem, the button will be activated. Using this button,
the user can check the projection information of the dem (Figure 4).
The DEM properties box lets you verify that the DEM map properties
are correct and make any needed changes. The DEM properties
should correctly define the grid size and units. The user can change
the Z unit to reflect the real situation.

Step 2: Focusing on Watershed Area

The interface allows users to import or create a grid map that masks out a part Figure 4 Projection properties of DEM
of the DEM grid and/or a shape map that defines the stream network. These
maps are not required but might speed the processing time for the GIS operations.

R. Srinivasan, [email protected] Texas

30 A & M University 6
1. Click beside Mask, then three options are available to specify the
watershed area of interest (Figure 5): Load mask grid from disk,
manually delineate and Select boundary theme from basins view.
Select the Load mask grid from disk option and click OK. Select the
mask grid located in your disk, and click ADD. The selected dem will be
resaved as <Project Directory>\RasterStore.mdb\Mask, which is loaded
into the ArcMap (Figure 7).
2. If you selected the Manually Delineate option, then a toolset will appear and
allow you draw polygon mask (Figure 6). A toolbar with three buttons, DRAW, EDIT
VERTEX and DELETE appears. These tools provide the capability similar to
drawing and editing a polygon theme in ArcMap. Figure 5 Options to define Mask
Click on the DRAW button to manually trace the boundary of the watershed area
of interest using the mouse. Double click the mouse left button to complete the
polygon. EDIT Vertex will allow you to adjust the boundary of the mask.
The DELETE Button will allow you to select the entire polygon for deletion using
the “Delete” function in your Keyboard. Click APPLY after you are satisfied with
the manually drawn masking area. This will create a masking grid (MASK) and
add the theme to the Watershed View.
Figure 6 Toolset for Manually
Mask Delineation

Note:
1. If a mask grid already exists, you can select Load mask grid from the disk option. The grid will be added to the Basins view and
used for delineation.
2. A polygon theme already in the Basins View can be selected as a mask, using the Select boundary theme from the Basins view option.

R. Srinivasan, [email protected] Texas

31 A & M University 7
 Masked area

Figure 7 DEM grid added to ArcMap View with a mask area

Note:
 You can exit the main Watershed Delineation Dialog anytime and open it again by selecting AUTOMATIC
DELINEATION from the WATERSHED DELINEATOR menu.

R. Srinivasan, [email protected] Texas

32 A & M University 8
Step 3: Burning in a stream network
A stream network theme such as Reach File (V1 or V3) or National Hydrography Dataset (NHD) can be superimposed onto the DEM
to define the location of the stream network.

Step 4: Stream Definition

(Note: For the Stream Definition function of the Watershed Delineator, There are two ways to define the watershed and stream
network. In this section, the method based threshold area will be introduced, while another method based on pre-defined watershed
will be introduced in Appendix I.)

1. In order to use the threshold method to delineate the watershed and stream network, the Flow Direction and Accumulation

needs to be calculated by clicking the button. Stream definition defines both the stream network and subbasin outlets.
A minimum, maximum, and suggested sub watershed area (in hectares) is
shown in the drainage area box (Figure 8). You have the option of changing the
size of the subbasins within the specified range of values. This function plays
an important role in determining the detail of the stream network and the size and Figure 8 Threshold area for stream
and subbasin definition
number of subbasins created. The threshold area defines the drainage area
required to form the beginning of a stream.
2. After setting the threshold value of subbasin, then the user can delineate the stream network and outlets through clicking the

button. The drainage network and stream juncture points, used to define subbasin outlets, are displayed on the DEM
map grid (Error! Reference source not found.).

Note:
 NHD is an enhanced stream network at the scale of 1:100,000. It is based on USGS Digital Line Graph (DLG)
hydrography data integrated with reach-related information from the EPA River Reach File version 3 (RF3).

R. Srinivasan, [email protected] Texas

33 A & M University 9
 Figure 9 Drainage network and subbasin outlets

R. Srinivasan, [email protected] Texas

34 A & M University 10
 Next step is Outlet and Inlet Definition (

3. Figure 10). Before proceeding, you have a number of options: (1) change the threshold area and rerun the stream and outlet
definition routine, (2) add outlet points by importing a table that contains the locations, (3) add outlet points manually, and (4)
remove outlet points. Assuming the outlet and stream definition to be acceptable proceed to inlet definition for the study area.
Inlets represent any point source loading into the study area or the inlets of drainage into the watershed from an upstream
area.

Figure 10 Options for defining Outlet and Inlet

R. Srinivasan, [email protected] Texas

35 A & M University 11
Note
 By specifying the threshold area, we define the stream network for networking.
 It means that a minimum number of cells are required to start delineating the stream.
 The minimum threshold area is for the entire watershed, not for each sub watersheds that are going to be delineated.
 The suggested area given in this window is the average are that could be used.

R. Srinivasan, [email protected] Texas

36 A & M University 12
Step 5: Main Watershed Outlet(s) Selection and Definition

In this step the users will select one or more outlet locations to define the boundary of the main watershed.

Click on the SELECT button to choose the watershed outlet. Draw a box covering the desired outlet locations will set the main
Watershed Outlets. In this example, select 1 outlet at the downstream edge of the masked area (

1. Figure 11) and click the Delineate Watershed button . Select YES in the following dialog to continue with the delineation
of main watershed and subbasins. A prompt box will appear to announce completion of the watershed and subbasin
delineation.
2. The delineated watershed with subbasins will be added to the View. If the delineation is not satisfactory or if the user wants to
select a different outlet for the watershed, click on the Cancel Selection button and repeat.

Click on the Calculate Subbasin Parameters button to estimate the subbasin parameters. This function calculates
basic watershed characteristics from the DEM and sub-watershed themes. It also assigns the necessary subbasin
identification. The results of the calculations are stored as additional fields in the streams and subbasins theme database
files. Click OK to completion of watershed delineation dialog box.

3. Figure 12 shows the delineated watershed with subbasins.

4. Open the Reach or Watershed attribute tables to view the calculated characteristics.

R. Srinivasan, [email protected] Texas

37 A & M University 13
 ** By holding the SHIFT key in your keyboard you can select more than one outlet. This feature allows adjacent watershed to be
simulated at the same time using SWAT. Do not select an outlet at the upstream of another outlet. At least one outlet must be
selected for delineation.

Selected outlet

R. Srinivasan, [email protected] Texas

38 A & M University 14
 Figure 11 Main watershed outlet selection

R. Srinivasan, [email protected] Texas

39 A & M University 15
 Figure 12 Delineated watershed and subbasins

R. Srinivasan, [email protected] Texas

40 A & M University 16
Step 6: Reservoirs

The user can add or remove reservoirs to refine the delineation process. The procedure is
similar to the process of manually adding or removing an outlet. In this exercise, we will
not focus on adding or removing a reservoir. However the general procedure is given
below for reference.

1. To add a reservoir, click on the ADD button and with the cross appearing as mouse pointer, click on the subbasin where the
reservoir will be added. The reservoir will be placed at the outlet of the subbasin. A prompt box will be displayed asking for
verification of the reservoir placement. If the wrong subbasin is listed, click No and repeat step 7(1). If the correct subbasin is
listed, click Yes.
2. To remove a reservoir click on the REMOVE button. Draw a square around the reservoir to be removed by holding the left
mouse button and moving the mouse. A prompt box will be displayed asking for verification of the reservoir removal. If the
wrong subbasin is listed, click No and repeat step 7(2). If the correct subbasin is listed, click Yes.

Step 7: Exit the Watershed Delineation Dialog

Click EXIT in the watershed delineation main dialog
Step 8: View Topographic Report
Open the Topographic Reports through click
the Watershed Reports Button under the
Watershed Delineator menu from Reports of
the project windows. You can view the area and
percentage wise distribution of elevation for the entire watershed and each subbasin.

R. Srinivasan, [email protected] Texas

41 A & M University 17
 Land use, Soil and Slope Definition

The Land Use, Soil and Slope Definition option in the HRU Analysis menu allows the user to specify the land use, soil and slope
themes that will be used for modeling using SWAT and NPSM. These themes are then used to determine the hydrologic response
unit (HRU) distribution in each sub-watershed.

Both NPSM and SWAT require land use data to determine the
area of each land category to be simulated within each subbasin.
In addition to land use information, SWAT relies on soil data to
determine the range of hydrologic characteristics found within
each subbasin. Land Use, Soil and Slope Definition option
guides the user through the process of specifying the data to be
used in the simulation and of ensuring that those data are in the
appropriate format. In particular, the option allows the user to
select land use or soil data that are in either shape or grid
format. Shapefiles are automatically converted to grid, the
format required by ArcGIS to calculate land use and soil
distributions within the subbasins of interest. Select the Land
Use / Soil / Slope Definition option from the HRU Analysis
menu. The Land Use / Soil / Slope Definition dialog box
(Figure 13) will open. The detailed procedures on how to use the
functions contained in this dialog were introduced below:
Figure 13 Dialog for Land Use / Soil / Slope Definition

R. Srinivasan, [email protected] Texas

42 A & M University 18
Step 1: Define Land use theme

1. Select the land use data layer by clicking on the open file folder button next to “Land Use Grid.”
2. A “Set the LandUse Grid” dialog box will appear (Figure 14). You will have the option to “Select Land use layer(s) from the
Map” or “Load Land Use dataset(s) from disk”. Select the Load Land Use
dataset(s) from disk option and click Open. Click Yes for the projection
information dialog box.
3. Select the Landuse grid file in the work directory and click Select. A message
box will indicate the successful loading of landuse theme.

Figure 14 Dialog for options of selecting landuse data

4. After loading the Landuse file into the map, choose the grid field which will
be used as index to define different landuse types. In this example, the
“Value” field is selected. Click OK, then a table titled “SWAT LandUse
Classification Table” will be created automatically by the interface (Figure
15). The first column contains the unique values in the Grid Field chosen
above. The second column contains the area of each type of landuse. And
the third column contains the landuse names in the SWAT database
corresponding to each index value.

Figure 18 SWAT LandUse Classification Table

R. Srinivasan, [email protected] Texas

43 A & M University 19
 5. In order to fill correct values in the third column, the land use grid codes must be assigned a land cover/plant description. You
may import a look-up table or manually assign a land cover/plant code. The
interface includes tables that convert the USGS land use/land cover
classification codes to SWAT land cover/plant codes. If the land use grid being
used is classified by an alternate method, you must create a look-up table or
enter the information manually.
6. Select the land use look-up table by clicking on the open “Look-up Table”
button next to “Table Grid Values –> Land cover classes.” A “Land Cover
Lookup Table” dialog box will appear (Figure 19 ).
7. When the user has land use data and the corresponding four character land
use code, the “User Table” option should be selected. Figure 19 Land cover lookup table options
8. Select User defined lookup table in .txt or .dbf format (the ASCII .txt lookup
table was shown in(Figure 21), then the SWAT LandUse Classification Table will
be populated (Figure 16).
9. Click Reclassify button. Then Landuse map will be reclassified according the
Lookup table. The reclassified landuse map is shown in Figure 18.

Figure 16 Populated SWAT LandUse Classification Table

R. Srinivasan, [email protected] Texas

44 A & M University 20
 Accessing “User Table” from the Landuse and soil definition dialog box:
 The user defined table has to be created in the format as shown in figure below:

Figure 17 ASCII (.txt) table format of lookup table

Note:
1. To manually create a look-up table, double click on the “LandUseSwat” field next to the first category number in the dialog. A
dialog box will appear listing the two database files from which a SWAT land type may be selected: Land Cover/Plant and Urban.
Select the desired database file by clicking on it. Click OK. A dialog box will appear listing the available SWAT land cover codes or
the available SWAT urban land type codes. Select the desired code from the list and click ok. Repeat this procedure for all the
values in the grid.
2. If you do not find the desired land cover in the database, you will have to add the land cover class to the database too.

R. Srinivasan, [email protected] Texas

45 A & M University 21
 Figure 18 Reclassified land use grid

R. Srinivasan, [email protected] Texas

46 A & M University 22
Step 2: Specify soils theme
1. Select the soils data layer by clicking on the open file folder button
under “Soil Grid” in the “Soil Data” tab (Figure 19).
2. A “Select Soils Data” dialog box appears. You have the option to Select
Soil layer(s) from the Map or Load Soils dataset(s) from disk or Load
ArcSWAT US STATSGO from disk. Select the Soil map from the disk
and load it to the map. A message box will indicate the successful loading
of the soil grid theme.
3. After loading the Soil file into the map, choose the field which will be used
as the index to define different soil types. In this example, the “Value” field
is selected. Click OK, then a table titled “SWAT Soil Classification Table”
will be created automatically by the interface. The first column contains the
unique values in the Grid Field chosen above. The second column
contains the area of each soil type. And the third column contains the soil
names in the SWAT database corresponding to each index value, which
need to be filled through a Lookup table.

The soil map grid must be linked to the U.S. soils database (provided with the
interface) or to the User Soils (customized) database. Under “Options,” click the
button beside the method used to link the soils map grids to the soils data. There
are five options available: Figure 19 Interface for setting the soil data
 Stmuid: State STATSGO polygon number, dominant soil phase
 S5id: Soils5ID number for USDA soil series data

R. Srinivasan, [email protected] Texas

47 A & M University 23
  Name: Name of soil in User Soils database
 Stmuid + Seqn: State STATSGO polygon number and sequence number of soil phase
 Name + Stmuid: State STATSGO polygon number and soil series name

4. Select Stmuid, then load look up values for the soil grid file and click the Reclassify button for soils grid. The reclassified
soils grid (Figure 20) is shown in the map.

Note:

 SSURGO soil data can also be used with SWAT. SWAT – SSURGO processing tool is available in
http://lcluc.tamu.edu/ssurgo/ .

R. Srinivasan, [email protected] Texas

48 A & M University 24
 Figure 20 Soils grid theme reclassified by State MUID class

Note:
1. You can manually reclass the soils grid using a procedure similar to the manual reclassification of land
use grid.
2. User defined lookup tables in database or text format can also added by click on the “OPEN FOLDER”

R. Srinivasan, [email protected] Texas

49 A & M University 25
Step 3: Specify slope theme

Slope is an important factor that determines the water, sediment and

nutrients movement. In this new version interface, a new function is
added to take slope into account to delineate HRU. The general
procedures for defining a slope theme will be introduced below.

1. Click the “Slope” tab, then the interface for defining slope theme
appear (Figure 21). There are two options for slope discretization:
“Single Slope” and “Multiple Slope” The “Single Slope” option
denotes that the mean value of slope will be used for for the whole
watershed, whilethe “Multiple Slope” option will allow slope be
classified into several level.

If the “Multiple Slope” option was selected, the user need to define
the “Number of Slope Classes”, and the range (Lower Limit and
Upper Limit) of each slope class. At last the slope classification
results will be stored in the “SWAT Slope Classification Table”.
Click “Reclassify”, the classified slope layer will be added to the map
(Figure 26).

Figure 21 Populated SWAT Slope Classification Table

R. Srinivasan, [email protected] Texas

50 A & M University 26
 Figure 22 Slope grid theme reclassified by user defined criteria

R. Srinivasan, [email protected] Texas

51 A & M University 27
Step 4: Overlay land use, soil and
slope layers

When the land use, soil and slope

layers have all been classified, the
OVERLAY button will be activated.
Click this button; all the three grid
layers will be overlaid. When the
overlay process is completed, an “Info”
message box will appear. Click OK to
proceed the delineation of hydrologic
response units (HRUs). Figure 23
shows the units with unique
combination of landuse, soil, and slope
within each subbasin. This layer will be
used later in defining HRU.

Step 5: View land use, soil and slope

distribution report
Figure 23 Shape file layer of the units with unique combination of landuse, soil and slope
within each subbasin
View the area and percentage distribution
of each land use and soil class within each subbasin by clicking the HRU Analysis Reports button under the HRU analysis menu

R. Srinivasan, [email protected] Texas

52 A & M University 28
 Hydrologic Response Unit Definition
Before executing SWAT, the distribution of hydrologic response units (HRUs) within
the watershed must be determined based on the land use, soil and slope layers
specified in the previous step. The interface allows the user to specify criteria to be
used in determining the HRU distribution. One or more unique land use/soil/slope
combination(s) (hydrologic response units or HRUs) can be created for each subbasin.

Subdividing the watershed into areas having unique land use, soil and slope
combinations enables the model to reflect differences in evapotranspiration for various
crops and soils. Runoff is predicted separately for each HRU and routed to obtain the
total runoff for the watershed. This increases accuracy and gives a much better
physical description of the water balance.

Click the HRU Definition button under the HRU Analysis menu. This will open a
“HRU Definition” dialog (Figure 24). The user has two options in determining the HRU
distribution: assign a single HRU to each subbasin or assign multiple HRUs to each
subbasin. If a single HRU per subbasin is selected, the HRU is determined by the
dominant land use/soil/slope combination within each subbasin. If multiple HRUs are
selected, the user may specify the several criteria for the land use, soil and slope data
that will be used to determine the number and type of HRUs in each subbasin. The
Figure 24 HRU definition dialog
procedures on how to use this tool are introduced below:

R. Srinivasan, [email protected] 29 53
Step 1: Hydrologic Response Unit Definition
1. Select the Multiple Hydrologic Response Units option. Modify the
threshold land use/soil/slope area percentage (%) over subbasin
area to define how detailed the watershed will be represented.
Select the desired threshold values for soil, land use and slope (for
example 10% for land use/soil/slope).
2. Now you can click the “Create HRUs” button to delineate HRU
distribution. Also you can go to the Land Use Refinement tap to
specify more detailed criteria. There are two ways to refine the
definition of HRU. The first one is to split one land use type into two
or several sub land use types. For example the AGRL land use type
can be split into two sub types: AGRL (50%) and AGRR (50%). The
other option is to set one land use type exempt, which mean that this
type of land use will exempt of the area threshold value set in
previous steps. In this example, the settings of these two functions
are shown in Figure 25.
3. After creating HRUs, the distribution report for the HRUs can open
through clicking the HRU Analysis Reports button under the HRU Figure 25 Interface for Land use refinement
Analysis menu.
4. If the distribution is not satisfactory, repeat the preceding steps,
altering the land use and soil sensitivities, until a satisfactory
distribution is obtained.
Figure 26 HRU definition dialog, land use refinement

R. Srinivasan, [email protected] Texas

54 A & M University 30
 Note for HRU distribution
 Selecting Multiple Hydrologic response units option allows us to eliminate minor land uses in each subbasin.
 For example, if we set the threshold for Landuse (%) over subbasin area to 15% landuses that occupy less that 15% of
subbasin area would be eliminated and the HRU will be created for landuses that occupy greater than 15% of the subbasin
area.
 The same holds for Soil and slope layer.

R. Srinivasan, [email protected] Texas

55 A & M University 31
 Soil and Water Assessment Tool (SWAT)

The following are the key procedures necessary for modeling using SWAT.

 Create SWAT project

 Delineate the designated watershed for modeling
 Define land use/soil/slope data grids
 Determine the distribution of HRUs based on the land use and soil data
 Define rainfall, temperature and other weather data
 Write the SWAT input files- requires access to data on soil, weather, land cover, plant growth, fertilizer and pesticide use,
tillage, and urban activities.
 Edit the input files – if necessary
 Setup and run SWAT – requires information on simulation period, PET estimation method and other options
 View SWAT Output

Now we have completed the first three procedures. In this tutorial we will concentrate on preparing the rest of the input data for
SWAT, running the model, and viewing the output from the model.

Note

 Spatial analyst is the main tool that will be used in SWAT. Without this, SWAT simply can’t be used.
 General info about SWAT: All SWAT input and output are in Metric units (MKS)

R. Srinivasan, [email protected] BRC/UT/EPA/BASINS 56

 Write Input Tables for SWAT
This menu contains functions to build database files that include information needed to generate default input for the SWAT model.
The commands on the menu need to be implemented only once for a project. However, if the user modifies the HRU distribution after
building the input database files, these commands must be reprocessed again.

Step 1: Define Weather data

1. Select the Weather Stations button under the Write Input
Tables menu. A “Weather Data Definition” dialog is opened
(Figure 27). This dialog will allow the user to define the input
data for rainfall, temperature and other weather data. For
weather data, you have the option of simulating the data in
the model or to read from data tables. If no observed weather
data is available, then information can be simulated using a
weather generator based on the data from 1041 weather
stations around the US or combination of WBAN (Weather
Bureau Army Navy) and Cooperative stations for four 30-year
Figure 27 Weather Data definition dialog
periods stored in a database or custom weather data can be
input through a database table. The weather generator data
must be defined before you can continue to define the other data, like precipitation and temperature.
2. Select the “WGEN_US_FirstOrder” option for Weather Generator Data to add the weather simulation database automatically.
3. Under the “Rainfall data” tab, select the “Raingages” option for rainfall data. Browse to the Work Directory and choose the
file pcpfork.txt and click Add.

R. Srinivasan, [email protected] Texas

57 A & M University 31
 4. Under the “Temperature data” tab, select the “Climate Stations” radio button option for temperature data. Browse to the Work
Directory and choose the file tmpfork.txt and click Add.

5. After selecting the rainfall, temperature and weather generator data, click OK to generate the SWAT weather input data files.
The locations of weather generator, rainfall and temperature gages will be displayed in the map view (Figure 28).
6. A message box will indicate successful generation of SWAT weather input database.

At this point you have the option to generate all the input data files using the WRITE ALL option under the INPUT menu or generate
each input file separately. The input files needed are:

 Watershed Configuration file (.fig)  Pond input (.pnd)

 Soil data (.sol)  Management Input (.mgt)
 Weather generator data (.wgn)  Main channel data (.rte)
 General HRU data (.sub)  Ground water data (.gw)
 Soil chemical input (.chm)  Water use data (.wus)
 Stream water quality input (.swq)

Note

In the new version of ArcSWAT (2012), the user can modify the weather data files later without rewriting the input
tables. In the previous version (2009) the input files needed to be rewritten after weather data modification and the
model parameterizations set to the default values.

R. Srinivasan, [email protected] Texas

58 A & M University 32
 Figure 28 Location of weather stations

R. Srinivasan, [email protected] Texas

59 A & M University 33
Step 2: Write All (Figure 29)
1. Select the Write SWAT Database Tables option under the Write Input Tables menu.
2. A “Current Status of Input Data” dialog appears. This window shows the
list of input files to be generated and the status of input file generation.
Click Select All and Create Tables.
3. For the management input file generation a message prompt verifies if
the US weather database is sufficient to estimate the Plant Heat Units.
Click Yes if the study area is within US.
4. A message box indicates the successful completion of input files
generation.

Figure 29 Write all input files status dialog

R. Srinivasan, [email protected] Texas

60 A & M University 34
 Edit SWAT Input

The commands listed under the Edit SWAT Input menu bring up dialog boxes that allow you to alter default SWAT input data. The
Edit SWAT Input menu can be used to make input modifications during the model calibration process. In this exercise you are not
required to edit any input information.

However a general procedure is given to familiarize you with the SWAT input files
and editing capabilities in ArcSWAT.

Step 1: Editing Databases

1. Select the Edit Databases button under the Edit SWAT Input menu. You
will be given a list of options to choose the databases to be edited (Figure 30).
Step 1-1: Edit Soils database (Figure 31) Figure 30 List of SWAT databases
1. Select User Soils option, and click OK.
2. “User Soils Edit” dialog box with a list of abbreviated soil names appears. Click on a soil name to edit the entire soil profile
data or individual soil layer information.
3. You can also add new soil into the database by clicking on the Add New button in the bottom of the “Add and Edit User Soil”
dialog box.
4. Click EXIT after completion of editing the database. A prompt box will give you the option to save or ignore the changes made
to soils database.
By using a procedure similar to editing soils database, the Database option under the Edit Input menu you can edit or add
information to the weather, land cover/plant growth, fertilizer, pesticide, tillage, and urban area databases.

R. Srinivasan, [email protected] Texas

61 A & M University 35
 Figure 31 Edit soils database dialog

Note:
 Moving the mouse pointer near an object (text box, radio button etc.,) in any of the edit input dialog box will display a short
description of the parameter contained in the object.

R. Srinivasan, [email protected] Texas

62 A & M University 36
Step 2: Edit Point Discharge Inputs
1. Select the Point Sources Discharges option under
the Edit SWAT Input menu. “Edit Point Source
Inputs” dialog box (Figure 32) with a list of subbasins
containing point discharges will appear.
2. Click on the subbasin number whose point
discharge database needs to be edited. A “Point
Discharges Data” dialog box appears with the list of
attributes of the point data. The dialog box allows
the input of point source data in one of four formats:
constant daily loadings, average annual loadings,
average monthly loadings, and daily loadings.
3. Choose a format by clicking the button next to the
format to be used. The default point source data
format option is constant daily loadings. If you
select this format, you will have the option of either
inputting average daily flow (m3/s), sediment loading Figure 32 Edit Point Discharge Inputs Dialog
(tons), and organic N, organic P, NO3, mineral P
loadings (all in kg), three conservative metals, and 2 categories of bacteria or load PCS data directly. If you select the
“Annual Records” option you will be prompted to load the data from disk by clicking on the open folder button or from PCS by
clicking on the Load PCS button. If you select the “Monthly Records” or “Daily Records” option you will be prompted to load
data from the disk. Click OK to complete the editing of point discharges database for a subbasin.

R. Srinivasan, [email protected] Texas

63 A & M University 37
 4. If you wish to edit the point sources in another subbasin, select it from the list in the “Edit Point Source Inputs” dialog box.
Click Exit to complete editing of point discharges database in all subbasins.

Step 3: Edit Inlet Dischargers Input

1. Select the Inlet Discharges option under the Edit
Input menu
2. If there are any inlet dischargers in the project, “Edit
Inlet Input” dialog box with a list of subbasins containing
inlet dischargers will appear (Figure 33).
3. You will be able to modify the inlet input information
using a procedure similar to editing the point
dischargers database
4. The dialog box allows the input of inlet discharge data
in one of four formats: constant daily loadings, average
annual loadings, average monthly loadings, and daily
loadings. Choose a format by clicking the button next to
the format to be used.
5. The default inlet discharge data format option is
constant daily loadings. If you select this format, you will be Figure 33 Edit Inlet Discharge Inputs Dialog
prompted to input average daily flow (m3), sediment loading (tons), and organic N, organic P, NO3 and mineral P loadings.
6. If you choose “Annual Records”, “Monthly Records” or “Daily Records” option you will be prompted to load the data from the
disk.
7. Click OK to complete the editing of inlet discharges for a subbasin.

R. Srinivasan, [email protected] Texas

64 A & M University 38
 8. If you wish to edit the inlets in another subbasin, select it from the list in the “Edit Inlet Discharger Input” dialog box. Click Exit
to complete editing of inlet dischargers input in all subbasins
9. Since there are no inlet dischargers defined in this tutorial you will get a message “No Inlet Discharges in the Watershed”

Step 4: Edit Reservoir Input

1. To edit the reservoirs, on the Edit Input menu, select Reservoirs. A dialog box will appear with a list of the subbasins
containing reservoirs.
2. To edit the reservoirs within a subbasin, click on the number of the subbasin in the “Edit Reservoirs Inputs” dialog box.
3. Since there are no reservoirs defined in this project you will get a message “No Reservoirs in the Watershed”.

Step 5: Edit Subbasins Data

1. To edit the subbasin input files, select the Subbasins Data
option under the Edit Input menu. “Edit Subbasin Inputs”
dialog box will appear (Figure 37).
This dialog box contains the list of subbasins, land uses, soil
types and slope levels within each subbasin and the input
files corresponding to each subbasin/land use/soil/slope
combination. To select an input file, select the subbasin, land
use, soil type and slope that you would like to edit. When you Figure 34 Edit Subbasin Inputs main dialog
select a subbasin, the combo box of land uses, soil types, and
slope levels will be activated in sequence. Specify the subbasin/land use/soil combination of interest by selecting each
category in the combo box.

R. Srinivasan, [email protected] Texas

65 A & M University 39
 2. To edit the soil physical data, click on the .sol extension, and select the subbasin number, landuse type, soil type and slope
level. Then the OK button is activated. Click OK; a new dialog box will appear (Figure 35). Click the Edit Values button; all the
boxes are activated and the user can revise the default values.
3. The interface allows the user to save the revision of current .sol file to other .sol files. Three options are available: 1) extend
edits to current HRU, which is the default setting, 2) extend edits to all HRUs, or 3) extend edits to selected HRUs. For the
third option, the user needs to specify the subbasin number, landuse type, soil type and slope levels for the HRUs that the
user wants to apply current .sol file parameters.

Figure 35 Edit soil input file dialog

R. Srinivasan, [email protected] Texas

66 A & M University 40
 4. To edit the weather generator data click on the .wgn extension. For the .wgn file you only need to select the subbasin
number, and then the OK button will be activated. Click OK; a new dialog box (Figure 36) will appear which will allow you to
modify the data in .wgn file. Similar to .sol file, the interface also allow the user to extend the current edits to other subbasins.
The user can select to 1) extend edits to current Subbasin, which is the default setting, 2) extend edits to all Subbasins, or 3)
extend edits to selected Subbasins.

Figure 36 Edit weather generator input file main dialog

R. Srinivasan, [email protected] Texas

67 A & M University 41
 5. To edit general subbasin data, click on the .sub
extension in the “Select Input File”. Select the subbasin
number and click OK, then a new dialog box will appear
displaying the existing general subbasin data for the
selected subbasin (Figure 37). To modify data, activate all
fields by clicking Edit Values button. For the elevation
band parameters, the user can choose ELEVB,
ELEVB_FR and SNOEB in the combo box beside the
Elevation Band, and then change the parameter values
for each band. In this version (ArcSWAT2012) the user
can modify the snow parameters over the elevation
bands while in the previous version the snow
parameters were constant across the entire watershed.
Also, the user can choose RFINC, TMPINC, RADINC,
HUMINC in the combo box aside of Weather
Adjustment, then change the parameter values for each
month. Once you have made all editing changes, click
the OK button. The interface will save all changes and
return you to the “Edit Subbasin Inputs” box. The user can Figure 37 Edit Subbasin (Sub) dialog
select to 1) extend edits to current Subbasin, which is the
default setting, 2) extend edits to all Subbasins, or 3) extend edits to selected Subbasins. For selecting multiple subbasins,
hold the Shift key when clicking the preferred subbasin numbers.

R. Srinivasan, [email protected] Texas

68 A & M University 42
 6. To edit general HRU data, click on the .hru
extension in the “Select Input File” section of the
“Edit Subbasin Inputs” dialog box (Figure 38). A
new dialog box will appear displaying the
existing general HRU data for the selected
subbasin. To modify data, activate a field by
positioning the cursor over the text box and
clicking. Once a cursor appears in the field,
make the desired changes. Once you have
made all editing changes, click the OK button.
If you do not want to copy the edited HRU
generator data to other data sets, click No on
the prompt dialog. In this version of ArcSWAT
two new modules: Carbon and Urban BMP, and
Drainage parameters can be modified in the
HRU scale by user.

Figure 38 Edit Hydrologic Response Units (HRU) dialog.

R. Srinivasan, [email protected] Texas

69 A & M University 43
 7. To edit the main channel input file, click on the .rte in the “Select Input file” section of “Edit Subbasin Inputs” dialog box. A new
dialog box (Figure 39) will appear with the existing main channel data for the selected subbasin. Click Edit Values button to
activate all the textboxes to all user’s modification. Also the user can extend current edits to other basins with three types of
options.

Figure 39 Edit Main channel input data dialog

R. Srinivasan, [email protected] Texas

70 A & M University 44
 8. To edit the ground water input file, click on the .gw in the “Select Input file” section of “Edit Subbasin Inputs” dialog box. In the
dialog box that opens (Figure 40) with the existing data, make modifications by clicking Edit Values button to activating all
textboxes. Similarly, after the modification, the user has three options to Save Edits to other HRUs.

9. To edit the consumptive water use input data, click on the .wus in the “Select Input file” section of “Edit Subbasin Inputs”
dialog box. In the dialog box (Figure 41) that opens with the existing data, click Edit Values button, then the user can modify
the data. Also the current edits can be saved to other subbasins.

Figure 40 Edit Ground Water input data dialog

R. Srinivasan, [email protected] Texas

71 A & M University 45
 Figure 41 Edit water use input data dialog

10. To edit the management file input data, click on the .mgt in the “Select Input file” section of “Edit Subbasin Inputs” dialog box.
A new dialog box (Figure 42) will appear and display the management data editor. This dialog has two tabs: General
Parameters and Operations. In the first tab the user can modify the general parameters concerned with Initial Plant Growth,
General Management, Urban Management, Irrigation Management, and Tile Darin Management. In the second tab, the user
can arrange the detailed management options on the current HRU. The management operations can be scheduled by Date
or by Heat Units. The settings of the management operations can also be extended to other HRUs that the user has defined.

R. Srinivasan, [email protected] Texas

72 A & M University 46
 Figure 42 Edit management input file main dialog

R. Srinivasan, [email protected] Texas

73 A & M University 47
 11. To edit the soil chemical data click on the .chm in the “Select Input file” section of “Edit Subbasin Inputs” dialog box. A new
dialog box (Figure 43) will appear displaying the Soil Chemical data editor. To modify the displayed data, click the Edit Values
button to activate all the textboxes. After the modification of Soil Chemical Data, the user also can extend the modification to
other user specified HRUs.

Figure 43 Soil chemical input data editor

R. Srinivasan, [email protected] Texas

74 A & M University 48
 12. To edit pond data click on the .pnd in the “Select Input file” section of “Edit Subbasin Inputs” dialog box. A new dialog box
(Figure 44) will appear displaying the pond data editor. To modify the displayed data, click the Edit Values button to activate all
the textboxes. After the modification of pond data, the user also can extend the modification to other user specified
Subbasins.

Figure 44 Dialog to edit pond input data file

R. Srinivasan, [email protected] Texas

75 A & M University 49
 13. To edit stream water quality input data file click on the .swq in the “Select Input file” section of “Edit Subbasin Inputs” dialog
box. A new dialog box (Figure 45) will appear displaying the stream water quality input data editor. To modify the displayed
data, click the Edit Values button to activate all the textboxes. After the modification of stream water quality input data, the
user also can extend the modification to other user specified Subbasins.

Figure 45 Stream water quality input data editor

R. Srinivasan, [email protected] Texas

76 A & M University 50
 14. To edit septic input data file click on the .sep in the “Select Input file” section of “Edit Subbasin Inputs” dialog box. A new
dialog box (Figure 45) will appear displaying the septic input data editor. To modify the displayed data, click the Edit Values
button to activate all the textboxes. After the modification of septic input data, the user also can extend the modification to
other user specified Subbasins.

Figure 46 Septic input data Editor

R. Srinivasan, [email protected] Texas

77 A & M University 51
 15. To edit Operations input data file click on the .sep in the “Select Input file” section of “Edit Subbasin Inputs” dialog box. A new
dialog box (Figure 45) will appear displaying the septic input data editor. To modify the displayed data, click the Edit Values
button to activate all the textboxes. After the modification of septic input data, the user also can extend the modification to
other user specified Subbasins.

Figure 50 Operations parameters Editor

R. Srinivasan, [email protected] Texas

78 A & M University 52
 Step 6. Edit Watershed Data
1. Go to the Watershed Data item under the Edit SWAT Input menu, and click the General Data (.BSN) button, then
a new dialog (Error! Reference source not found.) will appear.

This interface allows you to modify the parameters concerned with three major groups:
1) Water Balance, Surface Runoff, and Reaches,
2) Nutrients and Water Quality, and
3) Basin-wide Management. After revision of the parameters, click Save Edits.

Figure 51 General Watershed Parameters Editor

R. Srinivasan, [email protected] Texas

79 A & M University 53
 2. Go to the Watershed Data item under the Edit SWAT Input menu, and click the Water Quality Data (.WWQ)
button, then a new dialog (Figure ) will appear. This interface allows you to modify the parameters concerned with
Watershed Water Quality Simulation. After revision of the parameters, click Save Edits.

Figure 52 Watershed Water Quality Parameters

R. Srinivasan, [email protected] Texas

80 A & M University 54
 3. Go to the Watershed Data item under the Edit SWAT Input menu, and click the Land Use Update (.LUP) button, then
a new dialog (Figure ) will appear. This interface allows you to modify the parameters concerned with Watershed
Water Quality Simulation. After revision of the parameters, click Save Edits.

Figure 53 Watershed Land Use Update

R. Srinivasan, [email protected] Texas

81 A & M University 55
Step 7. Integrate APEX Model
The Integrate APEX Model commands allow the user to specify subbasins within the current SWAT model that they wish
to simulate using an existing APEX model. To use this option, the APEX model must have been set up with the ArcGIS
APEX interface using the option to link integrate the APEX model with an existing SWAT model. In addition the APEX
model must be run independently of the SWAT model. Choosing the Integrate APEX Model option will open a new dialog
that will allow the user to define the subbasins and associated APEX model inputs (Figure 54).

1. Select the subbasind that you want to

simulate using APEX form the select
APEX Subbasin list box and click Add
Subbasin (Figure 54)
2. Brows the appropriate APEX TxtInOut
folder that contain the input and output of
the APEX model and click the OK button.
3. Browse to the appropriate folder and click
Figure 54. APEX Model input file main dialog
the OK button on the APEX Input dialog. The
folder path selected will be written into the Current APEX Subbasin table.
4. To re-write the SWAT model fig.fig file to incorporate the APEX model for the selected subbasin, click the Update to
APEX Integrated Model button.
5. To reset the fig.fig for the current SWAT model back to original model click to Reset to Original SWAT Model button.

R. Srinivasan, [email protected] Texas

82 A & M University 56
 SWAT Simulation Setup
SWAT simulation menu contains commands that setup and run SWAT simulation. To build SWAT input files and run the simulation,
proceed as follows:

Step 1: Setup data and Run SWAT

1. Select the Run SWAT command under the SWAT Simulation menu. It will open a dialog box (Figure 55) that will allow you to
set up the data for SWAT simulation.

Figure 55 SWAT data setup and simulation dialog

R. Srinivasan, [email protected] Texas

83 A & M University 57
 2. Select the 1/1/1977 for the “Starting date” and 12/31/1978 for the “Ending date” option. If you are using simulated rainfall
and temperature data, both these fields will be blank and you have to input the information manually.
3. Choose “monthly” option for Printout Frequency
4. Keep the rest at the default selections.
5. After all the parameters have been set, click the Setup SWAT Run button in the “Run SWAT” dialog box (Figure ) to build the
SWAT CIO, COD, PCP.PCP and TMP.TMP input files. Once all input files are setup, the Run SWAT button is activated in the
bottom right of the Run SWAT dialog.
6. Click the button labeled Run SWAT. This will run the SWAT executable file. A message box will indicate the successful
completion of SWAT run.

Step 2: Read SWAT Output

1. On the SWAT Simulation menu, click Read SWAT Output (Figure 56).
2. A dialog box will be brought up (Figure 57)

Figure 56. SWAT Simulation menu

R. Srinivasan, [email protected] Texas

84 A & M University 58
 3. To view the “Output.std” file, click the Open output.std button.
4. To view the “Output.std” file, click the Open output.std button.
5. To import selected SWAT files into an Access database, check
the checkboxes associated with the output files of interest, and
then click the Import Files to Database button. The selected
output files will be converted to Access database tables in the
“workdirectory\SWAT_Demo\Scenarios\Default\TablesOut\SwatO
utput.mdb”.
6. To perform simple check to identify potential model problems,
click the “SwatCheck” button.
7. Now, save the current SWAT simulation as “Sim1”
8. Type in a name (e.g. Sim1) for the current run
9. The button Save Simulation is activated now. You need to save
the current setting of the SWAT project to another folder. Click
the Save Simulation button, and input “Sim1” as the name of
Figure 57. Dialog of saving SWAT run

current SWAT Run. Click OK. Then the interface will copy the files
under “workdirectory\SWAT_Demo\Scenarios\Default” to
“workdirectory\SWAT_Demo\Scenarios\Sim1”. And a dialog will
appear to notify you that the current SWAT run has been saved as
“Sim1” (Error! Reference source not found.).

Figure 58. Dialog of successfully saved SWAT run

R. Srinivasan, [email protected] Texas

85 A & M University 59
 Step 3: Set Default Simulation
To choose saved simulation (Sim1) to be set as the current default simulation and input dataset, proceed as follows:
1. On the SWAT Simulation menu, click Set Default Simulation (Figure 59).
2. A dialog box will be displayed (Figure 60)
3. Choose a saved simulation “Sim1” from the Select Simulation box to set as the
default simulation and click the Copy to Default button. The current SWAT project
geodatabase and the contents of the
“Default\TxtInOut\” folder are overwritten with the geodatabase found in the
Figure 59. SWAT Simulation menu
“TablesIn\” folder and the “TxtInOut\” folder contents from the selected
simulation respectively. A dialog will appear to notify you that the current SWAT run will be replaces by the saved simulation
“Sim1” (Error! Reference source not found.).

Figure 60. SWAT Simulation menu Figure 61. Replacing the default simulation with the saved simulation

R. Srinivasan, [email protected] Texas

86 A & M University 60
 Step 4: Manual Calibration Helper
This dialog provides a simple method for making adjustments to parameters across a
user-defined group of HRUs or subbasins during the manual calibration process.
1. On the SWAT Simulation menu, click Manual Calibration Helper (Figure 62).
2. A dialog box will be displayed (Figure 63)

3. A parameter to adjust is selected by choosing a parameter from the Select

Figure 62. SWAT Simulation menu
Parameter combo box.
4. Select a mathematical operation (Multiply By, Add, Replace Value)
in the Mathematical Op combo box.
5. Enter in a value to multiply by, add or replace in the Value box.
6. Specify the subbasins or HRUs that the parameter adjustment will
apply. In Figure 63, the parameter adjustment has been chosen to
be applied to all “PAST” HRUs of Subbasin 1. Click Update
Parameter.
7. When completed a message will appear indicating the changes
have been made.

Note:
 The user is responsible for keeping track of the changes made to
parameters when using the Manual Calibration Helper. It is possible to
make unrealistic parameter modifications using this tool, particularly
for a novice user. It is always recommended to save a SWAT
simulation prior to making significant changes to parameter values
Figure 63. Dialog of Manual Calibration
during calibration.

R. Srinivasan, [email protected] Texas

87 A & M University 61
 Appendix: Installing ArcSWAT
System Requirements

The SWAT2012/ArcSWAT 2012.10_0.1 Beta8 Interface requires:

Hardware:

 Personal computer using a Pentium IV processor or higher, which runs at 2 gigahertz or faster
 1 GB RAM minimum
 500 megabytes free memory on the hard drive for minimal installation and up to 1.25 gigabyte for a full installation
(including sample datasets and US STATSGO data)

Software (ArcSWAT 2012.10_0.1 for ArcGIS 10 version):

 Microsoft Windows XP, or Windows 2000 operating system with most recent kernel patch*
 ArcGIS-ArcView 10 with service pack 5 (Build 4400)
 ArcGIS Spatial Analyst 10 extension
 ArcGIS Developer Kit (usually found in C:\Program Files\ArcGIS\DeveloperKit\)
 ArcGIS DotNet support (usually found in C:\Program Files\ArcGIS\DotNet\)
 Microsoft .Net Framework 2.0
 Adobe Acrobat Reader version 7 or higher

Microsoft constantly updates the different versions of windows. This interface was developed with the latest version of Windows
and may not run with earlier versions. Patches are available from Microsoft.

R. Srinivasan, [email protected] Texas

88 A & M University 62
 Using the ArcSWAT Setup Wizard:

After downloading the ArcSWAT program, open the ArcSWAT_Install_1.0.0 folder. Click the icon to begin
installation. Follow the installation wizard instructions.

Select the appropriate folder location for the program, preferably the computer’s main hard drive. Click the Disk Cost button to ensure enough disk space
for installation.
Indicate if program access will be for everyone who uses the computer or just the installer.

R. Srinivasan, [email protected] Texas

89 A & M University 63
 Follow the wizard instructions until installation is completed successfully.

Make sure the SWAT_US_Soils.zip file is also downloaded and unzipped under ArcSWAT\databases directory

Note:
 After installation is complete, check that the downloaded data is located in the correct location:
C:\Program Files\ArcSWAT\Databases\ and contains both the SWAT2012.mdb and SWAT_US_Soils.mdb databases.
R. Srinivasan, [email protected] Texas
90 A & M University 64
Additional information on ArcSWAT installation:
What are the build numbers for all the recent releases of ArcGIS?

http://support.esri.com/index.cfm?fa=knowledgebase.techArticles.articleShow&d=30104

You need ArcGIS Desktop 10 Service Pack 5:

http://support.esri.com/en/downloads/patches-ServicePacks/view/productid/159/metaid/1892

How to install ArcGIS 10 with .NET support?

1. Insert the ArcView or ArcGIS Desktop installation disk.

2. Select Install ArcGIS Desktop.

3. Select Modify.

4. Expand Applications; verify that '.NET Support' is installed. If

you see a red X, click on the X and select 'Entire feature will be
installed' and then follow the rest of the wizard.

R. Srinivasan, [email protected] Texas

91 A & M University 65
92
 Section 3

SWAT Calibration Techniques

93
94
 SWAT Calibration Techniques

Calibration, Validation & Verification

 CALIBRATION: model testing with known input
and output used to adjust or estimate factors

 VALIDATION: comparison of model results with

an independent data set (without further
adjustment).

 VERIFICATION: examination of the numerical

technique in the computer code to ascertain that it
truly represents the conceptual model and that
there are no inherent numerical problems

Calibration/Validation Periods
• distinct time period
• similar range of
conditions
• adequate time period to
simulate conditions

Time

Setup Calibration Validation

95
 Model Configuration
 Land use categories
– land use types in watershed, existing and future land
uses, management techniques employed, management
questions
 Subwatersheds
– location, physical characteristics/soils, gaging station
locations, topographic features, management questions.
 Reaches
– topographic features, stream morphology, cross-section
data available

Calibration Issues:
• individual land use parameter determination
• location of gaging station data
• location of water quality monitoring information
• available information on stream systems

Model Configuration
Calibration Points Example

LEGEND

Calibration/Validation
Procedures
 Hydrology - first and foremost
 Sediment - next
 Water quality - last (nitrogen, phosphorus,
pesticides, DO, bacteria)

 Check list for model testing

 water balance - is it all accounted for?
 time series
 annual total - stream flow & base flow
 monthly/seasonal total
 frequency duration curve
 sediment and nutrients balance

96
 Calibration Time Step

 Calibration sequence
– annual water balance
– seasonal variability
– storm variability
 time series plot
 frequency duration curve
– baseflow
– overall time series

Calibration/Validation
Statistics

– Mean and standard deviation of the

simulated and measured data
– Slope, intercept and regression
coefficient/coefficient of determination
– Nash-Suttcliffe Efficiency

Calibration/Validation
Common Problems
 too little data - too short a monitoring period
 small range of conditions
– only small storms
– only storms during the spring...

 prediction of future conditions which are

outside the model conditions
 calibration/validation does not adequately
test separate pieces of model
– accuracy of each land use category prediction
 calibration adjustments destroy physical
representation of system by model
 adjustment of the wrong parameters

97
 Calibration/Validation
Suggested References
 Neitsch, S. L., J. G. Arnold, J. R. Kiniry and J. R. Willams. 2001. Soil and Water
Assessment Tool – Manual, USDA-ARS Publications. pp: 341-354.
http://www.brc.tamus.edu/swat/manual.

 Santhi, C., J. G. Arnold, J. R. Williams, W. A. Dugas, R. Srinivasan and L. M. Hauck.

2001. Validation of the SWAT Model on a Large River Basin with Point and Nonpoint
Sources. J. American Water Resources Association 37(5): 1169-1188.

 Srinivasan, R., T. S. Ramanarayanan, J. G. Arnold and S. T. Bednarz. 1997. Large area

hydrologic modeling and assessment: Part II - Model application. J. American Water
Resources Association 34(1): 91-102.

 Arnold, J.G., R. S. Muttiah, R. Srinivasan and P. M. Allen. 2000. Regional estimation of

baseflow and groundwater recharge in the upper Mississippi basin. J. Hydrology
227(2000): 21-40.

Hydrology Calibration
Summary
 Key considerations
– Water balance
 overall amount
 distribution among hydrologic components
– Storm sequence
 time lag or shifts
– time of concentration, travel time
 shape of hydrograph
– peak
– recession
– consider antecedent conditions

Example Calibration Plot

98
 Example Calibration Plot

Hydrologic Calibration
Scenario 1

Simulated
Observed
Flow (cfs)

Time (hours)

Hydrologic Calibration
Model failed to simulate some peak
flows
 Rainfall station is not
Simulated
representative Observed
Flow (cfs)

 Localized storm -no

response
 Malfunctioning gages
(precipitation or flow) Time (hours)

Solutions
 Use precipitation data from representative
meteorological stations
 Carefully review precipitation and flow data for the
particular duration

99
 Hydrologic Calibration
Scenario 2

Simulated
Observed
Flow (cfs)

Time (hours)

Hydrologic Calibration
Model consistently over predicts the
flow
 High Surface flow Simulated
Observed
Flow (cfs)

Time (hours)
Solutions
 Decrease curve number for different land uses (CN in
.mgt)
 Soil available water (SOL_AWC in .sol)
 Soil evaporation compensation factor (ESCO in *.bsn)

Hydrologic Calibration
Model consistently over predicts the
flow
 High base flow Simulated
Observed

 Too little
Flow (cfs)

evapotranspiration

Time (hours)
Solutions
 Increase deep percolation loss (Adjust threshold depth of
water in shallow aquifer required for the base flow to
occur) (GWQMN in .gw)
 Increase groundwater revap coefficient (GW_REVAP in
.gw)
 Decrease theshold depth of water in shallow aquifer for
revap to occur (REVAPMN in .gw)

100
 Hydrologic Calibration
Scenario 3

Simulated
Observed
Flow (cfs)

Time (hours)

Hydrologic Calibration
Simulated flow follows the observed pattern
but lags the actual flow consistently

 Time of concentration is
Simulated
too long Observed
Flow (cfs)

 Less than actual slope

for overland flow
 Over estimated surface
roughness Time (hours)

Solutions
 Adjust slope for over land flow (SLOPE in .hru)
 Adjust Manning’s roughness coefficient (OV_N in
.sub or .rte)
 Adjust the value of overland flow length
(SLSUBBSN in .sub or .hru), if necessary

Hydrologic Calibration
Scenario 4

Simulated
Observed
Flow (cfs)

Time (hours)

101
 Hydrologic Calibration
Simulated flow over predicts peak flows but
under predicts all other times

Simulated
 Too little base flow Observed

Flow (cfs)
 Too high surface runoff

Time (hours)

Solutions
 Adjust infiltration
 Adjust interflow
 Adjust base flow recession parameter

Sediment Calibration Summary

 Key considerations
– Sources of sediment loadings
 Loadings from HRUs/Subbasins
 Channel degradation/deposition
– Sediment loading distribution
 overall amount
 Seasonal loading
– distribution by storm sequence
• rising and falling limb of hydrograph
• peak concentration

Example Calibration Plot

102
 Sediment Calibration
Scenario 1

Sediment Calibration
Model consistently under predicts the
sediment
 Low sediment yield

Solutions
 Calibrate HRU/Subbasin Loadings
– Adjust USLE crop management factor (P) (USLE_P in .mgt)
– Adjust USLE slope length factor (LS) (SLSUBBSN in .sub or .hru)
– Adjust the slope of HRUs (SLOPE in .hru)
– Adjust crop practice factor (C) for land use (USLE_C in crop.dat)
– Verify tillage operations in *.mgt files and adjust crop residue
coefficient ( RSDCO) and bio-mixing efficiency (BIOMIX) in .bsn
 Calibrate Channel degradation/deposition
– Linear and exponential parameters used for channel sediment
routing (SPCON and SPEXP in .bsn)
– Channel erodibility facor (CH_EROD in .rte)
– Channel cover factor (CH_COV in .rte)

Nutrients Calibration Summary

 Key considerations
– Sources of nutrients loadings
 Loadings from HRUs/Subbasins
 In-stream processes
– Nutrient loading distribution
 overall amount
 Seasonal loading
– distribution by storm sequence
• rising and falling limb of hydrograph
• peak concentration

103
 Example Calibration Plot

Monthly calibration of nitrogen at Hico, Bosque Watershed, TX

Mineral Nitrogen Calibration

Scenario 1

Mineral Nitrogen Calibration

Model consistently under predicts the
mineral nitrogen
 Low mineral nitrogen
loading
Solutions
 Calibrate mineral nitrogen loadings
– Adjust initial concentration of the nutrient in soils (SOL_NO3 in
.chm)
– Verify fertilizer application rates and adjust fertilizer application
fraction to surface layer as 0.20 (FRT_LY1 in .mgt)
– Verify tillage operations in *.mgt files and adjust crop residue
coefficient ( RSDCO) and bio-mixing efficiency (BIOMIX) in .bsn
– Adjust nitrogen percolation coefficient (NPERCO in .bsn)
 Calibrate in-stream mineral nitrogen processes
– Adjust fraction of algal biomass that is as nitrogen for water
quality (AI1 in.wwq)

104
 Organic Nitrogen Calibration
Scenario 1

Organic Nitrogen Calibration

Model consistently under predicts the
organic nitrogen
 Low Organic nitrogen
loading

Solutions
 Calibrate organic nitrogen loadings
– Adjust initial concentration of the nutrient in soils (SOL_ORGN in
.chm)
– Verify fertilizer application rates and adjust fertilizer application
fraction to surface layer as 0.20 (FRT_LY1 in .mgt)
 Calibrate in-stream organic nitrogen processes
– Adjust fraction of algal biomass that is as nitrogen for water
quality (AI1 in.wwq)

Soluble Phosphorus Calibration

Scenario 1

105
Soluble Phosphorus Calibration
Model consistently under predicts the
soluble phosphorus
 Low soluble phosphorus
loading
Solutions
 Calibrate soluble phosphorus loadings
– Adjust initial concentration of the nutrient in soils (SOL_MINP in
.chm)
– Verify fertilizer application rates and adjust fertilizer application
fraction to surface layer as 0.20 (FRT_LY1 in .mgt)
– Verify tillage operations in *.mgt files and adjust crop residue
coefficient ( RSDCO) and bio-mixing efficiency (BIOMIX) in .bsn
– Adjust phosphorus percolation coefficient (PPERCO in .bsn)
– Adjust phosphorus soil partitioning coefficient (PHOSKD in .bsn)
 Calibrate in-stream soluble phosphorus processes
– Adjust fraction of algal biomass that is as phosphorus for water
quality (AI2 in.wwq)

Organic Phosphorus Calibration

Scenario 1

Organic Phosphorus Calibration

Model consistently under predicts the
organic phosphorus
 Low organic phosphorus
loading

Solutions
 Calibrate organic phosphorus loadings
– Adjust initial concentration of the nutrient in soils (SOL_ORGP in
.chm)
– Verify fertilizer application rates and adjust fertilizer application
fraction to surface layer as 0.20 (FRT_LY1 in .mgt)

 Calibrate in-stream organic phosphorus processes

– Adjust fraction of algal biomass that is as phosphorus for water
quality (AI2 in.wwq)

106
 Section 4

SWAT Publications
Validation of SWAT on a Large River Basin
with Point and Nonpoint Sources

Information to Run SWAT Model

SSURGO Processing for ArcSWAT
Model Evaluation Guidelines for
Systematic Quantification of Accuracy in
Watershed Simulations

107
 JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION
VOL. 37, NO. 5 AMERICAN WATER RESOURCES ASSOCIATION OCTOBER 2001

VALIDATION OF THE SWAT MODEL ON A LARGE RWER

BASIN WITH POINT AND NONPOINT SOURCES1

C. Santhi, J. G. Arnold, J. R. Williams, W. A. Dugas, R. Srinivasan, and L. M. Hauck2

ABSTRACT: The State of Texas has initiated the development of a (USEPA, 1998). To restore the quality of these water
Total Maximum Daily Load program in the Bosque River Water-
shed, where point and nonpoint sources of pollution are a concern.
bodies, the Total Maximum Daily Load (TMDL) pro-
Soil Water Assessment Tool (SWAT) was validated for flow, sedi- cess was established by section 303(d) of the Clean
ment, and nutrients in the watershed to evaluate alternative man- Water Act (CWA). A TMDL quantifies pollutant
agement scenarios and estimate their effects in controlling sources and maximum allowable loads to the con-
pollution. This paper discusses the calibration and validation at tributing point and nonpoint sources so that the
two locations, Hico and Valley Mills, along the North Bosque River.
Calibration for flow was performed from 1960 through 1998. Sedi-
water quality standards are attained to protect drink-
ment and nutrient calibration was done from 1993 through 1997 at ing water, aquatic life, and other water uses (USEPA,
Hico and from 1996 through 1997 at Valley Mills. Model validation 1998). Once the necessary pollutant reduction levels
was performed for 1998. Time series plots and statistical measures are identified through establishment of TMDLs, con-
were used to verify model predictions. Predicted values generally trol measures such as best management practices
matched well with the observed values during calibration and vali- (BMPs) are to be implemented.
dation (R2 0.6 and Nash-Suttcliffe Efficiency 0.5, in most
instances) except for some underprediction of nitrogen during cali- The Texas Natural Resource Conservation Com-
bration at both locations and sediment and organic nutrients dur- mission (TNRCC) is implementing TMDL projects in
ing validation at Valley Mills. This study showed that SWAT was Texas in cooperation with the Texas State Soil and
able to predict flow, sediment, and nutrients successfully and can Water and Conservation Board (TSSWCB). Of the 386
be used to study the effects of alternative management scenarios.
(KEY TERMS: watershed management; total maximum daily load;
classified water bodies, 147 have been listed on the
erosion; sedimentation; phosphorus loading; dairy manure manage-
1998 303(d) list. One of these projects is in the Bosque
ment.) River Watershed in North Central Texas, where
phosphorus from confined animal-feeding operations
is a concern (TNRCC, 1999). Runoff from cropland
and urban areas (nonpoint sources) and effluent from
INTRODUCTION wastewater treatment plants (WWTPs) (point
sources) are other pollution sources in this watershed.
The United States Environmental Protection Agen- Water quality simulation models can assist with
cy (USEPA) reported nutrient enrichment as the TMDL development through simulating loads to
major cause for impairment of lakes and other water receiving water bodies under various BMPs. Models
bodies in the United States (USEPA, 1994). USEPA's in combination with observational data from histori-
water quality inventory report of 1996 indicated that cal and current monitoring programs will provide the
40 percent of the surveyed rivers, lakes, and estuaries information for TMDL waste/load allocations and
were polluted relative to their designated uses implementation strategies. In this study, the hydro-
logic/water quality model, Soil Water Assessment Tool

1Paper No. 00044 of the Journal of the American Water Resources Association. Discussions are open until June 1, 2002.
2Respectively, Post Doctoral Research Scientist, Blackland Research Center, 720 East Blackland Road, Temple, Texas 76502-9622;
Hydraulic Engineer, USDA/ARS, 808 East Blackland Road, Temple, Texas 76502; Research Scientist, Blackland Research Center, 720 East
Blackland Road, Temple, Texas 76502-9622; Professor and Resident Director, Blackland Research Center, 720 East Blackland Road, Temple,
Texas 76502-9622; Associate Professor and Director of Spatial Sciences Lab, Blackland Research Center, 700 East University Drive, Suite
104, College Station, Texas 77840; and Assistant Director, Texas Institute for Applied Environmental Research, P.O. Box T-0410, Stephenville,
Texas 76402 (E-MaillSanthi: [email protected]).

JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION 1169 JAWRA

 Santhi, Arnold, Williams, Dugas, Srinivasan, and Hauck

(SWAT) was used to study the effect of different BMPs The nitrogen processes modeled by SWAT and the
in reducing the pollution. The objective of this paper various pools of nitrogen in the soil are shown in Fig-
is to describe the calibration and validation of the ure 1 (Arnold et al., 1998; USDA-ARS, 1999). Plant
SWAT model for flow, sediment, organic, and mineral use of nitrogen is estimated using the supply and
nutrients in the Bosque River Watershed. The model demand approach (Williams et al., 1984). Daily plant
was then used to study the impact of different BMPs demand is a function of plant biomass and biomass N
on the Bosque River Watershed and identify suitable concentration. Available nitrogen in the soil (root
BMPs for achieving water quality targets (Santhi et depth) is supplied to the plant. When demand exceeds
al., 2000). supply, there is a nutrient stress. Masses of N03-N
contained in runoff, lateral flow and percolation are
estimated as products of the volume of water and the
average concentration of nitrate (N03-N) in the soil
MATERIALS AND METHODS layer. Organic N transport with sediment is calculat-
ed with a loading function developed by McElroy et
SWAT Model Description al. (1976) and modified by Williams and Hann (1978)
for application to individual runoff events. The load-
SWAT is a hydrologic/water quality model devel-
ing function estimates daily organic N runoff loss
based on the concentration of organic N in the top soil
oped by United States Department of Agriculture - layer, the sediment yield, and an enrichment ratio
Agricultural Research Service (USDA-ARS) (Arnold (that is, the ratio of organic N in sediment to organic
et al., 1998). It is a continuous time model that oper-
N in soil and typically ranging from two to four). The
ates on a daily time step. The objective in model phosphorus processes modeled by SWAT and the vari-
development was to predict the impact of manage- ous pools of phosphorus in the soil are depicted in Fig-
ment on water, sediment, and agricultural chemical
ure 2 (USDA-ARS, 1999). Plant use of phosphorus is
yields in large ungaged basins. To satisfy the objec- estimated using the supply and demand approach
tive, the model (a) is physically based (calibration is
similar to nitrogen. The loss of dissolved phosphorus
not possible on ungaged basins); (b) uses readily in surface runoff is based on the concept of partition-
available inputs; (c) is computationally efficient to ing pesticides into solution and sediment phases as
operate on large basins in a reasonable time; and
(d) is continuous in time and capable of simulating described by Leonard and Wauchope (1980). The
amount of soluble P removed in :runoff is predicted
long periods for computing the effects of management using labile P concentration in the top 10 mm of the
changes. SWAT uses a command structure for routing
runoff and chemicals through a watershed. Com- soil, the runoff volume and a phosphorus soil parti-
tioning factor (that is, the ratio of P attached to sedi-
mands are included for routing flows through streams
ment to P dissolved in soil water and typical values
and reservoirs, adding flows, and inputting measured range from 100 to 175 depending on the soil). Sedi-
data from wastewater treatment plants. The sub- ment transport of P is simulated with a loading func-
basin/subwatershed components of SWAT can be
tion as described in organic N transport.
placed into eight major components — hydrology, Instream nutrient dynamics have been incorporat-
weather, erosion/sedimentation, soil temperature, ed into SWAT (Ramanarayanan et al., 1996) using the
plant growth, nutrients, pesticides, and land manage- kinetic routines from an instream water quality
ment. Erosion and sediment yield are estimated for
each subbasin with the Modified Universal Soil Loss model, QUAL2E (Brown and Barnwell, 1987). Howev-
er, the QUAL2E routines were not explicitly used in
Equation (MUSLE) (Williams, 1975). The channel this study. The North Bosque River is a phosphorus
sediment routing equation uses a modification of Bag-
limited system and dominated by large amounts of
nold's sediment transport equation (Bagnold, 1977)
that estimates the transport concentration capacity as attached algae (periphyton) as opposed to suspended
algae. The in-stream kinetics formulation for SWAT
a function of velocity.
does not include a periphyton component, a common
CY = SPCON*VSPEXP (1) limitation of many in-stream water quality models. In
addition, inadequate monitoring information exists to
where, CY is sediment transport concentration quantify the periphyton biomass distribution in the
capacity in g/m3; SPCON is the concentration capaci- river. To overcome these limitations, a first-order
ty in g/m3 at a velocity of 1 mIs; V is flow velocity in decay kinetic function was developed for the nutrient
mis; and SPEXP is a constant in Bagnold's equation. of concern in the study (phosphorus) and to character-
The model either deposits excess sediment or re- ize the microbial transformation of in-stream soluble
entrains sediment through channel erosion depending P to organic P rather than the more traditional
on the sediment load entering the channel. approach involving algae growth and phosphorus
JAWRA 1170 JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION
 Validation of the SWAT Model on a Large River Basin With Point and Nonpoint Sources

NITROGEN

Mineral N Organic N
Denitrification
Volatilization Fertilizer Fertilizer
Plant

Nitrification

i Residue Mineralization

Figure 1. Nitrogen Processes Modeled in SWAT (USDA-ARS, 1999).

PHOSPHORUS

Mineral Organic
Plant Uptake P fertilizer I Humic Substances Residue

I Organic P
fertilizer Plant residue
I
I

Residue Mineralization

Figure 2. Phosphorus Processes Modeled in SWAT (USDA-ARS, 1999).

uptake and cycling. The transformation is particular- ASP=SP-SP0 (3)

ly important during low flow periods because
extremely low stream velocities create long travel OP0 = OP1 + LSP (4)
times. Equations for phosphorus are given below:
where, PC1 and PC0 are reach soluble P inflow and
PC0 = PC1 *expa(chllq) (2) outflow concentrations in g/m3; a is the degradation
coefficient; chi is the channel length in 1cm; q is the

JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION 1171 JAWRA

 Santhi, Arnold, Williams, Dugas, Srinivasan, and Hauck

flow rate in mmlday; SP and SP0 are the inflow and Agricultural Policy Environmental eXtender (APEX)
outflow soluble P loads in kg; and OP1 and OP0 are and input to SWAT. In this project, dairy management
the inflow and outflow organic P loads in kg. Similar practices were directly input to and simulated by
equations were developed for nitrogen. SWAT.
SWAT has been extensively validated across the
U.S. for stream flow and sediment yields (Arnold et
al., 1999). Limited validation of the SWAT nutrient Watershed Description
simulation has been attempted (Engel et al., 1993;
Jacobson et al., 1995; Alexander et al., 2000; Saleh et The SWAT model was applied to the 4277 km2
al., 2000). Saleh et al. (2000) applied SWAT to subwa- Bosque River Watershed that flows into Lake Waco
tersheds in the Bosque River basin upstream of Hico.
(Figure 3) which is the source of drinking water to the
However, in that project, dairy waste application city of Waco, Texas. Landuses in this watershed is
areas were simulated by a multi-field scale model,
mostly range and pasture in the upper portion while

Figure 3. Bosque River Watershed in Texas.

JAWRA 1172 JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION

 Validation of the SWAT Model on a Large River Basin With Point and Nonpoint Sources

TABLE 1. Percentage of Major Landuses in Bosque River Watershed.

Row and Dairy Waste

Rangel Non-Row Application
Location Pasture Forest Crops Urban Fields

Upstream of Hico 15.4 69.2 6.5 1.7 7.2

Upstream of Valley Mills 13.6 73.0 8.9 1.4 3.1

Entire Watershed 15.0 64.0 17.0 2.0 2.0

cropland is wide spread in the southern portion near positive correlation between elevated levels of phos-
Middle Bosque River (Table 1). Major soil series are phorus, the number of cows and the total acreage of
Windthorst (sandy loam, deep, hydrologic group C), manure application fields (McFarland and Hauck,
Purves (clay, fine, shallow, hydrologic group D), Den- 1997; McFarland and Hauck, 1999). Other sources of
ton (silty clay, fine, deep, hydrologic group D), Brack- pollution include runoff from cropland and urban
ett (clay loamy, gravel, deep, hydrologic group C), areas and effluent from WWTPs.
Crawford (silty clay, fine, deep, hydrologic group C),
Eckrant (clay, fine, shallow, hydrologic group D), and
Aledo (clay loam, gravel, shallow, hydrologic group C). Model Inputs for Bosque River Watershed
The percentages of shallow soils above Hico, above
Valley Mills, and in the entire watershed are about A Geographic Resource Analysis Support System-
5.5, 22.5 and 28.0, respectively. Geographic Information System (GRASS-GIS) inter-
There are about 100 dairies with 40,450 cows in face (Srinivasan and Arnold, 1994) was used to
this watershed mainly located upstream of Hico. develop SWAT input files for the watershed (Table 2).
Dairy manure is applied over an area of about 9,450 Initially, the watershed was delineated into subbasins
ha. Application of manure to pasture or cropland may using the digital elevation map. The delineated sub-
contribute to the nonpoint source pollution. The Texas basin map, landuse, soils and waste application field
Institute for Applied Environmental Research maps were overlaid to identify the manure-applica-
(TIAER) has monitored stream runoff, sediment, and tion and nonapplication areas in each subbasin.
nutrients for several areas in the watershed since SWAT simulates different landuses in each subbasin.
the early 1990s (McFarland and Hauck, 1995). In The following crops were used on the manure!
the validation process, the model is operated with waste application area: summer pasture/bermuda
input parameters set during the calibration process grass (40 percent of the area); summer pasture grass
without any change and the results are compared to with winter wheat in rotation (30 percent); sorghum
the remaining Observational data. There is a strong

TABLE 2. Model Input Data Sources for Bosque River Watershed.

Data Type Scale Source Data Description/Properties

Topography 1:24,000 USGS Elevation, overland, and channel slopes, lengths

Soils 1:24,000 USDA-NRCS Soil physical properties like bulk density, texture, saturated
conductivity, etc.

Landuse 1:24,000 USGS Landuse classifications

Manure Application Field Map 1:24,000 TIAER Location and area of the waste application fields

Weather 12 stations NWS Daily precipitation and temperature

Waste Water Treatment Plant Information 8 plants TIAER Daily outflow, sediment, organic and mineral nutrients

Land Management Information — TIAER Fertilizer and manure application rates and timing, tillage,
planting, and harvesting dates for different crops

JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION 1173 JAWRA

 Santhi, Arnold, Williams, Dugas, Srinivasan, and Hauck

hay (5 percent); and sorghum hay-winter wheat in Equation's cover and management factor, or C factor,
rotation (25 percent). Four to five hay cuttings per may be adjusted to provide a better fit.
year were simulated. For waste application area, the Water quality data from TIAER monitoring sta-
annual manure application rate was determined tions (McFarland and Hauck, 1997) were used to cali-
based on the number of animals and the annual brate and validate SWAT for the Bosque River
amount of manure generated per animal (ASAE, Watershed. Both long-term (wherever measured data
1988). Manure nutrient concentrations were taken available) and short-term simulation results were
from experimental studies (Gassman, 1997). compared with measured data at six locations in the
Nonwaste application pasture was modeled as watershed. The results of the two most important
improved pasture, with typical nutrient application gaging stations (Hico and Valley Mills on the North
rate and four grass cuttings per year. Corn, winter Bosque River) are discussed in this paper. Hico is
wheat, and grain sorghum were the major crops on located immediately below the majority of the dairy
cropland. Urban land inputs had pervious (to repre- industry and the largest WWTP (Stephenville). Val-
sent lawn/garden) and impervious (street sweeping) leys Mills is the last monitoring station along the
land management. Other typical management inputs North Bosque River near Lake Waco. Watershed
on tillage operations, irrigation, fertilizer application areas above Hico and Valley Mills are 926 and 2,997
dates and amounts were used wherever applicable. )p2, respectively.
Water quality data from WWTPs of the eight towns
in the watershed were input into the SWAT-GIS inter- Calibration Procedure. The procedure for cali-
face as point sources (Figure 3). brating the SWAT model for flow, sediment and nutri-
ents is shown in Figure 4. Initially, base flow was
separated from surface flow for both observed and
Evaluation of Model Prediction simulated stream flows using an automated digital
filter technique (Nathan and McMahon, 1990; Arnold
Mean, standard deviation, coefficient of determina- and Allen, 1999). Calibration parameters for various
tion (R2), Nash-Suttcliffe simulation efficiency (ENS) model outputs were constrained within the ranges
(Nash and Suttcliffe, 1970) and sorted efficiency or shown in Table 3. Model outputs were calibrated to
prediction efficiency (RE) were used to evaluate model fall within a percentage of average measured values
prediction. The R2 value is an indicator of strength of and then monthly regression statistics (R2 and ENS)
relationship between the observed and simulated val- were evaluated. If measured and simulated means
ues. Nash-Suttcliffe simulation efficiency (ENS) indi- met the calibration criteria and monthly R2 and ENS
cates how well the plot of observed versus simulated did not, then additional checking was performed to
value fits the 1:1 line. The prediction efficiency (RE) ensure that rainfall variability and plant growing sea-
indicates the model's ability to describe the probabili- sons were properly simulated over time. If all param-
ty distribution of the observed results. If the R2, ENS, eters were pushed to the limit of their ranges for a
and E values are less than or very close to zero, the model output (i.e., flow or sediment or nutrient) and
model prediction is considered "unacceptable or poor." the calibration criteria were still not met, then cali-
If the values are one, then the model prediction is bration was stopped for that output.
"perfect." However, there are no explicit standards Stream flow was the first output calibrated (Figure
specified for assessing the model prediction using 4). Surface runoff was calibrated until average mea-
these statistics. sured and simulated surface runoff was within 15
percent and monthly R2 > 0.6 and ENS > 0.5. The
same criteria were applied to base flow, and surface
Model Calibration runoff was continually rechecked as the base flow cal-
ibration variables also effect surface runoff. Sediment
The SWAT model was built with state-of-the-art was calibrated after flow calibration and continued
components with an attempt to simulate the process- until average measured and simulated sediment loads
es physically and realistically. Most of the model were within 20 percent. Organic and mineral nitrogen
inputs are physically based (that is, based on readily and phosphorus were calibrated to within 25 percent
available information). It is important to understand after flow and sediment calibration were completed
that SWAT is not a "parametric model" with a formal (Figure 4).
optimization procedure (as part of the calibration There are numerous potential errors that can occur
process) to fit any data. Instead, a few important in the measured input data and data used for calibra-
variables that are not well defined physically such as tion, including: (1) spatial variability errors in rain-
runoff curve number and Universal Soil Loss fall, soils and land use; (2) errors in measuring flow;

JAWRA 1174 JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION

 Validation of the SWAT Model on a Large River Basin With Point and Nonpoint Sources

— Nash Suttcliffe Efficiency

ENS
R2 — Coefficient of Determination
conc. — Concentration

Figure 4. Calibration Procedure for Flow, Sediment, and Nutrients in the SWAT Model.

1175 JAWRA
JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION
 Santhi, Arnold, Williams, Dugas, Srinivasan, and Hauck

TABLE 3. Inputs Used in Model Calibration.

Variable Model Model Actual Value/

Name Processes Description* Range Change Used
CN2 Flow Curve number ±10% + 10%
REVAPC Flow Ground water revap coefficient 0.00 to 1.00 0.03
ESCO Flow Soil evaporation compensation factor 0.00 to 1.00 0.60
EPCO Flow Plant uptake compensation factor 0.00 to 1.00 0.10
C FACTOR Sediment Cover or management factor 0.003 to 0.45 Pasture: 0.006
Range: 0.10
Cropland: 0.20
SPCON Sediment Linear factor for channel sediment routing 0.0001 to 0.01 0.0008
SPEXP Sediment Exponential factor for channel sediment routing 1.0 to 1.5 1.00

SOL_ORGN Organic N Initial organic N concentration in the upper soil Manure Area: 5000 ppm
layer for a particular land use Pasture/Range: 850 ppm
Cropland: 1100 ppm
Urban: 2000 ppm
SOL_ORGP Organic P Initial organic P concentration in the upper soil Manure Area: 700 ppm
layer for a particular land use Pasture/Range: 150 ppm
Cropland: 200 ppm
Urban: 400 ppm
NPERCO Mineral N Nitrogen percolation coefficient 0.2 to 0.6 0.5
SOL_MINP Mineral P Initial mineral P concentration in the upper soil Manure Area: 250 ppm
layer for a particular land use Pasture/Range: 5 ppm
Cropland: 20 ppm
Urban: 5 ppm
PPERCO Mineral P Phosphorus percolation coefficient 10.0 to 17.5 10
PHOSKD Mineral P Phosphorus soil partitioning coefficient 100-175 100
BIOMIX Sediment, Organic, and Biological mixing efficient 0.2 to 0.5 0.40
Mineral Nutrients
RSDCO Sediment, Organic, and Residue decomposition coefficient 0.01 to 0.05 0.05
Mineral Nutrients

*Detailed descriptions are available at http:llwww.brc.tamus.edulswatJmanual (USDA-ARS, 1999).

and (3) errors caused by sampling strategies. Winter were calibrated (Figure 4) for the period from 1960
(1981) suggests errors in annual estimates of precipi- through 1997 at Hico and Valley Mills. For flow cali-
tation, stream flow, and evaporation ranged from 2 to bration, the runoff curve numbers (CN2) were adjust-
15 percent whereas monthly rates could range from 2 ed within 10 percent from the tabulated curve
to 30 percent. Errors in sampling strategies can also numbers (Mockus, 1969) to reflect conservation tillage
be significant. Walling and Webb (1988) determined practices and soil residue cover conditions of the
that using continuous turbidity and daily flow data watershed (Table 3). Other flow related model param-
resulted in errors of 23 to 83 percent when calculating eters such as re-evaporation coefficient for ground
annual sediment loads. Robertson and Roerish (1999) water [REVAPC represents the water that moves
found median absolute errors in annual phosphorus from the shallow aquifer back to the soil profile/root
loads up to 30 percent depending on sampling strate- zone and plant uptake from deep roots (Arnold et al.,
gies. The calibration criteria for this study evolved 1993)], soil evaporation compensation factor (ESCO),
based on these potential errors shown in literature. and plant evaporation compensation factor (EPCO)
were adjusted from SWAT initial estimates to match
Flow. Stream flows from the United States Geolog- the simulated and observed flows (Table 3). However,
ical Survey's (USGS) stations, and SWAT simulation long-term calibration for sediment and nutrients
JAWRA 1176 JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION
 Validation of the SWAT Model on a Large River Basin With Point and Nonpoint Sources

could not be done, as observed data were not avail- adjusted (Table 3). The initial values of most of these
able. variables are already set at default values in the
The model was then run from 1990 through 1998 SWAT model. They were refined during the calibra-
for calibrating sediment and nutrients as per avail- tion to suit the Bosque River Watershed conditions.
ability of the monitored data. Calibration of sediment
and nutrient loads (kg/ha) were conducted for Hico
from 1993 through 1997 and for Valley Mills from Model Validation
1996 through 1997 on a monthly basis. The simula-
tion was started from 1990 to reduce errors in initial In the validation process, the model is operated
estimates of state variables such as soil water content with input parameters set during the calibration pro-
and surface residue. cess without any change and the results are compared
to the remaining observational data to evaluate the
Sediment. The cover, or C factor, of the Universal model prediction. Measurements for January through
Soil Loss Equation for range and pasture was adjust- December 1998 were used to validate the model for
ed to match observed and simulated sediment loads Hico and Valley Mills. The same statistical measures
(Figure 4). The C factor was adjusted (Table 3) to rep- were used to assess the model prediction.
resent the surface cover better in the range and pas-
ture lands. Channel sediment routing variables such
as the linear factor for calculating the maximum Best Management Practices
amount of sediment reentrained during channel sedi-
ment routing (SPCON) and the exponential factor for The calibrated model was used to study the long-
calculating the sediment reentrained in the channel term effects of various BMPs related to dairy manure
sediment routing (SPEXP) were also adjusted (Table management and municipal wastewater treatment
3) during the sediment calibration. These two vari- plant loads in this watershed. Among several scenar-
ables were adjusted to represent the cohesive nature ios studied, four scenarios are discussed in this paper.
of the channels in this watershed (Allen et al., 1999). Detailed description of the BMPs can be found in San-
thi et al. (2000). The existing condition scenario simu-
Organic Nutrients. Initial concentrations of lates the watershed under existing conditions such as
organic nitrogen and phosphorus in the soil upper lay- present dairy herd size, present waste application
ers (SOL_ORGN and SOL_ORGP) for different lan- fields (WAF) areas, average manure application rate
duses are available in the soil database table (13 tfha/yr), present discharge volumes from WWTPs
(USDA-ARS, 1999). However, these values were with current median concentrations for nutrients, and
refined/modified to suit each landuse condition in this present urban and cropland areas (Table 4). The
watershed (Table 3 and Figure 4). future condition scenario reflects the projected condi-
tions of the watershed in year 2020 such as projected
Mineral Nutrients. SWAT estimates the initial dairy herd size (67,000), manure application in WAFs
concentration for mineral N (Mm N) based on the soil at the crop nitrogen requirement rate (N rate at 46
properties at individual subbasins. Therefore, no tlha/yr), WAF area calculated at N rate requirement,
adjustments were made to initial concentration of mm maximum permitted discharge volumes from WWTPs
N. Nitrogen percolation coefficient (NPERCO) was using nutrient concentrations defined by current
adjusted from initial value of SWAT to match this median values, urban area increased by 30 percent to
watershed condition (Table 3 and Figure 4). Initial reflect the projected population growth in 2020, and
concentration for mineral P (mm P) in the soil upper cropland area at current levels (due to no increase in
layer (SOL_MINP) for different landuse was set at cropland over last two decades) (Table 4). Three addi-
appropriate levels based on available literature tional WWTPs with 1 mg/l concentration of total P
(Gassman, 1997; A. N. Sharpley, USDA-ARS, Durant, were input into the model as point sources along the
Oklahoma, and L. M. Hauck, TIAER, Stephenville, North Bosque River to account for possible industrial
Texas, personal communication, 1998). Other mm P future growth outside existing communities.
related coefficients such as phosphorus percolation Several management practices on dairy manure
coefficient (PPERCO) and phosphorus soil partition- and WWTP effluents were simulated to study the
ing coefficient (PHOSKD) were initialized at accept- impact in reducing the mm P loadings. Dairy manage-
able levels (Table 3 and Figure 4) to suit the ment practices such as hauling of the solid manure
conditions of the watershed (L. M. Hauck, TIAER, from watershed, application of manure at crop P
Stephenville, Texas, personal communication, 1998). requirement rate (P rate) at 6.3 t/halyr (as N rate
The residue decomposition coefficient (RSDCO) and allows excess application of P than crop required),
biological mixing efficiency (BIOMIX) were also
JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION 1177 JAWRA
 Santhi, Arnold, Williams, Dugas, Srinivasan, and Hauck

TABLE 4. Assumptions of BMP Scenarios.

Dairy
Manure
WWTP Flow WWTP Application Reduced Manure
Scenario Period P Limit Rate P in Diet Haul Off
Existing 1997-1998 (actual) Median Concentration Btw N&P Rate No No
Future 2020 (permitted) Median Concentration N Rate No No
Scenario E 1997-1998 All WWTPs at Median Concentration P Rate Yes Yes
and Stephenville WWTP — 1 mg/l

Scenario F 2020 All WWTPs with Loads Equal to P Rate Yes Yes
Scenario E and Stephenville WWTP
— With load equal to 1 mg/i of future

and reduction of the dairy diet P to 0.4 percent observed and simulated flows reveal that hydrologic
(results in reducing the dairy manure P content by 29 processes in SWAT are modeled realistically and the
percent as suggested by Keplinger, 1999) were concentrations at all flow regimes are realistic.
imposed. The concentrations of total P in WWTP
effluents were reduced to 1 mg/l. Sediment. The temporal variations of sediment
Scenario E was a modification of the existing con- loading at Hico and Valley Mills are represented in
dition scenario with additional conditions imposed on Figures 6b ad 7b. There is a significant R2 relation-
manure application rate (P rate), hauling off 38 per- ship between observed and simulated sediments both
cent of the manure, P diet reduction in animal feed, at Hico and Valley Mills (R2 0.81). The values of
and 1 mgfl limits of P in WWTPs (Table 4). Scenario F ENS for Hico and Valley Mills above 0.69 indicate that
was a modification over the future scenario with the simulated sediment is closer to the observed sedi-
manure applied at P rate, hauling off 38 percent of ment. The E values at Hico and Valley Mills indicate
the manure, P diet reduction, and 1 mg/I P limits on that model explained the probability distribution of
all WWTPs. the observed sediments well (Tables 5 and 6).

Organic Nutrients. Monthly loadings of organic

nitrogen and phosphorus (Org N and org P) for Hico
RESULTS AND DISCUSSION
and Valley Mills are shown in Figures 6c, 6d, 7c and
7d. Mean and standard deviation for observed and
Model Calibration simulated org N are within 10 percent of difference at
Hico (Table 5). The R2 value indicated that observed
Flow. Measured and simulated annual flows at and simulated values were closer and ENS indicated
Hico and Valley Mills matched well (Figures 5a and that the observed versus simulated plot was close to
5b). Monthly simulated and observed flows during the 1:1. There were under prediction of loads during a few
nutrient calibration at the two locations matched well months in early 1993 and middle of 1997 (Figure 6c).
(Figures 6a and 7a). Means and standard deviations At Valley Mills, simulated org N was much closer to
of the observed and simulated flows for long-term and observed org N (Table 6). At Hico and Valley Mills,
short-term at Hico and Valley Mills were within a dif- the values of simulated and observed org P were clos-
ference of 10 percent (Tables 5 and 6). Further agree- er (Tables 5 and 6). The deviations of the simulated
ment between observed and simulated flows at Hico values were closer to their mean compared to the
and Valley Mills are shown by the coefficient of deter- observed values.
minations greater than 0.6, ENS and E values
greater than 0.72 (Tables 5 and 6). The estimated pro- Mineral Nutrients. The temporal patterns of
portions of base flow from the observed flows at Hico observed and simulated mm N and P for Hico
and Valley Mills were 30 percent and 34 percent from (Figures 6e and 6f) and statistics indicated that pre-
the filter technique. They were 28 percent and 27 per- dictions of Mm N and P at Hico corresponded well
cent for the same locations for SWAT simulated flows. with the observed values (Table 5). However, the
These results for surface runoff and base flow for model overpredicted the mm N at Valley Mills during

JAWRA 1178 JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION

 Validation of the SWAT Model on a Large River Basin With Point and Nonpoint Sources

350
£ Observed - - U- - Simulated a) Hico
300
(5

250

200
E
150 _____________ ________ ________________________
0
> ______________________ __________________________
0
LL. 50

0
C . CD
CD CD CD
CD
CD l-C C1
1'- r-..
CD CD
N..
C
CD
C'l
CD CD
CD
CD
CD 0 C
CD CD CD CD
CD
CD
——
CD
CD CD
——1
CD CD CD CD CD
-
CD CD CD CD CD
— .-
CD CD CD CD
1- 1
CD

Time

350

300

250

.200
150

100
0
LI.
50

0
ON CDN.0 N CD 0 N CD 0 N
r,
CD CD CD CD
CD CD CD CD CD F— F-- N.. N.. CD CD CD CD CD CD CD CD CD
,
CD
——
CD
CD CD
I
CD CD
1
CD CD CD CD
— 1CD
CD CD
1
CD
1
CD CD
1
CD CD CD

Time

Figure 5. Observed and Simulated Annual Flows at (a) Hico and (b) Valley Mills.

February 1997 (Figure 7e) and as a result the ENS loadings. The trends for mm P were relatively closer
and prediction efficiency showed poor correspondence to the observed values at Valley Mills (Figure 7f) and
of observed versus simulated values, even though the statistics indicated that the prediction of mm p was
R2 value of 0.72 (Table 6) indicated the strong linear reasonable (Table 6).
relationship between the observed and simulated

JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION 1179 JAWRA

 Santhi, Arnold, Williams, Dugas, Srinivasan, and Hauck

(ati Flow (b) Sediment

£ Observed - - .- - Simulated . Observed - - .- - Simulated
80 0.60
(U
.c 0.50
60
• 50 ___ _______ fl———
0.40

0
40
-H
_______ _______________
— 0.30
I
0.20 _______ _______________ ____

aAJ.f
'U

20

1
1LJ I . • 0.10
Cl)
0.00
CO1U)
CJCJ 0)C')(0 C) LI)0 N.
C\JU)

Month (1 993-97) Month (1993-97)

(c ) Organic N (d) Organic P

£ Observed - - .- - Simulated £ Observed - - .- - Simulated
2.00 0.35
0.30
1.60 (U ____
-C
1.20
Z 0.80 a- 0.15
0.10
0.40
0.05
0.00 0.00 inii.
(0 '- CJ C'i0) C')
LI)
C'J
(0 C') U)ON.
LI) N-.
1 0)C')C')
C') U) N- C')LI)
0) U)

Month (1 993-97) Month (1993-97)

(e) Mineral N (f) Mineral P

Observed - - - - Simulated
• Observed -. u Simulated a.

0.60 0.30

.
_.
'U
0.50
0.20
0.30
.E 0.20 . 0.10

0.10
0.00 0.00
CO
'- CJ 0) C')(DC') U)
LI)
C'J C'J
ON.U) N-
i-i- 0)
C')
C\JU),-C')N.C') C') 0) U)
U)

Month (1993-97) Month (1 993-97)

Figure 6. Observed and Simulated Monthly Flow and Sediment and Nutrient Loadings at Hico During Calibration Period.

JAWRA 1180 JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION

 Validation of the SWAT Model on a Large River Basin With Point and Nonpoint Sources

(a) Flow (b) Sediment

.' Observed - - - - 'Simulated £ Observed - - .- - Simulated
120 1.50
-C
C
0
60
0 0.60
> 40 E A
220
U- Cl)
0.30

0 0.00 •,___Urhh1r/9jUj
'- N- 0
1 C) (0
¶ 0) C.J
C%J
. N- 0'- C) '— 0)- cJ
'—
CD CJ

Month (1996-97) Month (1996-97)

(C ) Organic N (d) Organic P

£ Observed - - .- - 'Simulated £ Observed - - .- - 'Simulated
3.00 0.60
2.50 ——-—— —________ 0.50
2.00
..1.50
•
-C
0.40
0.30
1.00 :' 0.20
00.50 0 0.10
• 0.00
0.00 N- 0
'- N- 0
I

1 C)
I

• • (0T
I I

C'J
0)
I IUIAII1'
CJ .-C)1 • • CJ
(0 0) CsJ

Month (1996-97) Month (1 996-97)

(e) Mineral N (t) Mineral P

£ Observed - - - Simulated a Observed - - .- - 'Simulated
1.20 0.25
1.00 _. 0.20
0.80
0.15
' 0.60
z 0.10
. 0.40
0.05
0.20 r
— _k__.•__ !
0.00 0.00
'- N- 0 C)_ 1(0 • ('10) Ci '- N- 0
'- C)
'- CD
,_
0)
¶
CJ
C.J

Month(1 996-97) Month(1 996-97)

Figure 7. Observed and Simulated Monthly Flow and Sediment and Nutrient Loadings at Valley Mills During Calibration Period.

JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION 1181 JAWRA

 Santhi, Arnold, Williams, Dugas, Srinivasan, and Hauck

TABLE 5. Monthly Calibration Results at Hico for the Period 1993 to 1997.

Variable Mean Standard Deviation

(units) Observed Simulated Observed Simulated R2 ENS PE
Flow Volume (mm)* 64.460 72.090 62.280 52.970 0.88 0.86 0.92
Flow Volume (mm) 12.110 11.790 15.060 14.820 0.80 0.79 0.99
Sediment (tiha) 0.045 0.038 0.092 0.077 0.81 0.80 0.94
Organic N (kg/ha) 0.213 0.161 0.322 0.260 0.61 0.58 0.87
Organic P (kg/ha) 0.036 0.031 0.057 0.049 0.71 0.70 0.93
Mineral N (kg/ha) 0.090 0.065 0.105 0.073 0.64 0.59 0.82
Mineral P (kg/ha) 0.026 0.021 0.042 0.030 0.60 0.59 0.83

*kmual flow between 1962 and 1997.

TABLE 6. Monthly Calibration Results at Valley Mills for the Period 1993 to 1997.

Variable Mean Standard Deviation

(units) Observed Simulated Observed Simulated R2 ENS E
Flow Volume (mm)* 86.500 92.310 77.050 53.230 0.66 0.72 0.89
Flow Volume (mm) 15.220 14.800 25.340 17.420 0.89 0.83 0.85
Sediment (t/ha) 0.086 0.069 0.214 0.112 0.87 0.69 0.72
Organic N (kg/ha) 0.236 0.308 0.522 0.418 0.60 0.57 0.70
Organic P (kg/ha) 0.045 0.048 0.104 0.065 0.61 0.59 0.67
Mineral N (kg/ha) 0.081 0.117 0.130 0.222 0.72 -0.08 0.03
Mineral P (kg/ha) 0.017 0.014 0.042 0.019 0.66 0.53 0.64

flow between 1960 and 1997.

Model Validation simulated loading for both nutrients at Hico corre-

sponded reasonably well (Table 7), while results indi-
Flow. Observed and simulated flows at Hico cate that the model underpredicted these nutrients at
matched well (Figure 8a and Table 7). The model Valley Mills (Table 8).
underpredicted the flow in the month of March 1998
and slightly overpredicted in late 1998 at Valley Mills Mineral Nutrients. Observed and simulated mm
(Figure 9a). The difference might be due to the spatial N and P at Hico and Valley Mills showed similar
variability of precipitation. However, the prediction trends (Figures 8e, 8f, 9e, and 90. Predicted and
statistics were high at both locations (Tables 7 and 8). observed mm N and P at Hico and Valley Mills
matched fairly well except for a slight overprediction
Sediment. Observed and simulated sediment load- of mm in early 1998 at Hico and at the end of 1998
ing at Hico matched well except for March 1998 when at both locations. Statistics show simulated and
the flow was also underpredicted by the model (Fig- observed mm N and P at both locations were closer
ure 8b). Again, statistics show prediction at Hico was (Tables 7 and 8).
satisfactory (Table 7). Predicted sediment was very
low for March 1998 at Valley Mills (Figure 9b) and BMP Analysis
statistical parameters (Table 8) show poor agreement
between observed and simulated results.
Mm p loadings are displayed as probability
Organic Nutrients. The time series plots for exceedance plots to analyze the effectiveness of BMPs.
organic N and P at Hico and Valley Mills (Figures In these exceedance plots, annual mm P loadings
8c, 8d, 9c, and 9d) show model underprediction in (y-axis) for the simulation period (1960 through 1998)
March 1998. Being sediment bound nutrients, were ordered and plotted with their associated
they follow the trends of sediment. Observed and exceedance probability values (x-axis) for Hico and

JAWRA 1182 JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION

 Validation of the SWAT Model on a Large River Basin With Point and Nonpoint Sources

(a) Flow (b) Sediment

£ Observed - •- - Simulated A Observed - - - - Simulated
80 0.60

6050 0.40
40 0.30
0
0.20
20
10 G)
U)
0.10
0
1 2 3 4 5 6 7 8 9 101112
0.00
- (Y) ) N- C) i-
1
Month (1998) Month (1998)

(C) Organic N (d) Organic P

A Observed - - .-- Simulated A Obsened - - s--- Simulated
2.00
0.35
1.60 . 0.30
2 0.25
1.20
0.20
z 0.80 0. 0.15
o 0.40
0.00
cv) U) N- C)
o.io
0.05
0.00
, ,'.
- C') U) N- 0) 1
Month (1998) Month (1998)

Figure 8. Observed and Simulated Monthly Flow and Sediment and Nutrient Loadings at Hico During Validation Period.

JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION 1183 JAWRA

 Santhi, Arnold, Williams, Dugas, Srinivasan, and Hauck

(a) Flow (b) Sediment

.i Observed - - .- - Simulated .1 Observed -. .- - Simulated
120 1.50
1 E
(l 1.20
80 C
0.90
E
60
0> 0.60
40 E

20 0
Cl)
0.30

0.00
1 2 3 4 5 6 7 8 9 10 1112 1 2 3 4 5 6 7 8 9 1011 12
Month (1998) Month (1998)

(C ) Organic N (d) Organic P

£ Observed - - .- - Simulated •' Observed - - - - Simulated
3.00 0.60
2.50 0.50
2.00 0.40

. 1.50 0.30

1.00 0;
I- 0.20
0 0.50 O 0.10
jIrn. _.
! 1—11'
uU
0.00
0.00
1 2 3 4 5 6 7 8 9 101112 1 2 3 4 5 6 7 8 9 1011 12
Month (1998)
Month (1998)

(e) Mineral N (f) Mineral P

£ Observed - - - - Simulated £ Observed - - .- - Simulated
1.20 0.25
1.00 0.20
0.80
0.15
0.60
z 0.10
. 0.40
0.20
0.00
tp . I.. •.........
.i
0.05

0.00
A
. -.-U-.-. . . •-,i--u-,-I-,-i
1 2 3 4 5 6 7 8 9 10 11 1 2 3 4 5 6 7 8 9 10 11
Month (1998) Month (1998)

Figure 9. Observed and Simulated Monthly Flow and Sediment and Nutrient Loadings at Valley Mills During Validation Period.

JAWRA 1184 JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION

 Validation of the SWAT Model on a Large River Basin With Point and Nonpoint Sources

TABLE 7. Monthly Validation Results at Hico for the Period 1998.

Variable Mean Standard Deviation

(units) Observed Simulated Observed Simulated R2 ENS E
Flow Volume (mm) 11.580 9.000 13.120 11.290 0.92 0.87 0.91
Sediment (tlha) 0.045 0.025 0.114 0.056 0.98 0.70 0.70
OrganicN(kg/ha) 0.194 0.114 0.296 0.191 0.92 0.73 0.79
Organic P (kg/ha) 0.043 0.023 0.074 0.042 0.95 0.72 0.73
Mineral N (kg/ha) 0.060 0.046 0.089 0.054 0.89 0.75 0.81
Mineral P (kg/ha) 0.024 0.012 0.033 0.017 0.83 0.53 0.62

TABLE 8. Monthly Validation Results at Valley Mills for the Period 1998.

Variable Mean Standard Deviation

(units) Observed Simulated Observed Simulated R2 ENS E
Flow Volume (mm) 9.760 10.980 20. 150 9.470 0.80 0.62 0.63
Sediment (tlha) 0.132 0.039 0.386 0.064 0.95 0.23 0.23
Organic N (kg/ha) 0.257 0.211 0.715 0.224 0.71 0.43 0.44
Organic P (kg/ha) 0.055 0.034 0.144 0.039 0.80 0.39 0.41
Mineral N (kg/ha) 0.050 0.068 0.097 0.063 0.72 0.64 0.73
Mineral P (kg/ha) 0.006 0.008 0.016 0.010 0.93 0.81 0.83

Valley Mills (Figure 10). These plots provide informa- CONCLUSIONS

tion on the probability of achieving a particular load
of mm P through a BMP at a particular location. Mm The objective of this study was to calibrate and val-
P loading curves for the various scenarios varied from idate the SWAT model for flow, sediment, and nutri-
10,000 kg to 40,000 kg at 10 percent probability at ents for a watershed where water quality problems
Hico, whereas it varied from 20,000 kg to 80,000 kg at exist that involve dairy manure application fields,
Valley Mills. These curves showed loadings within wastewater treatment plants, croplands, and urban
10,000 kg at Hico at 90 percent probability and they areas. Monthly simulated flow, sediment, and nutri-
showed loadings within 20,000 kg at Valley mills for ent (organic and mineral) loadings were compared
the same probability. In general, the loading curves with observed values at Hico and Valley Mills gauging
were wider at lower probabilities and they become stations for the calibration and validation period.
closer as they reach higher probabilities. The mm p In most instances, simulated flow, sediment, and
loadings were increased by about 27 percent at Hico nutrients were closer to the measured values during
and 29 percent at Valley Mills in the future condition the calibration period. Model validation was done for
scenario as compared to the existing condition sce- one year, 1998. In general, simulated flow and nutri-
nario. These increases were predominantly caused by ents were closer to the measured values during the
projected conditions for dairy and WWTPs in the validation except for some underprediction of
future scenario (Table 4). Scenario E showed reduc- simulated sediment and organic nutrients in March
tion in mm P loadings of about 67 percent at Hico and 1998 at Valley Mills. These underpredictions caused
57 percent at Valley Mills from the future scenario. significant degradation in model prediction statistics
Scenario F showed reduction in mm P loadings of in validation. SWAT predictions were acceptable,
about 54 percent at Hico and 48 percent at Valley especially given the approximations and spatial vari-
Mills from the future scenario. Scenario E indicated
ability involved in simulating a large complex system
that at existing conditions implementation of the or watershed.
BMPs (Table 4) would come closer in achieving the The calibrated model was used to study the effects
desired water quality goals. However, with year 2020 of various BMPs on nutrients (Santhi et al., 2000).
growth (future) conditions, more stringent controls Scenario E, a scenario with added BMP conditions
are required to meet the water quality goals. over the existing scenario, showed higher percentage
reductions in mm loadings. SWAT is integrated into

JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION 1185 JAWRA

 Santhi, Arnold, Williams, Dugas, Srinivasan, and Hauck

Hico

80000

70000

60000

50000
U)
V
0 40000
a.
C
30000

20000

10000

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Exceedance probability

Valley Mills

180000

160000

140000

.C)120000

. 100000
0
a. 80000
60000

40000

20000

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Exceedance probability
• Existing - - +- - Future £ Scenario-E • Scenario-F

Figure 10. Exceedance Probability of Mineral P Loadings for BMP Scenarios at Hico and Valley Mills.

JAWRA 1186 JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION

 Validation of the SWAT Model on a Large River Basin With Point and Nonpoint Sources

the USEPA's watershed modeling framework, called Keplinger, K. 0. 1999. Cost Savings and Environmental Benefits of
Dietary P Reductions for Dairy Cows in the Bosque River
BASINS and is expected to be used by many state and Watershed. PR99-09, Texas Institute for Applied Environmental
federal agencies. The SWAT calibration and valida- Research(TIAER), Tarleton State University, Stephenville,
tion procedures presented in this paper will be useful Texas.
to researchers and engineers involved in studying Leonard, R. A. and R. D. Wauchope, 1980. CREAMS: A Field-Scale
Model for Chemicals, Runoff and Erosion from Agricultural
water quality problems. Management Systems. In: The Pesticide Submodel, W. G. Kin-
sel, (Editor). USDA Conservation Research Report No. 2, Chap-
ter 5.
ACKNOWLEDGMENTS McElroy, A. D., S. Y. Chiu, J. W. Nebgen et al., 1976. Loading Func-
tions for Assessment of Water Pollution from Nonpomt Sources.
The authors would like to acknowledge the USDA and Texas Environmental Protection Technical Services, EPA 600/2-76-
Institute for Applied Environmental Research, Stephenville, Texas, 151.
for providing funding for this research work. McFarland, A. M. S. and L. M. Hauck, 1995. Livestock and the
Environment: Scientific Underpinnings for Policy Analysis.
Report No. PR95-01, TIAER, Stephenville, Texas.
McFarland, A. M. S. and L. M. Hauck, 1997. Livestock and the
LITERATURE CITED Environment: A National Pilot Project. Report No. PR97-02,
TIAER, Stephenville, Texas.
Alexander, R. B., R. A. Smith, G. E. Schwarz, S. D. Stephen Pre- McFarland, A. M. S. and L. M. Hauck, 1999. Relating Agricultural
ston, J. W. Brakebill, R. Srinivasan, and P. Percheo, 2000. Atmo- Landuses to In-Stream Stormwater Quality. Journal of Environ-
spheric Nitrogen Flux from the Fluvial Drainages of Major mental Quality 28:836-844.
Estuaries in the United States and Application of the SPAR- Mockus, V., 1969. Hydrologic Soil-Cover Complexes. In: SCS
ROW Watershed Model. Draft of the chapter to be published by National Engineering Handbook, Section 4, Hydrology. USDA-
the American Geophysical Union on Assessing the Relative Soil Conservation Service, Washington, D.C.
Nitrogen Inputs to Coastal Waters from the Atmosphere. Nash, J. E. and J. V. Suttcliffe, 1970. River Flow Forecasting
Allen, P. M., J. G. Arnold, and E. Jakubowski, 1999. Prediction of Through Conceptual Models, Part I. A Discussion of Principles.
Stream Channel Erosion Potential. Environmental and Engi- Journal of Hydrology 10(3):282-290.
neering Geoscience 3:339-35 1. Nathan, R. J. and T. A. McMahon, 1990. Evaluation of Automated
Arnold, J. G. and P. M. Allen, 1999. Automated Methods for Esti- Techniques for Baseflow and Recession Analysis. Water
mating Baseflow and Groundwater Recharge from Stream Flow Resources Research 26(7):1465-1473.
Records. Journal of American Water Resources Association Ramanarayanan, T. S., R. Srinivasan, and J. G. Arnold, 1996. Mod-
eling Wister Lake Watershed Using a GIS-Linked Basin Scale
35(2):411-424.
Arnold, J. G., P. M. Allen, and G. Bernhardt, 1993. A Comprehen- Hydrologic/Water Quality Model. In: Third International
sive Surface-Groundwater Flow Model. Journal of Hydrology Conference/Workshop on Integrating Geographic Information
142(1993):47-69.
Systems and Environmental Modeling, National Center for
Arnold, J. G., R. Srinivasan, R. S. Muttiah, and P. M. Allen, 1999. Geographic Information and Analysis, Santa Fe, New Mexico.
Continental Scale Simulation of the Hydrologic Balance. Jour- Robertson, D. M. and E. R. Roerish, 1999. Influence of Various
nal of American Water Resources Association 35(5):1037-1051. Water Quality Sampling Strategies on Load Estimates for Small
Arnold, J. G., R. Srinivasan, R. S. Muttiah, and J. R. Williams, Streams. Water Resources Research (35)3747-3759.
1998. Large Area Hydrologic Modeling and Assessment Part I: Saleh, A., J. G. Arnold, P. W. Gassman, L. M. Hauck, W. D. Rosen-
Model Development. Journal of American Water Resources thal, J. R. Williams, and A. M. S. McFarland, 2000. Application
Association 34(1):73-89. of SWAT for the Upper North Bosque Watershed. Transactions
ASAE. 1988. Manure Production and Characteristics. ASAE of the ASAE 43(5):1077-1087.
Standards, ASAE Data: ASAE D384. 1, ASAE, St. Joseph, Michi- Santhi, C., J. G. Arnold, J. R. Williams, L. M. Hauck, and W. A.
Dugas, 2000. Application of a Watershed Model to Evaluate
gan. Management Effects on Point and NonPoint Pollution. Transac-
Bagnold, R. A., 1977. Bedload Transport in Natural Rivers. Water
Resources Research 13(2):303-312. tions of ASAE (in press).
Brown, L. C. and T. 0. Barnwell, 1987. The Enhanced Water Quali- Srinivasan, R. and J. G. Arnold, 1994. Integration of Basin-Scale
ty Models: QUAL2E and QUAL2E-UNCAS Documentation and
Water Quality Model with GIS. Water Resources Bulletin
User Manual. EPA/600/3-87/007, USEPA. Athens, Georgia. 30(3):453-462.
Engel, B. A., R. Srinivasan, J. G. Arnold, C. Rewerts, and S. J. TNRCC, 1999. State of Texas 1999 Clean Water Act Section 303 (d)
Brown, 1993. Nonpoint Source Pollution Modeling Using Models List. TNRCC, Texas, available at http://www.tnrcc.state.tx.us/
Integrated with Geographic Information Systems. Water Science adminltopdoc/sfr/058-99/99_303d.pdf.
USDA-ARS, 1999. Soil and Water Assessment Tool — Model Docu-
Technology 28(3-5):685690.
Gassman, P. W., 1997. The National Pilot Program Integrated Mod- mentation. Available at http://www.brc.tamus.edulswatimanual.
USEPA, 1994. The Quality of Our Nation's Water: 1994. Executive
eling System: Environmental Baseline Assumptions and Results
for the APEX Model. Livestock Series Report 9, CARD Publica- Summary of the National Water Quality Inventory: 1994 Report
tions. to United States Congress. EPA841-S-94-002, USEPA, Washing-
Jacobson, B. M., J. Feng, G. D. Jennings, and K. C. Stone, 1995. ton, D.C.
Watershed Scale Non-Point Source Model Evaluation for the USEPA, 1998. National Water Quality Inventory: 1996 Report to
North Carolina Coastal Plain. In: Symposium on Water Quality United States Congress. EPA841-R-97-008, USEPA, Washing-
Modeling. ASAE Paper No. 05-95. Orlando, Florida, pp. 186- ton, D.C.
191.
Walling, D. E. and B. W. Webb, 1988. The Reliability of Rating
Curve Estimates of Suspended Sediment Yield: Some Further
Comments. In: Sediment Budgets. IAHS Publ. 174:337-350.

JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION 1187 JAWRA

 Santhi, Arnold, Williams, Dugas, Srinivasan, and Hauck

Williams, J. R., 1975. Sediment Routing for Agricultural Water-

sheds. Water Resources Bulletin 11(5): 965-974.
Williams, J. R. and R. W. Hann, 1978. Optimal Operation of Large
Agricultural Watersheds with Water Quality Constraints. Thxas
Water Resources Institute, Texas A&M University, Technical
Report No. 96.
Williams, J. R., C. A. Jones, and P. A. Dyke, 1984. A Modeling
Approach to Determine the Relationship Between Erosion and
Soil Productivity. Transactions of ASAE 27:129-144.
Winter, T. C., 1981. Uncertainties in Estimating the Water Bal-
ances of Lakes. Water Resources Research (17) 82-115.

JAWRA 1188 JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION

 Information to Run SWAT Model

Type Information File Note

Model Objectives - Have overall objectives of modeling (e.g. nitrogen discharge study)
Decide Spatial and temporal scale (How large area, what is the term e.g. daily, monthly,
Model scale -
or yearly)
Basic Model Info
Modeling period - Decide from when to when for model
Watershed size - SWAT also provide watershed size after delineation

DEM GRID Required. Usually 10m to 30m, sometimes coarser resolution (e.g. USGS DEM or NED)
Landuse GRID, .shp Required.
GIS data Soil .shp Required. STATSGO or SSURGO
Channel network .shp Useful to compare with SWAT delineation

Weather station .shp Required. Weather station location

Flow gage station .shp Needed for model calibration. Flow gage station location (e.g. USGS gage station)
Observation Sampling points .shp If any.
Location Point sources .shp Point sources, if any, in watershed. Either existing one or can be manually input.
Other flow input .shp Flow input from outside of watershed, if any

Precipitation .pcp Required. Daily precipitation

Min. Temperature .tmp Required. Daily temperature
Weather Max Temperature .tmp Required. Daily temperature
Other weather info .wgn Not required. Simulated solar energy, wind velocity etc can be used.

SWAT generates default value of all parameters depending on the types of landuse. For
Landuse (in HRU) example, Initial residue cover, Manning’s n for overland flow, Sediment concentration in
Landuse general .hru
parameters lateral flow, Enrichment ratio for sediment and nutrients, Pothole storage, etc. Values can
be adjusted as necessary.
SWAT generates default value of all parameters depending on the types of management.
Management (in For example, Initial LAI, Initial biomass, Curve number, USLE P, Filter strip width,
.mgt
HRU) parameters Irrigation amount, etc. Values can be adjusted as necessary. If tile is simulated, Time to
drain soil to field capacity and Drain tile lag time are needed.
Management
Management schedule including plant, harvest, fertilizer and pesticide application, and
Management tillage operation can be input in SWAT, if needed. Auto application can also be
.mgt
schedule established. Information needed includes operation dates, fertilizer amount, tillage depth
etc.
 SWAT offers default value of all parameters for each crop type. For example, LAI, Heat
unit, Optimum and minimum temperature, Harvest index, Intercepted PAR, Radiation
Crop crop.dat
efficiency, USLE C, Biomass energy ratio, Plant decomposition rate, and etc. However,
values can be adjusted as necessary.
SWAT offers default value of all parameters for each tillage type. For example, Mixing
Tillage till.dat
efficiency and Tillage depth. However, values can be adjusted as necessary.
SWAT offers default value of all parameters for pesticide. For example, Name,
SWAT default
Pesticide pest.dat Absorption rate, Wash off, Degradation, Solubility, etc. However, values can be adjusted
data
as necessary.
SWAT offers default value of all parameters for fertilizer. For example, Name, Fraction of
Fertilizer fert.dat
each nutrient and bacteria. However, values can be adjusted as necessary.
SWAT offers default value of all parameters for urban. For example, Fraction of
impervious land, Maximum amount of solid built-up in impervious area, Concentration of
Urban urban.dat
each nutrient, Curve number for impervious land, etc. However, values can be adjusted
as necessary.
Usually soil properties are used as default from soil data. However, if necessary,
Soil properties .sol
properties can be adjusted such as hydraulic conductivity and USLE K, etc
Soil
Chemical properties .chm Optional. For example, initial nutrients and pesticide in soil

Ground water SWAT offers default value. If necessary, parameter value can be adjusted for Depth of
Ground Water .gw
properties aquifers, Ground water delay time, recharge, and soluble phosphorus, etc
SWAT generates channel properties based on DEM and some default value. However, if
Channel Channel properties .rte necessary, properties such as Manning’s n, channel erodibility, hydraulic conductivity,
channel cover factor, and so on can be adjusted.
If pond is included in simulation, some parameters are necessary from user input (usually
physical properties) and most of parameter values are offered as default. Parameters are
Pond Pond information .pnd Fraction of subbasin area that draining into the pond(s), Surface area, Volume of water,
Initial volume of water, Initial sediment concentration, hydraulic conductivity at the bottom,
nutrients settling rate, etc.
SWAT offers default value for all parameters. It can be specified here for watershed wide
Watershed-wide
.wwq parameters such as uptake of nutrients by algae and relationship between nutrients,
water quality
oxygen, and algae.
Sub-watershed water SWAT offers default value for all parameters. It can be adjusted for each sub-basin
Water quality .swq
quality separately for decay and settling of nutrients, bacteria, algae relationship.
Reservoir .res Similar to pond information, some physical properties are necessary for user to input.
Lake .lwq Similar to pond information, some physical properties are necessary for user to input.
Point source Point source including flow, sediment, and nutrients can be input as hourly, daily,
Point source .dat
discharge monthly, yearly discharge.
 c:\my documents\arcswat\SSURGO_Processing_Geospatial_gateway.ppt
SSURGO Processing for ArcSWAT (Geospatial Gateway website)
Go to following website address:
http://datagateway.nrcs.usda.gov/Gatewayhome.html

Get Data

Quick Counties

1
SSURGO Processing for ArcSWAT (Geospatial Gateway website)

3. Choose state
Get Counties
1. Get Data
Choose county
Continue

2. Quick Counties

2
SSURGO Processing for ArcSWAT (Geospatial Gateway website)

Check Soil Survey Spatial and Tabular Data

SSURGO 2.2
Continue to Step 3

Vector projection – UTM Zone NAD83

Standard Extent
ESRI Shape Files
Continue to Step 4

3
 SSURGO Processing for ArcSWAT (Geospatial Gateway website)

Fill out form to order file(s)

Continue to Step 5

Click on Place Order

You will receive an email from Geospatial Gateway with links

4
to the files you ordered.
SSURGO Processing for ArcSWAT (Geospatial Gateway website)
Preparing files for ArcSWAT interface

Open ArcMap and add ‘soilmu_a_*.shp’

in ‘spatial’ directory

Click on Spatial Analyst/Convert/

Features to Raster

5
SSURGO Processing for ArcSWAT (Geospatial Gateway website)

Input features: soilmu_a file

Field: MUKEY
Output cell size: 30
Output raster: newfilename

OK – (this process may take a few minutes)

Right-mouse click on shapefile

Open Attribute table

6
SSURGO Processing for ArcSWAT (Geospatial Gateway website)

Click on ‘Options’/Add field

Name: mukey_i
Type: Long integer
Ok

Right-mouse click on new column

7
SSURGO Processing for ArcSWAT (Geospatial Gateway website)

Click on Field Calculator

Click yes on next window

Double click on “MUKEY”

Ensure type is “Number”
Click OK

8
SSURGO Processing for ArcSWAT (Geospatial Gateway website)

After calculations, a new column

is added to table that is right-justified

Minimize this window

Go to Spatial Analyst/Convert/
Features to Raster

Input features: soilmu_a

Field: mukey_i
Output cell size: 30
Output raster: newfilename
Click Ok
9
SSURGO Processing for ArcSWAT (Geospatial Gateway website)

To Create lookup table file -

Open comp.txt (in tabular directory)
in Excel (as “|” delimited file)

Copy Column D and DD to new

workbook
DD – First column (add value header)
D – Second column (add name header)

10
SSURGO Processing for ArcSWAT (Geospatial Gateway website)

Autofilter the ‘Unnamed’ soils

and delete

Save file as *.csv type

Browse in Windows Explorer and rename

*.csv to *.txt extension.
11
If two or more counties need to be merged

Open ArcMAP/ArcToolBox
Click on Data Management Tools
General/Merge

Use the browser and add the two (or

more) counties that you need merged.

Input the name of the newly merged file

Click OK.

You also need to merge the tables.

12
MERGING TWO OR MORE TABLES:
Combine the comp.txt files for both counties in Excel. Delete the
unnamed Soils. Add headings, save as *.csv file. Rename the .csv to
.txt.
(See Pages 10-11)

You must load the states SSURGO soils (get from Srini) into the
SWAT2005.mdb file before using in ArcSWAT. To do this, see the
following document:
C:\mydocuments\arcswat_instructions\arcswat_ssugo_import.ppt

13
 MODEL EVALUATION GUIDELINES FOR SYSTEMATIC
QUANTIFICATION OF ACCURACY IN WATERSHED SIMULATIONS
D. N. Moriasi, J. G. Arnold, M. W. Van Liew, R. L. Bingner, R. D. Harmel, T. L. Veith

ABSTRACT. Watershed models are powerful tools for simulating the effect of watershed processes and management on soil and
water resources. However, no comprehensive guidance is available to facilitate model evaluation in terms of the accuracy
of simulated data compared to measured flow and constituent values. Thus, the objectives of this research were to:
(1) determine recommended model evaluation techniques (statistical and graphical), (2) review reported ranges of values and
corresponding performance ratings for the recommended statistics, and (3) establish guidelines for model evaluation based
on the review results and project-specific considerations; all of these objectives focus on simulation of streamflow and
transport of sediment and nutrients. These objectives were achieved with a thorough review of relevant literature on model
application and recommended model evaluation methods. Based on this analysis, we recommend that three quantitative
statistics, Nash-Sutcliffe efficiency (NSE), percent bias (PBIAS), and ratio of the root mean square error to the standard
deviation of measured data (RSR), in addition to the graphical techniques, be used in model evaluation. The following model
evaluation performance ratings were established for each recommended statistic. In general, model simulation can be judged
as satisfactory if NSE > 0.50 and RSR < 0.70, and if PBIAS + 25% for streamflow, PBIAS + 55% for sediment, and PBIAS
+ 70% for N and P. For PBIAS, constituent-specific performance ratings were determined based on uncertainty of measured
data. Additional considerations related to model evaluation guidelines are also discussed. These considerations include:
single-event simulation, quality and quantity of measured data, model calibration procedure, evaluation time step, and project
scope and magnitude. A case study illustrating the application of the model evaluation guidelines is also provided.
Keywords. Accuracy, Model calibration and validation, Simulation, Watershed model.

C
omputer-based watershed models can save time Sensitivity analysis is the process of determining the rate
and money because of their ability to perform long- of change in model output with respect to changes in model
term simulation of the effects of watershed pro- inputs (parameters). It is a necessary process to identify key
cesses and management activities on water quality, parameters and parameter precision required for calibration
water quantity, and soil quality. These models also facilitate (Ma et al., 2000). Model calibration is the process of estimat-
the simulation of various conservation program effects and ing model parameters by comparing model predictions (out-
aid policy design to mitigate water and soil quality degrada- put) for a given set of assumed conditions with observed data
tion by determining suitable conservation programs for par- for the same conditions. Model validation involves running
ticular watersheds and agronomic settings. In order to use a model using input parameters measured or determined dur-
model outputs for tasks ranging from regulation to research, ing the calibration process. According to Refsgaard (1997),
models should be scientifically sound, robust, and defensible model validation is the process of demonstrating that a given
(U.S. EPA, 2002). site-specific model is capable of making “sufficiently accu-
rate” simulations, although “sufficiently accurate” can vary
based on project goals. According to the U.S. EPA (2002), the
process used to accept, reject, or qualify model results should
Submitted for review in May 2006 as manuscript number SW 6494; ap-
proved for publication by the Soil & Water Division of ASABE in March be established and documented before beginning model eval-
2007. uation. Although ASCE (1993) emphasized the need to clear-
The authors are Daniel N. Moriasi, ASABE Member Engineer, ly define model evaluation criteria, no commonly accepted
Hydrologist, USDA-ARS Grazinglands Research Laboratory, El Reno, guidance has been established, but specific statistics and per-
Oklahoma; Jeffrey G. Arnold, Supervisory Agricultural Engineer, USDA-
ARS Grassland Soil and Water Research Laboratory, Temple, Texas; Mi-
formance ratings for their use have been developed and used
chael W. Van Liew, ASABE Member Engineer, Environmental Science for model evaluation (Donigian et al., 1983; Ramanarayanan
Specialist, Water Quality Planning Bureau, Montana Department of Envi- et al., 1997; Gupta et al., 1999; Motovilov et al., 1999; Saleh
ronmental Quality, Helena, Montana; Ronald L. Bingner, ASABE Mem- et al., 2000; Santhi et al., 2001; Singh et al., 2004; Bracmort
ber Engineer, Agricultural Engineer, USDA-ARS Watershed Physical Pro- et al., 2006; Van Liew et al., 2007). However, these perfor-
cesses Research Unit, Oxford, Mississippi; R. Daren Harmel, ASABE
Member Engineer, Agricultural Engineer, USDA-ARS Grassland Soil and mance ratings are model and project specific. Standardized
Water Research Laboratory, Temple, Texas; and Tamie L. Veith, ASABE guidelines are needed to establish a common system for judg-
Member Engineer, Agricultural Engineer, USDA-ARS Pasture Systems ing model performance and comparing various models
and Watershed Management Research Unit, University Park, Pennsylva- (ASCE, 1993). Once established, these guidelines will assist
nia. Corresponding author: Daniel N. Moriasi, USDA-ARS Grazinglands
Research Laboratory, 7207 W. Cheyenne St., El Reno, OK 73036-0000;
modelers in preparing and reviewing quality assurance proj-
phone: 405-262-5291, ext. 263; fax: 405-262-0133; e-mail: dmoria- ect plans for modeling (U.S. EPA, 2002) and will increase ac-
[email protected]. countability and public acceptance of models to support

Transactions of the ASABE

Vol. 50(3): 885−900 2007 American Society of Agricultural and Biological Engineers ISSN 0001−2351 885
scientific research and to guide policy, regulatory, and man- statistics were selected based on the following factors: (1) ro-
agement decision-making. bustness in terms of applicability to various constituents,
A number of publications have addressed model evalua- models, and climatic conditions; (2) commonly used, accept-
tion statistics (Willmott, 1981; ASCE, 1993; Legates and ed, and recommended in published literature; and (3) identi-
McCabe, 1999), but they do not include recently developed fied strengths in model evaluation. The trade-off between
statistics (e.g., Wang and Melesse, 2005; Parker et al., 2006). long-term bias and residual variance was also considered, as
More importantly, none of the previous publications provide recommended by Boyle et al. (2000). Bias measures the aver-
guidance on acceptable ranges of values for each statistic. age tendency of the simulated constituent values to be larger
Borah and Bera (2004) present an excellent review of values or smaller than the measured data. Residual variance is the
for various statistics used in hydrologic and nonpoint-source difference between the measured and simulated values, often
model applications, but more elaborate analysis is needed to estimated by the residual mean square or root mean square er-
aid modelers in determining performance ratings for these ror (RMSE). According to Boyle et al. (2000), optimizing
statistics. RMSE during model calibration may give small error vari-
In most watershed modeling projects, model output is ance but at the expense of significant model bias. The com-
compared to corresponding measured data with the assump- pilation of recommended statistics was also constrained by
tion that all error variance is contained within the predicted the recommendation of Legates and McCabe (1999) to in-
values and that observed values are error free. In discussions clude at least one dimensionless statistic and one absolute er-
of model evaluation statistics, Willmott (1981) and ASCE ror index statistic with additional information such as the
(1993) do recognize that measured data are not error free, but standard deviation of measured data, and to include at least
measurement error is not considered in their recommenda- one graphical technique as well.
tions perhaps because of the relative lack of data on measure-
ment uncertainty. However, uncertainty estimates for REPORTED VALUE RANGES AND PERFORMANCE RATINGS
measured streamflow and water quality data have recently FOR RECOMMENDED STATISTICS
become available (Harmel et al., 2006) and should be consid- Additional literature review was conducted to determine
ered when calibrating, validating, and evaluating watershed published ranges of values and performance ratings for rec-
models because of differences in inherent uncertainty be- ommended model evaluation statistics. Reported daily and
tween measured flow, sediment, and nutrient data. monthly values during the calibration and validation periods
The importance of standardized guidelines for model for streamflow, sediment, and nutrients are recorded along
evaluation is illustrated by the Conservation Effects Assess- with the model used for evaluation in tables A-1 through A-9
ment Project Watershed Assessment Study (CEAP-WAS, in the Appendix. All the reported data were analyzed and
2005). The CEAP-WAS seeks to quantify the environmental compiled into a summary of daily and monthly value ranges
benefits of conservation practices supported by USDA in the for different constituents during calibration and validation
2002 Farm Bill, also known as the Farm Security and Rural (table 1). The summary values in table 1 include the sample
Investment Act. One of the CEAP-WAS goals is to formulate size of the reported values (n) and the minimum, maximum,
guidelines for calibration, validation, and application of and median of the values reported for streamflow, surface
models used in CEAP to simulate environmental effects of runoff, sediment, organic, mineral and total nitrogen, and or-
conservation practices. Thus, based on the need for standard- ganic, mineral, and total phosphorus.
ized model evaluation guidelines to support watershed mod-
eling in CEAP-WAS and other projects, the objectives for the MODEL EVALUATION GUIDELINES
present research were to: (1) determine recommended model General model evaluation guidelines that consider the rec-
evaluation techniques (statistical and graphical), (2) review ommended model evaluation statistics with corresponding
reported ranges of values and corresponding performance performance ratings and appropriate graphical analyses were
ratings for the recommended statistics, and (3) establish then established. A calibration procedure chart for flow, sedi-
guidelines for model evaluation based on the review results ment, and nutrients, similar to the one proposed by Santhi et
and project-specific considerations. In addition, a case study al. (2001), is included to assist in application of the model
illustrating the application of the model evaluation guide- evaluation guidelines to manual model calibration. It is
lines was provided. This research focuses on watershed mod- noted, however, that these guidelines should be adjusted by
el evaluation guidelines for streamflow, sediments, and the modeler based on additional considerations such as:
nutrients. Throughout this article, “model evaluation” refers single-event simulation, quality and quantity of measured
to the applicable steps of sensitivity analysis, calibration, val- data, model calibration procedure, evaluation time step, and
idation, uncertainty analysis, and application. project scope and magnitude. Additionally, a brief discussion
of the implications of unmet performance ratings is provided.

METHODS
MODEL EVALUATION TECHNIQUES RESULTS AND DISCUSSION
To determine recommended techniques for watershed MODEL EVALUATION TECHNIQUES
model evaluation, an extensive review was conducted on Both statistical and graphical model evaluation tech-
published literature related to calibration, validation, and ap- niques were reviewed. The quantitative statistics were divid-
plication of watershed models. Specifically, the information ed into three major categories: standard regression,
compiled focused on the strengths and weaknesses of each dimensionless, and error index. Standard regression statistics
statistical and graphical technique and on recommendations determine the strength of the linear relationship between sim-
for their application. The recommended model evaluation ulated and measured data. Dimensionless techniques provide

886 TRANSACTIONS OF THE ASABE

a relative model evaluation assessment, and error indices represents the ratio between the mean square error and the
quantify the deviation in the units of the data of interest (Leg- “potential error” (Willmott, 1984). The author defined poten-
ates and McCabe, 1999). A brief discussion of numerous tial error as the sum of the squared absolute values of the dis-
model evaluation statistics (both recommended statistics and tances from the predicted values to the mean observed value
statistics not selected for recommendation) appears subse- and distances from the observed values to the mean observed
quently; however, the relevant calculations are provided only value. The index of agreement can detect additive and pro-
for the recommended statistics. portional differences in the observed and simulated means
Several graphical techniques are also described briefly be- and variances; however, d is overly sensitive to extreme val-
cause graphical techniques provide a visual comparison of ues due to the squared differences (Legates and McCabe,
simulated and measured constituent data and a first overview 1999). Legates and McCabe (1999) suggested a modified in-
of model performance (ASCE, 1993) and are essential to ap- dex of agreement (d1) that is less sensitive to high extreme
propriate model evaluation (Legates and McCabe, 1999). values because errors and differences are given appropriate
Based on recommendations by ASCE (1993) and Legates weighting by using the absolute value of the difference
and McCabe (1999), we recommend that both graphical tech- instead of using the squared differences. Although d1 has
niques and quantitative statistics be used in model evalua- been proposed as an improved statistic, its limited use in the
tion. literature has not provided extensive information on value
ranges.
MODEL EVALUATION STATISTICS (STANDARD REGRESSION) Nash-Sutcliffe efficiency (NSE): The Nash-Sutcliffe ef-
Slope and y-intercept: The slope and y-intercept of the ficiency (NSE) is a normalized statistic that determines the
best-fit regression line can indicate how well simulated data relative magnitude of the residual variance (“noise”)
match measured data. The slope indicates the relative rela- compared to the measured data variance (“information”)
tionship between simulated and measured values. The y- (Nash and Sutcliffe, 1970). NSE indicates how well the plot
intercept indicates the presence of a lag or lead between of observed versus simulated data fits the 1:1 line. NSE is
model predictions and measured data, or that the data sets are computed as shown in equation 1:
not perfectly aligned. A slope of 1 and y-intercept of 0 indi-
 n 
 ∑ (Yi − Yi ) 
cate that the model perfectly reproduces the magnitudes of obs sim 2
measured data (Willmott, 1981). The slope and y-intercept  
are commonly examined under the assumption that measured NSE = 1 −  in=1  (1)
∑ (Y obs − Y mean) 
2
and simulated values are linearly related, which implies that i
all of the error variance is contained in simulated values and  i =1 
 
that measured data are error free (Willmott, 1981). In reality,
measured data are rarely, if ever, error free. Harmel et al. where Yi obs is the ith observation for the constituent being
(2006) showed that substantial uncertainty in reported water evaluated, Yi sim is the ith simulated value for the constituent
quality data can result when individual errors from all proce- being evaluated, Ymean is the mean of observed data for the
dural data collection categories are considered. Therefore, constituent being evaluated, and n is the total number of ob-
care needs to be taken while using regression statistics for servations.
model evaluation. NSE ranges between −∞ and 1.0 (1 inclusive), with NSE =
Pearson’s correlation coefficient (r) and coefficient of 1 being the optimal value. Values between 0.0 and 1.0 are
determination (R2): Pearson’s correlation coefficient (r) generally viewed as acceptable levels of performance,
and coefficient of determination (R2) describe the degree of whereas values <0.0 indicates that the mean observed value
collinearity between simulated and measured data. The cor- is a better predictor than the simulated value, which indicates
relation coefficient, which ranges from −1 to 1, is an index of unacceptable performance.
the degree of linear relationship between observed and simu- NSE was recommended for two major reasons: (1) it is
lated data. If r = 0, no linear relationship exists. If r = 1 or −1, recommended for use by ASCE (1993) and Legates and
a perfect positive or negative linear relationship exists. Simi- McCabe (1999), and (2) it is very commonly used, which pro-
larly, R2 describes the proportion of the variance in measured vides extensive information on reported values. Sevat and
data explained by the model. R2 ranges from 0 to 1, with high- Dezetter (1991) also found NSE to be the best objective func-
er values indicating less error variance, and typically values tion for reflecting the overall fit of a hydrograph. Legates and
greater than 0.5 are considered acceptable (Santhi et al., McCabe (1999) suggested a modified NSE that is less sensi-
2001, Van Liew et al., 2003). Although r and R2 have been tive to high extreme values due to the squared differences, but
widely used for model evaluation, these statistics are over- that modified version was not selected because of its limited
sensitive to high extreme values (outliers) and insensitive to use and resulting relative lack of reported values.
additive and proportional differences between model predic- Persistence model efficiency (PME): Persistence model
tions and measured data (Legates and McCabe, 1999). efficiency (PME) is a normalized model evaluation statistic
that quantifies the relative magnitude of the residual variance
MODEL EVALUATION STATISTICS (DIMENSIONLESS) (“noise”) to the variance of the errors obtained by the use of
Index of agreement (d): The index of agreement (d) was a simple persistence model (Gupta et al., 1999). PME ranges
developed by Willmott (1981) as a standardized measure of from 0 to 1, with PME = 1 being the optimal value. PME val-
the degree of model prediction error and varies between 0 and ues should be larger than 0.0 to indicate “minimally accept-
1. A computed value of 1 indicates a perfect agreement be- able” model performance (Gupta et al., 1999). The power of
tween the measured and predicted values, and 0 indicates no PME is derived from its comparison of model performance
agreement at all (Willmott, 1981). The index of agreement with a simple persistence forecast model. According to Gupta
et al. (1999), PME is capable of clearly indicating poor model

Vol. 50(3): 885−900 887

performance, but it has been used only occasionally in the lit- The optimal value of PBIAS is 0.0, with low-magnitude val-
erature, so a range of reported values is not available. ues indicating accurate model simulation. Positive values in-
Prediction efficiency (Pe): Prediction efficiency (Pe), as dicate model underestimation bias, and negative values
explained by Santhi et al. (2001), is the coefficient of deter- indicate model overestimation bias (Gupta et al., 1999).
mination (R2) calculated by regressing the rank (descending) PBIAS is calculated with equation 2:
of observed versus simulated constituent values for a given
n 
time step. Pe determines how well the probability distribu- ∑ (Yi − Yi )* (100) 
obs sim
tions of simulated and observed data fit each other. However,  
it has not been used frequently enough to provide extensive PBIAS =  i =1 n  (2)
information on ranges of values. In addition, it may not ac-  ∑ (Yi obs ) 
count for seasonal bias.  
 i =1 
Performance virtue statistic (PVk): The performance
virtue statistic (PVk) is the weighted average of the Nash- where PBIAS is the deviation of data being evaluated, ex-
Sutcliffe coefficients, deviations of volume, and error func- pressed as a percentage.
tions across all flow gauging stations within the watershed of Percent streamflow volume error (PVE; Singh et al.,
interest (Wang and Melesse, 2005). PVk was developed to as- 2004), prediction error (PE; Fernandez et al., 2005), and per-
sess if watershed models can satisfactorily predict all aspects cent deviation of streamflow volume (Dv) are calculated in
(profile, volume, and peak) of observed flow hydrographs for a similar manner as PBIAS. The deviation term (Dv) is used
watersheds with more than one gauging station (Wang and to evaluate the accumulation of differences in streamflow
Melesse, 2005). PVk can range from −∞ to 1.0, with a PVk volume between simulated and measured data for a particular
value of 1.0 indicating that the model exactly simulates all period of analysis.
three aspects of observed flow for all gauging stations within PBIAS was selected for recommendation for several rea-
the watershed. A negative PVk value indicates that the aver- sons: (1) Dv was recommended by ASCE (1993), (2) Dv is
age of observed streamflow values is better than simulated commonly used to quantify water balance errors and its use
streamflows (Wang and Melesse, 2005). PVk was developed can easily be extended to load errors, and (3) PBIAS has the
for use in snow-fed watersheds; therefore, it may be neces- ability to clearly indicate poor model performance (Gupta et
sary to make adjustments to this statistic for rain-fed wa- al., 1999). PBIAS values for streamflow tend to vary more,
tersheds. PVk was not selected for recommendation because among different autocalibration methods, during dry years
it was developed for streamflow only. In addition, PVk was than during wet years (Gupta et al., 1999). This fact should
only recently developed; thus, extensive information on val- be considered when attempting to do a split-sample evalua-
ue ranges is not available. tion, one for calibration and one for validation.
Logarithmic transformation variable (e): The logarith- RMSE-observations standard deviation ratio (RSR):
mic transformation variable (e) is the logarithm of the pre- RMSE is one of the commonly used error index statistics
dicted/observed data ratio (E) that was developed to address (Chu and Shirmohammadi, 2004; Singh et al., 2004;
the sensitivity of the watershed-scale pesticide model error Vasquez-Amábile and Engel, 2005). Although it is common-
index (E) to the estimated pesticide application rates (Parker ly accepted that the lower the RMSE the better the model per-
et al., 2006). The value of e is centered on zero, is symmetri- formance, only Singh et al. (2004) have published a guideline
cal in under- or overprediction, and is approximately normal- to qualify what is considered a low RMSE based on the ob-
ly distributed (Parker et al., 2006). If the simulated and servations standard deviation. Based on the recommendation
measured data are in complete agreement, then e = 0 and E = by Singh et al. (2004), a model evaluation statistic, named the
1.0. Values of e < 0 are indicative of underprediction; values RMSE-observations standard deviation ratio (RSR), was de-
> 0 are indicative of overprediction. Although e has great po- veloped. RSR standardizes RMSE using the observations
tential as an improved statistical technique for assessing standard deviation, and it combines both an error index and
model accuracy, it was not selected because of its recent de- the additional information recommended by Legates and
velopment and limited testing and application. McCabe (1999). RSR is calculated as the ratio of the RMSE
and standard deviation of measured data, as shown in equa-
MODEL EVALUATION STATISTICS (ERROR INDEX) tion 3:
MAE, MSE, and RMSE: Several error indices are com-  n 
 ∑ (Yi obs − Yi sim ) 
2
monly used in model evaluation. These include mean abso-
 i =1 
lute error (MAE), mean square error (MSE), and root mean RMSE  
square error (RMSE). These indices are valuable because RSR = = (3)
STDEVobs  n 
they indicate error in the units (or squared units) of the con- 2
stituent of interest, which aids in analysis of the results.  ∑ (Yi obs − Y mean ) 
 i =1 
RMSE, MAE, and MSE values of 0 indicate a perfect fit.  
Singh et al. (2004) state that RMSE and MAE values less than
half the standard deviation of the measured data may be con- RSR incorporates the benefits of error index statistics and
sidered low and that either is appropriate for model evalua- includes a scaling/normalization factor, so that the resulting
tion. A standardized version of the RMSE was selected for statistic and reported values can apply to various constitu-
recommendation and is described later in this section. ents. RSR varies from the optimal value of 0, which indicates
Percent bias (PBIAS): Percent bias (PBIAS) measures zero RMSE or residual variation and therefore perfect model
the average tendency of the simulated data to be larger or simulation, to a large positive value. The lower RSR, the low-
smaller than their observed counterparts (Gupta et al., 1999). er the RMSE, and the better the model simulation perfor-
mance.

888 TRANSACTIONS OF THE ASABE

 Daily root-mean square (DRMS): The daily root-mean techniques, hydrographs and percent exceedance probability
square (DRMS), which is a specific application of the RMSE, curves, are especially valuable. Other graphical techniques,
computes the standard deviation of the model prediction er- such as bar graphs and box plots, can also be used to examine
ror (difference between measured and simulated values). The seasonal variations and data distributions.
smaller the DRMS value, the better the model performance A hydrograph is a time series plot of predicted and mea-
(Gupta et al., 1999). Gupta et al. (1999) determined that sured flow throughout the calibration and validation periods.
DRMS increased with wetness of the year, indicating that the Hydrographs help identify model bias (ASCE, 1993) and can
forecast error variance is larger for higher flows. According identify differences in timing and magnitude of peak flows
to Gupta et al. (1999), DRMS had limited ability to clearly and the shape of recession curves.
indicate poor model performance. Percent exceedance probability curves, which often are
daily flow duration curves, can illustrate how well the model
GRAPHICAL TECHNIQUES reproduces the frequency of measured daily flows throughout
Graphical techniques provide a visual comparison of sim- the calibration and validation periods (Van Liew et al., 2007).
ulated and measured constituent data and a first overview of General agreement between observed and simulated fre-
model performance (ASCE, 1993). According to Legates and quencies for the desired constituent indicates adequate simu-
McCabe (1999), graphical techniques are essential to ap- lation over the range of the conditions examined (Singh et al.,
propriate model evaluation. Two commonly used graphical 2004).

Table 1. Summary statistics from the literature review of reported NSE and PBIAS values.[a]
Calibration Validation
NSE PBIAS NSE PBIAS
Constituent Statistic Daily Monthly Daily Monthly Daily Monthly Daily Monthly
n 92 33 72 0 128 70 82 0
Minimum -0.23 0.14 -91.70 na -1.81 -3.35 -155.60 na
Streamflow
Maximum 0.95 0.91 26.50 na 0.89 0.93 47.18 na
Median 0.89 0.79 -1.30 na 0.67 0.63 -1.90 na
n 0 2 0 0 0 2 0 0
Minimum na 0.35 na na na 0.63 na na
Surface runoff
Maximum na 0.62 na na na 0.77 na na
Median na 0.49 na na na 0.70 na na
n 2 6 0 0 2 6 0 0
Minimum -2.50 0.49 na na -3.51 -2.46 na na
Sediment
Maximum 0.11 0.86 na na 0.23 0.88 na na
Median -1.20 0.76 na na -1.64 0.64 na na
n 0 2 0 0 0 2 0 0
Organic nitrogen Minimum na 0.57 na na na 0.43 na na
(organic N) Maximum na 0.58 na na na 0.73 na na
Median na 0.58 na na na 0.58 na na
n 0 2 0 0 0 2 0 0
Mineral nitrogen Minimum na -0.08 na na na 0.64 na na
(NO3-N) Maximum na 0.59 na na na 0.75 na na
Median na 0.26 na na na 0.70 na na
n 0 0 0 0 1 6 0 0
Total nitrogen Minimum na na na na 0.19 0.10 na na
(organic N + NO3-N) Maximum na na na na 0.19 0.85 na na
Median na na na na 0.19 0.76 na na
n 0 2 0 0 0 2 0 0
Organic phosphorus Minimum na 0.59 na na na 0.39 na na
(organic P) Maximum na 0.70 na na na 0.72 na na
Median na 0.65 na na na 0.56 na na
n 0 3 0 0 0 3 0 0
Mineral phosphorus Minimum na 0.53 na na na 0.51 na na
(PO4-P) Maximum na 0.78 na na na 0.81 na na
Median na 0.59 na na na 0.53 na na
n 0 1 0 0 0 1 0 0
Total phosphorus Minimum na 0.51 na na na 0.37 na na
(organic P + PO4-P) Maximum na 0.51 na na na 0.37 na na
Median na 0.51 na na na 0.37 na na
[a] n = number of reported values for the studies reviewed (sample size), NSE = Nash-Sutcliffe efficiency, PBIAS = percent bias,
and na = not available (used when n = 0).

Vol. 50(3): 885−900 889

REPORTED RANGES OF VALUES FOR RECOMMENDED STATIS- REPORTED PERFORMANCE RATINGS FOR RECOMMENDED
TICS STATISTICS
A review of published literature produced ranges of daily A review of published literature resulted in the perfor-
and monthly NSE and PBIAS values for both model calibra- mance ratings for NSE and PBIAS shown in tables 2 and 3.
tion and validation for surface runoff, streamflow, and se- Because RSR was developed in this research, similar re-
lected constituents, including: sediment, organic nitrogen, ported performance ratings are not available. Therefore, RSR
mineral nitrogen, total nitrogen, organic phosphorus, and ratings were based on Singh et al. (2004) recommendations
mineral phosphorus (tables A-1 through A-9). A few weekly that RMSE values less than half the standard deviation of
values reported by Narasimhan et al. (2005) are also included measured data can be considered low. In this study, we used
to indicate ranges of values for a weekly time step. It is impor- the recommended less than 0.5 RSR value as the most strin-
tant to note that calibration and validation were performed for gent (“very good”) rating and suggested two less stringent
different time periods; thus, the reported values reflect this ratings of 10% and 20% greater than this value for the “good”
difference. A summary of reported values for each constitu- and “satisfactory” ratings, respectively.
ent appears in table 1. Median values instead of means are in-
cluded because medians are less sensitive to extreme values MODEL EVALUATION GUIDELINES BASED ON PERFORMANCE
and are better indicators for highly skewed distributions. RATINGS
Most of the reviewed studies performed model calibration General model evaluation guidelines, for a monthly time
and validation based on streamflow (table 1). This is attrib- step, were developed based on performance ratings for the
uted to the relative abundance of measured long-term stream- recommended statistics and on project-specific consider-
flow data compared to sediment or nutrient data. Based on the ations. As stated previously, graphical techniques, especially
summary information for streamflow calibration and valida- hydrographs and percent exceedance probability curves, pro-
tion, daily NSE values tended to be higher than monthly val- vide visual model evaluation overviews. Utilizing these im-
ues, which contradicts findings from some individual studies portant techniques should typically be the first step in model
(e.g., Fernandez et al., 2005; Singh et al., 2005; Van Liew et evaluation. A general visual agreement between observed
al., 2007). This anomaly is potentially due to the increased and simulated constituent data indicates adequate calibration
sample sizes (n) for daily data. As expected, NSE and PBIAS and validation over the range of the constituent being simu-
values for streamflow were better for the calibration periods lated (Singh et al., 2004).
than the validation periods.

Table 2. Reported performance ratings for NSE.

Performance
Model Value Rating Modeling Phase Reference
HSPF >0.80 Satisfactory Calibration and validation Donigian et al. (1983)
APEX >0.40 Satisfactory Calibration and validation (daily) Ramanarayanan et al. (1997)
SAC-SMA <0.70 Poor Autocalibration Gupta et al. (1999)
SAC-SMA >0.80 Efficient Autocalibration Gupta et al. (1999)
DHM >0.75 Good Calibration and validation Motovilov et al. (1999)[a]
DHM 0.36 to 0.75 Satisfactory Calibration and validation Motovilov et al. (1999)[a]
DHM <0.36 Unsatisfactory Calibration and validation Motovilov et al. (1999)[a]
SWAT >0.65 Very good Calibration and validation Saleh et al. (2000)
SWAT 0.54 to 0.65 Adequate Calibration and validation Saleh et al. (2000)
SWAT >0.50 Satisfactory Calibration and validation Santhi et al. (2001); adapted by Bracmort et al. (2006)
SWAT and HSPF >0.65 Satisfactory Calibration and validation Singh et al. (2004); adapted by Narasimhan et al. (2005)
[a] Adapted by Van Liew et al. (2003) and Fernandez et al. (2005).

Table 3. Reported performance ratings for PBIAS.

Performance
Model Value Rating Modeling Phase Reference
HSPF < 10% Very good Calibration and validation Donigian et al. (1983)[a]
HSPF 10% to 15% Good Calibration and validation Donigian et al. (1983)[a]
HSPF 15% to 25% Fair Calibration and validation Donigian et al. (1983)[a]
SWAT <15% Satisfactory Flow calibration Santhi et al. (2001)
SWAT <20% Satisfactory For sediment after flow calibration Santhi et al. (2001)
SWAT <25% Satisfactory For nitrogen after flow and sediment calibration Santhi et al. (2001)
SWAT 20% Satisfactory Calibration and validation Bracmort et al. (2006)
SWAT <10% Very good Calibration and validation Van Liew et al. (2007)
SWAT <10% to <15% Good Calibration and validation Van Liew et al. (2007)
SWAT <15% to <25% Satisfactory Calibration and validation Van Liew et al. (2007)
SWAT >25% Unsatisfactory Calibration and validation Van Liew et al. (2007)
[a] Adapted by Van Liew et al. (2003) and Singh et al. (2004).

890 TRANSACTIONS OF THE ASABE

 Table 4. General performance ratings for recommended statistics for a monthly time step.
Performance PBIAS (%)
Rating RSR NSE Streamflow Sediment N, P
Very good 0.00 < RSR < 0.50 0.75 < NSE < 1.00 PBIAS < ±10 PBIAS < ±15 PBIAS< ±25
Good 0.50 < RSR < 0.60 0.65 < NSE < 0.75 ±10 < PBIAS < ±15 ±15 < PBIAS < ±30 ±25 < PBIAS < ±40
Satisfactory 0.60 < RSR < 0.70 0.50 < NSE < 0.65 ±15 < PBIAS < ±25 ±30 < PBIAS < ±55 ±40 < PBIAS < ±70
Unsatisfactory RSR > 0.70 NSE < 0.50 PBIAS > ±25 PBIAS > ±55 PBIAS > ±70

Separate surface runoff (SR) and ranges of values (table 1), were used to establish general per-
baseflow (BF) for measured daily flow
formance ratings, which appear in table 4. As shown in
table 4, the performance ratings for RSR and NSE are the
same for all constituents, but PBIAS is constituent specific.
Run model, then initial
graphical analysis
This difference is due to the recent availability of information
(PBIAS) on the uncertainty of measured streamflow and wa-
ter quality. Harmel et al. (2006) used the root mean square er-
ror propagation method of Topping (1972) to calculate the
If average of
Sim SR is ±10%
cumulative probable error resulting from four procedural
Adjust appropriate
Yes of average Meas SR No
parameters categories (discharge measurement, sample collection, sam-
and
NSE > 0.65 ple preservation and storage, and laboratory analysis) associ-
RSR < 0.60 ated with water quality data collection. Under typical
scenarios with reasonable quality control attention, typical
Run model, then
graphical analysis financial and personnel resources, and typical hydrologic
conditions, cumulative probable error ranges (in similar units
If average of to PBIAS) were estimated to be 6% to 19% for streamflow,
Sim BF is ±10%
Yes of average Meas BF No Adjust appropriate 7% to 53% for sediment, and 8% to 110% for N and P. These
parameters
and results were used to establish constituent-specific perfor-
NSE > 0.65
RSR < 0.60 mance rating for PBIAS. Constituent-specific ratings for
RSR and NSE can be established when similar information
Run model, then becomes available.
graphical analysis
Based on table 4, model performance can be evaluated as
If average of
“satisfactory” if NSE > 0.50 and RSR < 0.70 and, for mea-
Sim Sed is ±15%
Adjust appropriate
sured data of typical uncertainty, if PBIAS ± 25% for stream-
Yes of average Meas Sed No
and parameters flow, PBIAS ± 55% for sediment, and PBIAS ± 70% for N and
NSE > 0.65 P for a monthly time step. These ratings should be adjusted
RSR < 0.60
to be more or less strict based on project-specific consider-
Run model, then ations discussed in the next section.
graphical analysis A general calibration procedure chart (fig. 1) for flow, sed-
iment, and nutrients is included to aid with the manual model
If average of
Sim Org N&P is ±25%
calibration process. The recommended values for adequate
Yes of average Meas Org N&P No Adjust appropriate
parameters
model calibration are within the “good” and “very good” per-
and
NSE = 0.65 formance ratings presented in table 4. These limits for ade-
RSR = 0.60 quate manual calibration are stricter than the “satisfactory”
rating for general model evaluation because model parameter
Run model, then
graphical analysis values are optimized during calibration but not during model
validation or application. The importance of and appropriate
If average of
Sim Min N&P is ±25%
methods for proper model calibration are discussed in the
Yes of average Meas N&P No Adjust appropriate next section.
and parameters
NSE = 0.65
RSR = 0.60 ADDITIONAL CONSIDERATIONS
The model evaluation guidelines presented in the previous
section apply to the typical case of continuous, long-term
Calibration complete
NSE = Nash−Suttcliffe efficiency simulation for a monthly time step. However, because of the
RSR = RME/OBSTDEV ratio diversity of modeling applications, these guidelines should
be adjusted based on single-event simulation, quality and
Figure 1. General calibration procedure for flow, sediment, and nutrients quantity of measured data, model calibration procedure,
in the watershed models (based on calibration chart for SWAT from San- evaluation time step, and project scope and magnitude.
thi et al., 2001).
Single-Event Simulation
The next step should be to calculate values for NSE, When watershed models are applied on a single-event ba-
PBIAS, and RSR. With these values, model performance can sis, evaluation guidelines should reflect this specific case.
be judged based on general performance ratings (table 4). Generally, the objectives of single-event modeling are the de-
The reported performance ratings and corresponding values termination of peak flow rate and timing, flow volume, and
developed for individual studies, in addition to the reported recession curve shape (ASCE, 1993; Van Liew et al., 2003).

Vol. 50(3): 885−900 891

Accurate prediction of peak flow rate and time-to-peak is es- perturbing parameter values one at a time (Hamby, 1994). A
sential for flood estimation and forecasting (Ramírez, 2000). detailed review of sensitivity analysis methods is presented
Time-to-peak is affected by drainage network density, slope, by Hamby (1994) and Isukapalli (1999). According to Isuka-
channel roughness, and soil infiltration characteristics palli (1999), the choice of a sensitivity method depends on
(Ramírez, 2000), and peak flow rate is typically affected by the sensitivity measure employed, the desired accuracy in the
rainfall intensity and antecedent soil moisture content, estimates of the sensitivity measure, and the computational
among other factors. cost involved. Detailed information on how these factors af-
One of the recommended (ASCE, 1993) model evaluation fect the choice of sensitivity analysis method is given by Isu-
statistics for the peak flow rate is a simple percent error in kapalli (1999). After key parameters and their respective
peak flow rates (PEP) computed by dividing the difference required precision have been determined, manual or auto-
between the simulated peak flow rate and measured peak matic calibration is done.
flow rate by the measured peak flow rate and expressing the Conventionally, calibration is done manually and consists
result as a percentage. Model performance related to time-to- of changing model input parameter values to produce simu-
peak can be determined similarly. In addition, for single- lated values that are within a certain range of measured data
event simulation, Boyle et al. (2000) recommended that the (Balascio et al., 1998). The calibrated model may be used to
hydrograph be divided into three phases (rising limb, falling simulate multiple processes, such as streamflow volumes,
limb, baseflow) based on differing watershed behavior dur- peak flows, and/or sediment and nutrient loads. In such cases,
ing rainfall and dry periods. A different performance rating two or more model evaluation statistics may be necessary in
should be applied to each hydrograph phase. If the perfor- order to address the different processes (Balascio et al.,
mance ratings are similar for each phase, then a single perfor- 1998). However, when the number of parameters used in the
mance rating can be applied to the overall hydrograph in manual calibration is large, especially for complex hydrolog-
subsequent evaluation. Otherwise, a different performance ic models, manual calibration can become labor-intensive
rating should be used to evaluate each hydrograph phase. (Balascio et al., 1998). In this, case automatic calibration is
more appropriate.
Quality and Quantity of Measured Data
Automatic calibration involves computation of the predic-
Although it is commonly accepted that measured data are
tion error using an equation (objective function) and an auto-
inherently uncertain, the uncertainty is rarely considered in
matic optimization procedure (search algorithm) to search for
model evaluation, perhaps because of the lack of relevant
parameter values that optimize the value of the objective func-
data. Harmel et al. (2006) suggested that the uncertainty of
tion (Gupta et al., 1999). One of the automatic calibration meth-
measured data, which varies based on measurement condi-
ods is the shuffled complex evolution global optimization
tions, techniques, and constituent type, must be considered to algorithm developed at the University of Arizona (SCE-UA)
appropriately evaluate watershed models. In terms of the
(Duan et al., 1993) and used by vanGriensven and Bauwens
present guidelines and performance ratings, modeled stream-
(2003) to develop a multi-objective calibration method for
flow may be rated “good” if it is within 10% to 15% of mea- semi-distributed water quality models. Other automatic calibra-
sured streamflow data of typical quality (table 4), for
tion methods include the multilevel calibration (MLC) semi-
example. In contrast, performance ratings should be stricter
automated method (Brazil, 1988), the multi-objective complex
when high-quality (low uncertainty) data are available. In evolution algorithm (MOCOM-UA) (Yapo et al., 1998), and pa-
this case, PBIAS for modeled streamflow may be required to
rameter estimation by sequential testing (PEST) (Taylor and
be <10% to receive a “good” rating. If data collected in worst-
Creelman, 1967), among others.
case conditions are used for model evaluation, then perfor- Data used to calibrate model simulations have a direct ef-
mance ratings should be relaxed to reflect the extreme
fect on the validation and evaluation results. Ideal calibration
uncertainty. It can be argued in this case, however, that highly
should use 3 to 5 years of data that includes average, wet, and
uncertain data offer little value and should not be used in dry years so that the data encompass a sufficient range of
model evaluation.
hydrologic events to activate all model constituent processes
Finally, in situations when a complete measured time se-
during calibration (Gan et al., 1997). However, if this is not
ries does not exist, for instance when only a few grab samples possible, then the available data should be separated into two
per year are available, the data may not be sufficient for anal-
sets, i.e., “above-mean” flows (wet years) and “below-mean”
ysis using the recommended statistics. In such situations,
flows (dry years), and then evaluated with stricter perfor-
comparison of frequency distributions and/or percentiles mance ratings required for wet years (Gupta et al., 1999).
(e.g., 10th, 25th, 50th, 75th, and 90th) may be more appropri-
Moreover, if the goal is to test the robustness of the model ap-
ate than the quantitative statistics guidelines.
plications under different environmental conditions, then dif-
Model Calibration Procedure ferent datasets can be used during model calibration and
Proper model calibration is important in hydrologic mod- validation.
eling studies to reduce uncertainty in model simulations (En- In addition, a good calibration procedure uses multiple
gel et al., 2007). Ideal model calibration involves: (1) using statistics, each covering a different aspect of the hydrograph,
data that includes wet, average, and dry years (Gan et al., so that the whole hydrograph is covered. This is important be-
1997), (2) using multiple evaluation techniques (Willmott, cause using a single statistic can lead to undue emphasis on
1981; ASCE, 1993; Legates and McCabe, 1999; Boyle et al., matching one aspect of the hydrograph at the expense of other
2000), and (3) calibrating all constituents to be evaluated. aspects (Boyle et al., 2000). For manual calibration, each sta-
The calibration procedure typically involves a sensitivity tistic should be tracked while adjusting model parameters
analysis followed by manual or automatic calibration. The (Boyle et al., 2000) to allow for balancing the trade-offs in the
most fundamental sensitivity analysis technique utilizes par- ability of the model to simulate various aspects of the hydro-
tial differentiation, whereas the simplest method involves graph while recognizing potential errors in the observed data.

892 TRANSACTIONS OF THE ASABE

 Finally, ideal model calibration considers water balance 2007), although a few used annual time steps (Gupta et al.,
(peak flow, baseflow) and sediment and nutrient transport be- 1999; Shirmohammadi et al., 2001; Reyes et al., 2004), and
cause calibrating one constituent will not ensure that other one used weekly time steps (Narasimhan et al., 2005). There-
constituents are adequately simulated during validation. fore, the time steps considered in this article are the daily and
Even though a complete set of hydrologic and water quality monthly. Typically, model simulations are poorer for shorter
data are rarely available, all available data should be consid- time steps than for longer time steps (e.g., daily versus
ered. To calibrate water balance, it is recommended to sepa- monthly or yearly) (Engel et al., 2007). For example, Yuan
rate baseflow and surface flow (surface runoff) from the total et al. (2001) reported an R2 value of 0.5 for event comparison
streamflow for both the measured and simulated data using of predicted and observed sediment yields, and an R2 value
a baseflow filter program. The baseflow filter developed by of 0.7 for monthly comparison. The NSE values were 0.395
Arnold et al. (1995) and modified by Arnold and Allen (1999) and 0.656 for daily and monthly, respectively, for
is available at www.brc.tamus.edu/swat/soft_baseflow .html. DRAINMOD-DUFLOW calibration, and 0.363 and 0.664
Baseflow and recharge data from this procedure have shown for daily and monthly, respectively, for DRAINMOD-W cal-
good correlation with those produced by SWAT (Arnold et ibration (Fernandez et al., 2005). Similarly, the NSE values
al., 2000). With estimated baseflow data, the baseflow ratio were 0.536 and 0.870 for daily and monthly, respectively, for
can be computed for measured and simulated data by divid- DRAINMOD-DUFLOW validation, and 0.457 and 0.857 for
ing baseflow estimates by the total measured or simulated daily and monthly, respectively, for DRAINMOD-W valida-
streamflow. The calibration and validation process can be tion (Fernandez et al., 2005). Additional research work that
considered satisfactory if the estimated baseflow ratio for supports the described findings includes that of Santhi et al.
simulated flow is within 20% of the measured flow baseflow (2001), Van Liew et al. (2003), and Van Liew et al. (2007) us-
ratio (Bracmort et al., 2006). Because plant growth and bio- ing SWAT. The performance ratings presented in table 4 for
mass production can have an effect on the water balance, rea- RSR and NSE statistics are for a monthly time step; therefore,
sonable local/regional plant growth days and biomass they need to be modified appropriately. Generally, as the
production may need to be verified during model calibration. evaluation time step increases, a stricter performance rating
Annual local/regional evapotranspiration (ET) may also is warranted.
need to be verified or compared with measured estimates dur-
Project Scope and Magnitude
ing model calibration.
The scope and magnitude of the modeling project also af-
Stricter performance ratings should generally be required
during model calibration than during validation. This differ- fects model evaluation guidelines. The intended use of the
model is an indication of the seriousness of the potential con-
ence is recommended because parameter values are opti-
mized during model calibration, but parameters are not sequences or impacts of decisions made based on model re-
adjusted in validation, which is possibly conducted under dif- sults (U.S. EPA, 2002). For instance, stricter performance
rating requirements need to be set for projects that involve
ferent conditions than those occurring during calibration. Al-
though the importance of model calibration is well potentially large consequences, such as congressional testi-
established, performance ratings can be relaxed if improper mony, development of new laws and regulations, or the sup-
port of litigation. More modest performance ratings would be
calibration procedures are employed.
It is necessary to note that although proper model calibra- acceptable for technology assessment or “proof of principle,”
tion is important in reducing error in model output, experi- where no litigation or regulatory action is expected. Even
lower performance ratings will suffice if the model is used for
ence has shown that model simulation results may contain
substantial errors. Therefore, rather than provide a point esti- basic exploratory research requiring extremely fast turn-
mate of a given quantity of model output, it may be preferable around or high flexibility.
Finally, according to the U.S. EPA (2002), if model simu-
to provide an interval estimate with an associated probability
that the value of the quantity will be contained by the interval lation does not yield acceptable results based on predefined
(Haan et al., 1998). In other words, uncertainty analysis needs performance ratings, this may indicate that: (1) conditions in
the calibration period were significantly different from those
to be included in model evaluations. Uncertainty analysis is
defined as the process of quantifying the level of confidence in the validation period, (2) the model was inadequately or
in a given model simulation output based on: (1) the quality improperly calibrated, (3) measured data were inaccurate,
(4) more detailed inputs are required, and/or (5) the model is
and amount of measured data available, (2) the absence of
measured data due to the lack of monitoring in certain loca- unable to adequately represent the watershed processes of in-
tions, (3) the lack of knowledge about some physical pro- terest.
cesses and operational procedures, (4) the approximate
nature of the mathematical equations used to simulate pro-
cesses, and (5) the quality of the model sensitivity analysis A CASE STUDY
and calibration. Detailed model uncertainty analysis is be- PROJECT DESCRIPTION
yond the scope of this research, but more model output uncer- As a component of the CEAP-WAS, the Soil and Water
tainty information can be obtained from published literature. Assessment Tool (SWAT2005; Arnold et al., 1998) was ap-
plied to the Leon River watershed in Texas. The watershed
Evaluation Time Step
Most of the literature reviewed used daily and/or monthly drains into Lake Belton, which lies within Bell and Coryell
counties and provides flood control, water supply, and public
time steps (Saleh et al., 2000; Santhi et al., 2001; Yuan et al.,
recreation. The lake has a surface area of approximately
2001; Sands et al., 2003; Van Liew et al., 2003; Chu and Shir-
mohammadi, 2004; Saleh and Du, 2004; Singh et al., 2004; 12,300 acres with a maximum depth of 124 feet. The total
conservation storage is 372,700 ac-ft. The Lake Belton wa-
Bracmort et al., 2006; Singh et al., 2005; Van Liew et al.,
tershed (Leon River) covers approximately 2.3 million acres

Vol. 50(3): 885−900 893

within five counties in central Texas. The majority of the land procedures, refer to the previous discussion in the Additional
use in the watershed is pasture, hay, and brushy rangeland Considerations section.
(63%). Cropland comprises about 10% of the watershed area.
The northwestern (upper) half of the watershed contains nu- MODEL EVALUATION RESULTS BASED ON THE DEVELOPED
merous animal feeding operations, mainly dairies. Currently, GUIDELINES
there are approximately 60 permitted dairies and 40 smaller When these model performance ratings were applied to
dairies (not requiring permits), with approximately 70,000 the SWAT2005 modeling in the Leon River (Lake Belton)
total cows. The main goal of the study is to use SWAT2005 watershed, the following results were obtained (figs. 2 and 3,
to predict the impact of land management on the watershed table 5). Graphical results during calibration (fig. 2) and val-
over long periods of time, with special focus on waste man- idation (fig. 3) indicated adequate calibration and validation
agement practices. over the range of streamflow, although the calibration results
To accomplish this goal, SWAT2005, with its many input showed a better match than the validation results. NSE values
parameters that describe physical, chemical, and biological for the monthly streamflow calibration and validation ranged
processes, required calibration for application in the study from 0.66 to 1.00. According to the model evaluation guide-
watershed. Proper model calibration is important in hydro- lines, SWAT2005 simulated the streamflow trends well to
logic modeling studies to reduce uncertainty in model predic- very well, as shown by the statistical results, which are in
tions. For a general description of proper model calibration agreement with the graphical results. The RSR values ranged

Figure 2. Monthly discharge (CMS) calibration for the Leon River sub-basin 6 WS outlet.

Figure 3. Monthly discharge (CMS) validation for the Leon River sub-basin 13.

894 TRANSACTIONS OF THE ASABE

 Table 5. Results of SWAT2005 average monthly streamflow model output, Leon
River watershed, Texas, based on the developed model evaluation guidelines.
Evaluation Statistic
NSE RSR PBIAS
Sub-basin Calibration Validation Calibration Validation Calibration Validation
6 1.00 (very good) 1.00 (very good) 0.03 (very good) 0.06 (very good) -2.86 (very good) -3.85(very good)
13 0.66 (good) 0.69 (good) 0.58 (good) 0.56 (good) -4.89 (very good) -29.04 (unsatisfactory)
21 0.81 (very good) 0.84 (very good) 0.43 (very good) 0.40 (very good) -3.62 (very good) 0.41 (very good)
36 0.93 (very good) 0.85 (very good) 0.26 (very good) 0.39 (very good) -0.28 (very good) -2.94 (very good)
44 1.00 (very good) 0.78 (very good) 0.06 (very good) 0.46 (very good) -1.58 (very good) 12.31 (good)
50 0.78 (very good) -- 0.46 (very good) -- -1.10 (very good) --
58 0.69 (very good) -- 0.55 (very good) -- 2.15 (very good) --

from 0.03 to 0.58 during both calibration and validation. sion apply to evaluation of model simulation related to
These values indicate that the model performance for stream- streamflow, sediments, and nutrients (N and P).
flow residual variation ranged from good to very good. The Based on previous published recommendations, a com-
PBIAS values varied from −4.89% to 2.15% during calibra- bination of graphical techniques and dimensionless and error
tion and from −29.04% to 12.31% during validation. The av- index statistics should be used for model evaluation. In addi-
erage magnitude of simulated monthly streamflow values tion to hydrographs and percent exceedance probability
was within the very good range (PBIAS < ±10) for each sub- curves, the quantitative statistics NSE, PBIAS, and RSR
basin during calibration (table 5). However, simulated values were recommended. Performance ratings for the recom-
fell within unsatisfactory, good, and very good ranges during mended statistics, for a monthly time step, are presented in
validation for various sub-basins. Aside from one indication table 4. In general, model simulation can be judged as “satis-
of unsatisfactory model performance, SWAT2005 simulation factory” if NSE > 0.50 and RSR < 0.70, and if PBIAS ± 25%
of streamflow was “good” to “very good” in terms of trends for streamflow, PBIAS ± 55% for sediment, and PBIAS
(NSE), residual variation (RSR), and average magnitude ± 70% for N and P for measured data of typical uncertainty.
(PBIAS). As apparent from this evaluation of the Leon River These PBIAS ratings, however, should be adjusted if mea-
watershed, situations might arise that generate conflicting surement uncertainty is either very low or very high. As indi-
performance ratings for various watersheds and/or output cated by these PBIAS ratings, it is important to consider
variables. measured data uncertainty when using PBIAS to evaluate
In situations with conflicting performance ratings, those watershed models. In addition, general guidelines for manual
differences must be clearly described. For example, if simu- calibration for flow, sediment, and nutrients were presented
lation for one output variable in one watershed produces un- (fig. 1). Additional considerations, such as single-event sim-
balanced performance ratings of “very good” for PBIAS, ulation, quality and quantity of measured data, model cal-
“good” for NSE, and “satisfactory” for RSR, then the overall ibration procedure considerations, evaluation time step, and
performance should be described conservatively as “satisfac- project scope and magnitude, which affect these guidelines,
tory” for that one watershed and that one output variable. were also discussed. The guidelines presented should be ad-
However, it would be preferable to describe the performance justed when appropriate to reflect these considerations. To il-
in simulation of average magnitudes (PBIAS) as “very lustrate the application of the developed model evaluation
good,” in simulation of trends (NSE) as “good,” and in simu- guidelines, a case study was provided.
lation of residual variation (RSR) as “satisfactory.” Similarly, Finally, the recommended model evaluation statistics and
if performance ratings differ for various watersheds and/or their respective performance ratings, and the step-by-step de-
output types, then those differences must be clearly de- scription of how they should be used, were presented together
scribed. to establish a platform for model evaluation. As new and im-
proved methods and information are developed, the recom-
mended guidelines should be updated to reflect these
developments.
SUMMARY AND CONCLUSIONS
Most research and application projects involving wa-
ACKNOWLEDGEMENTS
tershed simulation modeling utilize some type of predefined,
project-specific model evaluation techniques to compare The USDA-NRCS is acknowledged for providing funds for
this work through the support of the Conservation Effects As-
simulated output with measured data. Previous research has
sessment Project − Watershed Assessment Studies. The authors
produced valuable comparative information on selected
model evaluation techniques; however, no comprehensive are grateful to Tim Dybala for allowing the use of part of his
work for the case study. The authors also thank Alan Verser for
standardization is available that includes recently developed
additional information on model calibration procedure.
statistics with corresponding performance ratings and appli-
cable guidelines for model evaluation. Thus, the present re-
search selected and recommended model evaluation
techniques (graphical and statistical), reviewed published REFERENCES
ranges of values and corresponding performance ratings for Amatya, D. M., G. M. Chescheir, G. P. Fernandez, R. W. Skaggs,
the recommended statistics, and established guidelines for and J. W. Gilliam. 2004. DRAINWAT-based methods for esti-
model evaluation based on the review results and project- mating nitrogen transport in poorly drained watersheds. Trans.
specific considerations. These recommendations and discus- ASAE 47(3): 677-687.

Vol. 50(3): 885−900 895

Arnold, J. G., and P. M. Allen. 1999. Automated methods for esti- Hamby, D. M. 1994. A review of techniques for parameter sensitiv-
mating baseflow and ground water recharge from streamflow re- ity analysis of environmental models. Environ. Monitoring and
cords. J. American Water Resources Assoc. 35(2): 411-424. Assessment 32(2): 135-154.
Arnold, J. G., P. M. Allen, R. Muttiah, and G. Bernhardt. 1995. Au- Harmel, R. D., R. J. Cooper, R. M. Slade, R. L. Haney, and J. G. Ar-
tomated base flow separation and recession analysis techniques. nold. 2006. Cumulative uncertainty in measured streamflow and
Ground Water 33(6): 1010-1018. water quality data for small watersheds. Trans. ASAE 49(3):
Arnold, J. G., R. Srinivisan, R. S. Muttiah, and P. M. Allen. 1998. 689-701.
Large-area hydrologic modeling and assessment: Part I. Model Isukapalli, S. S. 1999. Uncertainty analysis of transport-
development. J. American Water Resources Assoc. 34(1): 73-89. transformation models. Unpublished PhD diss. New Brunswick,
Arnold, J. G., R. S. Muttiah, R. Srinivasan, and P. M. Allen. 2000. N.J.: Rutgers, The State University of New Jersey, Department
Regional estimation of base flow and groundwater recharge in of Chemical and Biochemical Engineering.
the upper Mississippi River basin. J. Hydrology 227(1-2): 21-40. Legates, D. R., and G. J. McCabe. 1999. Evaluating the use of
ASCE. 1993. Criteria for evaluation of watershed models. J. Irriga- “goodness-of-fit” measures in hydrologic and hydroclimatic
tion Drainage Eng. 119(3): 429-442. model validation. Water Resources Res. 35(1): 233-241.
Balascio, C. C., D. J. Palmeri, and H. Gao. 1998. Use of a genetic Ma, L., J. C. Ascough II, L. R. Ahuja, M. J. Shaffer, J. D. Hanson,
algorithm and multi-objective programming for calibration of a and K. W. Rojas. 2000. Root zone water quality model sensitiv-
hydrologic model. Trans. ASAE 41(3): 615-619. ity analysis using Monte Carlo simulation. Trans. ASAE 43(4):
Borah, D. K., and M. Bera. 2004. Watershed-scale hydrologic and 883-895.
nonpoint-source pollution models: Review of applications. Motovilov, Y. G., L. Gottschalk, K. England, and A. Rodhe. 1999.
Trans. ASAE 47(3): 789-803. Validation of distributed hydrological model against spatial ob-
Boyle, D. P., H. V. Gupta, and S. Sorooshian. 2000. Toward im- servations. Agric. Forest Meteorology 98-99: 257-277.
proved calibration of hydrologic models: Combining the Narasimhan, B., R. Srinivasan, J. G. Arnold, and M. Di Luzio.
strengths of manual and automatic methods. Water Resources 2005. Estimation of long-term soil moisture using a distributed
Res. 36(12): 3663-3674. parameter hydrologic model and verification using remotely
Bracmort, K. S., M. Arabi, J. R. Frankenberger, B. A. Engel, and J. sensed data. Trans. ASAE 48(3): 1101-1113.
G. Arnold. 2006. Modeling long-term water quality impact of Nash, J. E., and J. V. Sutcliffe. 1970. River flow forecasting through
structural BMPS. Trans. ASAE 49(2): 367-384. conceptual models: Part 1. A discussion of principles. J. Hydrol-
Brazil, L. E. 1988. Multilevel calibration strategy for complex ogy 10(3): 282-290.
hydrologic simulation models. Unpublished PhD diss. Fort Col- Parker, R., J. G. Arnold, M. Barrett, L. Burns, L. Carrubba, C.
lins, Colo.: Colorado State University, Department of Civil En- Crawford, S. L. Neitsch, N. J. Snyder, R. Srinivasan, and W. M.
gineering. Williams. 2006. Evaluation of three watershed-scale pesticide
CEAP-WAS. 2005. Conservation effects assessment project: Wa- fate and transport models. J. American Water Resources Assoc.
tershed assessment studies. Available at: ftp://ftp- (in review).
fc.sc.egov.usda.gov/NHQ/nri/ceap/ceapwaswebrev121004.pdf. Peterson, J. R., and J. M. Hamlett. 1998. Hydrologic calibration of
Accessed 15 August 2005. the SWAT model in a watershed containing fragipan soils. J.
Chu, T. W., and A. Shirmohammadi. 2004. Evaluation of the SWAT American Water Resources Assoc. 34(3): 531-544.
model’s hydrology component in the piedmont physiographic Ramanarayanan, T. S., J. R. Williams, W. A. Dugas, L. M. Hauck,
region of Maryland. Trans. ASAE 47(4): 1057-1073. and A. M. S. McFarland. 1997. Using APEX to identify alterna-
Donigian, A. S., J. C. Imhoff, and B. R. Bicknell. 1983. Predicting wa- tive practices for animal waste management. ASAE Paper No.
ter quality resulting from agricultural nonpoint-source pollution via 972209. St. Joseph, Mich.: ASAE.
simulation − HSPF. In Agricultural Management and Water Quali- Ramírez, J. A. 2000. Chapter 11: Prediction and modeling of flood
ty, 200-249. Ames, Iowa: Iowa State University Press. hydrology and hydraulics. In Inland Flood Hazards: Human,
Duan, Q. Y., V. K Gupta, and S. Sorooshian. 1993. Shuffled com- Riparian and Aquatic Communities. E. Wohl, ed. Cambridge,
plex evolution approach for effective and efficient global mini- U.K.: Cambridge University Press.
mization. J. Optimization Theory and Appl. 76(3): 501-521. Reyes, M. R., R. W. Skaggs, and R. L. Bengtson. 2004. GLEAMS-
Engel B., D. Storm, M. White, and J. G. Arnold. 2007. A hydrolog- SWT with nutrients. Trans. ASAE 47(1): 129-132.
ic/water quality model application protocol. J. American Water Refsgaard, J. C. 1997. Parameterisation, calibration, and validation
Resources Assoc. (in press). of distributed hydrological models. J. Hydrology 198(1): 69-97.
Engelmann, C. J. K, A. D. Ward, A. D. Christy, and E. S. Bair. Saleh, A., and B. Du. 2004. Evaluation of SWAT and HSPF within
2002. Applications of the BASINS database and NPSM model BASINS program for the upper North Bosque River watershed
on a small Ohio watershed. J. American Water Resources Assoc. in central Texas. Trans. ASAE 47(4): 1039-1049.
38(1): 289-300. Saleh, A, J. G. Arnold, P. W. Gassman, L. M. Hauk, W. D. Rosen-
Fernandez, G. P., G. M. Chescheir, R. W. Skaggs, and D. M. Ama- thal, J. R. Williams, and A. M. S. MacFarland. 2000. Applica-
tya. 2005. Development and testing of watershed-scale models tion of SWAT for the upper North Bosque River watershed.
for poorly drained soils. Trans. ASAE 48(2): 639-652. Trans. ASAE 43(5): 1077-1087.
Gan, T. Y., and G. F. Biftu. 1996. Automatic calibration of concep- Sands, G. R., C. X. Jin, A. Mendez, B. Basin, P. Wotzka, and P.
tual rainfall-runoff models: Optimization algorithms, catchment Gowda. 2003. Comparing the subsurface drainage flow predic-
conditions, and model structure. Water Resources Res. 32(12): tion of the DRAINMOD and ADAPT models for a cold climate.
3513-3524. Trans. ASAE 46(3): 645-656.
Gan, T. Y, E. M. Dlamini, and G. F. Biftu. 1997. Effects of model Santhi, C, J. G. Arnold, J. R. Williams, W. A. Dugas, R. Srinivasan,
complexity and structure, data quality, and objective functions and L. M. Hauck. 2001. Validation of the SWAT model on a
on hydrologic modeling. J. Hydrology 192(1): 81-103. large river basin with point and nonpoint sources. J. American
Gupta, H. V., S. Sorooshian, and P. O. Yapo. 1999. Status of auto- Water Resources Assoc. 37(5): 1169-1188.
matic calibration for hydrologic models: Comparison with mul- Sevat, E., and A. Dezetter. 1991. Selection of calibration objective
tilevel expert calibration. J. Hydrologic Eng. 4(2): 135-143. functions in the context of rainfall-runoff modeling in a Suda-
Haan, C. T., D. E. Storm, T. Al-Issa, S. Prabhu, G. J. Sabbagh, and nese savannah area. Hydrological Sci. J. 36(4): 307-330.
D. R. Edwards. 1998. Effect of parameter distributions on uncer- Shirmohammadi, A., T. W. Chu, H. Montas, and T. Sohrabi. 2001.
tainty analysis of hydrologic models. Trans. ASAE 41(1): 65-70. SWAT model and its applicability to watershed nonpoint-source

896 TRANSACTIONS OF THE ASABE

 pollution assessment. ASAE Paper No. 012005. St. Joseph, Van Liew, M. W., T. L. Veith, D. D. Bosch, and J. G. Arnold. 2007.
Mich.: ASAE. Suitability of SWAT for the conservation effects assessment
Singh, J., H. V. Knapp, and M. Demissie. 2004. Hydrologic model- project: A comparison on USDA-ARS experimental watersheds.
ing of the Iroquois River watershed using HSPF and SWAT. J. Hydrologic Eng. 12(2): 173-189.
ISWS CR 2004-08. Champaign, Ill.: Illinois State Water Survey. Varanou, E., E. Gkouvatsou, E. Baltas, and M. Mimikou. 2002.
Available at: www.sws.uiuc.edu/pubdoc/CR/ Quantity and quality integrated catchment modeling under cli-
ISWSCR2004-08.pdf. Accessed 8 September 2005. mate change with use of soil and water assessment tool model.
Singh, J., H. V. Knapp, J. G. Arnold, and M. Demissie. 2005. ASCE J. Hydrologic Eng. 7(3): 228-244.
Hydrologic modeling of the Iroquois River watershed using Vazquez-Amábile, G. G., and B. A. Engel. 2005. Use of SWAT to
HSPF and SWAT. J. American Water Resources Assoc. 41(2): compute groundwater table depth and streamflow in the Musca-
361-375. tatuck River watershed. Trans. ASAE 48(3): 991-1003.
Spruill, C. A., S. R. Workman, and J. L. Taraba. 2000. Simulation Wang, X., and A. M. Melesse. 2005. Evaluation of the SWAT mod-
of daily and monthly stream discharge from small watersheds el’s snowmelt hydrology in a northwestern Minnesota wa-
using the SWAT model. Trans. ASAE 43(6): 1431-1439. tershed. Trans. ASAE 48(4): 1359-1376.
Taylor, M. M., and C. D. Creelman. 1967. PEST: Efficient estimates Willmott, C. J. 1984. On the evaluation of model performance in
on probability functions. J. Acoustical Soc. America 41(4A): physical geography. In Spatial Statistics and Models, 443-460.
782-787. G. L. Gaile and C. J. Willmott, eds. Norwell, Mass.: D. Reidel.
Topping, J. 1972. Errors of Observation and Their Treatment. 4th Willmott, C. J. 1981. On the validation of models. Physical Geog-
ed. London, U.K.: Chapman and Hall. raphy 2: 184-194.
U.S. EPA. 2002. Guidance for quality assurance project plans for Yapo, P. O., H. V. Gupta, and S. Sorooshian, 1998. Multi-objective
modeling. EPA QA/G-5M. Report EPA/240/R-02/007. Wash- global optimization for hydrologic models. J. Hydrology 204(1):
ington, D.C.: U.S. EPA, Office of Environmental Information. 83-97.
vanGriensven, A., and W. Bauwens. 2003. Multiobjective autocal- Yuan, Y., R. L. Bingner, and R. A. Rebich. 2001. Evaluation of An-
ibration for semidistributed water quality models. Water Re- nAGNPS on Mississippi Delta MSEA watersheds. Trans. ASAE
sources Res. 39(12): 1348-1356. 44(5): 1183-1190.
Van Liew, M. W., J. G. Arnold, and J. D. Garbrecht. 2003. Hydro-
logic simulation on agricultural watersheds: Choosing between
two models. Trans. ASAE 46(6): 1539-1551.

APPENDIX
Reported Values of NSE and PBIAS for Various Constituents

Table A-1. Daily and monthly surface runoff calibration and validation value ranges.[a]
Watershed - Model Calibration Value Ranges Validation Value Ranges
(Reference) Statistic Daily Monthly Daily Monthly
Warner Creek, Maryland - SWAT NSE -- 0.35 -- 0.77
Chu and Shirmohammadi, 2004) PBIAS
Black Creek, Indiana - SWAT NSE -- 0.62 to 0.80 -- 0.63 to 0.75
(Bracmort et al., 2006) PBIAS
[a] In tables A-1 through A-9, a dash (--) indicates no value reported for the statistic used; a blank space indicates that the statistic was not used.

Table A-2. Daily and monthly sediment calibration and validation value ranges.
Watershed - Model Calibration Value Ranges Validation Value Ranges
(Reference) Statistic Daily Monthly Daily Monthly
Bosque River, Texas - SWAT NSE -- -- -- 0.81
(Saleh et al., 2000) PBIAS
Bosque River, Texas - SWAT NSE -- 0.69 to 0.80 -- 0.23 to 0.70
(Santhi et al., 2001) PBIAS
Bosque River, Texas - SWAT NSE -2.50 0.83 -3.51 0.59
(Saleh and Du, 2004) PBIAS
Bosque River, Texas - HSPF NSE 0.11 0.72 0.23 0.88
(Saleh and Du, 2004) PBIAS
Hellbranch Run, Ohio - HSPF NSE -- 0.49 -- -2.46
(Engelmann et al., 2002)[a] PBIAS
Black Creek, Indiana - SWAT NSE -- 0.86 to 0.92 -- 0.68 to 0.75
(Bracmort et al., 2006) PBIAS
[a] In Borah and Bera (2004).

Vol. 50(3): 885−900 897

 Table A-3. Daily and monthly organic N calibration and validation value ranges.
Watershed - Model Calibration Value Ranges Validation Value Ranges
(Reference) Statistic Daily Monthly Daily Monthly
Bosque River, Texas - SWAT NSE -- -- -- 0.78
(Saleh et al., 2000) PBIAS
Bosque River watershed, Texas - SWAT NSE -- 0.57 to 0.58 -- 0.43 to 0.73
(Santhi et al., 2001) PBIAS

Table A-4. Daily and monthly NO3−N calibration and validation value ranges.
Watershed - Model Calibration Value Ranges Validation Value Ranges
(Reference) Statistic Daily Monthly Daily Monthly
Bosque River, Texas - SWAT NSE -- -- -- 0.37
(Saleh et al., 2000) PBIAS
Plymouth, North Carolina - DRAINMOD-W NSE 0.36 0.66 0.46 0.86
(Fernandez et al., 2005) PBIAS
Plymouth, North Carolina - DRAINMOD-DUFLOW NSE 0.40 0.66 0.54 0.87
(Fernandez et al., 2005) PBIAS
Bosque River, Texas - SWAT NSE -- -0.08 to 0.59 -- 0.64 to 0.75
(Santhi et al., 2001) PBIAS

Table A-5. Daily and monthly total N (organic N + NO3−N) calibration and validation value ranges.
Watershed - Model Calibration Value Ranges Validation Value Ranges
(Reference) Statistic Daily Monthly Daily Monthly
Bosque River, Texas - SWAT NSE -- -- -- 0.86
(Saleh et al., 2000) PBIAS
Plymouth, North Carolina - DRAINWAT NSE -- -- 0.19 0.76
(Amatya et al., 2004) PBIAS

Table A-6. Daily and monthly PO4−P calibration and validation value ranges.
Watershed - Model Calibration Value Ranges Validation Value Ranges
(Reference) Statistic Daily Monthly Daily Monthly
Bosque River, Texas - SWAT NSE -- -- -- 0.94
(Saleh et al., 2000) PBIAS
Bosque River, Texas - SWAT NSE -- 0.53 to 0.59 -- 0.53 to 0.81
(Santhi et al., 2001) PBIAS
Black Creek, Indiana - SWAT NSE -- 0.78 to 0.84 -- 0.51 to 0.74
(Bracmort et al., 2006) PBIAS

Table A-7. Daily and monthly organic P calibration and validation value ranges.
Watershed - Model Calibration Value Ranges Validation Value Ranges
(Reference) Statistic Daily Monthly Daily Monthly
Bosque River, Texas - SWAT NSE -- -- -- 0.54
(Saleh et al., 2000) PBIAS
Bosque River, Texas - SWAT NSE -- 0.59 to 0.70 -- 0.39 to 0.72
(Santhi et al., 2001) PBIAS

Table A-8. Daily and monthly total P calibration and validation value ranges.
Watershed - Model Calibration Value Ranges Validation Value Ranges
(Reference) Statistic Daily Monthly Daily Monthly
Black Creek, Indiana - SWAT NSE -- 0.51 -- 0.37
(Bracmort et al., 2006) PBIAS

Table A-9. Daily and monthly streamflow calibration and validation value ranges (continued).
Watershed - Model Calibration Value Ranges Validation Value Ranges
(Reference) Statistic Daily Monthly Daily Monthly
Bosque River, Texas - SWAT NSE -- -- -- 0.56
(Saleh et al., 2000) PBIAS
Eight watersheds in southwest Oklahoma - SWAT NSE 0.56 to 0.58 0.66 to 0.79 -0.37 to 0.72 -1.05 to 0.89
(Van Liew et al., 2003) PBIAS
(continued)

898 TRANSACTIONS OF THE ASABE

 Table A-9 (continued). Daily and monthly streamflow calibration and validation value ranges.
Watershed - Model Calibration Value Ranges Validation Value Ranges
(Reference) Statistic Daily Monthly Daily Monthly
Eight watersheds in southwest Oklahoma - HSPF NSE 0.64 to 0.72 0.74 to 0.82 -1.37 to 0.87 -3.35 to 0.92
(Van Liew et al., 2003) PBIAS
Plymouth, North Carolina - DRAINMOD-W NSE 0.83 0.85 0.87 0.93
(Fernandez et al., 2005) PBIAS
Plymouth, North Carolina - DRAINMOD-DUFLOW NSE 0.68 0.76 0.81 0.92
(Fernandez et al., 2005) PBIAS
Bosque River, Texas - SWAT NSE -- 0.79 to 0.83 -- 0.62 to 0.87
(Santhi et al., 2001) PBIAS
Six watersheds in Texas - SWAT NSE -- 0.52 to 0.90 -- 0.55 to 0.81
(Narasimhan et al., 2005)[a] PBIAS
Bosque River, Texas - SWAT NSE 0.17 0.50 0.62 0.78
(Saleh and Du, 2004) PBIAS
Bosque River, Texas - HSPF NSE 0.72 0.91 0.70 0.86
(Saleh and Du, 2004) PBIAS
Plymouth, North Carolina - DRAINWAT NSE -- -- 0.71 to 0.84 0.85
(Amatya et al., 2004) PBIAS
Muscatatuck River, Indiana - SWAT NSE -0.23 to 0.28 0.59 to 0.80 -0.35 to 0.48 0.49 to 0.81
(Vazquez-Amábile and Engel, 2005) PBIAS
Warner Creek, Maryland - SWAT NSE -- 0.52 -- 0.63
(Chu and Shirmohammadi, 2004) PBIAS
Eight CRR-catchment test cases - XNJ (local simplex) NSE 0.89 to 0.94 -- 0.37 to 0.88 --
(Gan and Biftu, 1996) PBIAS -4.0 to 0.0 -- -16.0 to 34.0 --
Eight CRR-catchment test cases - NAM (SCE-UA) NSE 0.82 to 0.92 -- 0.41 to 0.88 --
(Gan and Biftu, 1996) PBIAS -16.7 to 15.5 -- -31.9 to 37.0 --
Eight CRR-catchment test cases - SMAR (SCE-UA) NSE 0.86 to 0.91 -- 0.86 to 0.91 --
(Gan and Biftu, 1996) PBIAS -20.9 to 12.0 -- -32.9 to 31.7 --
Eight CRR-catchment test cases - SMAR (local simplex) NSE 0.74 to 0.90 -- 0.00 to 0.85 --
(Gan and Biftu, 1996) PBIAS -6.0 to 0.0 -- -27.0 to 44.0 --
Eight CRR-catchment test cases - SMAR (local simplex) NSE 0.74 to 0.90 -- 0.00 to 0.85 --
(Gan and Biftu, 1996) PBIAS -6.0 to 0.0 -- -27.0 to 44.0 --
Eight CRR-catchment test cases - XNJ (SCE-UA) NSE 0.89 to 0.95 -- 0.45 to 0.88 --
(Gan and Biftu, 1996) PBIAS -13.9 to 24.4 -- -31.5 to 30.8 --
Eight CRR-catchment test cases - NAM (local simplex) NSE 0.81 to 0.92 -- 0.43 to 0.87 --
(Gan and Biftu, 1996) PBIAS -4.0 to 0.0 -- -16.0 to 44.0 --
Eight CRR-catchment test cases - SMA (SCE-UA) NSE 0.87 to 0.94 -- 0.31 to 0.89 --
(Gan and Biftu, 1996) PBIAS -7.6 to 1.3 -- -58.6 to 20.1 --
Eight CRR-catchment test cases - SMA (local simplex) NSE 0.85 to 0.93 -- 0.29 to 0.88 --
(Gan and Biftu, 1996) PBIAS -11.9 to 6.4 -- -54.8 to 47.2 --
Eight CRR-catchment test cases - SMA (local simplex) NSE 0.85 to 0.93 -- 0.29 to 0.88 --
(Gan and Biftu, 1996) PBIAS -11.9 to 6.4 -- -54.8 to 47.2 --
Eight CRR-catchment test cases - SMA (local simplex) NSE 0.85 to 0.93 -- 0.29 to 0.88 --
(Gan and Biftu, 1996) PBIAS -11.9 to 6.4 -- -54.8 to 47.2 --
Eight CRR-catchment test cases - SMA (local simplex) NSE 0.85 to 0.93 -- 0.29 to 0.88 --
(Gan and Biftu, 1996) PBIAS -11.9 to 6.4 -- -54.8 to 47.2 --
Iroquois River, Illinois and Indiana - HSPF NSE 0.81 0.88 0.70 0.82
(Singh et al., 2004) PBIAS
Iroquois River, Illinois and Indiana - SWAT NSE 0.79 0.89 0.73 0.83
(Singh et al., 2004) PBIAS
Ariel Creek, Pennsylvania - SWAT NSE 0.04 0.14 -- --
(Peterson and Hamlett, 1998)[b] PBIAS
Ali Efenti, Greece - SWAT NSE 0.62 0.81 -- --
(Varanou et al., 2002)[b] PBIAS
University of Kentucky Animal Research Center - SWAT NSE 0.19 0.89 -0.04 0.58
(Spruill et al., 2000)[b] PBIAS
Iroquois River, Illinois and Indiana - HSPF NSE 0.81 0.88 0.69 to 0.71 0.80 to 0.87
(Singh et al., 2005) PBIAS
(continued)

Vol. 50(3): 885−900 899

 Table A-9 (continued). Daily and monthly streamflow calibration and validation value ranges.
Watershed - Model Calibration Value Ranges Validation Value Ranges
(Reference) Statistic Daily Monthly Daily Monthly
Iroquois River, Illinois and Indiana - SWAT NSE 0.79 0.88 0.70 to 0.83 0.80 to 0.93
(Singh et al., 2005) PBIAS
Black Creek, Indiana - SWAT NSE -- 0.73 to 0.84 -- 0.63 to 0.73
(Bracmort et al., 2006) PBIAS
Five USDA-ARS experimental watersheds - SWAT NSE 0.30 to 0.76 0.48 to 0.90 -1.81 to 0.68 -2.50 to 0.89
(Van Liew et al., 2007) PBIAS 2.9 to -91.7 -- 2.7 to -155.6 --
[a] Weekly values.
[b] In Borah and Bera (2004).

900 TRANSACTIONS OF THE ASABE