4>EPA
United States
Environmental Protection
Agency
Impact of Best Management
Practices on Water Quality of
Two Small Watersheds in
Indiana: Role of Spatial Scale
-------
EPA/600/R-05/080
July 2004
Impact of Best Management Practices
on Water Quality of Two Small
Watersheds in Indiana:
Role of Spatial Scale
Prepared By
Mazdak Arabi
School of Civil Engineering
Purdue University
West Lafayette, Indiana 47907
Rao S. Govindaraju
School of Civil Engineering
Purdue University
West Lafayette, Indiana 47907
Contract No. 3C-R289-NAEX
Project Officer
Mohamed M. Hantush
National Risk Management Research Laboratory
Office of Research and Development
U.S. Environmental Protection Agency
Cincinnati, Ohio 45268
-------
Notice
The U.S. Environmental Protection Agency through its Office of Research and Development
funded the research described here. This report has been subjected to the Agency's peer and
administrative review and has been approved for publication as an EPA document. Mention
of trade names or commercial products does not constitute endorsement or recommendation
for use.
All research projects making conclusions or recommendations based on environmentally
related measurements and funded by the Environmental Protection Agency are required to
comply with the Agency Quality Assurance Program. This report has been subjected to
QA/QC review. The report presented a mathematical framework for modeling water quality
in eutrophic water bodies and did not involve collection and analysis of environmental
measurements.
-------
Foreword
The U.S. Environmental Protection Agency (EPA) is charged by Congress with protecting
the Nation's land, air, and water resources. Under a mandate of national environmental laws,
the Agency strives to formulate and implement actions leading to a compatible balance
between human activities and the ability of natural systems to support and nurture life. To
meet this mandate, EPA's research program is providing data and technical support for
solving environmental problems today and building a science knowledge base necessary to
manage our ecological resources wisely, understand how pollutants affect our health, and
prevent or reduce environmental risks in the future.
The National Risk Management Research Laboratory (NRMRL) is the Agency's center for
investigation of technological and management approaches for preventing and reducing risks
from pollution that threaten human health and the environment. The focus of the Laboratory's
research program is on methods and their cost-effectiveness for prevention and control of
pollution to air, land, water, and subsurface resources; protection of water quality in public
water systems; remediation of contaminated sites, sediments and ground water; prevention
and control of indoor air pollution; and restoration of ecosystems. NRMRL collaborates with
both public and private sector partners to foster technologies that reduce the cost of
compliance and to anticipate emerging problems. NRMRL's research provides solutions to
environmental problems by: developing and promoting technologies that protect and improve
the environment; advancing scientific and engineering information to support regulatory and
policy decisions; and providing the technical support and information transfer to ensure
implementation of environmental regulations and strategies at the national, state, and
community levels.
This publication has been produced as part of the Laboratory's strategic long-term research
plan. It is published and made available by EPA's Office of Research and Development to
assist the user community and to link researchers with their clients.
Sally Gutierrez, Director
National Risk Management Research Laboratory
11
-------
Abstract
Transport and fate of sediments and nutrients within watersheds have important implications
for water quality and water resources. Water quality issues often arise because sediments
serve as carriers for various pollutants such as nutrients, pathogens, and toxic substances. The
Clean Water Act provision (CWA) [Section 303(d)] requires all states to develop and
implement a Total Maximum Daily Load (TMDL) for their impaired water bodies, and water
bodies that are likely to join this list. Implementation of Best Management Practices (BMPs)
is a conventional approach for controlling nonpoint sources of sediments and nutrients.
However, implementation of BMPs has rarely been followed by a good long-term data
monitoring program in place to study how effective they have been in meeting their original
goals. Long-term data on flow and water quality within watersheds, before and after
placement of BMPs, is not generally available. Utility of mathematical models provides an
effective and powerful tool for evaluation of long-term performance of BMPs (especially
new ones that have had little or no history of use). In this study, a process-based modeling
framework is developed to evaluate the effectiveness of parallel terraces, field borders,
grassed waterways, and grade stabilization structures in reducing sediment and nutrient
yields in two small agricultural watersheds (<10 km2) in Indiana, with Soil and Water
Assessment Tool (SWAT) serving as the watershed model. Based on the functionality of
each BMP, appropriate model parameters are selected and altered to represent the effect of
the BMP on hydrologic and water quality processes. A sensitivity analysis is performed to
evaluate the sensitivity of model computations to selected parameters. Results indicated that
parallel terraces and field borders were effective at a field scale, while grassed waterways
and grade stabilization structures were the more effective BMPs at a watershed scale.
Distributed-parameter models partition the watershed into subunits (subwatersheds/hyrologic
response units/grids) during computations to represent heterogeneity within the watershed.
Homogeneous properties are assumed over each computational unit. Identification of the
stream network and partitioning of the study area into subunits may significantly affect
hydrologic and waters quality simulations of a distributed-parameter model. Because model
outputs are affected by geomorphologic resolution, the evaluation of performance of BMPs
based on model predictions will be influenced as well. Thus, examination of the efficacy of
in
-------
BMPs must be conducted in conjunction with studies performed at multiple spatial scales. In
this study, sediment and nutrient outputs from the calibrated SWAT model are compared at
various watershed discretization levels both with and without implementation of these BMPs.
Results indicated that evaluation of the impacts of these BMPs on sediment and nutrient
yields at the outlet of the two agricultural watersheds in Indiana was very sensitive to the
level of discretization that was applied for modeling. An optimal watershed discretization
level for representation of the BMPs was identified through numerical simulations. It would
appear that the average subwatershed area corresponding to approximately 4% of total
watershed area is needed to represent the influence of BMPs in a modeling effort.
It should be noted that the results of this study are location-dependent, and also depend on
the type of BMPs. However, the methodology can be utilized for similar studies in other
watersheds with different BMPs.
IV
-------
Table of Contents
SECTION 1.0 INTRODUCTION 1
1.1 BACKGROUND AND RATIONALE 1
1.2 OBJECTIVES 4
SECTION 2.0 WATERSHED DESCRIPTION AND MODEL SELECTION 7
2.1 INTRODUCTION 7
2.2 THE STUDY AREA AND AVAILABLE DATA 7
2.3 MODEL SELECTION 10
2.3.1 Background. 10
2.3.2 SWAT Mode I Description 15
2.3.3 Model Inputs 20
2.4 BASE FLOW SEPARATION MODEL 24
SECTION 3.0 ROLE OF WATERSHED DISCRETIZATION ON SWAT COMPUTATIONS 26
3.1 INTRODUCTION 26
3.2 OBJECTIVES 28
3.3 METHODOLOGY 29
3.4 EFFECTS OF WATERSHED DISCRETIZATION ON MODEL OUTPUTS 30
3.4.1 Stream/low 32
3.4.2 Sediment 33
3.4.3 Nutrients 36
3.5 IDENTIFICATION OF AN OPTIMAL WATERSHED DISCRETIZATION LEVEL 37
3.6 CONCLUSIONS 41
SECTION 4.0 MODEL CALIBRATION AND VALIDATION 43
4.1 INTRODUCTION 43
4.1 INDICATORS OF MODEL PERFORMANCE 44
4.2 SENSITIVITY ANALYSIS 45
4.2.1 Sensitivity Index 45
4.2.2 Additional analysis 48
4.2.3 Limitations 51
4.2.3 Conclusions 51
4.3 REPRESENTATION OF BEST MANAGEMENT PRACTICES (BMPs) WITH SWAT 52
4.4 MODEL CALIBRATION 56
4.6 DISCUSSION 60
SECTION 5.0 EVALUATION OF LONG-TERM IMPACT OF BEST MANAGEMENT PRACTICES
ON WATER QUALITY WITH A WATERSHED MODEL: ROLE OF SPATIAL
RESOLUTION 65
5.1 INTRODUCTION 65
5.2 METHODOLOGY 66
5.2.1 Watershed Discretization 66
5.3 IMPACT OF BEST MANAGEMENT PRACTICES ON WATER QUALITY 67
5.3.1 Effects of BMPs on Stream/low 67
5.3.2 Impact of BMPs on Sediment Yield. 70
5.3.3 Impact of BMPs on Nutrient Yields 73
5.4. FIELD SCALE VERSUS WATERSHED SCALE EVALUATION 76
5.5. CONCLUSIONS 77
SECTION 6.0 SOURCE IDENTIFICATION 80
6.1 INTRODUCTION 80
6.2 OBJECTIVE 81
v
-------
6.3 METHODOLOGY 82
6.3.1 Sediment and Nutrient Source Maps 82
6.3.2 Long-Term Performance ofBMPs 91
6.4 CONCLUSIONS 94
SECTION 7.0 CONCLUSIONS 96
7.1 MANAGEMENT IMPLICATIONS 97
7.2 MODELING IMPLICATIONS 98
7.3 CLOSING REMARKS 100
BIBLIOGRAPHY 101
VI
-------
List of Figures
Figure 2.1. (a) Land Use, (b) Digital Elevation Model (DEM) for the Dreisbach and Smith Fry Watersheds,
Allen County, Indiana
Figure 2.2. Type and Location of BMPs in the Dreisbach and Smith Fry Watersheds, Indiana 10
Figure 2.3. Phosphorus Processes Modeled in SWAT (USDA-ARS, 1999) 19
Figure 2.4. Nitrogen Processes Modeled in SWAT (USDA-ARS, 1999) 19
Figure 2.5. Monthly Precipitation Time Series from January 1970 to December 2002, Black Creek Watershed,
Indiana 21
Figure 2.6. Comparison of the Estimated Baseflow Using "ISEP" Model and the Model Adapted from Arnold
etal. (1999), Dreisbach Watershed, Indiana 25
Figure 2.7. Comparison of the Estimated Baseflow Using "ISEP" Model and the Model Adapted from Arnold
etal. (1999), Smith Fry Watershed, Indiana 25
Figure 3.1. Watershed Configurations Usedforthe Dreisbach Watershed 30
Figure 3.2. Watershed Configurations Used for the Smith Fry Watershed 31
Figure 3.3. Effects of Watershed Discretization on SWAT Streamflow Computations 33
Figure 3.4. Effect of Watershed Discretization on Weighted Average LS factor 34
Figure 3.5. Effects of Watershed Discretization on SWAT Sediment Computations 35
Figure 3.6. Effects of Watershed Discretization on Drainage Density (DD) and Average Slope of Channel
Network, Smith Fry Watershed 36
Figure 3.7. Effects of Watershed Discretization on SWAT Total P Computations 37
Figure 3.8. Effects of Watershed Discretization on SWAT Total N Computations 37
Figure 3.9. Effects of Watershed Discretization on Area Index (AT) 39
Figure 3.10. Correlation between the Erosion Index (El) and the Area Index (AI) 41
Figure 4.1. Sensitivity of SWAT Parameters Listed in Table 4. IDetermined Based on (a) Streamflow, (b)
Sediment, (c) Total P, and (d) Total N 49
Figure 4.2. Sensitivity of SWAT Parameters Listed in Table 4. IDetermined Based on Sediment Yield 50
Figure 4.3. Sensitivity of Streamflow Output of the SWAT Model at the Outlet of Dreisbach Watershed to
GWQMN Parameter 51
Figure 4.4. Schematic of Parallel Terraces 53
Figure 4.5. Schematic of Grade Stabilization Structures 55
Figure 4.6. Calibration Flowchart (Adapted from Santhietal., 2001a) 58
Figure 4.7. Measured and Simulated (a) Streamflow, (b) Surface Runoff, and (c) Plot 1:1 Streamflow,
Calibration and Validation Period, Dreisbach Watershed, Indiana 61
Figure 4.8. Measured and Simulated (a) Streamflow, (b) Surface Runoff, and (c) Plot 1:1 Streamflow,
Calibration and Validation Period, Smith Fry Watershed, Indiana 62
Figure 4.9. Measured and Simulated (a) Sediment, (b) Mineral P, and (c) Total P, (d) Total N, Calibration and
Validation Period, Dreisbach Watershed, Indiana 63
Figure 4.10. Measured and Simulated (a) Sediment, (b) Mineral P, and (c) Total P, (d) Total N, Calibration and
Validation Period, Smith Fry Watershed, Indiana 64
Vll
-------
Figure 5.1. Watershed Configurations Usedforthe Dreisbach Watershed 68
Figure 5.2. Watershed Configurations Used for the Smith Fry Watershed 69
Figure 5.3. Average Annual Sediment Yield at the Outlet of (a) Dreisbach Watershed, (b) Smith Fry
Watershed, (c) Percent Sediment Reduction. Scenario A: Simulations with No BMP; Scenario B:
Simulations with BMPs in Place 71
Figure 5.4. Average Annual Total P Yield at the Outlet of (a) Dreisbach Watershed, (b) Smith Fry Watershed,
(c) Percent Sediment Reduction. Scenario A: Simulations with No BMP; Scenario B: Simulations
with BMPs in Place 74
Figure 5.5. Average Annual Total N Yield at the Outlet of (a) Dreisbach Watershed, (b) Smith Fry Watershed,
(c) Percent Sediment Reduction. Scenario A: Simulations with No BMP; Scenario B: Simulations
with BMPs in Place 75
Figure 6.1. Simulated Average Annual Loads Generated at Upland Areas before Implementation of BMPs for
Dreisbach Watershed, 1971-2000: (a) Sediment, (b) Total P, and (c) Total N 83
Figure 6.2. Simulated Average Annual Loads Generated at Upland Areas after Implementation of BMPs for
Dreisbach Watershed, 1971 -2000: (a) Sediment, (b) Total P, and (c) Total N 84
Figure 6.3. Simulated Average Annual Loads Generated at Upland Areas before Implementation of BMPs for
Smith Fry Watershed, 1971-2000: (a) Sediment, (b) Total P, and (c) Total N 85
Figure 6.4. Simulated Average Annual Loads Generated at Upland Areas after Implementation of BMPs for
Smith Fry Watershed, 1971-2000: (a) Sediment, (b) Total P, and (c) Total N 86
Figure 6.5. Simulated Average Annual Loads from Channel Network before Implementation of BMPs for
Dreisbach Watershed, 1971-2000: a. Sediment, b. Total P, and c. Total N 87
Figure 6.6. Simulated Average Annual Loads from Channel Network after Implementation of BMPs for
Dreisbach Watershed, 1971-2000: a. Sediment, b. Total P, and c. Total N 88
Figure 6.7. Simulated Average Annual Loads from Channel Network before Implementation of BMPs for
Smith Fry Watershed, 1971-2000: 1971-2000: a. Sediment, b. Total P, and c. Total N 89
Figure 6.8. Simulated Average Annual Loads from Channel Network after Implementation of BMPs for Smith
Fry Watershed, 1971-2000: a. Sediment, b. Total P, and c. Total N 90
Vlll
-------
List of Tables
Table 2.1. Land Use in the Dreisbach and Smith Fry Watersheds, Indiana 9
Table 2.2. Major Soil Series in the Dreisbach and Smith Fry Watersheds, Indiana 9
Table 2.3. Corn-Soybean Rotation for the Dreisbach and Smith Fry Watersheds in 1975-1978 22
Table 2.4. Corn-Soybean-Winter Wheat Rotation for the Dreisbach and Smith Fry Watersheds in 1975-1978
23
Table 2.5. List of Available Input Data and Their Sources 24
Table 3.1. Properties of the Watershed Configurations Used for the Dreisbach Watershed, Indiana 29
Table 3.2. Properties of the Watershed Configurations Usedforthe SmithFry Watershed, Indiana 29
Table 4.1. List of SWAT Parameters Considered in Sensitivity Analysis 46
Table 4.2. Parameters Identified as Being Important from Sensitivity Analysis for Calibration 52
Table 4.3. Representation of Field Borders, Parallel Terraces, Grassed Waterways, and Grade Stabilization
Structures in SWAT 56
Table 4.4. Results of Calibration of SWAT for Streamflow, Sediment and Nutrient Simulations 59
Table 4.5. Results of Validation of SWAT for Streamflow, Sediment and Nutrient Simulations 60
Table 5.1. Properties of the Watershed Configurations Used for the Dreisbach Watershed, Indiana 67
Table 5.2. Properties of the Watershed Configurations Usedforthe SmithFry Watershed, Indiana 67
Table 5.3. Reduction of Sediment, Total P, and Total N loads Resulted from Implementation of Parallel
Terraces and Field Borders 76
Table 6.1. Impact of Field Borders on Sediment, Total P, and Total N Loads at a Field Scale 91
Table 6.2. Impact of Parallel Terraces on Sediment, Total P, and Total N Loads at a Field Scale 92
Table 6.3. Impact of Grassed Waterways on Sediment Loads (t/km) from Channel Segments 93
IX
-------
Acknowledgements
The U.S. Environmental Protection Agency through its Office of Research and Development
funded the research described here. This support is acknowledged. The report benefited from
the constructive review comments of Dr. M. M. Hantush, Dr. S.C. McCutcheon, Dr. L.
Kalin, and Dr. D.F. Lai.
x
-------
Section 1.0
Introduction
1.1 Background and Rationale
Sediment and nutrient yield from a watershed have important implications for water quality
and water resources. Water quality issues often arise because sediments serve as carriers for
various pollutants such as nutrients, pathogens, and toxic substances. Surface water quality is
important not only for protection offish and aquatic life, but it is often used as an indicator of
the environmental health of a watershed. Increased sediment load to a watershed can be
detrimental to an entire ecosystem. Land use changes over the years have had an enormous
effect on sediment levels in surface waters throughout the United States. Important sources
of sediments include erosion from agricultural fields, construction sites and reclaimed mining
areas. Estimates of sediment and nutrient yield are required for a wide spectrum of problems
dealing with dams and reservoirs, fate and transport of pollutants in surface waters, design of
stable channels, protection of fish and other aquatic life, watershed management and for
environmental impact statements.
Often, sediments in surface water bodies are contaminated with chemicals that sorb onto
fine-grained organic and inorganic soil particles. Sources of such contamination can result
from either existing point or non-point sources, historical spills, or discharges. When such
contamination exceeds critical levels, ecological and human health risks require appropriate
remedial actions. Such remedial measures take the form of isolating the contaminated
sediments, reducing their exposure to other parts of the ecosystem, complete removal of the
contaminated sediment, or some combination of the above. For all such measures, an
accurate understanding of the fate and transport of sediments/contaminants is crucial for
designing suitable remediation measures.
The Clean Water Act provision (CWA) [Section 303(d)] requires all states to develop and
implement a Total Maximum Daily Load (TMDL) for their impaired water bodies, and water
bodies that are likely to join this list. Implementation of the TMDL program is now
considered to be pivotal in securing the nation's water quality goals (NRC, 2001). A TMD is
-------
the maximum of point and nonpoint source loads that can enter a water body without
exceeding specified water quality standards. Over the past 30 years, some success has been
achieved by reducing pollution from point sources such as sewage treatment plants and
industrial discharges. However, controlling the pollution from nonpoint sources, which is
essential to the successful implementation of TMDL, still requires more study. According to
the most recent lists submitted to EPA, there are nearly 26,000 impaired water bodies in the
nation. Sediment/siltation and nutrients together are the major concern for approximately
11,000 of these water bodies, thus the most common impairments are sediment related.
Once a water body is listed as impaired and its type of impairment is classified as sediment
and nutrients, water quality modeling is required to make predictions that support the TMDL
process. Water quality modeling for TMDL development usually involves watershed
modeling and waterbody modeling. While the latter is necessary to determine pollutant
concentrations as a function of pollutant loads into the waterbody, the former is employed to
predict the pollutant loads into a waterbody as a function of watershed characteristics such as
slope of the watershed, land use, soil series, and management practices. Watershed models
are also utilized to evaluate the effectiveness of abatement strategies such as implementation
of Best Management Practices (BMPs). Watershed models have been classified into various
categories including empirical vs. physically-based, event-based vs. continuous, and lumped
vs. distributed-parameter models. Selection of a suitable model depends on several factors
such as capability to simulate design variables (runoff, groundwater, sediment yield, nutrient
yield, etc.), accuracy, available data, and temporal and spatial scales.
Spatial scale is an important consideration in watershed modeling. In large watersheds,
channel processes tend to become more important while in small watersheds hydrology is
usually dominated by overland flow. The validity of the predictions of a watershed model
depends on how well the spatially heterogeneous characteristics of the watershed are
represented by the model inputs. Lumped models consider a watershed as a single unit for
computations, and watershed parameters are averaged over this unit. The ability to represent
spatial variability inherent in watershed characteristics is the reason that distributed models
have been favored over lumped models. Distributed models partition a watershed into
-------
subunits (subwatersheds, HRUs, or grids) for simulation purposes, and homogeneous
properties are assumed for each subunit. Because model inputs are averaged over a subunit,
model simulations are greatly influenced by the size and number of subunits.
Currently, watershed delineation and extraction of stream networks are accomplished with
GIS databases of Digital Elevation Models (DEMs). The most common method for
extracting channel networks requires the a-priori specification of a critical source area that is
required for channel initiation. The nature of the channel network is very sensitive to this
critical source area, with drainage density decreasing exponentially with increasing critical
source area. Thus, the channel network could be viewed at multiple scales within the same
watershed. There are no established guidelines on how to select this critical source area.
Thus, for the same watershed and Digital Elevation Model (DEM), users may obtain
markedly different channel networks, and subsequently the watershed model results based on
the channel network could be affected as well. The challenge is to identify an optimal scale
of geomorphologic resolution such that further refinement in spatial scale does not contribute
to a significant improvement in predicting design parameters at the watershed outlet. Such an
optimal spatial scale, if identifiable, can be further used for identification of nonpoint sources
of sediments and nutrients.
Natural sources of sediment are primarily upland areas where erosion, including both sheet
and rill erosion, is dominated by overland flow, or in ephemeral gullies. Sheet erosion results
in removal of a fairly uniform layer of sediment from an area, while rill erosion is restricted
to concentrated channel flows. Large runoff events, like those that occur during a flood
event, can lead to mass sediment and nutrient removal. Anthropogenic activities may lead to
creation of important sources of sediments and nutrient, among which agricultural tillage has
the strongest influence. Highway construction, timber cutting, mining, urbanization, land
development for recreational use and animal grazing also contribute to varying degrees.
Large channels within a watershed not only serve as the source for movement of
contaminant-laden sediments, but may also act as a source because of erosion from
streambeds or banks. On the other hand, depending on the main channel geometry, sediment
particles could be deposited in the main channel. In the latter case, there is a significant
-------
difference between sediment and nutrient loads generated from upland areas and the ones
measured at the outlet of the watershed. Considering this phenomenon, implementation of
sediment and nutrient reduction plans will be highly affected by the control processes within
the watershed. For example, in a transport limited watershed, the transport capacity of the
watershed stream network is less than the sediment generated in upland areas (Keller et al.,
1997).
Identification of sources of sediment and nutrients within a watershed is necessary for
developing control measures. Most modeling strategies have focused on the forward problem
of predicting sediment and/or contaminant concentrations given the source locations and
strengths. While good geomorphologic data on stream networks and soil types are available
within watersheds from GIS databases, most monitoring programs are located at the
watershed outlet. Thus, detailed information within a watershed is rarely available at a
resolution that would enable proper identification of sediment or contaminant sources. This
problem is complicated because sediment and contaminants are carried along with the flow,
and water movement over a watershed tends to be fairly dynamic, behaving in a nonlinear
fashion. Previous studies do not provide a good modeling framework for identification of
sediment and nutrient sources within a watershed. Specifically, a methodology that could be
utilized to identify the control processes and management actions on sediment and nutrient
movement have not been developed.
1.2 Objectives
The overall objectives of this study are:
1. Evaluation of effectiveness of Best Management Practices (BMPS) in reduction
of sediment and nutrient yields: BMPs are conventional tools used widely as
sediment and nutrient reduction plans. While a few studies have addressed the
effectiveness of some BMPs (Mostaghimi et al., 1997; Williams et al., 2000;
Vache et al., 2002; Yuan et al., 2002a,b; Dybala, 2003; Santhi et al., 2003), the
importance of scale (i.e. watershed or farm scale) has been neglected in the
appraisal of the BMPs. In this research, the long term impacts of BMPs on water
-------
quality will be studied. The effectiveness of BMPs will be evaluated at the
watershed scale and farm (subwatershed) scale.
2. Investigation of role of watershed discretization on model simulations and
evaluation of effectiveness of BMPs in reduction of sediment and nutrient yields:
There are two sub-problems that result from multi-scale effects. First, an optimal
scale of geomorphologic resolution needs to be identified such that further
refinement in spatial scale does not contribute to a significant improvement in
simulating design quantities at the watershed outlet. This optimal geomorphologic
resolution, along with the associated drainage density, can then be utilized to
determine the appropriate critical source area. Second, the role of spatial scale on
evaluation of the efficacy of BMPs will be investigated. Because model results
are affected by the geomorphologic resolution, the predicted performance of
BMPs will be influenced by model parameters.
The remainder of this report is organized in six sections. Section 2.0 reviews the
characteristics of the study area and the watershed model that were used in this study. The
following criteria are utilized for selecting a watershed that will support the proposed
objectives: (i) The watershed should have been listed as and impaired waterbody by EPA or
the state authorities, (ii) BMPs must have been implemented for nonpoint source pollution
control and (iii) Daily water quality data (streamflow, and sediment and nutrient loads)
should have been collected at the outlet of the watershed for a reasonable period of time.
Various components of the selected watershed model are also discussed in this section. The
effect of watershed discretization on various hydrologic and water quality components of the
selected model is presented in Section 3.0. The possibility of identifying an optimal critical
source area for the model simulations and the conditions when such an identification is
relevant will be examined. A simple process-based index will be developed to help
identification of a proper watershed configuration prior to model calibration. The procedure
adopted for representation of Best Management Practices (BMPs) and model calibration is
described in Section 4.0. A discussion on the role of watershed discretization effects on
evaluation of the effectiveness of BMPs is provided in Section 5.0. Section 6.0 presents the
-------
utility of the watershed model in generation of sediment and nutrient source maps that can be
used in TMDL development. Overall conclusions of the study are summarized in Section 7.0.
-------
Section 2.0
Watershed Description and Model Selection
2.1 Introduction
To achieve the goals of this project, a suitable watershed model is selected, calibrated, and
validated for a study area where adequate water quality data are available. The availability of
a rather unique dataset for a particular watershed, and how this will be utilized to meet the
project goals will be described briefly.
2.2 The Study Area and Available Data
For the objectives of this research to be successfully fulfilled a watershed must be selected
where BMPs have been implemented and adequate hydrologic and water quality data
including rainfall, streamflow, and sediment and nutrient yields are available. Various
watersheds in the United States have been studied for evaluation of the effects of BMPs on
water quality (Batchelor et al., 1994; Park et al., 1994; Griffin, 1995; Edwards et al., 1996;
Saleh et al., 2000; Santhi et al., 2001; Saleh and Du, 2002; Vache et al. 2002; Santhi et al.,
2003). However none of these studies had the data needed for a thorough evaluation of the
influence of BMPs. After an exhausting search, the Black Creek watershed, northeast
Indiana, was identified as perhaps one of the very few watersheds with both daily measured
water quality data and with detailed information on various implemented BMPs. This
watershed is also preferred because daily water quality data were measured at two outlets
within the watershed (Figure 2.1). This allows for further validation of the conclusions of this
study.
A study on the Black Creek watershed, funded by EPA, was conducted in 1970s and early
1980s to examine the short-term effects of soil and water conservation techniques on
improving water quality by reducing sediment and nutrient loads leaving the watershed. This
watershed, located in Allen County, northeast Indiana (see Figure 2.1) is an approximately 50
km2 (12,000 acre) watershed in the Maumee River basin. In this previous study, detailed
water quality monitoring was carried out during the duration of the project. Nineteen major
-------
(a)
0
State of Indiana
Land Use
corn
soyabean
small grain
pastu re/g rass
woodland
UrbarAResidential
DEM (m)
211-220
220-230
230-240
240 - 250
250-260
260-270
B
8 Kilometers
Figure 2.1. (a) Land Use, (b) Digital Elevation Model (DEM) for the Dreisbach and Smith Fry Watersheds, Allen County, Indiana.
-------
monitoring stations were established within the watershed. However, data collected from
automated samplers located at Smith Fry and Dreisbach outlets were the most complete and
were used for most of the analysis reported in the project. The areas of the Smith Fry and
Dreisbach watersheds shown in Figure 2.1 are 7.3 km2 and 6.23 km2, respectively. Daily
precipitation, streamflow, and sediment and nutrient loads were recorded at the outlet of
these two watersheds. Land use in the Dreisbach watershed (Figure 2.la) is mostly pasture in
the upper portion, while cropland is wide spread in remainder of the watershed. Land use in
the Smith Fry watershed is mostly croplands (see Figure 2.la). Table 2.1 presents land use
distributions for the two watersheds.
The Digital Elevation Model (DEM) for the study area is shown in Figure 2.1(b). The
dominant hydrological soil group of soil series in both watersheds is type C. Major soil series
in the two watersheds are listed in Table 2.2.
Table 2.1 Land Use in the Dreisbach and Smith Fry Watersheds, Indiana.
Land Use
Pasture (PAST)
Corn (CORN)
Soybean (SOYB)
Winter Wheat (WWHT)
Forest (FRSD)
Residential- Low Density (URLD)
% Dreisbach Area
37.55
23.38
7.22
16.97
5.83
9.06
% Smith Fry Area
8.72
33.59
31.84
14.28
8.93
2.64
Table 2.2. Major Soil Series in the Dreisbach and Smith Fry Watersheds, Indiana.
Soil
WHITAKER
RENSSELAER
MORLEY
PEWAMO
NAPPANEE
HOYTVILLE
BLOUNT
% Dreisbach Area
3.77
9.1
40.88
13.31
3.68
5.52
14.87
% Smith Fry Area
11.25
20.64
8.35
5.77
4.23
10.7
22.21
Hydrologic group
C
B
C
C
D
C
C
-------
There were 26 best management practices (BMPs) installed in the Dreisbach watershed in
1974 while this number was 6 for the Smith Fry watershed. The BMPs were installed in the
Smith Fry watershed in 1975. The types and locations of the BMPs in the Dreisbach and
Smith Fry watersheds are shown in Figure 2.2.
2.3 Model Selection
2.3.1 Background
Watershed models are utilized to better understand the role of hydrological processes that
govern surface and subsurface water movement. Moreover, they provide assessment tools for
decision making in regard to water quality issues. Watershed models have been classified
into various categories including empirical vs. physically-based, event-based vs. continuous,
and lumped vs. distributed-parameter models. Selection of a suitable model depends on
several factors such as capability to simulate design variables (runoff, groundwater, sediment
yield, nutrient yield, etc.), accuracy, available data, and temporal and spatial scales.
t Field Border
$ Parallel Terrace
N Grade Stabilization Structure
N Grassed Waterway
8 Kilometers
Figure 2.2. Type and Location of BMPs in the Dreisbach and Smith Fry Watersheds, Indiana.
10
-------
Empirical models are developed based on statistical relationships between design parameters
and watershed characteristics. These relationships are obtained from regression analysis
using observed data. Application of these models will likely be limited to the same statistical
conditions over which the observed data were acquired. For example, the well-known
Universal Soil Loss Equation (USLE) (Wischmeier and Smith, 1978) was developed based
on statistical analysis of many years of rainfall, runoff, and soil loss data from many small
plots around the United States, and is suitable for estimation of average annual soil loss from
a field based on steepness, soil series, land use, and management practice. Application of the
USLE for daily and/or monthly estimation of soil loss may not yield realistic results. These
limitations do not hold for physically-based models as they are grounded in physical
principles of conservation of mass, energy, and momentum. These models are preferred
because they provide a better understanding of the processes in the watershed. Many models
utilize both empirical and physically-based relationships to represent hydrologic and water
quality processes within a watershed, and may be labeled as process-based models.
Lumped models consider a watershed as a single unit for computations, and watershed
parameters are averaged over this unit, while distributed-parameter models partition a
watershed into subunits (subwatersheds, HRUs, or grids) for simulation purposes, and
homogeneous properties are assumed for each subunit. As a result, the number of input
parameters increases significantly. However, the spatial variability of watershed parameters
such as land use, soil series, and management actions are more easily represented in
distributed-parameter models.
In addition to spatial scale, watershed models utilize different temporal scales for
computations. Event-based models usually require small time steps, at times in the order of
seconds. These models are suitable for analyzing influence of design storms. Larger time
steps, in the order of days, are usually sufficient for continuous models that are appropriate
for long term assessment of hydrological and land use change and watershed management
practices.
Water quality models estimate sediment and nutrients loads through prediction algorithms.
These are primarily empirical in nature and use versions of the Universal Soil Loss Equation
11
-------
(USLE) (Wischmeier and Smith, 1978) for sediment loads as in AGNPS (Young et al., 1987,
1989) and SWRRBQ (Renard et al., 1997). Particle detachment and wash equations are
utilized in HSPF (Bicknell et al., 1993), ANSWERS (Beasley et al., 1980), and other models.
AGNPS and ANSWERS evaluate sediment transport associated with individual events, while
models like HSPF and SWAT (Neitsch et al., 2001a,b) utilize hourly or daily time steps and
are better suited for long-term simulations. Some of these models can be used to estimate
sediment erosion and nutrient loads from multiple source categories and can track the fate
and transport of sediments and nutrients. Therefore, they are well-suited to providing useful
information on sediment and nutrient yields from different regions of a watershed (Reid and
Dunne, 1996). While these models can delineate sediment and nutrients sources at the point-
scale in principle, the problem would have to be posed in an inverse sense, and would entail
very substantial amounts of data requirements and computer effort.
Borah (2002) reviewed eleven continuous-simulation and single-event watershed scale
models including the ones mentioned above. The study provides a better understanding of the
mathematical bases of the models. Among all, the Soil and Water Assessment Tool (SWAT)
model is the only continuous/process-based/distributed-parameter model that contains both
sediment and nutrient components and is capable of representing BMPs at a watershed scale.
Implementation of the Best Management Practices (BMPs) is a conventional approach for
nonpoint source pollution control. Various watershed and field scale models have been used
to simulate the effectiveness of BMPs (Bachelor et al., 1994; Park et al., 1994; Edwards et al.
1996). The WEPP model (Flanagan and Livingston, 1995) has the most mechanistic
sediment transport component and can simulate various BMPs including agricultural
practices (e.g. tillage, contouring, irrigation, drainage, crop rotation, etc.), ponds, terraces,
culverts, filter fences and check dams (Kalin and Hantush, 2003). However, the application
of the model is limited to field scale studies or very small watersheds (<3 km2). Most of
models with good representation of BMPs, such as WEPP, are more applicable to field scale
studies.
Kalin and Hantush (2003) reviewed key features and capabilities of widely cited watershed
scale hydrologic and water quality models with emphasis on the ability of the models in
12
-------
representation of Best Management Practices (BMPs) and TMDL development. The review
indicated that the SWAT and AGNPS models offer the most management alternatives for
modeling of agricultural watersheds. In this study, the Soil and Water Assessment Tool
(SWAT) model is selected as the watershed model.
Saleh and Du (2002) compared the performances of the SWAT and HSPF models, both
integrated into BASINS framework, in predictions of streamflow, sediment yield, and
nutrient loads. The authors suggested that SWAT is more user-friendly and had a better
prediction of nutrient loads while HSPF streamflow and sediment predictions were closer to
the measured data. The SWAT model and HSPF model streamflow predictions were also
compared by Van Liew et al. (2003). They found that although the modeling errors were
smaller for HSPF in the calibration period, SWAT exhibited more robustness during the
calibration and validation periods. The robustness refers to more acceptable error statistics
during validation period. They also concluded that SWAT might be more suitable for long-
term assessment of the effects of climate variability on surface water resources.
The SWAT model has been widely used for streamflow, sediment yield, and nutrient load
predictions. The SWAT model development, operation, limitations, and assumptions were
discussed by Arnold et al. (1998). Srinivasan et al. (1998) reviewed the applications of the
SWAT model in streamflow prediction, sediment and nutrients transport, and effects of
management practices on water quality. Arnold and Allen (1996) evaluated the performance
of different hydrologic components of the SWAT model for three watersheds in Illinois (100-
250 km2). Comparing the model outputs to measured data, the calibrated model reasonably
simulated runoff, groundwater, and other components of hydrologic cycle for the study
watersheds. Most simulated average monthly outputs were within 5% of the historical data
and nearly all of them were within 25%. R2 (correlation coefficient) statistic was used to
compare the correlation between the observed and simulated average monthly variables. Also,
the interaction among various components of hydrologic budgets was recognized to be
realistic. SWAT was utilized in a study by Arnold et al. (2000) to compare the performance
of two baseflow and groundwater recharge models. The first model was the water balance
components of the SWAT model. A combination of a digital hydrograph separation tool and
13
-------
a modified hydrograph recession curve displacement technique composed the second model.
The results of the two models were in general agreement in the Upper Mississippi river basin.
A detailed procedure for calibration of SWAT was laid out by Santhi et al. (200la). Jha et al.
(2003) found curve Number (CN) as the most sensitive parameter in streamflow prediction.
A series of studies have been carried out with SWAT to model sediment and nutrients
transport within Upper North Bosque River watershed (4277 km2), TX (Saleh et al., 2000;
Santhi et al., 2001a; Santhi et al., 2001b; Saleh and Du, 2002; Santhi et al., 2003). Manure
application to pasture and cropland is the main nonpoint source pollution concern in this
large watershed. Dairy management practices have been utilized for phosphorus load control.
In conclusion, SWAT performance has been extensively validated for streamflow, and
sediment and nutrients yield predictions for different regions of United States.
SWAT has been applied to evaluate the effects of a number of BMPs such as waterways,
filter strips, and field boarders on streamflow and sediment and nutrients annual loads from
U.S. Corn Belt (Vache et al. 2002). The study indicated that implementation of BMPs
resulted in 30 to 60% reduction in sediment and nutrients loads. The SWAT model was
utilized by Kirsch et al. (2002) to appraise the effectiveness of BMPs on reduction of
sediment and phosphorus load over Rock River Basin (9708 km2), WI. The BMP practices
analyzed included modifications in tillage operations, and adoption of recommended nutrient
application rates. They concluded that implementation of modified tillage practices would
result in almost 20% sediment reduction. Additional in-stream modeling, and field scale
water quality screening was recommended.
In conclusion, SWAT performance has been extensively validated for streamflow, and
sediment and nutrients yield predictions for different regions of the United States. The model
has also been successfully utilized for representation of various management scenarios. In
this study, the Soil and Water Assessment Tool (SWAT) model is selected to simulate fate
and transport of sediments and nutrients in the Dreisbach and Smith Fry watersheds.
14
-------
2.3.2 SWAT Model Description
Soil and Water Assessment Tool (SWAT) (Neitsch et al., 2001a,b) has been widely used for
watershed scale studies dealing with water quantity and quality. SWAT is a process-based
distributed-parameter simulation model, operating on a daily time step. SWAT partitions the
watershed into subwatersheds, each of which is treated as an individual unit. The model has
also been integrated into USEPA's modeling framework, Better Assessment Science
Integrating Point and Nonpoint Sources (BASINS). This framework provides users with a
watershed delineation tool that enables users to automatically or manually delineate the
watershed based on a Digital Elevation Model (DEM). A stream definition value is required
by the delineation tool for watershed delineation. Selecting several different values for
stream definition and by comparing the predicted sediment and nutrient yields, the role of
subwatershed division on predicted responses of water and contaminant fluxes from the
watershed can be examined to address the issue of spatial resolution required for modeling
purposes. The SWAT model needs to be calibrated and validated for the study area to ensure
that model parameters are representative for the study region.
SWAT is a process-based based model, operating on a daily time step. The model was
originally developed to quantify the impact of land management practices in large, complex
watersheds with varying soils, land use, and management conditions over a long period of
time. SWAT uses readily available inputs and has the capability of routing runoff and
chemicals through streams and reservoirs, and allows for addition of flows and inclusion of
measured data from point sources. The model is capable of simulating long periods for
comparing the effect of management changes. Moreover, SWAT has the capability to
evaluate the relative effects of different management scenarios on water quality, sediment,
and agricultural chemical yield in large, ungaged basins. Major components of the model
include weather, surface runoff, return flow, percolation, evapotranspiration (ET),
transmission losses, pond and reservoir storage, crop growth and irrigation, groundwater
flow, reach routing, nutrient and pesticide loads, and water transfer.
For simulation purposes, SWAT partitions the watershed into subunits including subbasins,
reach/main channel segments, impoundments on main channel network, and point sources to
15
-------
set up a watershed. Subbasins are divided into hydrologic response units (HRUs) that are
portions of subbasins with unique land use/management/soil attributes.
SWAT uses a modification of the SCS curve number method (USDA Soil Conservation
Service, 1972) or Green and Ampt infiltration method (Green and Ampt, 1911) to compute
surface runoff volume for each HRU. The SCS curve number equation is:
( d°y a>
where <2OTr/is the accumulated runoff or rainfall excess (mm water), Rday is the rainfall depth
for the day (mm water), Ia is initial abstraction which includes surface storage, interception
and infiltration prior to runoff (mm water), and S is the retention parameter (mm water).
25.4(1^-10) (2.2)
^ CN
where CN is the SCS runoff curve number. The initial abstraction, Ia, is commonly
approximated as 0.2^:
(Rda-Q2S)2
(2.3)
Runoff will only occur when Rday > Ia.
Peak runoff rate is estimated using a modification of the rational method. Daily or sub-daily
rainfall data is used for calculations. The rational formula is:
Ci.Area ,_ „.
Vpeak = (2.4)
16
-------
where qpeak is the peak runoff rate (m3/s), C is the runoff coefficient, / is the rainfall intensity
(mm/hr), Area is the HRU area (km2), and 3.6 is a unit conversion factor. Flow is routed
through the channel using a variable storage coefficient method developed by Williams
(1969) or the Muskingum routing method.
Erosion and sediment yield are estimated for each HRU with the Modified Universal Soil
Loss Equation (MUSLE) (Williams, 1975):
sed = n.^(Qsurf.qpeak.areahrJ0-56XUSLE.CUSLE.PUSLE.LSUSLE.CFRG (2.5)
where sed is the sediment yield on a given day (metric tons), Qsurf is the surface runoff
volume (mm water), qpeak is the peak runoff rate (m3/s), area^u is the area of the HRU (ha),
KUSLE is the USLE soil erodibility factor, CUSLE is the USLE cover and management factor,
PUSLE is the USLE support practice factor, LSusis is the USLE topographic factor, and CFRG
is the coarse fragment factor.
Sediment deposition and degradation are the two dominant channel processes that affect
sediment yield at the outlet of the watershed. Whether channel deposition or channel
degradation occurs depends on the sediment loads from upland areas and transport capacity
of the channel network. If sediment load in a channel segment is larger than its sediment
transport capacity, channel deposition will be the dominant process. Otherwise, channel
degradation (i.e. channel erosion) occurs over the channel segment. SWAT estimates the
transport capacity of a channel segment as a function of the peak channel velocity:
Tch = a.vb (2.6)
where Tch (ton/m3) is the maximum concentration of sediment that can be transported by
streamflow (i.e. transport capacity), a and b are user defined coefficients, and v (m/s) is the
peak channel velocity. The peak velocity in a reach segment is calculated:
--R 2/3v 1/2 (7.7}
n
17
-------
where a is the peak rate adjustment factor with a default value of unity, n is Manning's
coefficient, Rch is the hydraulic radius (m), and Sch is the channel invert slope (m/m).
Channel degradation (Seddeg) and deposition (Seddep) in tons are computed as:
sedt > Tch : seddep = (sed1 - Tch) x Vch & sed^ = 0 (2.8)
sedl + Sedteg ~ Seddep) X -TT (2.10)
Vch
In (5), Vout is the volume of water leaving the channel segment (m3) at each time step.
Movement and transformation of several forms of nitrogen and phosphorus over the
watershed are accounted within the SWAT model. Nutrients are introduced into the main
channel and transported downstream through surface runoff and lateral subsurface flow.
Major phosphorous sources in mineral soil include organic phosphorus available in humus,
mineral phosphorus that is not soluble, and plant available phosphorus. Phosphorus may be
added to the soil in the form of fertilizer, manure, and residue application. Surface runoff is
the major carrier of phosphorous out of most catchments (Sharpley and Syers, 1979). The
transformation of phosphorus in the soil is controlled by the phosphorus cycle (see Figure
2.3). Unlike phosphorus that has low solubility, nitrogen is highly mobile. Major nitrogen
sources in mineral soil include organic nitrogen available in humus, mineral nitrogen in soil
colloids, and mineral nitrogen in solution. Nitrogen may be added to the soil in the form of
fertilizer, manure, or residue application. Plant uptake, denitrification, volatilization,
leaching, and soil erosion are the major mechanisms of nitrogen removal from a field. In the
18
-------
Mineral P
Pknt "Jji
A
Organic P
Humid fiiil>starv:*5
fell Ir.vr
M-*\ Ac I Ire )<—*»,, Solution
PLiri i milut!
' Aflici:'>«-*<'aiibli;sK—(" fittli ')
Figure 2.3. Phosphorus Processes Modeled in SWAT (USDA-ARS, 1999).
Mineral N
uryeiD MfnLlirj!
Phnl Ujitifce
.IT
\ NO,
i-¥
-( NHi
Organic N
4 Fresh
Figure 2.4. Nitrogen Processes Modeled in SWAT (USDA-ARS, 1999).
soil, transformation of nitrogen from one form to another is governed by the nitrogen cycle
(see Figure 2.4).
SWAT simulates pesticide movement into the stream network via surface runoff (in solution
and absorbed to sediment transported by runoff), and into the soil profile of the underlying
aquifer by percolation (in solution). The equations used to model the movement of the
pesticide were adopted from GLEAMS (Leonard et al., 1987).
The movement of water, sediment, and nutrients through the channel network of the
watershed to the outlet is simulated by routing in main channel and reservoirs.
Very detailed management input data are required for a SWAT simulation including general
management practices such as tillage, harvest and killing, pesticide application, fertilizer
19
-------
application, and irrigation management. Input data needed to run the SWAT model include
soil, land use, weather, rainfall, management conditions, stream network, and watershed
configuration. The summary output file (output.std), the HRU output file (.sbs), the subbasin
output file (.bsb), and the main channel or reach output file (.rch) are the primary output files
generated in every SWAT simulation. Users can refer to Soil and Water Assessment Tool
User's Manual and theoretical documentation version 2000 (Neitsch et al., 2001a,b),
published by the Agricultural Research Service and The Texas Agricultural Experiment
Station, Temple, Texas for a detailed description of SWAT model. Various documents are
also available at the SWAT website: www.brc.tamus.edu/swat/index.html.
SWAT has been integrated into USEPA's modeling framework, Better Assessment Science
Integrating Point and Non-point Sources (BASINS). This framework provides users with a
watershed delineation tool that allows automatic or manual watershed delineation based on
Digital Elevation Model (DEM) data. BASINS is available for free download at:
www.epa.gov/waterscience/basins.
2.3.3 Model Inputs
The SWAT model requires inputs on weather, topography, soil, land use, management,
stream network, ponds, and reservoirs. The BASINS framework is used to develop the input
parameters.
Climate Inputs
Daily precipitation from January 1974 to June 1977 was obtained from the monitoring station
located at the outlet of the Dreisbach and Smith Fry watersheds. The elevation of the outlet
of the Dreisbach watershed is 230 m above see level while the elevation of the outlet of the
Smith Fry watershed is 222 m. The recorded daily precipitation for the two watersheds was
published in the Black Creek project data report (Lake and Morrison, 1978). This
information was converted to a tabular form and is available at: http://pasture.ecn.purdue.edu
/ABE/blackcreek/original data/Weather.
20
-------
300
250 -
200 -
.& 150
22
•51 100 -
a
o
50 -
o
r-p
cS
o
^t- ^D
00 00
OO
OO
O
(N
O\
00
O\
cS
cS
O (N
9 9
a a
cS cS
Time (month)
Figure 2.5. Monthly Precipitation Time Series from January 1970 to December 2002, Black Creek
Watershed, Indiana.
Daily precipitation was obtained from the Fort Wayne disposal plant station (Station ID:
123037) monitored by Purdue University Applied Meteorology Group for 1902 to 1973 and
1978 to 2002. This station is located at 4F06'N / 85°07'W (LAT/LONG) which is
approximately 32 kilometers southwest of the outlet of the Black Creek watershed. The
elevation of Fort Wayne disposal plant station is 240 m above the sea level. The daily
precipitation and temperature data for this station from 1900 to 2003 are available at:
http://shadow.agry.purdue.edu/sc.index.html. Information on daily temperatures was
obtained from the Fort Wayne station. Figure 2.5 depicts monthly precipitation time series
for the 1970-2002 period at the outlet of the Black Creek watershed.
Elevation Map
A 30-m resolution, UTM NAD83 projected Digital Elevation Model (DEM) was obtained
from the National Elevation Dataset dated 2001. The DEM for the whole state of Indiana is
available at: http://pasture.ecn.purdue.edu/ABE/Indina.
21
-------
Soils
Soil data were obtained from the Soil Survey Geographic Database. Detailed digital
representation of County Soil Survey maps was published by the Natural Resources
Conservation Service (USDA-NRCS). The Soil Survey Geographical Database (SSURGO)
soil map dated 2002 for the whole state of Indiana is available at: http://pasture.ecn.purdue
. edu/ABE/Indina.
Land Use
Land use map was digitized into Arc View shapefile format from the Black Creek project
historical files. The land use maps for 1975, 1976, 1977, and 1978 were extracted from aerial
photos dated 1975, 1976, 1977, and 1978, respectively. This information is available at:
http:'//pasture, ecn.purdue. edu/ABE/blackcreek.
The information on the best management practices (BMPs) such as type, location, and date of
installment were obtained from the Black Creek project technical report (Lake and Morrison,
1977a,b). A BMP shapefile was built to locate the BMPs in the watersheds. Individual
subbasins were determined based on the location of the BMPs. The main goal was to locate
each BMP in a different subbasin, although in some cases there is more than one BMP in a
subbasin. The historical crop rotation in the Black Creek watershed is presented in Table 2.3
and Table 2.4.
Table 2.3. Corn-Soybean Rotation for the Dreisbach and Smith Fry
Watersheds in 1975-1978.
year
1
1
1
1
1
2
2
2
2
operation
tillage
fertilizer
plant/begin. Growing season
pesticide application
harvest and kill
plant/begin. Growing season
pesticide application
harvest and kill
tillage
crop
CORN
CORN
CORN
SOYB
SOYB
SOYB
date
month
May
May
May
May
October
May
June
October
October
day
3
6
10
10
15
20
15
1
10
22
-------
Table 2.4. Corn-Soybean-Winter Wheat Rotation for the Dreisbach and
Smith Frv Watersheds in 1975-1978.
year
1
1
2
2
2
3
3
3
3
3
4
4
4
4
operation
tillage
plant/begin. Growing season
fertilizer
harvest and kill
tillage
tillage
fertilizer
plant/begin. Growing season
pesticide application
harvest and kill
plant/begin. Growing season
pesticide application
harvest and kill
tillage
crop
WWHT
WWHT
WWHT
CORN
CORN
CORN
SOYB
SOYB
SOYB
date
month
October
October
April
July
July
May
May
May
May
October
May
June
October
October
day
12
15
5
15
30
O
6
10
10
15
20
15
1
10
Flow, Sediment, and Nutrient Data
Streamflow discharge, sediment, and nutrient yields were measured at the two monitoring
stations at the outlet of the Dreisbach and Smith Fry watersheds. Complete discussion of all
monitoring sites, laboratory methods, and supporting study designs were contained in the
Black Creek project technical report (Lake and Morrison, 1977a,b) and the Black Creek
project final report (Lake et al., 1981). The measured daily streamflow discharge, sediment,
and nutrient yields were reported in the Black Creek project data report (Lake and Morrison,
1978) and the Black Creek project final report (Lake et al., 1981). The available set of
measured data include daily streamflow discharge from January 1975 to December 1978,
sediment yield from April 1973 to June 1977, and nutrient yields from April 1973 to June
1977. All the above information was converted to tabular form and is available at:
http://pasture.ecn.purdue.edu/ABE/blackcreek/original data. The available data compiled
for use in SWAT along with their sources are summarized in Table 2.5.
23
-------
Table 2.5. List of Available Input Data and Their Sources.
Data Type
Digital Elevation
Model (DEM)
Soils
Land Use
Land Use
Weather
Weather
Crop Management
Streamflow
Water Quality
Source
National Elevation
Data
Soil Survey
Geographic Database
USDA-NRCS
Black Creek Project
Black Creek Project1
Purdue Applied
Meteorology Group
Engel&Lim(2001)
Black Creek Project2
Black Creek Project1
Date
2001
2002
2003
1975
1974-1977
1902-2002
1975
1975-1978
1974-May
1977
Description
30-m resolution, U.S.
Geological Survey
Digital representation of
County Soil Survey maps
Digitized into GIS from
aerial photos
Digitized into GIS from
aerial photos
Daily precipitation
graphs
Minimum and maximum
daily temperature and
Management scenarios
for crops
Daily Streamflow
Daily sediment, mineral
P, total P, and total N
Lake and Morrison (1978), Morrison and Lake (1981).
2.4 Base Flow Separation Model
An automated hydrograph separation model "ISEP" was used to determine the relative
contribution of surface runoff and ground water to total Streamflow. This model was
developed in the Department of Agricultural and Biological Engineering, Purdue University.
The "ISEP" program is available at: http://danpatch.ecn.purdue.edu/~sprawl/iSep. To further
validate the separation model, the determined hydrographs were confirmed with another flow
separation model (Arnold and Allen, 1999). The results of the two models were consistent in
their determinations of contributions to surface runoff and baseflow parts of the total stream
flow. Figures 2.6 and 2.7 show the baseflow volume (mm) estimated by the two flow
separation models for the Dreisbach and Smith Fry watersheds, respectively.
24
-------
0.08
0.06
0.04
0.02
o Arnold et al. (1999) model
e ISEP model
Time (Month)
Figure 2.6. Comparison of the Estimated Baseflow Using "ISEP" Model and the Model Adapted from
Arnold et al. (1999), Dreisbach Watershed, Indiana.
0.16
0.12
0.08
m
0.04
Arnold et al. (1999) model
ISEP model
Time (Month)
Figure 2.7. Comparison of the Estimated Baseflow Using "ISEP" Model and the Model Adapted from
Arnold et al. (1999), Smith Fry Watershed, Indiana.
25
-------
Section 3.0
Role of Watershed Discretization on SWAT Computations
3.1 Introduction
The ability of a nonpoint source pollution model to simulate design parameters including
streamflow, sediment yield, and nutrient loads depends on how well the watershed
characteristics are represented by the model inputs. The ability to represent spatial variability
inherent in watershed characteristics is the reason that distributed hydrological and water
quality models have been favored over the lump models. Distributed models partition a
watershed into subunits (subwatersheds, hydrologic response units, or grids) for simulation
purposes, and homogeneous properties are assumed for each subunit. As the model inputs are
averaged over a subunit, model simulations are greatly influenced by the size and number of
the computational units.
The question of spatial resolution can be posed in two ways. First, the spatial resolutions and
attributes of input data such as soil series, land use, and Digital Elevation Model (DEM)
might significantly influence the model computations. Utilizing finer resolutions of these
input data, if available, will result in more accurate simulations although it might be
computationally more demanding. Secondly, spatial resolution in the form of watershed
discretization is an important consideration in watershed modeling. Currently, watershed
delineation and extraction of stream networks are accomplished with GIS databases of
Digital Elevation Models (OEMs). The most common method for extracting channel
networks requires the a-priori specification of a critical source area (CSA) that is required for
channel initiation. For the same watershed and Digital Elevation Model (DEM), users may
obtain markedly different channel networks, and watershed configurations (i.e. the number
and size of subunits). The input parameters are averaged over the computational units.
Subsequently the watershed model computations based on the channel network and
watershed configuration could be affected as well. This study is an attempt to assess the latter
problem - that is given specific soil series, land use, management scenarios, and Digital
Elevation Model (DEM) how are model outputs affected by watershed discretization?
26
-------
Effect of watershed discretization on model outputs has been the motivation of several
studies in the past. Norris and Haan (1993) demonstrated that increasing the number of
subwatersheds beyond a certain threshold level did not improve runoff generation
significantly. Other studies established a threshold value for critical source area for channel
initiation (Goodrich, 1992; Zhang and Montgomery, 1994). Miller et al. (1999) concluded
that the hydrologic response of small watersheds was more sensitive to changes in
topography within the subwatersheds. Kalin et al. (2003) studied the effect of catchment
scale on runoff generation and sediment yield over small watersheds. They concluded that a
critical source area could be identified for particular combinations of rainfall events and
watershed characteristics.
Bingner et al. (1997) utilized the SWAT model to evaluate the impact of the number and size
of subwatersheds on runoff generation and fine sediment loads. They found simulated runoff
to be rather insensitive to the subwatershed scale. They could identify a critical source area
for fine sediment yield. In contrast, Mamillapalli (1998) found that the SWAT model runoff
simulations tended to be more accurate with finer discretization of the watershed into
subwatersheds or by increasing the number of hydrologic response units (HRUs) in the
watershed. It was concluded that the model accuracy does not improve beyond a certain level
of discretization. Further, land use and soil distributions were found to have a more
significant effect on streamflow simulation than topography. The simultaneous impacts of
watershed characteristics, channel parameters, and spatial resolution on sediment generation
were studied by FitzHugh and McKay (2000). They concluded that due to limited transport
capacity of the channel network downstream of the study area, the streamflow and sediment
yield simulated by the SWAT model were not sensitive to changes in the number and size of
the subwatersheds. Thus, the role of spatial discretization on SWAT outputs is still unclear,
with conflicting viewpoints being expressed by researchers. Effect of watershed
discretization on some nutrient components of the SWAT model has been addressed by Jha
et al (2004). The results indicated that simulated nitrate (NO3-N) at the outlet of the
watershed increased with the number of subwatersheds while mineral phosphorus (MIN P)
was unaffected. These authors recommended further research on evaluation of the effect of
watershed discretization on nutrient components of the SWAT model.
27
-------
3.2 Objectives
An optimal scale of geomorphologic resolution needs to be identified such that further
refinement in spatial scale does not contribute to a significant improvement in predicting
design quantities at the watershed outlet. This optimal geomorphologic resolution, along with
the associated drainage density, can then be utilized to determine the appropriate critical
source area prior to calibration and validation of the model.
The following questions are posed to address the effects of spatial resolution in the form of
watershed discretization on SWAT model simulations:
(i) To investigate how the number and size of subwatersheds impact SWAT simulations
of streamflow, sediment yield, and nutrient load.
(ii) To evaluate the possibility of identifying an optimal critical source area for these
quantities, and the conditions when it is available.
(iii) To develop a simple process-based index that is solely a function of the watershed
discretization level (i.e. does not require any information on soil, land use, and
management data, and HRU distribution level) to serve as a surrogate for sediment
and nutrient outputs in evaluation of the effect of watershed discretization level on
SWAT computations.
Previous studies have only partially addressed these objectives. Specifically, the impact of
watershed discretization on nutrient loads from upland areas has not been discussed at all.
With the exception of mineral phosphorus and nitrate, the impact on various pools of
phosphorus and nitrogen at the outlet has not been addressed either. The conditions when a
critical source area can be identified as in objective (ii) have not been studied, while
objective (iii) is completely novel.
28
-------
3.3 Methodology
The SWAT model integrated into the BASINS framework was utilized to evaluate the effect
of watershed discretization on SWAT computations. SWAT simulations were performed
with various watershed configurations for a 30 years time horizon from 1971 to 2000. The
characteristics of some of the watershed configurations that were utilized in this study (see
Figures 3.1-3.2) are summarized in Tables 3.1 and 3.2 for the Dreisbach and Smith Fry
watersheds, respectively. The tables include information on the applied critical source area
and corresponding number of subwatersheds, total number of HRUs, drainage density (the
ratio of total channel length over total watershed area), and average subwatershed area.
The SWAT model streamflow simulations are very sensitive to HRU distribution levels for
soil and land use areas (Mamillapalli, 1998). These user-specified thresholds control the
number of hydrologic response units (HRUs) in the watershed. For example, if a 10% soil
area is defined in FIRU distribution, only soils that occupy more than 10% of a subwatershed
area are considered in URU distributions. Subsequently, the number of URUs in the
watershed decreases with increasing threshold values. Since the goal of this study was to
evaluate only the effect of watershed discretization, and not the effects of spatial resolutions
of soil series, land use, and Digital Elevation Model (DEM), a 0% threshold value was
assigned for both soil area and land use area in HRU distribution.
Table 3.1. Properties of the Watershed Configurations Used for the Dreisbach Watershed.
Critical Source Area (km )
Number of Subwatersheds
Number of HRUs
Drainage Density (km/km2)
Average Subwatershed Area (km2)
0.03
103
647
3.91
0.06
0.035
81
587
3.57
0.08
0.045
59
502
3.19
0.11
0.06
45
445
2.83
0.14
0.10
23
314
2.28
0.27
0.30
13
231
1.55
0.48
0.40
5
135
1.30
1.25
2.5
1
73
0.91
6.23
Table 3.2. Properties of the Watershed Configurations Used for the Smith Fry Watershed.
Critical Source Area (km )
Number of Subwatersheds
Number of HRUs
Drainage Density (km/km )
Average Subwatershed Area (km )
0.03
89
676
4.09
0.08
0.050
63
577
3.28
0.12
0.060
49
522
3.06
0.15
0.10
33
429
2.55
0.22
0.25
15
308
1.89
0.49
0.40
9
248
1.56
0.82
0.60
5
198
1.35
1.47
2.9
1
93
0.65
7.30
29
-------
CSA= 0.03 (km2) CSA= 0.05 (km2) CSA= 0.1 (km2) CSA= 0.15 (km2) CSA= 0.36 (km2)
DD= 3.91 (km/km2) DD= 3.05 (km/km2) DD= 2.28 (km/km2) DD= 1.97 (km/km2) DD= 1.39 (km/km2)
CSA= 0.38 (km2)
DD= 1.3 (km/km2)
CSA= 0.5 (km2)
DD= 1.22 (km/km2)
CSA= 1.5 (km2)
CSA= 2.5 (km2)
DD= 0.94 (km/knT) DD= 0.91 (km/knT)
Figure 3.1. Watershed Configurations Used for the Dreisbach Watershed, Indiana.
30
-------
CSA= 0.03 (km2)
DD= 4.09 (km/km2)
CSA= 0.05 (km2)
DD= 3.27 (km/km2)
CSA= 0.1 (km2) CSA= 0.15 (km2)
DD= 2.54 (km/km2) DD=2.25 (km/km2)
CSA= 0.30 (km2)
DD= 1.76 (km/km2)
CSA= 0.5 (km2)
DD= 1.45 (km/km2)
CSA= 1.5 (km2)
CSA= 2.9 (km2)
DD= 0.96 (km/knT) DD= 0.65 (km/knT)
Figure 3.2. Watershed Configurations Used for the Smith Fry Watershed, Indiana.
-------
3.4 Effects of Watershed Discretization on Model Outputs
Spatial resolution in the form of watershed discretization might influence estimation of
streamflows, sediment, and nutrient loads generated from upland areas in a different way
from the ones computed at the outlet of the watershed. The difference between the two is that
the loads generated from upland areas do not include main channel processes such as channel
degradation and deposition. Here, the impacts of watershed discretization level on both
sediment and nutrient loads from upland areas and the ones at the outlet are discussed.
3.4.1 Streamflow
The effect of watershed discretization on simulated water yield for each HRU can be
evaluated by quantifying its effects on surface runoff and transmission losses. SWAT uses
the SCS curve number method to compute surface runoff for each HRU. If the HRU
distribution levels are set at 0 percent, the overall soil, land use, and management attributes
of the HRUs will be the same for various watershed configurations. Therefore, the number
and size of subwatersheds will not influence surface runoff computations. Bingner et al.
(1997) observed that simulated annual runoff varied by nearly 5% for various watershed
configurations. The small variations were perhaps because they applied nonzero threshold
levels for soil and land use areas. FitzHugh and Mackay (2000) reported that surface runoff
was practically identical for all watershed configurations although a 10 percent threshold
level was selected for both soil and land use areas. The results of our study revealed that
surface runoff computations were unaffected by the watershed discretization.
At a HRU level, transmission losses (i.e. water lost from ephemeral channels through the
bed) are the only mechanism in water yield simulations that may be affected by the
watershed discretization. The structure and properties of the ephemeral channels vary with
the number and size of subwatersheds that may affect computations of transmission losses.
FitzHugh and Mackay (2000) concluded that the 12 percent variation in streamflow
simulations at the outlet was due to the impact of watershed discretization on transmission
losses. Jha et al. (2004) also came to this conclusion. We observed that transmission losses
32
-------
1.2
-0.9
o
o.3
O O O O
»ooo o o
O Dreisbach Watershed
O Smith Fry Watershed
20 40 60 80
Number of Subwatersheds
100
Figure 3.3. Effects of Watershed Discretization on SWAT Streamflow Computations.
simulated by the SWAT model for various watershed configurations were identical for the
Dreisbach and Smith Fry watersheds.
The SWAT model employs Manning's equation to estimate flow velocity in a given main
channel. The variable storage or the Muskingum channel routing method is applied to route
water through the channel network. Flow losses through evaporation and channel losses are
the only processes that may result in a difference between water yield from upland areas (i.e.
FtRUs) and streamflow at the outlet of the watershed. The results of this study indicate that
there was no significant difference between the two. This aspect can be explained by
considering the size of the watersheds and the fact that the simulations were performed over
a 30 year period (1971-2000). The difference between streamflows computed for the coarsest
and finest watershed discretization levels was quite small. Figure 3.3 graphically depicts the
insensitivity of streamflow simulations to watershed discretization in the Dreisbach and
Smith Fry watersheds.
3.4.2 Sediment
An amalgamation of the studies on the impacts of watershed discretization on sheet erosion
computations and sediment routing components of SWAT is required for appraisal of the
effects on sediment yield at the outlet of the watershed. For this aspect, conflicting results
have been reported in previous studies. Bingner et al. (1997) and Jha et al. (2004) only
studied the effects of watershed delineation on sediment yield at the outlet without making a
distinction between the effects on sediment loads generated at upland areas and the effects on
33
-------
in-stream processes (i.e. channel deposition or degradation). Both studies indicated that
sediment yield at the outlet is very sensitive to the number and size of sub watersheds.
FitzHugh and Mackay (2000) observed that sediment loads (i.e. sheet erosion) from upland
areas decreased with the number of subwatersheds (the graphs presented in the paper show
that sediment generation increases with average subwatershed size) while sediment yield at
the outlet was almost unaffected by the number and size of subwatersheds.
SWAT model applies the Modified Universal Soil Loss Equation (Equation 2.5) at a HRU
level to compute sheet erosion from upland areas. In this study, a constant P factor was
applied to the whole watershed. C, K, and CFRG parameters were estimated by the BASINS
framework for each HRU based on its soil, land use, and management attributes. Since a 0%
threshold level was considered for both soil and land use areas in HRU distribution, unlike
the number of HRUs, their overall attributes were not affected by variations in the number
and size of subwatersheds. Thus, spatial average of P, C, K, and CFRG factors were identical
for various watershed configurations. The only parameter in Equation 2.5 that was influenced
by altering watershed configuration was USLE topographic factor, LS. SWAT calculates this
parameter for each subwatershed based on its slope and slope length, and applies it to all
HRUs located in that particular subwatershed. Figure 3.4 demonstrates the effect of
watershed discretization on weighted average LS factor. The weighted average of LS
decreased with the number of subwatersheds in both Dreisbach and Smith fry watersheds.
The rate of reduction plateaued once the number of subwatersheds was more than 20. This
level of watershed discretization corresponds to a 15 (ha) CSA that is approximately 2% of
o o
CS
^n
W16
i-l
C/2
u 4
60
^H
^ 2
"S
s
'53 n
-
3
-ct> o
- o0
• o o
000 00.
0 0 0 0 o
.
O Dreisbach Watershed
O Smith Fry Watershed
20 40 60 80 100
Number of Subwatersheds
Figure 3.4. Effect of Watershed Discretization on Weighted Average LS factor.
34
-------
Dreisbach and Smith Fry areas. Likewise, sediment loads from upland areas decreased by
28.9% and 22.7% in the Dreisbach and Smith Fry watersheds, respectively, between the
coarsest and finest watershed discretization levels (see Figure 3.5).
A comparison of sediment yields at the outlet of the watershed simulated for various
watershed configurations would reveal how channel processes are influenced by the CSA.
Further, comparing sediment loads from upland areas and sediment yield at the outlet for
each watershed discretization would be helpful to identify whether the watershed is "supply-
limited" or "transport-limited". Supply-limited refers to watersheds whose transport capacity
of the channel network is greater than sheet erosion from upland areas. In this type of
watersheds, channel deposition tends to be the overall dominant main channel processes
influencing sediment yield at the outlet. The results of this study revealed that sediment loads
from upland areas were larger than the sediment yields at the outlet indicating that Dreisbach
and Smith Fry are "transport-limited" watersheds. In a transport-limited watershed, sheet
erosion from upland areas is the major source of sediments in the watershed. Simulated
sediment yields at the outlet of the study watersheds were within 10% of the simulated
sediment loads from upland areas. The correlation coefficients between sheet erosion and
sediment yield at the outlet of the Dreisbach and Smith Fry watershed were respectively 0.97
and 0.99, indicating that simulated sediment yield at the outlet tended to behave in
accordance with sheet erosion from upland areas. The impact of watershed discretization on
both sediment loads, i.e. sheet erosion, from upland areas and sediment yield at the outlet of
the study watersheds is shown in Figure 3.5. It should be noted that channel deposition is the
1
^ 1.8
&
O 1.2
0
o o o o o
ooo
00
O Upland Sheet Erosion: Dreisbach Watershed
O Sediment Yield at the Outlet: Dreisbach Watershed
A Upland Sheet Erosion: Smith Fry Watershed
n Sediment Yield at the Outlet: Smith Fry Watershed
0
100
20 40 60 80
Number of Subwatersheds
Figure 3.5. Effects of Watershed Discretization on SWAT Sediment Computations.
35
-------
a 3.6
^
^
-&1
'1 2.4
53
Q
u
bo
J 1.2,
c3
i-H
Q
n
0
.
o
o
0 0
0 00
cS>o °
Os
r
> O Drainage Density
O Average Slope of Channel Network (Secondary Axis).
i ^_^
0
0.75 t3
^
rt
d
0.5 |
O
^M
0.25 «
5o
Number of Subwatersheds
Figure 3.6. Effects of Watershed Discretization on Drainage Density (DD) and Average Slope of Channel
Network, Smith Fry Watershed.
overall dominant main channel process in transport-limited watersheds. Dominance of
channel deposition indicates that channel erosion does not significantly contribute to
sediment yield at the outlet in a transport-limited, and thus sediment yield at the outlet does
not increase with drainage density. Although Drainage Density (DD) and average slope of
the channel network of the Dreisbach and Smith Fry watersheds increased with finer
watershed discretization (Figure 3.6), they did not influence sediment yield at the outlets.
3.4.3 Nutrients
Similar to sediment outputs, the effects of watershed discretization on nutrient outputs of
SWAT model were studied by examining the effects on nutrient loads from upland areas and
effects on in-stream processes. These relationships have been partly examined by Jha et al.
(2004). Here, we evaluate the effects of watershed discretization on total phosphorus (total P)
and total nitrogen (total N) loads from upland areas as well as at the outlets of the Dreisbach
and Smith Fry watersheds shown in Figures 3.7 and 3.8.
Total P (sum of all phosphorus pools) and total N (sum of all nitrogen pools) loads from
upland areas differ by nearly 30 percent between coarsest to finest watershed discretization
levels (Figures 3.7 and 3.8). These outputs were highly correlated to sheet erosion from
upland areas. A comparison of nutrient loads from upland areas and nutrient yields at the
outlet revealed that in-stream processes did not dramatically change the nutrient yields at the
outlet of the Dreisbach and Smith fry watersheds. Thus, nutrient yields at the outlet exhibited
36
-------
16
12
^ 4
O
H
° o
O Upland Total P Load: Dreisbach Watershed
O Total P Yield at the Outlet: Dreisbach Watershed
A Upland Total P Load: Smith FryWatershed
n Total P Yield at the Outlet: Smith Fry Watershed
20 40 60 80
Number of Subwatersheds
100
Figure 3.7. Effects of Watershed Discretization on SWAT Total P Computations.
48
36
60
| 24
6
12
O
H
o
O Upland Total N Yield: Dreisbach Watershed
O Total N Yield at the Outlet: Dreisbach Watershed
A Upland Total N Load: Smith FryWatershed
n Total N Yield at the Outlet: Smith Fry Watershed
0
100
20 40 60 80
Number of Subwatersheds
Figure 3.8. Effects of Watershed Discretization on SWAT Total N Computations.
trends similar to nutrient loads from upland areas. Total P and total N yields at the outlet
decreased by nearly 40 percent between coarsest to finest watershed discretization levels.
The rate of reductions were considerably smaller once the number of Subwatersheds was
more than 20 corresponding to 2 percent of the Dreisbach and Smith Fry watershed area. It
would appear that in-stream processes did not play a significant role in nutrient loads at the
outlet of the study watersheds.
3.5 Identification of an Optimal Watershed Discretization Level
A comparison of sediment and nutrient loads from upland areas of the Dreisbach and Smith
Fry watersheds for various watershed configurations revealed that 2 percent of the total
watershed area could be considered as the optimal critical source area. Furthermore, it was
shown that this optimal watershed discretization level could be applied to the sediment yield
37
-------
at the outlet as well. Similar results were reported by Jha et al. (2004). These results along
with the ones reported by Bingner et al. (1997), and FitzHugh and Mackay (2000) provide
modelers with valuable insight into effects of watershed discretization on SWAT
computations. We now examine the nature of averaging that the SWAT model does in order
to elucidate the role of sub-grid processes. Results indicated that sheet erosion estimates by
SWAT are affected by watershed discretization because USLE topographic factor, LS, is
averaged over sub watersheds. The effect of averaging the LSusiE over subwatersheds can be
assessed by rewriting Eq. (2.5):
sed =
i=\
p
7=1
xLS,
(3.1)
where N is the total number of subwatersheds, p is the total number of HRUs in
sub watershed /', A (ha) is total watershed area, Atj (ha) is the area of HRUy in sub watershed /',
LSj is the USLE topographic factor averaged over subwatershed /', and C!;7, Ki:j, PJJ, and
CFRGjj are soil erosion parameters for HRUy in subwatershed / as defined in Eq. (2.5). The
quantity/j for each HRU is computed as:
,0.56
(3.2)
In Eq. (3.2), all parameters are defined as in Eq. (2.5). The runoff volume for HRUy in
subwatershed /' (Qi,y) is not affected by watershed discretization as discussed in Section 3.4.1.
Rational Method (Equation 2.4) is applied for computation of peak runoff rate for each HRU.
The area of HRUs is the only parameter in this equation that varies with the number and size
of subwatersheds. Therefore the effect of watershed discretization on parameter ftj can be
sought through the effect on AjjLU (i.e. [Aitj * AtJ] °'56). Sheet erosion from upland areas by
SWAT can be represented with an Erosion Index (El) defined as:
N
£/ = Z<
1.12
7=1
xC^xK^.xP^.xCFRG^
xLS,
(3.3)
38
-------
The parameter El is essentially a weighted average of USLE topographic factor over the
whole watershed that can reasonably represent sediment generated from upland areas for
investigation of watershed discretization effects. In a watershed with one land
use/management, and a single soil type this weighted average would not depend on the soil
and land use attributes and only Digital Elevation Model (DEM) attributes would be
important. In that case parameter El can be written as:
xLS,
(3.4)
where K is a constant (Q x Kj * Pt * CFRGj). If all HRUs in sub watershed / have the same
size, and the number of HRUs in different subwatersheds are the same, El can be rewritten:
EI=Kx
(3.5)
More insight into sheet erosion computations would be provided by computing another
index, namely the Area Index (AT), defined as:
1.12
(3.6)
2=1
Figure 3.9 presents the Area Index for various watershed configurations for both the
X
1)
2000
1500
1000
500
0
O Dreisbach Watershed
O Smith Fry Watershed
0 20 40 60 80 100
Number of Subwatersheds
Figure 3.9. Effects of Watershed Discretization on Area Index (AI).
39
-------
watersheds. This index is simple to compute, and does not require any information on soil,
land use, and management data. Also, AI is computed at a subwatershed level and not a HRU
level. Thus, the important HRU distribution levels for soil and land use areas do not affect its
computation. AI enables users to identify an optimal critical source area for a given
watershed utilizing only Digital Elevation Model (DEM) data and is independent of soil, land
use, and management attributes.
The limitations of utilizing AI for identification of an optimal critical source area arise from
the assumptions that were made in arriving at Equations (3.3)-(3.6). As critical source area
decreases, subwatershed scale approaches HRU scale. It was assumed that the effect of soil,
land use, and management properties could be factored out in the watershed. The validity of
this assumption depends on the importance of topographic attributes of the watershed that are
represented by a Digital Elevation Model (DEM) versus the importance of soil, land use, and
management properties.
The results of this study indicated that the effect of Digital Elevation Model (DEM) attributes
of the study area on runoff term of MUSLE equation, parameter/(Equation 3.2), dominated
the heterogeneity of soil and land use attributes. Thus, AI could represent sediment loads
from upland areas in identification of an optimal critical source area. In addition, nutrient
loads from upland areas and sediment yield at the outlet of the Dreisbach and Smith Fry
watersheds were strongly correlated to sediment loads from upland areas for various
watershed configurations. If these assumptions do not hold, then the more complicated
Erosion Index (Equation 3.3) needs to be used. The high correlation between the Erosion
Index (El) and the Area Index (AI), depicted in Figure 3.10, indicates that these assumptions
were valid for both Dreisbach and Smith Fry watersheds.
The correlation between sediment loads from upland areas and sediment yield at the outlet
depends on whether the watershed is transport- or supply-limited. In a transport-limited
watershed, upland areas are the major source of sediments. Therefore, application of the Area
Index would be adequate for identification of a proper watershed discretization level. In a
supply-limited watershed, not only upland areas contribute to sediment yield at the outlet, but
channel degradation also serves as a major source of sediment. Channel degradation depends
40
-------
I
o
R=0.96
R=0.98
O Dreisbach Watershed
O Smith FryWatershed
8oc
1800
900 1200 1500
Area Index (AI)
Figure 3.10. Correlation between the Erosion Index (El) and the Area Index (AI).
on drainage density and slope of the channel network. Computation of the Area Index does
not include the effects of drainage density. Thus, application of AI would not be appropriate
if the watershed is supply-limited and channel degradation significantly contributes to
sediment yield at the outlet.
3.6 Conclusions
The main conclusions of examination of the effect of watershed discretization on un-
calibrated model computations for the Dreisbach and Smith Fry watersheds are as follows:
Surface runoff computations of the SWAT model were virtually unaffected by the number
and size of sub watersheds. Transmission losses and losses in the main channel were mostly
unchanged between the coarsest to finest watershed discretization levels.
Sediment loads from upland areas are affected by watershed discretization. In both Dreisbach
and Smith Fry watersheds these loads decreased with the number of subwatersheds. The rate
of reduction plateaued once the number of subwatersheds was more than 20 in both
watersheds. This watershed discretization level corresponded to a critical source area about 2
percent of total area of the watersheds. Nutrient loads from upland areas were highly
correlated to sheet erosion.
Identification of control processes and key management actions within a watershed is
essential to obtain an optimal watershed discretization level for SWAT computations at the
41
-------
outlet of the watershed. Substantially different conclusions can be drawn for transport-limited
versus supply-limited watersheds. Both Dreisbach and Smith Fry watersheds exhibited the
behavior associated with transport-limited watersheds. However, BMPs were not represented
in this phase of the study. In-stream processes did not significantly influence nutrient
predictions at the outlet of the watersheds. Total P and total N yields at the outlets were
highly correlated to the nutrient loads from upland areas of the Dreisbach and Smith Fry
watersheds.
Computation of the Area Index (AT) appears to be a useful alternative for identification of an
appropriate watershed discretization level prior to model calibration. However it is cautioned
that application of the Area Index might not be appropriate for supply-limited watersheds. If
channel degradation contributes to sediment and nutrient yields at the outlet, a more accurate
measure for estimation of optimal drainage density is required. To overcome this limitation,
we recommend that the drainage density corresponding to the optimal watershed
discretization level be based on the Area Index approach be computed initially. This drainage
density could be compared to the channel network defined by USGS 7.5-min quadrangle
maps. The watershed discretization level corresponding to the one providing more detailed
channel network should then be utilized for modeling purposes.
42
-------
Section 4.0
Model Calibration and Validation
4.1 Introduction
Application of simulation modeling in research and decision making requires establishing
credibility, i.e., "a sufficient degree of belief in the validity of the model" (Rykiel, 1996), for
model simulations. The term validity has been defined in so many different ways that no
single literature has been able to embrace all of the methods employed to address the issue of
validation. However, it is reasonable to agree on the three fundamental attributes of a valid
model as described by Beck et al. (1997): (i) soundness of mathematical representation of
processes, (ii) sufficient correspondence between model outputs and observations, and (iii)
fulfillment of the designated task.
Peer-review is commonly practiced to deal with the first attribute, and is often followed by
model calibration. Model calibration is the exercise of adjusting model parameters manually
or automatically for the system of interest until model outputs adequately match the observed
data. The credibility of model simulations is further evaluated by investigating whether
model predictions are satisfactory on different data sets. The semantic of appropriate
terminology (validation, verification, corroboration, confirmation, etc.) for this procedure has
been disputed, although in practice these terms have been used interchangeably. The bottom
line is that all of these terms refer to truth and accuracy of the model (Konikow and
Bredehoeft, 1992; Oreskes et al., 1994). Here, the term "validation" will be used with no
attempt to clarify the appropriateness of these words.
Although model simulations can be conducted on various temporal and spatial scales,
representation of natural processes through the device of a model will always be macroscopic
in comparison to reality. Models provide nothing beyond an approximation of reality. A
certain degree of confidence in model predictions can be obtained by minimizing the errors
associated with such approximation through a calibration procedure. Calibration of a
watershed model is essentially the exercise of adjusting model parameters such that model
43
-------
predictions sufficiently match observations. In this section, the calibration and validation of
the SWAT model for the study watersheds is discussed.
4.1 Indicators of Model Performance
Various measures including the coefficient of determination R2 and the coefficient of
efficiency EN-S (Nash and Sutcliffe, 1970) have been utilized to evaluate the accuracy of
model predictions (Srinivasan et al., 1998; Eckhardt and Arnold, 2001; Santhi et al., 2001a;
Chung et al., 2002). The coefficient of determination is the square of the Pearson's product-
moment correlation coefficient. This coefficient describes the proportion of the total
variances in the observed data that can be explained by the model, and is defined as:
Ri=^L
(4.1)
where Ot and Pt are observed and predicted data points, respectively. O is the average of
observed data and P is the average of predicted values.
R2 values range from 0 to 1. An R2 value equal to one is indicative of a perfect correlation
between measured data and model predictions. The coefficient of determination is insensitive
to additive and proportional differences between the predicted and observed values. On the
other hand, R is more sensitive to outliers than to the values near the mean. This
oversensitivity leads to a bias toward extreme streamflow values.
To overcome the limitations associated with using the coefficient of determination, the
coefficient of efficiency E^-shas been widely used to evaluate the performance of hydrologic
models. The coefficient of efficiency is defined as:
44
-------
EN_S=1.0-
N
Z
i=\
N
(4.2)
EN-S ranges from -oo to 1, with higher values indicating a better prediction. If EN-S is
negative or very close to zero the model prediction is considered "unacceptable" (Santhi et
al., 200la). The coefficient of efficiency is indicative of how well the plot of observed versus
predicted values fit the 1:1 line.
4.2 Sensitivity Analysis
Large complex watershed models contain hundreds of parameters that represent hydrologic
and water quality processes in watersheds. Model predictions are more sensitive to
perturbation of some input parameters than others, even though the insensitive parameters
may bear a larger uncertain range. Thereby, adjustment of all model parameters for a given
study area not only is cumbersome, but is not essential. The main objective of sensitivity
analysis is to explore the most sensitive parameters to facilitate model calibration procedure.
4.2.1 Sensitivity Index
The SWAT model outputs depend on many input parameters related to the soil, land use,
management, weather, channels, aquifer, and reservoirs. Table 4.1 summarizes the 36 SWAT
parameters selected out of for sensitivity analysis in this study. These parameters were
chosen based on the results of previous studies by Arnold et al. (2000), Eckhardt and Arnold
(2001), Santhi et al. (2001a), Vandenberghe (2001), Sohrabi et al. (2003), and Benaman and
Shoemaker (2004). Sensitivity of streamflow, sediment, and nutrient outputs of the SWAT
model to the selected parameters is sought by perturbing model parameters "one-at-a-time"
and determining a linear sensitivity parameter ($), defined as (adapted from Gu and Li,
2002):
45
-------
Table 4.1. List of SWAT Parameters Considered in Sensitivity Analysis
No.
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
Parameter
CN2
SLOPE
SLSUBBSN
ESCO
CH-N1
CH-S1
CH-K1
CH-N2
CH-S2
CH-K2
GWQMN
ALPHA-BF
GW-DELAY
GW-REVAP
SOL-AWC
CH_EROD
CH_COV
SPCON
SPEXP
PRF
USLE_P
USLE_C
SOL_LABP
SOL_ORGP
SOL_NO3N
SOL_ORGN
RSI
RS2
Description
Initial SCS runoff curve number for moisture condition II
Average Slope steepness
Average slope length
Soil evaporation compensation factor
Manning's "n" value for tributary channels
Average slope of tributary channels
Effective hydraulic conductivity in tributary channel alluvium
Manning's "n" value for the main channel
Average slope of the main channel along the channel length
Effective hydraulic conductivity in main channel alluvium
Threshold depth of water in shallow aquifer for return flow to occur
Baseflow alpha factor
Groundwater delay time
Groundwater "revap" time
Available water capacity of the soil layer
Channel credibility factor
Channel cover factor
Linear coefficient for calculating maximum sediment re-entrained
Exponent coefficient for calculating maximum sediment re-entrained
Peak rate adjustment factor for sediment routing in channel network
USLE equation support practice factor
Maximum value of USLE equation cover factor for water erosion
Initial soluble P concentration in soil layer
Initial organic P concentration in soil layer
Initial NO3 concentration in soil layer
Initial organic N concentration in soil layer
Local algae settling rate at 20OC
Benthic (sediment) source rate for dissolved P in the reach at 20OC
Min
35
0
10
0
0.008
0
0
0.008
0
0
0
0
0
0.02
0
0
0
0.001
1
0
0.1
0.001
0
0
0
0
0
0.001
Max
98
0.6
150
1
30
10
150
0.3
10
150
5000
1
500
0.2
1
0.6
1
0.01
1.5
2
1
0.5
100
4000
5
10000
2
0.1
Units
m/m
m
m/m
mm/hr
m/m
mm/hr
mm
days
days
mm/mm
cm/hr/Pa
mg/kg
mg/kg
mg/kg
mg/kg
m/day
mg/m2.day
SWAT input file
.MGT
.HRU
.HRU
.HRU
.SUB
.SUB
.SUB
.RTE
.RTE
.RTE
.GW
.GW
.GW
.GW
.SOL
.RTE
.RTE
.BSN
.BSN
.BSN
.MGT
CROP.DAT
.CHM
.CHM
.CHM
.CHM
.SWQ
.SWQ
46
-------
Table 4.1. (Continued)
29
30
31
32
33
34
35
36
RS4
RS5
BC4
AIO
All
AI2
RHOQ
K-P
Rate coefficient for organic N settling in the reach at 20OC
Organic P settling rate in the reach at 20OC
Rate constant for mineralization of P to dissolved P in the reach at 20OC
Ratio of chlorophyll-a to algae biomass
Fraction of algal biomass that is nitrogen
Fraction of algal biomass that is phosphorus
Algal respiration rate at 20OC
Michaelis-Menton half-saturation constant for phosphorus
0.001
0.001
0
0.001
0.07
0.01
0.05
0.001
0.1
0.1
1
0.01
0.09
0.02
0.5
0.5
I/day
I/day
I/day
ug/mg
mg N/mg
mg P/mg
I/day
mpP/1
.SWQ
.SWQ
.SWQ
.WWQ
.WWQ
.WWQ
.WWQ
.WWQ
47
-------
St = max
(P,2-+P/)/2
(4.3)
where 0/and O,2 are model outputs corresponding to perturbation of parameter / from P/ to
Pi2, respectively. A "+" sign corresponds to parameter changes in positive direction, i.e., P/
< P,2, whereas a "-"sign indicates parameter changes in negative direction, i.e., P/ > P,2. In
(4.3), it is assumed that the response of model outputs to parameter perturbation is linear. St
is essentially a normalized estimate of sensitivity of design variables (streamflow, sediment
yield, etc.) to a parameter perturbation, with higher values indicating higher sensitivity.
The sensitivity of various outputs of the SWAT model to the parameters listed in Table 4.1
for the study watersheds is depicted in Figure 4.1(a-d). The indices shown in the figure were
calculated by incorporating the results of the sensitivity analysis on both Dreisbach and
Smith Fry watersheds:
- 0.5 X
•0.5x5,.
i,Smith Fry
(4.4)
where S^Dreisbach and S^smuhFry are the sensitivity indices determined for parameter /' at the
outlet of the Dreisbach and smith Fry watersheds, respectively.
4.2.2 Additional analysis
The magnitude of the sensitivity index, St (4.3), corresponding to each model parameter is
rendered subjective to the initial set of parameters that are used in the analysis. Figure 4.3
illustrates sensitivity of sediment output of SWAT to various input parameters listed in Table
4.1 for two cases. In case one, corresponding to the results shown in Figure 4.1(b), the
default value was used for the USLE practice factor, i.e., USLE_P=1. It was observed that in
this case the parameters that affect the magnitude of channel degradation such as PRF,
CH_COV, and CH_EROD (see Table 4.1 for definitions) did not bear a high sensitivity for
sediment outputs. However, when the USLE practice factor was altered to 0.3, i.e.,
48
-------
(a)
1.2 y-
52 0.9 -
X
•d
>, °-6 -
>
1 0.3 -
$
0.0 --
(c)
3.2 -r
52. 2.4 -
X
•d
V, 1.6 -
1 °'8 "
0 0 -
Streamflow
1
g^|3g&$S!gSs|8*3
°<|gS<^|ggggSgg
j^1— 'C/^ffiQp^1— ' i^J |_J (J ^
g ° d § £ 1
^ o o
Parameter
Total P Yield
HIM
I— ' i— 3 3 i— JO ^ >— J rt PH ^~7 i ,__]
g g-^o <^|S
(b)
4.0 -T-
&S 3.0 -
X
u
•d
^ 2-° -
'>
W
| 1.0 -
0.0 --
(d)
2.8 j
© 2.1 -
X
•d
£> 1.4 -
1 0-7-
f/3
0.0 -
Sediment Yield
III
CN PH CN O *~7, ^ O W ^ ' — ' fe ' — ' PH ^Z ' — ' ^ fe O f^ P^ Q ^*
^ .'^ io°?SPL2o
«0«fe"^^PD ofeg"gffiwogwi
pp o2 ^^S iffi°
« ^ | ^ | 0
Parameter
Total N Yield
Hun,
g°l^ g j °^g0&00gQ° ^l0 ^1 °lS "
O S K° ^& ggg^
Parameter Parameter
Figure 4.1. Sensitivity of SWAT Parameters Listed in Table 4.1Determined Based on (a) Streamflow, (b) Sediment, (c) Total P, and (d) Total N.
49
-------
USLE_P=0.3, the parameters corresponding to sediment transport in channel network were
among the most sensitive parameters as demonstrated in Figure 4.3. The procedure that is
utilized within the SWAT code for representation of sediment transport in the channel
network is the primary reason that the sensitivity index (Equation 4.3) for channel sediment
parameters varied with USLE practice factor that is utilized for estimation of sheet erosion.
Channel sediment processes within the SWAT code are represented by Equations 2.6-2.10.
At each time step, for each channel segment, the initial sediment concentration that depends
on both sheet erosion from upland areas and sediment processes (degradation or deposition)
in the upstream channel segments is compared to the transport capacity of the channel
segment. When a USLE practice factor equal to 1.0 was utilized, initial sediment
concentration in the channel network was greater than transport capacity of the channel
network and channel deposition was dominant in the channel network. In this case, channel
degradation is set to zero by the model and therefore, the sediment output was not sensitive
to model parameters that correspond to channel degradation (Figure 4.2, USLE_P=1.0).
When USLE practice factor was set at 0.3, sheet erosion from upland areas and subsequently
initial sediment concentration in the channel network decreased and sediment degradation
was the dominant channel processes. The sensitivity of sediment output to the channel
sediment parameters such as CH_N2, CH_S2, CH_EROD, CH_COV, SPCON, SPEXP, and
4.0
@ 3.0 -
X
u
-d
a
2.0-
1 Sediment Yield
llll
• USLE_P=1 .0 D USLE_P=0.3
.^^.. no
CS PH CN O *7 ^ O W ^ ' — ' fe ' — ' PH >Z, ' — ' ^ fe O ^ P^ Q ^*
^n'^n'O^S&m^S^X^^^raO^^oO
ow.wyffi^qg^^,.iw^ffi:3So
K0^fe0j,^po "S5&°QffiU§wiffi'
^P §d °^S ^ K "
o
Figure 4.2. Sensitivity of SWAT Parameters Listed in Table 4.1Determined Based on Sediment Yield.
50
-------
PRF (see Table 4.1 for definition of the parameters) significantly increased as a result as
shown in Figure 4.2 for USLE_P=0.3.
4.2.3 Limitations
In computing the sensitivity index ($) (Eq. 4.3), the underlying assumption that the response
of the model to parameter perturbation is linear may not hold for all of model parameters. For
example, Figure 4.3 shows the response of streamflow computations of the SWAT model to
GWQMN (threshold depth of water in shallow aquifer for return flow to occur) parameter at
the outlet of the Dreisbach watershed. Streamflow output of the model is very sensitive to the
parameter changes in the range of 0-500 (mm), whereas changes beyond 1000 (mm) do not
result in any appreciable variation in model output.
Moreover, correlations between model parameters that should be elicited and encoded in a
comprehensive sensitivity analysis are neglected in (4.3). A change in one parameter would
result in a subsequent change in the correlated parameter. The combined changes perhaps
results in a different response in the design variable.
4.2.3 Conclusions
A linear sensitivity index was applied to the Dreisbach and Smith Fry watersheds in Indiana
to determine the most sensitive SWAT parameters for calibration purposes. The most
sensitive parameters identified for various design variables are listed in Table 4.2. It should
1.6
*• 1.2
5§ °'8
I
1)
^ 0.4
0
0
1000
2000 3000 4000 5000
GWQMN (mm)
Figure 4.3. Sensitivity of Streamflow Output of the SWAT Model at the Outlet of Dreisbach Watershed
to GWOMN Parameter.
51
-------
Table 4.2. Parameters Identified as Being Important from Sensitivity Analysis for Calibration.
Design Variable
Parameter
Streamflow
CN2
SOL-AWC
GWQMN
CH-K1
SLOPE
ALPHA-BF
GW-DELAY
GW-REVAP
CH-N2
CH-S2
Sediment
CN2
USLE P
CH-S2
USLE C
CH-N2
CH EROD
CH COV
PRF
SPCON
SOL AWC
Total P
CN2
USLE P
AIO
SOL ORGP
AI2
RHOQ
USLE C
RSI
SOL LABP
RS5
Total N
CN2
AIO
USLE P
SOL ORGN
RHOQ
All
USLE C
RSI
SOL-AWC
RS4
be noted that these results are location- and size-dependent and may vary for watersheds with
different characteristics.
4.3 Representation of Best Management Practices (BMPs) with SWAT
There were four different types of structural BMPs installed on the Dreisbach and Smith fry
watersheds, namely grassed waterways, field borders, parallel terraces, and grade
stabilization structures. The BMPs were implemented in 1974 and 1975 in the Dreisbach and
Smith Fry Watersheds, respectively. Figure 2.2 depicts the location of these BMPs in the
watersheds. SWAT has previously been used to model the impact of some structural BMPs
in good condition. Vache et al. (2002) simulated riparian buffers, grassed waterways, filter
strips and field borders by modifying the channel cover factor and channel erodibility factor
in SWAT to model the cover density and erosion resistant ability of the structures. Santhi et
al. (2003) simulated grade stabilization structures in SWAT by modifying the slope and soil
erodibility factor and used a program that simulates filter strips based on the filter strip's
ability to trap sediment and nutrients based on the strip's width.
For this study, a method was developed to evaluate the ability of grassed waterways, grade
stabilization structures, field borders and parallel terraces in SWAT to reduce sediment and
nutrients loads from non-gully erosion, based on published literature pertaining to BMP
simulation in hydrological models and considering the hydrologic and water quality
52
-------
processes simulated in SWAT. Based on the function of the BMPs and hydrologic and water
quality processes that are modified by their implementation, corresponding SWAT
parameters were selected and altered as discussed below.
Field Borders
Field borders are strips of vegetation established at the borders of a field where excessive
sheet and rill erosion is known to occur. The vegetative cover slows down surface runoff and
reduces sheet and rill erosion, and nutrient and pesticide loads in surface runoff. "FILTERW"
(width of edge-of-field filter strip) parameter in .hru input file is used in SWAT to calculate
the filter strip's trapping efficiency for sediment, nutrients, and pesticides. The default value
for this parameter is zero. The width of the field borders installed in the study watersheds
was 5 m. Therefore, FILTERW was modified to 5 m for the FtRUs where the field borders
have been implemented
Parallel Terraces
Parallel terraces are often used to reduce the peak runoff rate and soil erosion, decrease
sediment content of runoff water, and improve water quality. Figure 4.4 illustrates a
schematic of a parallel terrace. The horizontal spacing between terraces is determined as
(ASAE 2003):
(4.5)
o
where H (SLSUBBSN in Table 4.1) is horizontal spacing between terraces, S (SLOPE in
Original Ground Surface
H
Figure 4.4. Schematic of Parallel Terraces.
53
-------
Table 4.1) is the weighted average land slope of the land draining into the terrace, 7 is a
variable with values of 0.3, 0.6, 0.9, or 1.2 influenced by soil erodibility, cropping systems,
and crop management practices. Xis a variable with values from 0.12 to 0.24. This value for
the study area is 0.21 (ASAE 2003). Equation (4.5) with the slope (5) assigned by SWAT
based on the Digital Elevation Model (DEM) and 7=0.9 was used to determine the
SLSUBBSN parameter for the HRUs with parallel terraces.
Streamflow, sediment, and nutrient computations of the SWAT model are most sensitive to
the SCS curve number (CN2 in Table 4.1). CN2 and consequently simulated surface runoff
volume (Equation 2.1) decrease significantly for terraced conditions. The CN2 values for the
HRUs with parallel terraces were altered to the values for terraced condition obtained from
Neitschetal. (2001a,b).
USLE support practice factor (USLE_P) in (2.5) accounts for the impact of a specific support
practice on soil loss from a field. Support practices include contour tillage, strip cropping on
the contour, and terrace systems. Figures 4.1(b)-(c) indicate that sediment and nutrient
computations of the SWAT model are very sensitive to this parameter. While the default
value for USLE_P is unity, this value was altered to 0.2 (Neitsch et al., 2001a,b) for the
HRUs with parallel terraces.
Grassed Waterways
Grassed waterways are used to protect a stream from gully erosion, and act as a filter to
absorb some of chemicals and nutrients being carried in surface runoff. A natural stream is
graded and seeded by grass to form a parabolic shape channel covered by grass. Surface
runoff flows down across the grass rather than eroding soils from the channel perimeter. To
represent grassed waterways in the SWAT model three parameters— channel erodibility
factor (CH_EROD), channel cover factor (CH_COV), and channel Manning's "w" value
(CH_N2) - were modified.
SWAT uses Manning's equation to compute the velocity of flow in the channel segments.
Flow velocity decreases with channel Manning's "«" value (CH_N2). The sensitivity of
54
-------
sediment computations of SWAT to Manning's number is shown in Figure 4.2. The default
value for CH_N2 in SWAT is 0.014. This value was modified to 0.24 for the channel
segments with grassed waterways (Chow, 1956). These channel segments were considered
fully protected by the vegetative cover (CH_COV=0), and non-erosive (CH_EROD=0).
Grade Stabilization Structures
A dam or an embankment built across a waterway or an existing gully reduces water flow
and gully erosion. The height of the grade stabilization structures installed on the Dreisbach
and Smith Fry watersheds was 1.2 m. Figure 4.5 shows the schematic of a grade stabilization
structure. The new slope (S^ of channel segments with grade stabilization structures was
calculated as:
S =S -
0 mod *-" org j
(4.6)
where Sorg is the original channel slope, and L is the length of the channel segment in meters.
The channel segments with grade stabilization structure were also considered non-erosive
(CH_EROD=0).
The representation of BMPs discussed above is summarized in Table 4.3. Once the BMPs
were represented by fixing the corresponding parameters at the values shown in Table 4.3,
the rest of model parameters were calibrated for the study watersheds.
1.2m
Figure 4.5. Schematic of Grade Stabilization Structures.
55
-------
Table 4.3. Representation of Field Borders, Parallel Terraces, Grassed Waterways, and Grade
Stabilization Structures in SWAT.
BMP
Field
Border
Parallel
Terrace
Grassed
Waterway
Grade
Stabilization
Structure
Function
Increase sediment
trapping
Reduce overland
flow
Reduce sheet
erosion
Reduce slope
length
Increase channel
cover
Reduce channel
erodibility
Increase channel
roughness
Reduce gully
erosion
Reduce slope
steepness
Representing SWAT Parameter
Variable
(input file)
FILTERW
(.hru)
CN(2)
(•mgt)
USLE P
(•mgt)
SLSUBBSN
(.hru)
CH COV
(.rch)
CH EROD
(.rch)
CH N(2)
(.rch)
CH EROD
(.rch)
CH S(2)
(.rch)
Range
0-5 (m)
0-100
0-1
10-150
0-1
0-1
0-0.3
0-1
-
Value when BMP
implemented
5(m)
*
0.2
(terraced condition)
FromEq. (4.5)
0.0
(completely protected)
0.0
(non-erosive channel)
0.24
0.0
(non-erosive channel)
From Eq. (4.6)
*Estimated based on land use and hydrologic soil group of the HRU where it is installed for terraced
condition.
4.4 Model Calibration
The characteristics of a good calibration data set have been subject of much discussion and
debate (James and Burges, 1982; Gupta and Sorooshian, 1985; Beck, 1987; Sorooshian and
Gupta, 1995). However, there are only general, qualitative guidelines for the selection of the
calibration data set. A good calibration data set contains sufficient information to fulfill the
goals of the study. Sorooshian et al. (1983) showed that a single year of measured stream
flow data could be adequate to calibrate a hydrologic model if it contains the right
information. Typically three to five years of data are required in calibration of a hydrologic
model.
In this study, hydrologic components of the SWAT model were calibrate and validated on a
monthly basis for a time period from January 1975 to December 1978. Average, minimum,
56
-------
and maximum monthly precipitations at the outlet of the Black Creek watershed for this
period were 70, 7, and 184 mm, respectively. The calibration and validation period contains
the lowest precipitation in 1970-2000 time period (see Figure 2.6). Only 2.5% of monthly
precipitations during 1970-2000 period exceeded 184 mm. The average monthly
precipitation during 1970-2000 period was 77 mm, slightly larger than 70 mm. The monthly
precipitation time series depicted in Figure 2.6 shows that the 1975-1978 time period
encompasses adequate information for calibration and validation of SWAT that will be
utilized for long-term (1970-2000 period) evaluation of the objectives of the study. The
available data for calibration and validation of SWAT for the study watershed are
summarized in Table 2.5.
For calibration and validation of the SWAT model, three steps were implemented. First, the
optimal watershed discretization level obtained from Section 3.0 was utilized for watershed
subdivision and extraction of channel networks of the study watersheds. It was concluded
that application of 2 percent of the watershed area as critical source area is sufficient for the
Dreisbach and Smith Fry watersheds. Further, HRU distribution levels for soil and land use
areas were set at 0%. These user-specified thresholds control the number of FtRUs in the
watershed. For example, if a 10% soil area is defined in HRU distribution, only soils that
occupy more than 10% of a subwatershed area are considered in FtRU distributions.
Moreover, parameters of the FtRUs and channel segments where the BMPs have been
installed were accordingly set to the values specified in Table 4.3 and were not altered during
calibration. The rest of model parameters were calibrated for streamflow, sediment, and
nutrient yields.
For flow calibration, the measured daily stream flow series from January 1975 to December
1978 was split into two sets. The first set of streamflows from January 1975 to June 1977 (30
months) was utilized for calibration. The rest of the time series containing 18 months of
measured streamflow was used for validation of the model. Sediment and nutrient
components of SWAT were calibrated for the time period from January 1974 to December
1975, and validated from January 1975 to May 1977. Both calibration and validation
57
-------
procedures were performed on a monthly basis. A flowchart describing the procedure for
calibration of the SWAT model is shown in Figure 4.6 (adapted from Santhi et al., 2001a).
Separate surface runoff (S.R.) andbaseflow
(B.F.) for measured daily flow
Run SWAT
If average simulated S.R.* is within
±15% of average measured S.R.;
R2>0.6; and EN_S>0.5
If average simulated S.F.* is within
±15% of average measured S.F.;
R2>0.6; and EN_S>0.5
If average simulated sediment is within
±20% of average measured sediment;
R2>0.6; and EN_S>0.5
CO
If average simulated total P is within
±20% of average measured total P.;
R2>0.6; and EN_S>0.5
R2>0.6 & EN_S>0.5
If average simulated total N is within
±20% of average measured total N;
R2>0.6; and EN.S>0.5
R2>0.6 & EN_S>0.5
NO
Adjust CN
NO.
Adjust SOL_AWC,
and GWQMN
NO.
Adjust USLE_C, USLE_P, CH_N2,
CH COV, and CH EROD
NO.
Adjust SOL_ORGP, SOL_LABP,
AIO, AI2, andRHOQ
Adjust SOL_ORGN, SOL_NO3,
AIO, All, andRHOQ
Calibration complete
*S.R.: surface runoff, S.F.: streamflow, and B.F.: baseflow.
Figure 4.6. Calibration Flowchart (Adapted from Santhi et al., 2001a).
58
-------
Initially, baseflow was separated from surface runoff using the "ISEP" hydrograph separation
model. Surface runoff was calibrated until the average monthly simulated surface runoff was
within ±15% of average observed surface runoff, R2> 0.6, and EN.S> 0.5 for the calibration
period. The same criteria were used for the total streamflow. Sediment and nutrient yields
were calibrated until the average simulated quantities were within ±20% of average observed
ones, R2 > 0.6, and EN-S > 0.5. The results of the calibration procedure are summarized in
Table 4.4.
Once calibration of the model was completed, validation was performed to evaluate the
performance of the model for a data set different from the one used for calibration. The
optimal parameter values obtained from model calibration were used in model validation.
Predicted and observed data were compared using coefficient of efficiency (EN.S) and
coefficient of determination (R2) to test the validity of the model. The summary results of
model validation are summarized in Table 4.5.
Satisfactory model calibration and validation results were obtained for both watersheds
(Tables 4.4 and 4.5). In general, the calibrated model was able to adequately predict both low
and high streamflow, sediment, and nutrient yields in both watersheds. However,
streamflows for March 1978 were underpredicted and a low coefficient of efficiency was
obtained for total P in the Smith Fry watershed in the validation period. While the model
slightly overpredicts mineral and total phosphorus yields at the outlets for the months with
low phosphorus yield, the high yield months were underpredicted.
Table 4.4. Results of Calibration of SWAT for Streamflow, Sediment and Nutrient Simulations.
Variable1
Streamflow (m3/s)
Surface Runoff (mVs)
Suspended Solids (t/ha)
Mineral P (kg/ha)
Total P (kg/ha)
Total N (kg/ha)
Dreisbach
Obs2
0.039
0.035
0.027
0.070
0.077
1.35
Sim3
0.04
0.037
0.024
0.070
0.094
1.53
R2
0.92
0.91
0.97
0.92
0.93
0.76
EN-S
0.84
0.80
0.92
0.84
0.78
0.54
Smith Fry
Obs2
0.054
0.045
0.151
0.46
0.587
8.81
Sim3
0.052
0.049
0.16
0.55
0.708
7.29
R2
0.86
0.84
0.94
0.92
0.91
0.82
EN-S
0.73
0.62
0.86
0.73
0.82
0.64
1 Monthly simulations,2 Observed;3 Simulated.
59
-------
Table 4.5. Results of Validation of SWAT for Streamflow, Sediment and Nutrient Simulations.
Variable1
Streamflow (m /s)
Surface Runoff (m /s)
Suspended Solids (t/ha)
Mineral P (kg/ha)
Total P (kg/ha)
Total N (kg/ha)
Dreisbach
Obs2
0.042
0.038
0.032
0.067
0.074
1.227
Sim3
0.047
0.045
0.033
0.067
0.09
1.20
R2
0.87
0.88
0.86
0.86
0.90
0.75
EN-S
0.73
0.75
0.75
0.74
0.79
0.52
Smith Fry
Obs2
0.053
0.051
0.052
0.139
0.241
2.59
Sim3
0.069
0.065
0.073
0.133
0.159
2.45
R2
0.81
0.84
0.85
0.73
0.73
0.85
EN-S
0.63
0.63
0.68
0.51
0.37
0.72
1 Monthly simulations,2 Observed;3 Simulated.
The observed and simulated monthly surface runoff, Streamflow, sediment, mineral
phosphorus, total phosphorus, and total nitrogen for the calibration and validation period at
the outlet of the Dreisbach and Smith Fry watersheds are shown in Figures 4.7 to 4.10. Based
on thee results, it was assumed that the SWAT model was calibrated and validated for the
study watersheds.
4.6 Discussion
A total of 26 different BMPs were implemented in the Dreisbach watershed while only 6
were implemented in the Smith Fry watershed (see Figure 1). After application of the same
method to represent the BMPs in the watersheds, the same set of calibrated parameters was
obtained for each of the Dreisbach and Smith Fry watersheds, except for USLE practice
factor (USLE_P). This provided further confirmation for the calibration procedure and the
method that was utilized to represent the BMPs. The reason for different optimal (calibrated)
USLE_P parameter is that a major portion of the Dreisbach watershed is cultivated by a
community that practices a more traditional method for farming.
60
-------
(a) 0.4
^ 0.3
5 0.2
- Observed
— Simulated
Precipitation
Time (Month)
(b) 0.4
0.3
0.2
0.1
Observed
Simulated
Precipitation
f-
t-
C3
ft
t-
t-
ft
OO
r-
Time (Month)
(c)
0.2
SO. 15
o
E/3
-d
0.05
0.05
0.1
0.15
Observed Streamflow (m /s)
150
o
150
g
a
o
ta
0.2
Figure 4.7. Measured and Simulated (a) Streamflow, (b) Surface Runoff, and (c) Plot 1:1 Streamflow,
Calibration and Validation Period, Dreisbach Watershed, Indiana.
61
-------
(a) 0.5
0.4
o
0.3
Observed
Simulated
Precipitation
150
o
Time (Month)
(b)
0.5
^ 0.4
o
- Observed
— Simulated
Precipitation
150
a
o
ta
I
Time (Month)
Observed Streamflow fm /s)
Figure 4.8. Measured and Simulated (a) Streamflow, (b) Surface Runoff, and (c) Plot 1:1 Streamflow,
Calibration and Validation Period, Smith Fry Watershed, Indiana.
62
-------
(a)
0.36
0.27
0.18
0.09
uuuuuuuuuuuu
o Observed
Simulated
? Precipitation
150
,0
Ti
t-p
C3
t-
t-
C3
Time (Month)
— Observed
Simulated
0 Precipitation -
(b)
0.6
0.45
P- 0.3
S 0.15
(d)
CS
f
uuuuuuuuuuyuuuu
o Observed
— Simulated
. Precipitation
150
,0
r- t- r-
sic
r-j- t-p
c c
t^
t- t-
Time (Month)
Time (Month)
Time (Month)
Figure 4.9. Measured and Simulated (a) Sediment, (b) Mineral P, and (c) Total P, (d) Total N, Calibration and Validation Period, Dreisbach
Watershed, Indiana.
63
-------
(a) 1.2
Observed
Simulated
Precipitation
(°) 4.8
CS
f
3.6
2.4
1.2
T,
t~~
Time (Month)
o Observed
Simulated
Q Precipitation
150
o
2
PH
r-j-
i
t-
t-p
C3
t-
t-
Time (Month)
(b)
f>
s
3.6
2.4
1.2
(d) 48
36
24
12
W>
t~~
-x> Observed
G Precipitation
150
c
"S
r-
t-
Time (Month)
o
Observed
Simulated
Precipitation
150
o
2
PH
•n
t-
t-
r-j-
t-
t-
Time (Month)
Figure 4.10. Measured and Simulated (a) Sediment, (b) Mineral P, and (c) Total P, (d) Total N, Calibration and Validation Period, Smith Fry
Watershed, Indiana.
64
-------
Section 5.0
Evaluation of Long-Term Impact of Best Management Practices on
Water Quality with a Watershed Model: Role of Spatial Resolution
5.1 Introduction
Implementation of Best Management Practices (BMPs) is a conventional approach for
controlling nonpoint sources of sediments and nutrients. However, implementation of BMPs
is rarely followed by a good long-term data monitoring program in place to study how
effective they have been in meeting their original goals. Long-term data on flow and water
quality within watersheds, before and after placement of BMPs, is not generally available.
Therefore, evaluation of BMPs (especially new ones that have had little or no history of use)
must be necessarily conducted through watershed models. In this regard, various watershed
and field scale models have been used to asses the effectiveness of BMPs (Moore et al., 1992;
Batchelor et al., 1994; Park et al., 1994; Griffin, 1995; Edwards et al., 1996; Mostaghimi et
al., 1997). A number of studies have been performed with the Soil and Water Assessment
Tool (SWAT) model to study the effects of different BMPs on sediment and nutrient
transport within watersheds (Saleh et al., 2000; Santhi et al., 2001b; Kirsch et al., 2002; Saleh
and Du, 2002; Vache et al. 2002; Santhi et al., 2003).
Distributed models partition the watershed into smaller units (subwatersheds/hyrologic
response units, or grids) to represent heterogeneity within the watershed. Delineation of the
watershed, identification of the stream network, and partitioning of the study area into
smaller units is generally accomplished through Geographic Information System (GIS)
databases that help automate this process and make it convenient for modeling purposes.
However, division into subwatersheds and identification of stream networks are extremely
sensitive to spatial scale. The number and size of computational units varies with a user-
defined critical source area (CSA), the minimum area required for channel initiation. Results
of Section 3.0 indicate that the SWAT model sediment and nutrient simulations vary quite
dramatically with the number and size of subwatersheds. Because model outputs are affected
by geomorphologic resolution, the predicted performance of BMPs will be influenced as
well. Thus, examination of the efficacy of BMPs must be conducted in conjunction with
65
-------
studies performed at multiple spatial scales. Previous research on evaluation of the
effectiveness of BMPs has not incorporated the effects of geomorphologic resolution.
In this section, the long-term water quality impact of BMPs is analyzed through the device of
a watershed model. The analysis is conducted in conjunction with investigating the role of
spatial resolution effects resulting from watershed discretization.
5.2 Methodology
Calibration of hydrologic and water quality components of the SWAT model for the
Dreisbach and Smith Fry watersheds was discussed in Section 4.0. Calibrated model
simulations were performed for a 30 year period (1971-2000) for two scenarios (scenarios A
and B). Scenario A corresponded to model results without BMPs, while scenario B simulated
the design variables (sediment and nutrient yields) with BMPs in place. Scenarios A and B
were compared at various watershed discretization levels in order to determine the efficiency
of the BMPs at each watershed discretization level. All of the input parameters for the two
scenarios were exactly the same over the study watersheds with the exception of the
parameters of the hydrologic response units (HRUs) with parallel terraces and field borders,
and the parameters of the channel segments with grassed waterways and stabilization
structures. In scenario A, these parameters were assumed to be the same as the rest of the
study area for which calibrated values are available. The values specified for different BMPs
in Table 4.3 were utilized for these parameters in scenario B. A comparison of model
predictions for these two scenarios enabled the determination of the long-term impacts of the
BMPs on sediment, and nutrient yields at the outlet of the Dreisbach and Smith Fry
watersheds.
5.2.1 Watershed Discretization
SWAT simulations were performed with various watershed configurations for a 30 year time
horizon from 1971 to 2000. The characteristics of the watershed configurations that were
utilized in this part are summarized in Tables 5.1 and 5.2 for the Dreisbach and Smith Fry
watersheds, respectively. The tables include information on the applied critical source area
66
-------
Table 5.1. Properties of the Watershed Configurations Used for the Dreisbach Watershed.
Critical Source Area (km2)
Number of Subwatersheds
Number of HRUs
Drainage Density (km/km2)
Average Subwatershed Area (km2)
0.03
103
647
3.91
0.06
0.05
51
470
3.05
0.12
0.10
29
359
2.28
0.22
0.15
19
301
1.97
0.33
0.36
11
204
1.39
0.57
0.50
5
138
1.22
1.26
1.5
2
91
0.94
3.11
2.5
1
73
0.91
6.23
Table 5.2. Properties of the Watershed Configurations Used for the Smith Fry Watershed.
Critical Source Area (km2)
Number of Subwatersheds
Number of HRUs
Drainage Density (km/km2)
Average Subwatershed Area (km2)
0.03
89
676
4.09
0.08
0.05
63
577
3.27
0.12
0.10
33
429
2.54
0.22
0.15
20
358
2.25
0.37
0.30
12
278
1.76
0.61
0.50
8
239
1.45
0.92
1.5
4
159
0.96
1.83
2.9
1
95
0.65
7.30
(km2) and corresponding number of Subwatersheds, drainage density (km/km2), and average
Subwatershed area (km2). Drainage density is defined as the ratio of total channel length to
the total watershed area. Note that the some of the discretization levels in Tables 5.1 and 5.2
are different from the ones reported in Tables 3.1 and 3.2. The corresponding watershed
configurations used for the Dreisbach watershed are shown in Figures 5.1 and 5.2,
respectively.
5.3 Impact of Best Management Practices on Water Quality
5.3.1 Effects of BMPs on Streamflow
Simulated runoff volume and Streamflow at the outlet of the Dreisbach and Smith Fry
watersheds were not affected by implementation of the BMPs. This was anticipated, because
the BMP selection for the Black Creek project was targeted at sediment and phosphorus
reduction (Lake and Morrison, 1977a; Lake and Morrison, 1977b; Morrison and Lake, 1983).
Parallel terraces, the only type of BMPs in the study watersheds that influence runoff
parameters (see Table 4.3), cover less than 2% and 1% of the Dreisbach and Smith Fry
watersheds, respectively. Thus, their impact on simulated Streamflow at the outlet of the
study watersheds was negligible.
67
-------
CSA= 0.03 (km2)
DD= 3.91 (km/km2)
CSA= 0.05 (km2)
DD= 3.05 (km/km2)
CSA= 0.1 (km2)
DD= 2.28 (km/km2)
CSA= 0.15 (km2)
DD= 1.97 (km/km2)
CSA= 0.36 (km2)
DD= 1.39 (km/km2)
CSA= 0.38 (km2)
DD= 1.3 (km/km2)
CSA= 0.5 (km2)
DD= 1.22 (km/km2)
CSA= 1.5 (km2)
CSA= 2.5 (km2)
DD= 0.94 (km/knT) DD= 0.91 (km/knT)
Figure 5.1. Watershed Configurations Used for the Dreisbach Watershed, Indiana.
68
-------
CSA= 0.03 (km2)
DD= 4.09 (km/km2)
CSA= 0.05 (km2)
DD= 3.27 (km/km2)
CSA= 0.1 (km2)
CSA= 0.15 (km2)
DD= 2.54 (km/km2) DD=2.25 (km/km2)
CSA= 0.30 (km2)
DD= 1.76 (km/km2)
CSA= 0.5 (km2)
DD= 1.45 (km/km2)
CSA= 1.5 (km2)
CSA= 2.9 (km2)
DD= 0.96 (km/km2) DD= 0.65 (km/km2)
Figure 5.2. Watershed Configurations Used for the Smith Fry Watershed, Indiana.
69
-------
5.3.2 Impact of BMPs on Sediment Yield
The effect of watershed discretization on sediment output of the SWAT model at the outlet of
the study watersheds is depicted in Figure 5.3. Under scenario A without the BMPs, average
annual sediment yield at the outlet of the watersheds increased by nearly 200% between the
coarsest and the finest discretization levels. The increase could be due to two processes:
higher sheet erosion from upland areas and/or more intense channel erosion.
The SWAT model employs the Modified Universal Soil Loss Equation (MUSLE) (Eq. 2.5)
to estimate sheet erosion. All of the parameters in the MUSLE equation are estimated for
each HRU with the exception of USLE topographic factor, LS, which is determined for each
subwatershed and applied to the HRUs contained in the subwatershed. The results of this
study presented in Section 3.0 revealed that the weighted average USLE topographic factor,
LS, was reduced by nearly 25% between the coarsest and finest discretization levels. The rate
of reduction plateaued at finer discretization levels. Similar trends were observed for the
computed sheet erosion from upland areas. Consequently, the model predicted that variation
of sheet erosion was not the reason for higher sediment yield at the outlet due to finer
watershed discretization.
When the impacts of the BMPs were not included (scenario A), sediment yield at the outlet
of the watersheds was computed by SWAT to be larger than estimated sheet erosion from
upland areas. Because estimated transport capacity of the channel network (Equation 2.6)
exceeded sediment loads from upland areas. Thus, channel degradation was predicted by the
model to be the dominant channel process and contributed to the sediment yield at the outlet.
Dominance of channel degradation indicated that sediment yield at the outlet would increase
with drainage density, which increased with finer discretization levels (Figure 3.4). At finer
discretization levels, higher drainage density provided longer channel network that would be
subject to channel degradation. This resulted in significantly higher sediment yields at the
outlets. The correlation coefficient between sediment yield at the outlet and drainage density
of the Dreisbach and Smith Fry watersheds was 0.98 and 0.97, respectively. The correlation
was extremely poor for scenario B which simulates the presence of the BMPs.
70
-------
(a)
1.6
-a i-2
§
T3
•3 0.8
>
~£H
I °-4
•3
00
0
0 Scenario A
— — O — Scenario B
20 40 60 80
Number of Subwatersheds
100
(b)
3.6
2.7
2
£
-M
O
S 0.9
-3
00
0
- - -O
0 Scenario A
— — 0- — Scenario B
20 40 60
Number of Subwatersheds
100
(C)
I
-3
0)
t/2
§g
"o
100
75
50
Q-
0 Dreisbach Watershed
O - - Smith Fry Watershed
20 40 60 80
Number of Subwatersheds
100
Figure 5.3. Average Annual Sediment Yield at the Outlet of (a) Dreisbach Watershed, (b) Smith Fry
Watershed, (c) Percent Sediment Reduction. Scenario A: Simulations with No BMP; Scenario B:
Simulations with BMPs in Place.
71
-------
Predicted sediment yield at the outlet was comparatively stable at various discretization
levels when model simulations were performed under scenario B. Transport capacity of the
channel network is a function of the peak channel velocity as indicated in Equation 2.6.
Implementation of grade stabilization structures in the watersheds resulted in lower main
channel slopes while implementation of grassed waterways increased channel resistance,
both of which lowered the peak channel velocity. Subsequently, transport capacity of the
channel network was significantly lower after implementation of the grassed waterways and
grade stabilization structures. With the BMPs, both Dreisbach and Smith Fry watersheds
exhibited the characteristics of "transport-limited" watersheds. For such watersheds,
estimated transport capacity of the channel network is less than sediment loads from upland
areas and sediment deposition is the dominant main channel process. Dominance of channel
deposition indicated that sediment yield at the outlet did not increase with drainage density.
The results presented in Figure 5.3 confirm that sediment yield at the outlet was relatively
insensitive to finer watershed discretization under scenario B when influence of BMPs was
included in the model simulations.
As discussed above, several factors contribute to determine the impact of the BMPs on
abatement of sediment yield at the outlet of watersheds. An overall evaluation was therefore
made by estimating BMP efficacy at any particular discretization level as:
™ 1 • /n/N Model output from scenario A - Model output from scenario B ,_ ,.
Reduction(%) = (5.1)
Model output from scenario A
In the Dreisbach watershed, the efficacy of the BMPs for abating sediment yield was
evaluated to be only 7% at the coarsest discretization level, while the efficacy was nearly
70% at the finest discretization level. The corresponding efficacy values in the Smith Fry
watershed were nearly zero and 50 %, respectively (see Figure 5.3 (c)).
An optimal watershed discretization level for representation of the BMPs and their validity
could be identified from Figure 5.3 at a CSA corresponding to 2 % of the total watershed
areas. The average subwatershed area at this discretization level was approximately 4% of
the total watershed area. There are two major reasons for this recommendation. First, the
72
-------
estimated sheet erosion from upland areas did not vary significantly beyond this
discretization level (see Section 3.0). Second, the asymptotic behavior of the average slope of
channel network (Figure 3.4) indicated that channel degradation and its contribution to the
sediment yield at the outlet also tended to stabilize at finer discretization levels. These trends
are more apparent in the Smith Fry watershed where upstream channel network is relatively
flatter than the one in the Dreisbach watershed.
5.3.3 Impact of BMPs on Nutrient Yields
Figures 5.4 and 5.5 depict simulated total P and total N yields at the outlet of the Dreisbach
and Smith Fry watersheds, respectively. Without BMPs (scenario A), total P predictions by
the SWAT model were 200% higher at the finest discretization level in comparison to the
coarsest level utilized for both watersheds. However, the rate of change stabilized at finer
discretization levels (Figures 5.4(a) and 5.4(b)). Total N predictions of the model exhibited
similar trends as evidenced in Figure 5.5. The installed BMPs were estimated to effectively
reduce total P yield at the outlet of the Dreisbach watershed by 30% when the finest
discretization level was utilized. The reduction (predicted by the SWAT model)
corresponding to the coarsest discretization level was 0% (see Figure 5.4(c)). The results
presented in Figure 5.5(c) demonstrate that simulated impact of BMPs in alleviating total N
yield at the outlet of the Dreisbach watershed also depended on the utilized watershed
discretization level. A 25% reduction was obtained at the finest discretization level while the
reduction was negligible at the coarsest level. Similar trends were observed for simulated
reduction of total P and total N in the Smith Fry watershed as depicted in Figures 5.4(b) and
5.5(b), respectively. From Figures 5.4(c) and 5.5(c), an optimal critical source area
corresponding to 2% of total areas of the respective watersheds continues to serve as an
appropriate discretization level for evaluation of effectiveness of the BMPs for reduction of
total P and total N. This was partly anticipated because the same optimal discretization level
was identified earlier for sediment yield.
The reduction in total P load was consistent with the reduction of sediment yield at the outlet
of the watersheds. This was anticipated for two reasons. First, in relatively small watersheds
like Dreisbach and Smith Fry, the role of in-stream nutrient processes that are simulated by
73
-------
(a)
1 2.4
|P 1.8
| 1-2
PH
1 0.6
0
_ - -e- -
0 Scenario A
— — O — Scenario B
_ - - - -O
20 40 60 8(
Number of Subwatersheds
100
10
IT 7.5
i
2.5
— -o
0 Scenario A
— — G- — Scenario B
20 40 60
Number of Subwatersheds
100
(C)
PH
O
I
100
75
50
25
0 Dreisbach Watershed
O- - - Smith Fry Watershed
0 20 40 60 80
Number of Subwatersheds
100
Figure 5.4. Average Annual Total P Yield at the Outlet of (a) Dreisbach Watershed, (b) Smith Fry
Watershed, (c) Percent Sediment Reduction. Scenario A: Simulations with No BMP; Scenario B:
Simulations with BMPs in Place.
74
-------
(a)
20
J 15
3 10
I
0 Scenario A
Q Scenario B
20 40 60 80
Number of Subwatersheds
100
(b)
100
75
3 50
0)
>
I -
Scenario A
— — 0- — - ScenarioB
20 40 60
Number of Subwatersheds
100
(C)
100
75
50
0 Dreisbach Watershed
G- - - Smith Fry Watershed
_- -e-
_ - -o
20 40 60 80 100
Number of Subwatersheds
Figure 5.5. Average Annual Total N Yield at the Outlet of (a) Dreisbach Watershed, (b) Smith Fry
Watershed, (c) Percent Sediment Reduction. Scenario A: Simulations with No BMP; Scenario B:
Simulations with BMPs in Place.
75
-------
SWAT, such as algal decay on phosphorus yield, is negligible compared to soil loss from
upland areas and channel erosion. In such watersheds, it can be claimed that sediment and
nutrient yields are correlated. The correlation coefficient between observed sediment yield
and nutrient loads (Table 2.5) at the outlets of the study watersheds was 0.72. Moreover, the
BMPs installed in the study watersheds were basically sediment control structures. The
impact of the BMPs on nutrient loads was as a consequence of reduction of sediment yield.
5.4. Field Scale versus Watershed Scale Evaluation
The impacts of the BMPs in the Dreisbach and Smith Fry watersheds were examined at two
spatial scales based on their functionality. Parallel terraces and field borders are implemented
to reduce soil loss from upland areas. Therefore, their efficacy may be evaluated at a HRU
(or field) scale as well as a watershed scale. The effect of grassed waterways and grade
stabilization structures must be discussed at a larger watershed scale because they are
implemented in channels and their effects can not be felt on upland areas. Model predictions
at the finest discretization level, i.e., critical source area equal to 0.03 (km2) were applied to
compare the efficacy of the BMPs at watershed and field scales. The sediment, total P, and
total N reduction rates determined by comparing model simulations with and without
inclusion of parallel terraces and field borders are summarized in Table 5.3. In this table, the
presented results at HRU scale correspond to reduction rates of model outputs averaged over
the particular field-plots where the parallel terraces and field borders have been implemented
(shown in Figure 2.2). At a watershed scale, these BMPs did not contribute to appreciable
sediment, total P, and total N reductions. This was anticipated because they have been placed
Table 5.3. Reduction of Sediment, Total P, and Total N loads Resulted
from Implementation of Parallel Terraces and Field Borders.
Watershed
Dreisbach
Smith Fry
Scale
HRU1
Watershed
HRU1
Watershed
% Reduction
Sediment
57
2
45
1
Total P
50
2
30
1
Total N
55
2
35
1
Obtained by averaging the reduction rates over the HRUs (i.e. fields)
where the parallel terraces and field borders have been installed.
76
-------
to target small portions of the study watersheds. On the contrary, sediment, total P, and total
N loadings from the fields where the terraces and field borders were installed decrease by
nearly 57%, 50%, and 55% in the Dreisbach watershed and by 45%, 30%, and 35% in the
Smith Fry watershed, respectively, which implies that land owners would substantially
benefit from their implementation, if the regulation were to be imposed immediately
downstream of the upland area.
Grassed waterways and grade stabilization structures would likely be more beneficial to
development of a sediment and nutrient TMDL at the outlet of the study watersheds. As
illustrated in Figures 5.3(c), 5.4(c), and 5.5(c), sediment, total P, and total N yields at the
outlet of the Dreisbach watershed decreased by nearly 70, 25, and 30% as a result of the
installation of the waterways and stabilization structures. The corresponding values in the
Smith Fry watersheds were approximately 50, 30, and 35%.
Interestingly, although the number of the BMPs implemented in the Smith Fry watershed was
significantly less than that for the Dreisbach watershed, the estimated sediment and nutrient
reduction rates were comparable. This indicates that not only the number of the BMPs, but
also their location in the watershed plays a significant role. Our assessment of the impact of
individual BMPs revealed that the two grade stabilization structures at the downstream
portion of the channel network in the Smith Fry watershed were the primary reason for such
reduction rates. These structures lowered the transport capacity of upstream channel
segments that resulted in deposition of a large amount of the sediments and nutrients in the
channel network. Thus, the simulated sediment and nutrient yields at the outlet were
dramatically reduced. This would imply that for maximum benefits, the BMPs should be
placed as close upstream as possible to where the regulation will be imposed. It also suggests
that with proper implementation of BMPs, managers are able to exert enough control to
convert a supply-limited watershed to a transport-limited one.
5.5. Conclusions
For the study watersheds, sediment, total phosphorus, and total nitrogen outputs of the
SWAT model were highly influenced by watershed discretization before representation of
77
-------
the BMPs. Predicted dominance of channel degradation in simulations without BMPs
resulted in an increase of these outputs with drainage density, which increased with finer
discretization levels.
The implemented grassed waterways and grade stabilization structures appreciably reduced
the transport capacity of the channel network of the watersheds. After implementation of the
BMPs, sediment deposition was the dominant channel process in the study watersheds. The
predicted sediment yield at the outlet of the study watersheds was relatively stable and did
not vary with finer discretization.
The predicted reduction of sediment and nutrient yields as a result of implementation of the
BMPs were insignificant when more coarse levels of discretization were applied. Utilization
of the finer discretization levels resulted in substantial sediment and nutrient reductions
according to the model. An optimal discretization level at a critical source area corresponding
to 2% of the total watershed area was identified to be adequate for representation of the
BMPs and assessment of their validity. Study results indicated that a proper assessment of
the efficacy of the BMPs must be conducted in conjunction with multiple watershed
discretization levels.
The management implications of this study were found to be scale dependent.
Implementation of parallel terraces and field borders significantly alleviated estimated
sediment and nutrient loadings from the fields where they have been installed. The reduction
was negligible at the outlet of the study watersheds. While land owners may identify parallel
terraces and field borders as being very effective for controlling downstream discharges,
watershed managers may not appreciate their impact on water quality at the outlet of the
Dreisbach and Smith Fry watersheds. Based on the SWAT model simulations, at a watershed
scale, grassed waterways and grade stabilization structures appeared to more effectively
reduce sediment and nutrient yields at the outlets. In particular, grassed waterways and grade
stabilization structures located in the downstream portion of the channel network increased
channel deposition in upstream segments. It may be concluded that placement of the BMPs
plays an important role in improving the water quality at the outlet of the watersheds.
78
-------
Identification of the most appropriate locations for implementation of abatement strategies
requires a better understanding of control processes in a watershed.
Since different sets of calibrated parameters may be obtained from the calibration procedure,
applying sensitivity and uncertainty analysis techniques would be valuable for identification
of control processes and key management actions such as sheet erosion, channel degradation,
and channel deposition within a watershed. In a watershed where channel degradation is the
dominant main channel process, implementation of grassed waterways and grade
stabilization structures would be highly successful in reducing sediment and nutrient loads to
the extent of converting a supply-limited watershed to a transport-limited one. Application of
BMPs such as parallel terraces and field borders would be more successful for watersheds
where upland areas are the dominant sources of sediments and nutrients. Their role in
changing the overall nature of the watershed is likely to be minimal.
The results of this study, which was conducted on small watersheds, should be verified by
other studies focused on evaluation of effectiveness of BMPs at various watershed
discretization levels. Sediment and nutrient yields from larger watersheds may exhibit
different trends with watershed discretization. The method presented in this paper for
evaluation of effectiveness of BMPs at various discretization levels is recommended for other
watershed studies because uncertainties resulting from spatial resolution deserve more
attention than has been devoted to them in the past.
79
-------
Section 6.0
Source Identification
6.1 Introduction
Identification of sediment and nutrient sources within a watershed has many important
implications for watershed management. Once nonpoint sources of sediment and nutrients
are identified, managers will be able to examine whether abatement strategies such as
implementation of Best Management Practices (BMPs) effectively reduce sediment and
nutrient losses. The question of where to place BMPs for maximum benefits also depends on
being able to identify contaminant sources. Major sources of sediment and nutrients within a
watershed can be categorized into sheet erosion from upland areas, and channel degradation.
Anthropogenic activities are known to contribute to both of these sources of erosion.
Sheet erosion results in removal of a fairly uniform layer of sediment and agrichemicals that
adhere to sediment particles from upland areas. Tillage and fertilizer application are perhaps
the most important anthropogenic activities that directly increase nutrient loads from upland
areas. An accurate estimate of sediment and nutrient loads from upland areas will be
beneficial to land owners and field-scale managers as well as watershed-scale managers.
In addition to sheet erosion, channel degradation and other channel processes that influence
nutrient yield at the outlet have significant roles in watershed-scale management. Channels
within a watershed not only serve as a conduit for movement of contaminant-laden
sediments, but may also act as a source because of erosion from streambeds and bank
erosion. Channel erosion may significantly contribute to sediment and nutrient yields at the
outlet, especially in supply-limited watersheds where transport capacity of channel network
is larger than sediment concentration in channel flow due to sheet erosion from upland areas.
Algal growth, transformation and respiration rates, and other in-stream processes may
influence transport of organic and inorganic forms of phosphorus and nitrogen. A proper
assessment of sediment and nutrient sources should include the influence of these activities.
80
-------
Most of the available data from monitoring programs have been collected at the outlet of
watersheds. These data are not adequate to directly identify nonpoint sources within
watersheds since sediment and nutrient movement over a watershed tends to be fairly
dynamic, often behaving in a nonlinear fashion. Flow, sediment, and nutrient monitoring
programs should be conducted at both field and watershed scales to help decision making and
management at various spatial scales. In doing so, however, there are two issues that need to
be addressed. First, installation, maintenance, and operation of monitoring stations are
usually expensive and time consuming. It is almost impossible to conduct such a program for
every single system of interest. Second, analysis of historical data may not be adequate for
evaluation of the impact (s) of certain management actions on the system, especially the ones
that have not been implemented yet.
Modeling studies not only provide a versatile tool for assessing the future of a given system
under various scenarios, but can also be used to examine whether a certain future state is
attainable for the system. Thus, they can be used for development and implementation of a
TMDL for the design variable (s) of concern. In this study, the SWAT model was selected to
assess the impact of implementation of various best management practices (BMPs) in the
Dreisbach and Smith Fry watersheds in Indiana. Good soil, land use, and management data
are available for the study watersheds and can be utilized by BASINS to prepare the required
input files for SWAT simulations. Application of SWAT for source identification and
evaluation of performance of BMPs are discussed in this section.
6.2 Objective
In Section 3.0, a method was developed to obtain an optimal watershed discretization level
such that further refinement would not change SWAT computations for sediment and
nutrient loadings from upland areas. The conditions under which such an "optimal"
resolution would be available were also determined. The goal of this section is to develop
sediment and nutrient maps for the study area. These maps will be generated based on best
resolutions available for soil, land use and management data and are indicators of the sources
within the study area. Two scenarios are examined: sediment and nutrient sources before
implementation of BMPs (scenario A), and sediment and nutrient sources after
81
-------
implementation of BMPs (scenario B). Furthermore, the effectiveness of BMPs in reducing
sediment and nutrient loads will be evaluated.
6.3 Methodology
To fulfill the source identification objectives, SWAT was calibrated and validated for the
Dreisbach and Smith Fry watersheds. The results of calibration procedure were presented in
Section 4.0. These two watersheds are located in the Black Creek watershed almost 10 km
apart. Calibration of the SWAT model resulted in the same model inputs for both watersheds
except for USLE practice factor (USLE_P). In Section 3.0, an optimal watershed
discretization level was identified for both watersheds. It was shown that SWAT
computations are not sensitive to critical source area (CSA) for resolutions finer than 20 (ha).
The experience gained form previous sections provides more confidence in application of
SWAT computations for source identification purposes. The mechanisms utilized by SWAT
to compute sheet erosion and channel processes were explained in detail in Section 2.
6.3.1 Sediment and Nutrient Source Maps
SWAT model simulations were performed over a 30 year time period from January 1970 to
December 2000. Average annual quantities predicted for subwatersheds were utilized as
sediment and nutrient source indicators. Figures 6.1-6.2 depict the source maps for sediment,
total P and total N loads from upland areas of the Dreisbach watershed before and after
implementation of BMPs, respectively. Similar maps are presented in Figures 63-6.4 for the
Smith Fry Watershed.
Various segments of channel network can be sources of sediment and nutrient. Sediment and
nutrient loads from the channel networks of Dreisbach and Smith Fry before and after
implementation of BMPs are presented in Figures 6.5-6.8. A positive value indicates that the
channel segment serves as a source of sediment or nutrient while a negative value is an
indicator of sediment or nutrient deposition in the channel segment.
82
-------
(a)
(b)
Sediment (t/ha)
0 - 0.25
IBJ 0.25 - 0.5
1.5-3.0
(c)
Total P (kg/ha)
0-0.15
0.15-0.3
0.3 - 0.45
0.45 - 0.6
0.6-1.5
Total N (kg/ha)
20 - 40
6 Kilometers
Figure 6.1. Simulated Average Annual Loads Generated at Upland Areas before Implementation of BMPs for Dreisbach Watershed, 1971-2000:
(a) Sediment, (b) Total P, and (c) Total N.
83
-------
(a)
(b)
$ Parallel Terracei
# Field Border
Sediment (t/ha)
0 - 0.25
0.25 - 0.5
0.5 -1
1-1.5
1.5-3.0
(c)
$ Parallel Terracen
# Field Border
Total P (kg/ha)
0-0.15
0.15-0.3
0.3 - 0.45
0.45 - 0.6
0.6-1.5
$ Parallel Terracei]
# Field Border
Total N (kg/ha)
0-5
I5-8
8-13
13-20
20-40
N
6 Kilometers
Figure 6.2. Simulated Average Annual Loads Generated at Upland Areas after Implementation of BMPs for Dreisbach Watershed, 1971-2000:
(a) Sediment, (b) Total P, and (c) Total N.
84
-------
Sediment (t/ha)
0-0.4
0.4 - 0.8
0.8-1.2
1.2-1.6
1.6-2
Total N (kg/ha)
0-5
5-10
10-15
15-20
20-30
Total P (kg/ha)
0.2-0.3
0.3-0.4
0.4-0.5
0.5-0.6
0.6-0.9
Figure 6.3. Simulated Average Annual Loads Generated at Upland Areas before Implementation of BMPs for Smith Fry Watershed, 1971-2000:
(a) Sediment, (b) Total P, and (c) Total N.
85
-------
$ Parallel Terrace]
# Field Border
Sediment (t/ha)
0-0.4
0.4 - 0.8
0.8-1.2
1.2-1.6
1.6-2
Parallel Terrace
# Field Border
Total N (kg/ha)
0-5
5-10
10-15
15-20
20-30
$ Parallel Terrace
# Field Border
Total P (kg/ha)
0.2 - 0.3
0.3 - 0.4
0.4 - 0.5
0.5 - 0.6
0.6 - 0.9
6 Kilometers
Figure 6.4. Simulated Average Annual Loads Generated at Upland Areas after Implementation of BMPs for Smith Fry Watershed, 1971-2000:
(a) Sediment, (b) Total P, and (c) Total N.
86
-------
(a)
Sediment (t/km)
A/ -300 -0
0-50
A/50-100
A/100-300
A/300 - 550
1 I Dreisbach
Total P (kg/km)
^o-o
°-30
-70
yy/70-140
y^/
| I
140-310
Dreisbach
Total N (kg/km)
A/-1000-0
0-300
A/300-500
'500-1000
'1000-3500
Dreisbach N
6 Kilometers
Figure 6.5. Simulated Average Annual Loads from Channel Network before Implementation of BMPs for Dreisbach Watershed, 1971-2000:
a. Sediment, b. Total P, and c. Total N.
87
-------
(a)
(b)
N Grassed Waterway^
N Grade Stab. Struc.
Sediment(t/km)
A/ -300 - 0
0-50
A/50-100
A/100-300
A/300-550
| I Dreisbach
(c)
N Grassed Waterway^
N Grade Stab. Struc.
Total P (kg/km)
- 30
A/ 30 -70
/V/70-140
A/ 140 -310
| I Dreisbach
N Grassed WaterwayJ
N Grade Stab. Struc.
Total N (kg/km)
/\/-1000-0
0-300
A/300-500
500-1000
1000-3500
Dreisbach
N
6 Kilometers
Figure 6.6. Simulated Average Annual Loads from Channel Network after Implementation of BMPs for Dreisbach Watershed, 1971-2000:
a. Sediment, b. Total P, and c. Total N.
-------
(a)
Total N (kg/km)
0-750
A/750-1500
"A/1500-2250
^^2250-3000
3000-5000
Total P (kg/km
0-50
Sediment (t/km)
0-5
A/5-10
10-50
A/50-100
100-1500
/VY 100 - 200
/\/ 200 -300
Figure 6.7. Simulated Average Annual Loads from Channel Network before Implementation of BMPs for Smith Fry Watershed, 1971-2000:
1971-2000: a. Sediment, b. Total P, and c. Total N.
89
-------
(a)
(c)
N Grassed Waterway
N Grade Stab. Struc.
Total P (kg/km)
A/-10000-0
0-50
50-100
A/100-200
A/200-300
N Grassed Waterway
N Grade Stab. Struc.
Sediment (t/km)
/V-70-0
0-5
5-10
A/ 10-50
50-100
100 -1500
N Grassed Waterway;;
N Grade Stab. Struc
Total N (kg/km)
0-750
A/ 750 -1500
A/1500 - 2250
A/2250-3000
3000-5000
6 Kilometers
Figure 6.8. Simulated Average Annual Loads from Channel Network after Implementation of BMPs for Smith Fry Watershed, 1971-2000:
a. Sediment, b. Total P, and c. Total N.
90
-------
6.3.2 Long-Term Performance of BMPs
Long-term impacts of the field borders, parallel terraces, grassed waterways, and grade
stabilization structures on water quality of the Dreisbach and Smith Fry watersheds were
discussed in Section 5.0. Here these impacts are assessed for the particular HRUs and
channel segments where the BMPs have been installed.
Table 6.1 includes simulated sediment, total P, and total N loads from particular fields in the
Dreisbach and Smith Fry watersheds with field borders for scenarios A and B (defined in
Section 5.2). Scenario A corresponded to model results without BMPs, while scenario B
simulated the design variables (sediment and nutrient yields) with the particular BMP in
place. The location of the field borders and the number of the subwatersheds where they are
located is presented in Figures 6.2 and 6.4 for the Dreisbach and Smith Fry watersheds,
respectively. The results of model predictions presented in Table 6.1 indicate that
implemented field borders resulted in a nearly 60% reduction of sediment and nutrient loads
from the corresponding fields in the Dreisbach watershed. In Smith Fry watershed, there was
only one field border that reduced sediment, and total P loads by 50% and total N by 40%.
The impact of parallel terraces on sediment and nutrient loads from the field where they have
been installed is presented in Table 6.2. Based on model simulations, the average reduction
Table 6.1. Impact of Field Borders on Sediment, Total P, and Total N Loads at A Field Scale.
Watershed
Dreisbach
Smith Fry
Location
(subwatershed)
4
27
30
31
35
37
38
20
Sediment (t/ha)
Seen. A
0.262
0.136
1.321
0.746
0.645
0.111
0.540
1.038
Seen. B
0.102
0.056
0.532
0.304
0.251
0.046
0.212
0.532
Total P (kg/ha)
Seen. A
0.3
0.3
0.7
0.5
0.5
0.2
0.4
0.5
Seen. B
0.1
0.1
0.3
0.2
0.2
0.1
0.2
0.3
Total N (kg/ha)
Seen. A
4.0
2.4
20.1
15.0
10.7
1.8
9.3
16.5
Seen. B
1.6
1.0
8.1
6.1
4.2
0.7
3.7
9.6
91
-------
Table 6.2. Impact of Parallel Terraces on Sediment, Total P, and Total N Loads at A Field Scale.
Watershed
Dreisbach
Smith Fry
Location
(subwatershed)
8
11
16
2
3
Sediment (t/ha)
Seen. A
0.719
0.347
2.957
1.910
1.313
Seen. B
0.350
0.173
1.561
0.812
1.073
Total P (kg/ha)
Seen. A
0.4
0.3
1.3
1.0
0.6
Seen. B
0.3
0.2
0.8
0.5
0.5
Total N (kg/ha)
Seen. A
11.7
5.2
36.2
27.1
19.1
Seen. B
6.3
3.1
21.2
14.3
15.7
of sediment, total P, and total N loads from the corresponding fields in the Dreisbach
watershed was approximately 50%, 30%, and 40%, respectively. The corresponding
reduction of sediment, total P, and total N loads in the Smith Fry watershed were nearly 40%.
It was discussed in Section 5.0 that based on SWAT computations, grassed waterways and
grade stabilization structures were the effective BMPs in the study watersheds, mainly
because they significantly lower the transport capacity of the channel segments where they
have been implemented. Table 6.3 shows the predicted sediment loads from the channel
segments before and after inclusion of the grassed waterways and grade stabilization
structures in the study watersheds. The values in the table were computed by subtracting
sediment loads at the beginning of the channel segment from the ones at the end. It is evident
that based on SWAT computations, sediment erosion was the dominant channel process in
most of these segments prior to implementation of the BMPs. After implementation of the
grassed waterways and grade stabilization structures (see Figures 4.6 and 4.8 for their
locations), channel deposition was dominant in most of these segments. In Table 6.3, a
positive value refers to channel degradation (erosion), while a negative value indicates
channel deposition. It is observed that implementation of grade stabilization structures almost
in all of the cases (except for the one installed on channel segment 29 in the Dreisbach
watershed) resulted either in channel deposition or a significant reduction of channel erosion
in both watersheds. Model predictions imply that implementation of grassed waterways and
stabilization structures on the channel segments in equilibrium, i.e., no channel degradation
and/or channel erosion in the segment, did not change channel characteristics. This indicated
that sediment loads in and out of the channel segment were the same even after installation of
these BMPs.
92
-------
Table 6.3. Impact of Grassed Waterways on Sediment Loads (t/km) from Channel Segments.
BMP
Grassed
Waterway
Grade
Stabilization
Structure
Dreisbach
Channel
Segment
12
23
24
25
29
3
17
19
22
26
28
29
32
33
36
37
Seen. A
2.403
0
0
0
0
25.684
6.042
31.210
432.699
60.436
286.374
0.000
115.939
-164.403
185.232
522.489
Seen. B
-39.779
0
0
0
0
-24.766
-3.241
-240.008
16.986
-121.837
-9.483
0.000
11.069
-231.304
14.407
-283.529
Smith Fry
Channel „ . „ _
„ ^ Seen. A Seen. B
Segment
6 1.944 -6.389
20 28.522 -18.412
27 1378.12 238.327
It should be noted that implementation of sediment reduction BMPs at a particular point of a
channel network does not only affect the upstream channel segments. If implementation of a
grassed waterway or grade stabilization structure result in significant reduction in
concentration of sediments in channel flow, cannel degradation may happen in downstream
segments. This can be observed in channel segment 35 in the Dreisbach watershed (refer to
Figures 6.5a and 6.6a). Slope of channel network in this part of the watershed is very small.
Therefore, estimated transport capacity of this segment is very low. Before implementation
of the grassed waterways and grade stabilization structures upstream of this channel segment,
simulated sediment concentration in channel flow was more than its transport capacity.
However, after implementation of these BMPs, sediment concentration in channel flow was
significantly reduced and was smaller than transport capacity of the channel segment. Thus,
model simulations indicated that channel degradation occurred after implementation of the
BMPs. This would imply that for maximum benefits, the BMPs should be placed as close
upstream as possible to where the regulation will be imposed.
93
-------
6.4 Conclusions
A calibrated and validated SWAT model was utilized to identify sediment and nutrient
sources within the Dreisbach and Smith Fry watersheds. Average annual quantities predicted
by the model were used to generate sediment and nutrient source maps. The results of this
study based on model simulations revealed that before implementation of BMPs, channel
processes, namely streambed or/and bank erosion, would contribute to sediment, total P and
total N loads at the outlet of the study watersheds. It was observed that implementation of the
BMPs in the Dreisbach watershed would result in a significant reduction of sediment and
nutrient yields. These reductions would be mainly due to implementation of grade
stabilization structures and grassed waterways that appreciably reduce the transport capacity
of channel network. Parallel terraces and field borders would be more effective at a farm
scale (i.e. subwatershed scale) while their effect on the sediment and nutrient yields at the
outlet would be relatively small. Also, the results indicate that spatial scale has a significant
role in the appraisal of the effectiveness of BMPs.
The attributes of the selected watershed and the watershed model, upland sediment and
nutrient loading, in-stream processes, or a combination thereof will control estimates of
sediment and nutrient yield at the outlet of the watershed. Calibration of a model while
essential to sediment and nutrient source identification, may not be sufficient since it usually
does not result in a unique set of input parameters. Performing an uncertainty analysis will be
critical for accurate interpretation of model results. Key control processes and management
actions can be identified by applying a proper sensitivity analysis.
In conclusion, application of watershed models such as SWAT in identification of sediment
and nonpoint sources requires two major steps. First, the model should be calibrated and
validated for the study area. Model simulations are performed to predict sediment and
nutrient loads from upland areas and at the outlet. A comparison of the two will provide a
good assessment of control processes in the watershed. Furthermore, a detailed sensitivity
and uncertainty analysis will be useful in confirmation and interpretation of the results from
the previous step. In this study, sediment and nutrient sources within Dreisbach and Smith
Fry watersheds were identified by utilizing SWAT simulations after model calibration. A
94
-------
thorough sensitivity analysis is required for further verification and interpretation of the
results.
95
-------
Section 7.0
Conclusions
The regulations stipulated for the Total Maximum Daily Load (TMDL) program require all
of the states to identify impaired water bodies within the states, and develop abatement
strategies for the impairment (s) of concern. NRC (2001) reported that implementation of
TMDL program is pivotal in securing the nation's water quality goals and should be the
target of management and decision making in watershed systems. Successful development of
the TMDL program depends to a large extent on the ability of managers and analysts (i) to
understand the transport and fate of contaminants within watersheds, and (ii) to evaluate the
outcome (s) of a certain management action on water quality of the system. Modeling proves
to be a useful tool for such purposes. Simulation models not only facilitate contemplating the
future of a given system under various management scenarios, but can also be used to
examine whether a certain future state is attainable for the system.
According to the latest list submitted to EPA, sediment and nutrients are the most
encountered cases of impairment in watersheds. Natural sources of sediment and nutrients
are primarily upland areas, including both sheet and rill erosion, and channel segments under
streambed and/or bank erosion. Anthropogenic activities are known to contribute to both of
these sources of sediments and nutrients. Over the past 30 years, Best Management Practices
(BMPs) have been installed in watersheds to reduce sediment and nutrient from various
sources. However, their implementation has been rarely followed by a long-term monitoring
program to study the performance of the BMPs. In the absence of good measured data,
watershed models can be utilized for such an evaluation. In this study, performance of
various BMPs in abatement of sediments and nutrients in two agricultural watersheds in
Indiana was investigated through the device of a watershed model. The management
implications of the study are site-specific and may not hold for other watershed systems.
However, the developed methodology for evaluation of the efficacy of BMPs can be applied
for other watersheds.
96
-------
7.1 Management Implications
Four different types of agricultural BMPs were installed in the Dreisbach and Smith Fry
watersheds, including field borders, parallel terraces, grassed waterways, and grade
stabilization structures in the early 1970's. A modeling framework was developed to
represent the BMPs with the Soil and Water Assessment Tool (SWAT) model and evaluate
their long-term impact (s) on the water quality of the study watersheds. First, the soundness
of various components of the SWAT model was evaluated through peer review. SWAT has
been widely used for streamflow, sediment, and nutrient simulations for a variety of
watersheds of different sizes (5-100,000 km2) throughout the world. Second, a certain level
of credibility for model computations was established by calibration of model parameters for
the study watersheds based on a set of observed data, and further confirming the validity of
model simulations for another dataset. Based on the function of the BMPs and hydrologic
and water quality processes that are modified by their implementation, corresponding model
parameters were altered to encode the impact of the BMPs on flow, sediment, and nutrient
simulations of the model. Finally, the calibrated model was used for comparison of two
scenarios, scenario A and scenario B. Scenario A represented model predictions over 1971-
2000 time period without inclusion of the BMPs, while scenario B reflected model
predictions for the same period with BMPs. These scenarios were compared at a field scale
as well as a watershed scale to evaluate the impact of the BMPs on sediment and nutrient
yields.
Field borders and parallel terraces were installed on the upland areas and were intended to
reduce sheet erosion from upland areas. Based on model predictions, implementation of these
BMPs would reduce sediment and nutrient loads from the fields where they have been
installed by nearly 50%. However, their impacts would not be felt at the outlet of the study
watersheds, primarily because they have been installed to influence less than 2% of total area
of the Dreisbach and Smith Fry watersheds.
Grassed waterways and grade stabilization structures would be the more effective BMPs at
watershed scales. Comparison of scenarios A and B revealed that implementation of these
BMPs would significantly reduce sediment and nutrient yields at the outlets of the
97
-------
watersheds. Under scenario A, the watersheds tended to behave like a supply-limited
watershed, i.e., simulated sediment and nutrient loads from upland areas were less than
estimated transport capacity of the channel network. Thus, the channel network would
undergo bed and bank erosion. The transport capacity of the channel networks would be
significantly lowered due to implementation of grassed waterways and grade stabilization
structures. Under scenario B, the study watersheds would show the characteristics of a
transport-limited watershed, i.e., simulated sediment and nutrient loads from upland areas
would be more than estimated transport capacity of the channel network. Thus, channel
deposition would be the overall dominant main channel process in the watersheds, indicating
that the channel network would not contribute to the sediment and nutrient yields at the
outlets. It was also observed that the grade stabilization structures that have been placed at
the downstream portion of the channel network would be the most effective ones. This would
imply that for maximum benefits, these BMPs should be placed as close upstream as possible
to where the regulation will be imposed
In a watershed where channel degradation is the dominant main channel process,
implementation of grassed waterways and grade stabilization structures would be highly
successful in reducing sediment and nutrient loads, perhaps to the extent of converting a
supply-limited watershed to a transport-limited one. Application of BMPs such as parallel
terraces and field borders would be more successful for watersheds where upland areas are
the dominant sources of sediments and nutrients.
7.2 Modeling Implications
Utility of a distributed-parameter watershed model for simulating sediments and nutrients
was discussed in this study. Also, a process-based method for representation of Best
Management Practices (BMPs) was developed. SWAT model was selected not only because
the model has both sediment and nutrient components in addition to hydrologic components,
but because of the model structure that allows representation of BMPs in a process-based
fashion. Similar to other distributed-parameter models, SWAT subdivides the watershed into
sub-units including subwatersheds and channel segments for computations. Further,
subwatersheds are partitioned into hydrologic response units (HRUs) that are used for
98
-------
computation of runoff, sheet erosion, and nutrient loads from upland areas. Thus,
representation of BMPs such as field borders and parallel terraces that are installed in a
particular field to reduce runoff, sediment, and nutrient loads can be easily done within
SWAT by altering appropriate model parameters for the corresponding HRU (field). These
estimated loads are routed through the channel network that is divided into various segments
for computation purposes. Subdivision of the channel segment into smaller segments
provides the option for alteration of model parameters for the particular segments with BMPs
such as grassed waterways and grade stabilization structures.
Evaluation of the performance of BMPs can be facilitated by utilizing distributed-parameter
watershed models that partition the watershed into fields (HRUs) and channel segments for
computations. In doing so, however, model computations are rendered subjective to the level
of watershed discretization. The results of this study revealed that sediment and nutrient
simulations of the SWAT model may be very sensitive to the number and size of
subwatersheds as well as the drainage density of the channel network (drainage density of the
channel network is defined as the ratio of length of channel network to the total watershed
area). As a result, a proper assessment of the efficacy of the BMPs must be conducted in
conjunction with multiple watershed discretization levels.
While size of subwatersheds influences sediment and nutrient loads from upland areas,
drainage density of the channel network is important in computing these loads eroded from
the bed and bank of the channel network. In Section 3.0, two indices i.e., Erosion Index and
Area Index were recommended to be applied for estimation of a proper watershed
discretization level for sheet erosion computations. These indices were derived based on the
Modified Universal Soil Loss Equation (MUSLE), which is used by SWAT for estimation of
sheet erosion from upland areas. The applicability of the two indices was confirmed for the
Dreisbach and Smith Fry watersheds. It was concluded that in transport-limited watersheds
where channel network does not contribute to the sediment and nutrient yields at the outlet,
application of the Erosion Index and Area Index is likely adequate for obtaining a proper
watershed discretization level. However, when the channel network contributes to the
sediment and nutrient loads at the outlet, an accurate estimation of the length and
99
-------
characteristics of the channel network is required. Examination of the effect of watershed
discretization on average slope of channel network was found useful for such an estimation.
7.3 Closing Remarks
A methodological framework for representation of BMPs with a watershed model was
developed in this study. A watershed model was selected and calibrated for the study area,
and was utilized for predicting the impact(s) of implementation of BMPs on water quality.
Calibration procedure is often used for establishing credibility for simulations of a model.
This common practice embraces the critical issue of non-uniqueness of the optimal
(calibrated) set of model parameters. The hydrological and water quality processes that are
represented by the model parameters may be affected by the choice of the calibrated
parameter data set. More credibility in the developed methodology could be established by
employing uncertainty techniques. The uncertainty of input parameters should be elicited and
encoded in the modeling approach to provide a better understanding of the processes that
control transport and fate of sediments and nutrients in a watershed for a comprehensive
management and decision making.
In addition to including uncertainty of input parameters in the modeling approach, an
accurate estimation of drainage density of the channel network is required. This problem is
complex because drainage density varies with different storm events. Application of remote
sensing techniques for extraction of the characteristics of the channel network from aerial
photos and satellite images at the time of large storm events as well as low flow conditions
would be helpful. Also, hydrologic and water quality monitoring programs at various
locations of the channel network should be conducted for such purposes.
100
-------
BIBLIOGRAPGY
Arnold, J.G., Muttiah, R.S., Srinivasan, R., Allen, P.M., 2000. Regional estimation of base
flow and groundwater recharge in the Upper Mississippi river basin, Journal of Hydrology,
227(2000): 21-40.
Arnold, J.G. and P.M. Allen, 1999. Automated methods for estimating base/low and ground
water recharge from streamflow records, Journal of the American Water Resources
Association, 35(2): 411-424.
Arnold, J.G., Srinivasan, R., Muttiah, R.S., Williams, J.R., 1998. Large area hydrologic
modeling and assessment part I: Model development, Journal of the American Water
Resources Association, 34(1): 73-89.
Arnold, J.G. and P.M. Allen, 1996. Estimating hydrologic budgets for three Illinois
watersheds, Journal of Hydrology, 176(1996): 51-11.
ASAE, 2003. Design, layout, construction and maintenance of terrace systems, ASAE
Standards S268.4 FEB03, St. Joseph, MI.
Batchelor, W.D., Diliaha, T.A., Wolf, M.L., Heatwole, C.D., Mostaghimi, S., 1994. A farm
scale water quality planning system for evaluating Best Management Practices, 1994 ASAE
International Summer Meeting.
Beasley, D.B., Huggins, L.F., Monke, E.J., 1980. ANSWERS: A model for watershed
planning, Transactions of the ASAE, 23(4): 938-944.
Beck, M.B., Ravetz, J.R., Mulkey, L.A., Barnwell, T.O., 1997. On the problem of model
validation for predictive exposure assessments, Stochastic Hydrology and Hydraulics, 11(3):
229-254.
Beck, M.B., 1987. Water quality modeling: A review of analysis of uncertainty, Water
Resources Research, 23(8): 1393-1442.
Benaman, J. and C.A. Shoemaker, 2004. Methodology for analyzing range of uncertain
model parameters and their impact on total maximum daily load process, Journal of
Environmental Engineering, ASCE, 130(6): 648-656.
Bicknell, B.R., Imhoff, M.C., Kittle, J.K., Donigian, A.S., Jr., Johanson, R.C., 1993.
Hydrologic Simulation Program FORTRAN (HSPF): User's manual for release 101 Report
No. EPA/600/R-93/174. Athens, Ga.: US EPA Environmental Research Lab.
Bingner, R.L., Garbrecht, J., Arnold, J.G., Srinivasan, R., 1997. Effect of watershed
subdivision on simulation runoff and fine sediment yield, Transactions of the ASAE, 40(5):
1329-1335.
101
-------
Borah, D.K., 2002. Watershed scale nonpoint source pollution models: Mathematical bases,
2002 ASAE International Meeting/CIGR World Congress, Chicago, IL. Paper no. 022091.
Chow, V.T., 1956. Open channel hydraulics, McGraw-Hill, New York.
Chung, S.W., Gassman, P.W., Gu, R., Kanwar, R.S., 2002. Evaluation of EPIC for assessing
tile flow and nitrogen losses for alternative agricultural management systems, Transactions
of the ASAE, 45(4): 1135-1146.
Dybala, T. J., 2003. Part II-Farm scale modeling using the Agricultural Policy Environmental
Extender, USDA-NRCS Water Resources Assistant Team, Temple, TX.
Eckhardt, K. and J.G. Arnold, 2001. Automatic calibration of a distributed catchment model,
Journal of Hydrology, 251(2001): 103-109.
Edwards, D.R., Daniel, T.C., Scott, H.D., Murdoch, J.F., Habiger, M.J., Burkes, H.M., 1996.
Stream quality impacts of Best Management Practices in a Northwestern Arkansas basin,
Journal of the American Water Resources Association, 32(3): 499-509.
Engel, B. and K.J. Lim, 2001. Estimating nutrient losses within Indiana, Purdue University,
Available at http://danpatch.ecn.purdue.edu/~napra/NAPRA_NUTRIENT_RESULTS_ FOR
_INDIANA/nut_input.htm.
FitzHugh, T.W. and D.S. Mackay, 2000. Impacts of input parameter spatial aggregation on
an agricultural nonpoint source pollution model, Journal of Hydrology, 236(2000): 35-53.
Flanagan, D.C. and M.A. Nearing (eds.), 1995. USDA-Water Erosion Prediction Project:
Hillslope Profile and Watershed Model Documentation. NSERL Report No. 10, USDA-ARS
National Soil Erosion Research Laboratory, West Lafayette, Indiana.
Goodrich, D.C., 1992. An overview of USDA-ARS climate change and hydrology program
and analysis of model complexity as a function of basin scale, Proceeding of the Workshop
on the Effects of Global Climate Change on Hydrology and Water Resources at a Catchment
Scale, 233-242, Tsukuba, Japan, 3-6 Feb.
Green, W.H. and G.A. Ampt, 1911. Studies on soil physics: 1. The flow of air and water
through soils, Journal of Agricultural Sciences 4:11-24.
Griffin, C.B., 1995. Uncertainty analysis of BMP effectiveness for controlling nitrogen from
urban nonpoint sources, Journal of the American Water Resources Association, 31(6): 1041-
1050.
Gu R. and Y. Lee, 2002. River temperature sensitivity to hydraulic and meteorological
parameters, Journal of Environmental Management, 66(1): 43-56.
102
-------
Gupta, V.K. and S. Sorooshian, 1985. The relationship between data and the precision of
parameter estimates ofhydrologic models, Journal of Hydrology 81:57-77.
James L.D. and S.J. Burges, 1982. Selection, calibration and testing of hydrologic models, in
Hydrologic Modeling of Small Watersheds (Eds. Haan, Johnson, and Brakensiek), ASAE
Monograph #5, St. Joseph, MI.
Jha, M., Gassman, P.W., Secchi, S., Gu, R., Arnold, J.G., 2004. Effects of watershed
subdivision on SWAT flow, sediment, and nutrient predictions, Journal of the American
Water Resources Association, 40(3): 811-825.
Jha, M., Gassman, P.W., Secchi, S., Gu, R., Arnold, J.G., 2003. Hydrologic simulation of the
Maquoketa River watershed with SWAT, AWRA 2003 Spring Specialty Conference.
Kalin. L. and M.M. Hantush, 2003. Evaluation of sediment transport models and
comparative application of two watershed models, National Risk Management Research
Laboratory, Office of Research and Development, US EPA, Cincinnati, OH 45268.
Kalin, L., Govindaraju, R.S., Hantush, M.M., 2003. Effect of geomorphologicalresolution on
modeling of runoff hydrograph and sedimentograph over small watersheds, Journal of
Hydrology, 276(2003): 89-111.
Keller, E.A., Valentine, D.W., Gibbs, D.R., 1997. Hydrologic response of small watersheds
following the Southern California Painted Cave Fire of June 1990, Hydrologic Processes,
H(4): 401-414.
Kirsch, K., Kirsch, A., Arnold, J.G., 2002. Predicting sediment and phosphorus loads in the
Rock River basin using SWAT, Transactions of the ASAE, 45(6): 1757-1769.
Konikow, L.F. and J.D., Bredehoeft, 1992. Ground-water models can not be validated,
Advanced in Water Resources, 15: 75-83.
Lake, J. and J. Morrison, 1981. Environmental impact of land use on water quality- Final
report on the Black Creek project- Phase II, EPA-905/9-81-003, U.S. Environmental
Protection Agency, Chicago, IL.
Lake, J. and J. Morrison, 1978. Environmental impact of land use on water quality- Final
report on the Black Creek project- Technical report, EPA-905/9-77-007-C, U.S.
Environmental Protection Agency, Chicago, IL.
Lake, J. and J. Morrison, 1977a. Environmental impact of land use on water quality- Final
report on the Black Creek project- Technical report, EPA-905/9-77-007-B, U.S.
Environmental Protection Agency, Chicago, IL.
103
-------
Lake, J. and J. Morrison, 1977b. Environmental impacts of land use on water quality- Final
report on the Black Creek project- Project data, EPA-905/9-77-007-C., U.S. Environmental
Protection Agency, Chicago, IL.
Leonard, R.A., Knisel, W.G., Still, D.A. 1987. GLEAMS: Groundwater loading effects on
agricultural management systems, Transactions of the ASAE 30(5): 1403-1428.
Mamillapalli, S., 1998. Effect of spatial variability on river basin stream flow modeling.,
Ph.D. Thesis, Purdue University, West Lafayette, IN 47907.
Miller, S.N., Guertin, D.P., Kamran, H.S., Goodrich, D.C., 1999. Using high resolution
synthetic aperture radar for terrain mapping, Influence of hydrologic and geomorphic
investigation, Proceeding of Wildland Hydrology, AWRA Summer Specialty Conference,
Bozeman, Montana.
Moore, L.W., Chew, C.Y., Smith, R.H., Sahoo, S., 1992. Modeling of best management
practices on North Reelfoot Creek, Tennessee., Water Environment Research, 64(3): 241-247.
Morrison, J. and J. Lake, 1983. Environmental impacts of land use on water quality- Black
Creek project- Final report, Allen County Soil and Water Conservation District, Fort Wayne,
IN.
Mostaghimi, S., Park, S.W., Cook, R.A., Wang, S.Y., 1997. Assessment of management
alternatives on a small agricultural watershed, Water Research, 31(8): 1867-1878.
Nash, I.E. and J.V. Sutcliffe, 1970. River flow forecasting through conceptual models part 1-
A discussion of principles, Journal of Hydrology, 10(1970): 282-290.
Neitsch, S.L., Arnold, J.G., Kiniry, J.R., Williams, J.R., King ,K.W., 2001a. Soil and Water
Assessment Tool theoretical documentation, version 2000, Grassland, Soil and Water
Research Laboratory, Agricultural Research Service, Temple. TX.
Neitsch, S.L., Arnold, J.G., Kiniry, J.R., Williams, J.R., King, K.W., 2001b. Soil and Water
Assessment Tool user's manual, version 2000, Grassland, Soil and Water Research
Laboratory, Agricultural Research Service, Temple, TX.
Norris J. and C.T. Hann, 1993. Impact of subdividing watersheds on estimated hydrographs,
Applied Engineering in Agriculture, 9(5):443-445.
NRC, 2001. Assessing the TMDL approach to water quality management, Committee to
Access the Scientific Basis of the Total Maximum Daily Load Approach to Water Pollution
Reduction, Water Science and Technology Board, Division on Earth and Life Studies,
National Research Council, Washington, D.C.
Oreskes N., Shrader-Frechette, K., Belitz, K., 1994. Verification, validation, and
confirmation of numerical models in the earth science, Science, 263: 641-646.
104
-------
Park, S.W., Mostaghimi, S., Cooke, R.A., McClellan, P.W., 1994. BMP impacts on
watershed runoff, sediment, and nutrient yields, Journal of the American Water Resources
Association, 30(6): 1011-1023.
Reid, L.M. and T. Dunne, 1996. Rapid evaluation of sediment budgets, Catena Verlag,
Reiskirchen, Germany.
Renard, K.G., Foster, G.R., Weesies, G.A., McCool, O.K., Yoder, D.C. (coordinator), 1997.
Predicting soil by water: A guide to conservation planning with the Revised Universal Soil
Loss Equation (RUSLE), U.S. Department of Agriculture, Agriculture Handbook No. 703,
404 pages.
Rykiel, E.J., 1996.Testing ecological models: the meaning of validation, Ecological
Modeling, 90: 229-244.
Saleh, A., Arnold, J.G, Gassman, P.W., Hauch, L.M., Rosenthal, W.D., Williams, J.R.,
McFarland, A.M.S., 2000. Application of SWAT for the Upper North Basque River
watershed, Transactions of the ASAE, 43(5): 1077-1087.
Saleh, A. and B. Du, 2002. Application of SWAT and HSPF within BASINS program for the
Upper North Basque River watershed, 2002 ASAE Annual International Meeting.
Santhi, C., Arnold, J.G., Williams, J.R., Dugas, W.A., Srinivasan, R., Hauck, L.M., 2001a.
Validation of the SWAT model on a large river basin with point and nonpoint sources,
Journal of the American Water Resources Association, 37(5): 1169-1188.
Santhi, C., Arnold, J.G., Williams, J.R., Hauck, L.M., Dugas, W.A., 2001b. Application of a
watershed model to evaluate management effects on point and nonpoint source pollution,
Transactions of the ASAE, 44(6): 1559-1570.
Santhi, C., Srinivasan, R., Arnold, J.G., Williams, J.R., 2003. A modeling approach to
evaluate the impacts of water quality management plans implemented in the Big Cypress
Creek Watershed, Second Conference on Watershed Management to Meet Emerging TMDL
Environmental Regulations, Albuquerque, NM, pp. 384-394.
Sharpley, A.N. J.K. Syers, 1979. Phosphorus input into a stream draining an agricultural
watershed: II. Amounts and relative significance of runoff types, Water, Air and Soil
Pollution, 11:417-428.
Sohrabi, T.M., Shirmohammadi, A., Chu, T.W., Montas, H., Nejadhashemi, A.P., 2003.
Uncertainty analysis ofhydrologic and water quality predictions for a small water shed using
SWAT2000, Environmental Forensics, 4(2003): 229-238.
105
-------
Sorooshian S., Gupta, V.K., Fulton, J.L. 1983. Evaluation of maximum likelihood parameter
estimation techniques for conceptual rainfall-runoff models: Influence of calibration data
variability and length on model credibility, Water Resources Research, Vol. 19(1): 251-259.
Sorooshian S. and V.K. Gupta, 1995. Model calibration, in Computer models of watershed
hydrology (Ed. V.P. Singh), Water Resources Publications, Highlands Ranch, CO.
Srinivasan, R., Ramanarayanan, T.S., Arnold, J.G., Bednarz, S.T., 1998. Large area
hydrologic modeling and assessment part II: model application, Journal of the American
Water Resources Association, 34(1): 91-101.
USDA-ARS 1999. Soil and Water Assessment Tool, model documentation, Available at
http://www.brc.tamus.edu/swat/manual.
USDA Soil Conservation Service. 1972. National engineering handbook section 4-
Hydrology, Chapters 4-10.
Vache, K.B., Eilers, J.M., Santelmann, M.V., 2002. Water quality modeling of alternative
agricultural scenarios in the U.S. Corn Belt, Journal of the American Water Resources
Association, 38(3): 773-787.
Van Liew, M.W., Arnold, J.G., Gardbercht, J.D., 2003. Hydrologic simulation on
agricultural watersheds: choosing between two models, Transactions of the ASAE, 46(6):
1539-1551.
Vandenberghe, V.A., Van Griensvan, A., Bauwens, W., 2001. Sensitivity analysis and
calibration of the parameters ofESWAT: Application to the River Denver, Water Science and
Technology, 43(7): 295-301.
Williams, J.R. 1969. Flood routing with variable travel time or variable storage coefficients,
Transactions of the ASAE 12(1): 100-103.
Williams, J.R., 1975. Sediment-yield prediction with universal equation using runoff energy
factor, p. 244-252, in Present and prospective technology for predicting sediment yield and
sources: Proceedings of the sediment yield workshop, USDA Sedimentation Lab., Oxford,
MS, November 28- 30, 1972, ARS-S-40.
Williams, J.R., Arnold, J.G., Srinivasan, R., 2000. The APEX model, BRC Report No. 00-06,
Texas Agricultural Experiment Station, Temple, TX.
Wilson, B.N., 2002. Good calibration and validation practices for hydrologic and water
quality models, 2002 ASAE International Meeting/CIGR World Congress, Chicago, Illinois,
ASAE, Paper No. 022294.
Wischmeier, W.H. and D.D. Smith. 1978. Predicting Rainfall Erosion Losses: A Guide to
Conservation Planning, USDA Agricultural Handbook, No 537.
106
-------
Young, R.A., Onstad, C.A., Bosch, D.D., Anderson, W.P., 1989. A nonpoint-sourcepollution
model for evaluating agriculturalwatersheds, Journal of Soil and Water Conservation, 44(2):
168-173.
Young, R.A., Onstad, C.A., Bosch, D.D., Anderson, W.P., 1987. AGNPS, Agricultural
nonpoint-source pollution model: A watershed analysis tool, Conservation Research Report
No. 35. Washington, D.C.: USD A.
Yuan, Y., Bingner, L., Rebich, R.A., 2002a. Application of AnnAGNPS for analysis of
nitrogen loadings from a small agricultural watershed in the Mississippi delta, Total
Maximum Daily Load (TMDL) Environmental Regulations: Proceedings of the March 11-
13, 2002 ASAE Conference, pp. 268-279.
Yuan, Y., Dabney, S.M., Bingner, R.L., 2002b. Cost effectiveness of agricultural BMP s for
sediment reduction in the Mississippi delta, Journal of Soil and Water Conservation, 57(5):
259-267.
Zhang, W. and D.R., Montgomery, 1994. Digital elevation model grid size, landscape
representation, andhydrologic simulation, Water Resources Research, 30(4): 1019-1028.
107
-------
SEPA
United States
Environmental Protection
Agency
National Risk Management
Research Laboratory
Cincinnati, OH 45268
Official Business
Penalty for Private Use
$300
EPA/600/R-05/080
July 2004
------- |