DEVELOPING NEW MODELING TECHNIQUES TO EVALUATE THE
ENVIRONEMENTAL AND ECONOMIC IMPACTS OF INDIVIDUAL MANAGEMENT
PRACTICES AT THE FIELD AND WATERSHED SCALES
By
Andrew Richard Sommerlot

A THESIS
Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of
MASTER OF SCIENCE
Biosystems Engineering
2012

ABSTRACT
DEVELOPING NEW MODELING TECHNIQUES TO EVALUATE THE
ENVIRONEMENTAL AND ECONOMIC IMPACTS OF INDIVIDUAL MANAGEMENT
PRACTICES AT THE FIELD AND WATERSHED SCALES
By
Andrew Richard Sommerlot
Excess sediment yield as nonpoint source pollution from agricultural lands is a major
threat to water quality in lakes and streams in the United States. Watershed modeling can provide
data about the source and amount of excess sediment yield. Identifying sediment yields at the
field and watershed scales allows watershed planners to make better decisions concerning water
quality. The specific objectives of this study were to: 1) compare three watershed-scale models
(Soil and Water Assessment Tool (SWAT), Field SWAT, and High Impact Targeting (HIT))
against a calibrated field-scale model (RUSLE2) in estimating sediment yields from fields in the
River Raisin Watershed; 2) evaluate the statistical significance among models; 3) assess the
watershed models’ capabilities in identifying the areas of concern at the field level; 4) evaluate
the reliability of the watershed-scale models for field-scale analysis; 5) design and test multiple
methods for quantifying the impacts of field-scale management changes the watershed outlet; 6)
compare the true costs of BMPs and those from government programs. SWAT was the only
model found to be not significantly different from the calibrated RUSLE2. All the models were
incapable of identifying priorities areas similar to RUSLE2. SWAT provided the most estimates
within the uncertainty bounds of RUSLE2 (51%). A hybrid RUSLE2-SEDMOM-SWAT model
proved to be the best method to predict the effects of field-scale management decisions at the
watershed outlet. The true costs of sediment reduction at the field and watershed scales were
greater than government defined program costs in five out of six BMP categories.

Copyright by
ANDREW RICHARD SOMMERLOT
2012

This thesis is dedicated to my family and friends, thank you.

iv

ACKNOWLEDGMENTS

First of all I would like to thank my major professor, Dr. Amir Pouyan Nejadhashemi for
giving me the chance to attend graduate school at the Michigan State University Biosystems
Engineering Department, and of course for all his valuable advice, input, and guidance
throughout this entire process.
I would also like to thank my committee members, Dr. Harrigan for his time and effort,
and especially for his writing advice, and Dr. Surbrook for all his professional and life advice
over the last 4 years.
Thanks to my mom and dad, Steve and Evelyn Sommerlot for being understanding and
loving parents. Thank you also to the rest of my family and friends for making all 6 years of my
college education a great time.
I would also like to thank my lab-mates for all of their work and making my experience
in the Environmental Modeling Lab one to remember.

Thank you all!

v

TABLE OF CONTENTS

LIST OF TABLES ........................................................................................................................ ix
LIST OF FIGURES ........................................................................................................................x
LIST OF ABBREVIATIONS ...................................................................................................... xii
1. INTRODUCTION .......................................................................................................................1
1.1 Background ..........................................................................................................................1
2.2 Objectives ............................................................................................................................3
2. LITURATURE REVIEW ...........................................................................................................5
2.1 The Importance of Agriculture ............................................................................................5
2.2 Environmental Problems Due to Agricultural Practices ......................................................5
2.3 Water Pollution ....................................................................................................................6
2.3.1 Sediment Pollution .......................................................................................................7
2.3.2 Nutrient Pollution ........................................................................................................9
2.3.3 Pesticide Pollution .......................................................................................................9
2.4 Methods to Reduce Pollution from Agricultural Practices ................................................10
2.4.1 Best Management Practices .......................................................................................11
2.4.2 Common Sediment BMP Descriptions Specific to Michigan ...................................12
2.5 Existing BMP (conservation practice) Public Programs ...................................................21
2.5.1 Main USDA BMP Programs .....................................................................................21
2.5.2 Other USDA Programs ..............................................................................................23
2.6 Status of BMP’s in the United States and Michigan ..........................................................27
2.7 Attitude of Producers .........................................................................................................28
2.8 Incorporation of Best Science and Understanding for BMP Implementation ...................30
2.8.1 Water Quality Trading ...............................................................................................30
2.8.2 BMP Auction .............................................................................................................32
2.9 Available Modeling Tools .................................................................................................33
2.9.1 Spreadsheet Tool for Estimating Pollutant Load (STEPL) ....................................... 34
2.9.2 GIS Pollutant Load Application (PLOAD) ............................................................... 35
2.9.3 Long Term Hydrologic Impact Analysis (L-THIA) ................................................. 36
2.9.4 Hydrologic Simulation Program-FORTRAN (HSPF) .............................................. 37
2.9.5 Annualized Agriculture Non-Point Source Model (AnnAGNPS) ............................ 38
2.9.6 Soil and Water Assessment Tool (SWAT) ............................................................... 40
2.9.7 Revised Soil Loss Equation 2 (RUSLE2) ................................................................. 41
2.9.8 High Impact Targeting (HIT) .................................................................................... 42
2.9.9 Agriculture Policy/Environmental eXtender (APEX) .............................................. 43
3. INTRODUCTION TO METHODOLOGY AND RESULTS ..................................................48

vi

4. EVALUATING THE CAPABILITIES OF WATERSHED-SCALE MODELS IN
ESTIMATING SEDIMENT YIELD AT FIELD-SCALE ...............................................49
4.1 ABSTRACT .......................................................................................................................49
4.2 INTRODUCTION .............................................................................................................50
4.3 METHODOLOGY ............................................................................................................52
4.3.1 Study Area .................................................................................................................52
4.3.2 Data Inventory ...........................................................................................................54
4.3.3 Model Descriptions ....................................................................................................55
4.3.4 Model Calibration/Validation ..........................................................................................60
4.3.5 Field-Scale Sediment Yield Estimations ...................................................................60
4.3.6 Data Analysis .............................................................................................................65
4.3.7 Uncertainty Analysis ..................................................................................................65
4.4 RESULTS AND DISCUSSION ........................................................................................67
4.4.1 Model Calibration/Validation ...................................................................................67
4.4.2 Evaluating Field-Scale Sediment Yield .....................................................................68
4.4.3 Comparing the Models’ Capabilities in Identifying Sediment Yield Areas of Concern
................................................................................................................................71
4.4.4 Uncertainty Analysis ..................................................................................................74
4.5 CONCLUSION ..................................................................................................................77
5. EVALUATING THE IMPACT OF FIELD-SCALE MANAGEMENT STRATEGIES AT
THE WATERSHED OUTLET .........................................................................................80
5.1 ABSTRACT .......................................................................................................................80
5.2 INTRODUCTION .............................................................................................................81
5.3 MATERIALS AND METHODS .......................................................................................84
5.3.1 Study Area .................................................................................................................84
5.3.2 Data Inventory ...........................................................................................................86
5.3.3 Models Used ..............................................................................................................87
5.4 SWAT Model Calibration ..................................................................................................91
5.5 Evaluating the Impact of Field-Scale Management Strategies at the Watershed Outlet ...93
5.5.1 Predefined Field-Scale Subbasin and Reach Layers in SWAT Model (Method1) ....94
5.5.2 Subbasin-Scale Sediment Delivery Ratio (Method 2) .............................................. 99
5.5.3 SWAT without Field-scale Delineation (Method 3) ................................................102
5.5.4 Hybrid Solution Combining Analysis from the RUSLE2, SEDMOD, and SWAT
(Method 4) ......................................................................................................... 103
5.5.5 Best Management Practices (BMPs) Implementation Scenarios .............................105
5.5.6 Economic Analysis ..................................................................................................107
5.6 RESULTS AND DISCUSSION ......................................................................................107
5.6.1 SWAT Model Calibration/Validation ......................................................................107
5.6.2 Predefined Field-Scale Subbasin and Reach Layers in SWAT
Model (Method 1) ....................................................................................107
5.6.3 Subbasin-Scale Sediment Delivery Ratio (Method 2) ............................................ 108
5.6.4 SWAT without Field-scale Delineation (Method 3) ............................................... 109
5.6.5 Hybrid Solution Combining Analysis from the RUSLE2, SEDMOD, and SWAT
(Method 4) ......................................................................................................... 110
vii

5.6.6 Overall Method Comparison ....................................................................................110
5.6.7 Overall Method Comparison in Evaluating BMP Implementation Scenarios .........112
5.6.8 Economic Analysis ..................................................................................................115
5.7 CONCLUSION ................................................................................................................121
6. CONCLUSIONS .....................................................................................................................124
7. RECOMMENDATIONS FOR FUTURE RESEARCH .........................................................126
APPENDIX .................................................................................................................................128
REFERENCES ...........................................................................................................................139

viii

LIST OF TABLES

Table 2-1. Common BMPs Applied in Michigan as Describe in the NRCS Technical Guide ....13
Table 2-2. Comparison Some Available Watershed and Field Scale Models ..............................46
Table 4-1. Comparison of SWAT and RULSE2 Management Operations for field number 1 ....62
Table 4-2. Combined calibration and validation for SWAT and Field_SWAT ...........................68
Table 4-3. P-values for differences of least squares means between models ...............................71
Table 4-4. Total number of the fields and the percentage of correctly identified fields within
each category compared to RUSLE2 ................................................................................73
Table 4-5. P-factors and R-factors for Field_SWAT, SWAT and the HIT Model .......................76
Table 5-1. Overall SWAT model calibration and validation for all four Methods .....................107
Table 5-2. Overview of the four methods ...................................................................................111
Table 5-3. Median average $/ha for producer requested and EQIP defined prices ....................116
Table 5-4. Producer Requested and EQIP Median Prices per Ton Sediment Reduction at the
Field Outlet .....................................................................................................................117
Table A-1. Residuals (data not transformed), P-values ...............................................................132
Table A-2. Residuals (log transformed raw data), P-values ........................................................133
Table A-3. Residuals (square root transformed raw data), P-values ...........................................134
Table A-4. Levene's test for equality of variances.......................................................................136
Table A-5. Reject null hypothesis (alpha = 0.05). The residuals are not homogeneous .............136
Table A-6. ANOVA Table ...........................................................................................................137
Table A-7. Input data for statistical analysis ...............................................................................138

ix

LIST OF FIGURES

Figure 4-1. Saginaw River watershed. For Interpretation of the references to color in this and all
other figures, the reader is referred to the electronic version of this thesis ......................53
Figure 4-2. Comparison of four modeling strategies for estimating sediment runoff from 41
randomly selected fields in the River Raisin Watershed. In each box plot, the bold line is
drawn at the median of the sample set, the top and bottom box ends are drawn at the 75
and 25 percentile of the sample set, respectively. RUSLE2 estimations were considered
benchmark values ..............................................................................................................70
Figure 4-3.Spatial distribution of fields identified as ―high,‖ ―medium,‖ and ―low‖ priority
within the 41 fields randomly selected for analysis in the River Raisin watershed for (a)
Field_SWAT, (b) SWAT, (c) HIT, and (d) RUSLE2 .......................................................72
Figure 4-4. Estimations from all four models organized by increasing RUSLE2 sediment
runoff predictions. Uncertainty bounds are provided for the RUSLE2 estimations ........75
Figure 5-1. Study Area - River Raisin Watershed ........................................................................85
Figure 5-2: (a) Automatically delineated subbasin and (b) predefined subbasin map .................96
Figure 5-3: Comparing the NHDPlus stream network (a) versus the manual stream network
delineation(b) ....................................................................................................... 97
Figure 5-4. Subbasin-scale delivery ratio map ...........................................................................101
Figure 5-5. Grid-based delivery ratio map at 10 m resolution using the SEDMOD model .......104
Figure 5-6. Comparison of four modeling strategies (methods) a is cover crop, b is cover crop
and filter strip, c is filter strip d is residue management, e is residue management and
cover crop, and f is residue management and filter strip ............................................... 114
Figure 5-7. Comparison of dollar spent per ton sediment reduction at the watershed outlet a is
cover crop, b is cover crop and filter strip, c is filter strip d is residue management, e is
residue management and cover crop, and f is residue management and filter strip ........120
Figure A-1. Schematic of logical flow path for statistical analysis .............................................131
Figure A-2. Histogram of residuals (data not transformed) .........................................................132
Figure A-3. Residuals (log transformed raw data).......................................................................133
Figure A-4. Residuals (square root transformed raw data) ..........................................................134
x

Figure A-5. Residual Variances for each model ..........................................................................135

xi

LIST OF ABBREVIATIONS

AIC

Akaike Information Criterion

AMA

Agricultural Management Assistance Program

AnnAGNPS

Annualized Agriculture Non-Point Source Model

ANOVA

analysis of variance

APEX

the Agricultural Policy Environmental Extender

ArcSWAT

The Soil and Water Assessment Tool for ArcGIS Software

AWEP

Agricultural Water Enhancement Program

BIC

Bayesian Information Criterion

BMP(s)

Best Management Practice(s)

BOD5

5-day biological oxygen demand

CBWI

Chesapeake Bay Watershed Initiative

CCPI

Cooperative Conservation Partnership Initiative

CDL

Crop Data Layer

CIG

Conservation Innovation Grants

CLU

Common Land use Unit

CPGL

Conservation of Private Grazing Land Program
iv

CRP

Conservation Reserve Program

CSP

The Conservation Stewardship program

CWA

Clean Water Act

EPA

United States Environmental Protection Agency

EQIP

Environmental Quality Incentives Program

FRPP

Farm and Ranch Lands Protection Program

GIS

Geographic Information Systems

GRP

Grassland Reserve Program

ha

hectare

HIT

High Impact Targeting

HRU

Hydraulic Response Unit

HSPF

Hydrologic Simulation Program-FORTRAN

km

Kilometer

km2

Square Kilometers

L-THIA

the Long-Term Hydrologic Impact Assessment model

LSD

Least Significant Difference

m

Meter

v

MI

Michigan

NASS

National Agricultural Statistics Service

NED

National Elevation Dataset

NED10

10m Resolution National Elevation Dataset

NHDPLUS

National Hydrology data set PLUS

NLCD

National Land Cover Database

NPS

Nonpoint Source

NRCS

Natural Recource Conservation Service

NSE

Nash-Sutcliffe efficiency

PBIAS

Percent Bias

PLOAD

GIS Pollutant Load Application

RSR

Root Mean Square Error to the Standard Deviation

RUSLE2

the Revised Universal Soil Loss Equation

SAS

Statistical Analysis Software

SCS

Soil Conservation Service

SEDMOD

Spatially Explicit Delivery Model

SDR

Sediment Delivery Ratio

vi

SOM

Soil Organic Matter

SSURGO

The Soil Survey Geographic Database

STEPL

Spreadsheet Tool for Estimating Pollutant Load

STORET

Storage and Retrieval database

SWAT

Soil and Water Assessment Tool

t

Tons

TSS

Total Suspended Solids

USDA

the United States Department of Agriculture

USGS

United States Geological Survey

USLE

Universal Soil Loss Equation

WEPP

the Water Erosion Prediction Project

WRP

Wetlands Reserve Program

WSS

Web Soil Survey

WQT

Water Quality Trading

yr

Year

vii

1. INTRODUCTION

1.1 Background
Non-point source (NPS) pollution from agricultural lands is a significant problem that
threatens water quality throughout the United States (EPA, 2005). Excess sediment load is a
major portion of this NPS pollution and directly links agricultural land use to water quality
(Bossio et al., 2010). Publically sponsored programs exist that have a goal of reducing NPS
pollution from agricultural lands (Shortle, 2012). These programs usually experience mixed
success, and have difficulty meeting their goals of NPS pollution reduction, in part due to a lack
of quality data about the sources and amounts of excess sediment loads (Thomas and Froemke,
2012). By collecting more data from a watershed, water quality programs can be improved and
experience a higher level of effectiveness. Monitoring sediment loads in a watershed is not
always feasible due to economic and time constraints. Watershed models can fill the gap in
available data for decision makers without being a heavy economic or time burden.
Government sponsored water quality improvement projects often rely on watershed scale
analysis. Watershed models such as the Soil and Water Assessment Tool (SWAT), the Water
Erosion Prediction Project (WEPP), the Annualized Agricultural Non-Point Source
(AnnAGNPS) model, the Long-Term Hydrologic Impact Assessment model (L-THIA), the
PLOAD model, and the HSPF model have been extensively used to quantify NPS and the effects
of BMP implementation (Shen et al., 2009; Parajuli et al., 2009; Im et al., 2009; Nejadhashemi et
al., 2011; Giri et al., 2012) are often used for these purposes. However, the main concern with
the above modeling approach is the scaling issue. Scaling data from watershed models to the

1

individual field level is a challenge, however most government programs require this type of data
for targeting conservation practices. The scaling challenge arises because the majority of the
watershed models outputs are generated at the subbasin level or smaller scales but lumped
together based on the physiographical characteristics.
A series of watershed model interfaces were developed in the past few years claiming to
provide field-scale information from watershed models, such as High Impact Targeting-HIT
(O'Neil, 2010; Bartholic, 2009) and Field SWAT (Pai et al., 2011). The results of these new
downscaling techniques have not been tested against calibrated, detailed field-scale models such
as the Revised Universal Soil Loss Equation, Version 2 (RUSLE2). The first of the main
objectives of this study was to compare sediment yield estimations from Field SWAT, HIT, and
a detailed SWAT model that was delineated according to the field boundary map in the River
Raisin Watershed in Michigan.
The modeling exercises mentioned above provide important information for making
informed watershed management decisions. However, execution of a large-scale BMP
implementation plan is infeasible due to a lack of enforcement of NPS regulations and loosely
defined contract requirements in conservation programs. BMPs are most often implemented on
individual fields, and due to the voluntary nature of most government programs, installation of
many BMPs covering a significant portion of a watershed is unlikely. Understanding the true
cost and effectiveness of individual BMPs both at the field and watershed scales is important to
guide informed, realistic decision making for conservation programs such as the BMP Auction
(Smith et al., 2009).
Field-scale models are available for the evaluation of BMP effectiveness, such as the
Revised Universal Soil Loss Equation 2 (RUSLE2) and Agricultural Policy Environmental

2

Extender (APEX). These models are useful for field scale analysis but lack the ability to quantify
the effects field scale management changes have at the watershed outlet. Results obtained from
watershed scale models such as SWAT can be unreliable at the field scale due to the limitations
of land use, topography, and soil input data resolutions for field-scale study (Daggupati et al.,
2011). There is need for a modeling framework that is capable of assessing the impact of fieldscale management strategies have at the watershed scale.
1.2 Objectives
The main objective of this research is to evaluate the effectiveness of BMPs at the field
and watershed scales. The Raisin River watershed was chosen as the study area, which located in
southeast Michigan roughly 100 km south west of Detroit, MI. First, the three watershed scale
models capable of providing field scale sediment yield estimations were tested against a
calibrated, detailed field scale model, RUSLE2. The information gained from this exercise was
used to advise the design of 4 methods aimed at quantifying field scale management changes at
the watershed outlet. Four techniques were proposed and tested with the goal of providing
realistic sediment yield savings at the watershed outlet based on field scale management changes.
Watershed scale sediment reduction loads from 80 field scale BMP scenarios defined by actual
producers were evaluated at the field scale and at the watershed outlet in the Raisin River
watershed. The applicability, advantages, and disadvantages of these approaches are discussed in
this study. Finally, an economic analysis was performed to compare producer requested prices
versus the prices defined by the USDA’s Environmental Quality Incentives Program (EQIP) for
BMP implementation. The specific objectives of this research were to:

3

compare three watershed-scale models (SWAT, Field SWAT, and HIT) against calibrated
field-scale model (RUSLE2) for estimating sediment yield from 41 randomly selected
farms within the River Raisin watershed
evaluate the statistical significance among models
assess the watershed models’ capabilities in identifying the areas of concern at the field
level
evaluate the reliability of the watershed-scale models to predict sediment erosion values
within the uncertainty of a calibrated field-scale model
design and evaluate multiple methods to evaluate the effects of individual BMPs at the
watershed outlet
perform an economic analysis comparing the true cost of sediment reduction defined by
producers and the EQIP program at the field and watershed-scales

4

2. LITERATURE REVIEW

2.1 The Importance of Agriculture
Agriculture is becoming increasingly important as world population swells. The United
2

States is the 3rd largest country in the world, totaling 9,826,675 km , (CIA, 2012). Agriculture
land takes up about 40% of this total area (Encyclopedia of the Nations, 2012). The United States
is the second largest agricultural producer, and the largest agricultural exporter in the world
(Alston et al., 2010). Agriculture is an important part of the United States economy, growing
from a $17 billion industry in the late 1920’s, to $98 billion early this century (Alston et al.,
2010). There are many benefits due to the high level of agricultural activity in this country
including food security. Food security is relatively high in the United States compared to some
other countries, with 85.3% of households being food-secure; a main reason for this level of
security is a successful agricultural sector (Nord, 2009). Expanding the amount of available
cropland and increasing crop yields will be necessary to keep food available and affordable to
the growing population (Tweeten, 2008; Von Braun, 2008).
2.2 Environmental Problems Due to Agricultural Practices
Substances used as inputs in agricultural activities can be harmful to the environment.
Pesticides caused an estimated $12 billion in environmental and social damages in recent years
(Pimentel, 2009). Agricultural activities can be damaging to drinking water supplies, even in
industrialized countries (Olmstead, 2010). Agricultural systems have the potential to severely
damage freshwater ecosystems (Moss, 2008). In comparison to land in its natural state,

5

agricultural land significantly increases water pollution by allowing, on average, a far greater
amount of sediment and nutrients to enter surface waters (Moss, 2008).
External costs, or externalities, are involuntary consequences from agricultural practices
for which there is no formal market trading; these consequences are put through the process of
valuation, assigning monetary values to events and actions that exist outside the marketplace
(Tegtmeier and Duffy, 2004). Agricultural activities usually involve high external costs (Porter et
al., 2009). Examples of agricultural externalities include decreased stream health due to excess
nutrient runoff, decreased biodiversity due to mono-cropping, decreased water quality from
pesticide drift, decreased soil fertility from poorly planned crop rotations, and increased sediment
runoff due to tillage practices, among many others. External costs due to agricultural practices
have been estimated to be between $5.7 and $16.9 billion annually in the United States
(Tegtmeier and Duffy, 2004).
2.3 Water Pollution
In lakes and streams, impaired water quality is due mostly to agricultural nonpoint source
(NPS) pollution (EPA, 2005). NPS pollution is defined as: ―…pollution that reaches receiving
waters through diffuse and complex pathways…‖ (Shortle, 2012) Agricultural NPS pollution
also contributes significantly to impaired ground water (EPA, 2005). Pollutants from
agricultural practices include excess sediment, nitrogen, phosphorus, and pesticide chemicals
(EPA, 2005). Environmental concerns from these pollutants include eutrophication, fish kills,
habitat destruction, and ground water contamination (EPA, 2005). Downstream pollution
specifically from sediment and nutrient runoff is a significant problem (Vitousek et al., 2009).
Marine ecosystem impacts from these pollutant sources include eutrophication in fresh water and

6

hypoxic zones marine systems (Gordon et al., 2010). Direct and indirect impacts from
agricultural water pollution cause dangerous changes in natural ecosystems that have negative
social and economic effects (Gordon et al., 2010). Instances of impaired drinking water,
reduction or destruction of natural fisheries, wetland and coastal ecosystem loss, decline in
waterfront property values, and loss of revenue from water related recreation can all be traced
back to agricultural water pollution (Gordon et al., 2010). There are three main categories for
agricultural NPS pollution: excess sediment, nutrient, and pesticide runoff (Dowd et al., 2008).
2.3.1 Sediment Pollution
Though agricultural NPS water pollution comes in many forms, sediment run off stands
out as an important link between land use and water quality (Bossio et al., 2010). As a general
rule, soil erosion increases with the application of agricultural practices (Bossio et al., 2010).
Agricultural practices put excess stress on the land in comparison to native grass stands and
during a rain event sediment can reach a stream by rain splash or rill erosion, and can be
accelerated by the presence of drain field tile, or ditches (Deasy et al., 2009).
An important component of agricultural land is soil organic matter (SOM), which is
likely to be degraded due to mismanagement or lost in a rainfall event do to runoff (Bossio et al.,
2010). As SOM decreases in agricultural soil, runoff begins to increase at an accelerated rate due
to increased soil compaction (Bossio et al., 2010). Thus, a decreasing amount of SOM
concentration in soil increases the rate of removal (Bossio et al., 2010).
Excess sediment in streams and lakes can be deposited a great distance from the source,
causing far reaching problems including increased turbidity, and disturbing aquatic ecosystems
(Deasy et al., 2009). In addition, sediment runoff harbors a host of associated pollutants that bind

7

to particles such as metals, pesticide and herbicide chemicals, and particulate nutrients including
nitrogen and phosphorus (Deasy et al., 2009). These sediment bonded contaminants reduce
benthic macro invertebrate species diversity, lowering food availability for fish and other
organisms, thus causing disruption in aquatic the ecosystem environment (De Lange et al.,
2004). Maintenance cost of dams, reservoirs, and lakes may increase due to excess sediment
runoff by causing shorted lifespan of components or more frequent dredging (Deasy et al., 2009).
Increased turbidity can also contribute to a loss of revenue from businesses associated with
recreation centered on a water body (Gordon et al., 2010). Consequences like these can be
quantified by externalities. The external costs due to excess sedimentation have been estimated at
13.4 billion per year in the United States (Tegtmeier and Duffy, 2004). Examples of external
costs from sediment pollution that have not yet been mentioned include damages to recreational
activities, fisheries, and wildlife preservation areas among many others.
In the Great Lakes Region, increased levels of sediment loads into lakes and streams
poses a major threat to water quality (Steinman et al., 2009). The great lakes legacy act allocated
$270 million dollars towards efforts in reduction contaminated sediment erosion in the great
lakes region (EPA, 2012a). According the EPA (2012), 14 of the 31 key sediment erosion
problem areas around the great lakes basin are located in Michigan. The United States Army
Corps of Engineers spends about $40 million in the removal of 3 million m3 of sediment from
navigable waterways in the Great Lakes Basin (Miller, 2012).
Excess sediment runoff is dangerous and has potentially far reaching environmental,
social, and economic consequences. Since the main source of sediment runoff is associated with
agricultural land use, management practices on these lands make a large impact on total sediment
pollution in a watershed (Tegtmeier and Duffy, 2004).

8

2.3.2 Nutrient Pollution
The cycle of nutrients in a terrestrial system is greatly disturbed by human activities
(Howarth et al., 2002). Nutrient pollution, namely nitrogen (N) and phosphorus (P) from
agricultural sources into lakes and streams reduces environmental quality and the well being of
humans (Vitousek et al., 2009).
The most common source of excess nutrients from agricultural lands is fertilizer in
synthetic, organic, or manure form (Vitousek et al., 2009). There has been a significant global
increase in (N) and (P) ground application of fertilizers in the recent past (Vitousek et al., 2009).
Over fertilization contributes to a host of environmental problems downstream of the source
(Vitousek et al., 2009). Significant environmental costs including rising levels of nitrous oxide, a
greenhouse gas, poor water quality, and photochemical smog are all consequences of excessive
nutrient balances on agricultural land (Vitousek et al., 2009).
Increases in N and P have lead to 60% of coastal rivers showing some level of water
quality impairment, making it one of the largest pollution problems in the united states (Howarth
et al., 2002). In the northeastern United States, nitrogen in streams has increased by a factor of
eight in the 40 years prior to 2002 (Howarth et al., 2002). Field scale nutrient losses contribute
significantly to overall fertilizer pollution through leaching (Howarth et al., 2002).
2.3.3 Pesticide Pollution
Agricultural pesticides are dangerous to fish and aquatic invertebrates environmental
even in low concentrations, (Schulz, 2004; Shortle, 2012). Over sixty studies reporting pesticide
presence in surface water have been published over the last 20 years (Schulz, 2004). These
studies combine to form a general theme and conclusion, stated by Darbowski and Schulz (2003)
9

as: ―nonpoint-source pesticide pollution from agricultural areas is widely regarded as the one of
the greatest threats to contamination of natural surface waters…‖ Pesticides, especially
insecticides cause problems in aquatic populations including fish, macro invertebrates,
amphibians and birds by affecting reproductive cycles, growth and development (Schulz, 2004).
There are three main application techniques used to apply agricultural pesticides: soil
incorporation, spraying, and fumigation (Gregoire et al., 2009). Once applied, pesticides enter
waterways from agricultural lands by runoff or spray drift (Dabrowski and Schulz, 2003). Of
these two, runoff is a greater threat to environmental heath, due to the far reaching consequences
pesticides can have in solution in moving bodies of water compared to atomization in the air
(Dabrowski and Schulz, 2003). Runoff from fields can result in 1% to 10% of the total amount
of pesticide application, and as much as 60% can drift away during spraying (Dabrowski and
Schulz, 2003). Once these chemicals are mobilized in stream networks, their removal is difficult
(Gregoire et al., 2009).
2.4 Methods to Reduce Pollution from Agricultural Practices
Sediment, nutrient and pesticide NPS pollutants from agricultural lands pose a threat to
human and aquatic environmental health; addressing this issue is important to ensure continuing
protection of water recourses. Common logic points out are two basic ways to achieve
unimpaired water recourses: removing pollutants after they have entered waterways, or, to
control pollutants to prevent them from entering waterways in the first place. Prevention is a
much more attractive solution to the problem than remedial efforts (Kay et al., 2009). Although
remediation techniques are necessary in waterways that are already polluted, the approach for a

10

permanent solution to agricultural NPS pollution in the future is to mitigate this pollution at the
source.
There are various techniques available that are employed to reduce NPS pollution from
agricultural lands (Kay et al., 2009). These practices can be applied at three different levels:
reduction of pollutant input into the system, reducing the transportation ability of applied
pollutants, and by collecting and degrading mobilized pollutants before they leave the source site
(Kay et al., 2009).
2.4.1 Best Management Practices
Best Management Practices (BMPs) are actions aimed at reducing agricultural NPS
pollution and improving water quality; they can be structural or non-structural practice
(Kaplowitz and Lupi, 2009). A non-structural practice is one that does not require construction,
but modifies actions taken on agricultural land, such as reducing the number of tillage passes in a
field (Kaplowitz and Lupi, 2009). Structural BMP’s require construction and more permanent
land use change, such as a filter strip or an artificial wetland to capture runoff (Kaplowitz and
Lupi, 2009). Recently, some of the most common BMPs specifically mentioned in the literature
are: native grass, terraces, filter strips, grassed waterways, crop rotation modifications,
conservation tillage, no till, nutrient management plans, cover crops, detention ponds, riparian
buffer, stream bank stabilization, grade stabilizations, constructed wetlands, and contour
farming. (Duncan and Bradshaw, 2007; Kaini et al., 2012; Maringanti, et al., 2009; Hsieh et al.,
2010; Maringanti et al., 2011; Makarewicz, 2009; Woznicki and Nejadhashemi, 2009; Chen et
al., 2010; Baumgart-Getz et al., 2012; Diebel et al., 2008; Georgas et al., 2009; Herendeen and
Glazier, 2009; Maxted et al., 2009; Short, 2011; Sharpley et al., 2009; Tuppad et al., 2010;

11

NRCS, 2012a). The following practices have the specific goal of reducing NPS sediment
pollution: terraces, native grass, filter strips, grassed waterways, riparian buffers, no till,
conservation till, contour farming, stream bank stabilization, and grade stabilizations.
2.4.2 Common Sediment BMP Descriptions Specific to Michigan
The Natural Resource Conservation Service in Michigan has a technical guide describing
the purpose and specifications of many BMPs (NRCS, 2012a; NRCS, 2012b). Definitions of
these common sediment focused BMPs in addition to a few others not mentioned above are
presented in Table 2-1. The main source of these definitions is the NRCS Technical Guide.

12

Table 2-1: Common BMPs Applied in Michigan as Describe in the NRCS Technical Guide
BMP

Definition

Terraces

Terraces are repeating, horizontal,
earthen ridges in a hillside across the
slope. There are three types of terraces,
broadbase, steep backslope, and
impound. A broadbase terrace is built
on gentle slope. Steep backslope terraces
have a sodden backslope; these are the
most common on extreme slopes.
Terraces with underground outlets are
given the term "impound". (NRCS,
2010f) (Nejadheshemi)

Filter Strip

Herbaceous vegetation in a strip or area
between agriculturally managed land
and sensitive areas that removes
contaminates in overland flow (NRCS,
2010b)

NRCS
CODE
CODE
600

Purpose

CODE
393

*Reduce sediment,
suspended solids, and
dissolved contaminants in
runoff
*Reduce contaminates
associated with irrigation
(NRCS, 2010b)

13

* Reduce erosion of soil
*conserve moisture by
retaining runoff
(NRCS, 2010f)

Condition of
Application
*Excess soil
erosion due to
extreme slope
length
*Excess runoff
causes
problems
*Water
conservation is
important
*Reasonable
farming can
continue after
construction
*Reasonable
outlet is
available
(NRCS, 2010f)
*Sensitive areas
that need
protection from
sediment and
other runoff
associated
pollutants

Specifications
*Multiple
terraces are
parallel
*Capacity to
control a 10
year, 24 hour
storm event
*Capacity to
continue
effective
operation
through 10
years of
sediment
accumulation

*10 year life
span
*minimum
flow length
equal to 20
feet
*Located
immediately
down slope
from source of
contaminants

Table 2-1 (cont'd)

Irrigation Water Determining and controlling frequency
Management
and volume of irrigation water in a
planned manner
(NRCS, 2010h)

CODE
449

14

*Manage soil moisture
*Optimize use of available
water
*Decrease NPS pollution
into surface and
groundwater(NRCS,
2010h)

*Applicable to
all irrigated
land

*Slope above
filter strip is
no more that
1%
*Vegetation
must tolerate
partial burial
from sediment
and herbicides
used in nearby
agricultural
practices
(NRCS,
2010b)
*Applied in
accordance
with federal,
state, and local
laws
*Water cannot
be applied in
excess of need
*An
―Irrigation
Water
Management
Plan‖ must be
developed to
assist in
consistent
decision
making

Table 2-1 (cont'd)
Grassed
Waterways

Riparian
Buffers

A channel that is established with
vegetation and graded so as to transport
water at an appropriate velocity to be
non-erosive
(NRCS, 2010i)

CODE
412

*Transfer runoff away
from water concentrations
while avoiding flooding or
erosion
*Combat gully erosion
*Improve and/or protect
water quality
(NRCS, 2010i)

*Areas of water
concentrations
where added
vegetation and
designed
grading could
reduce erosion
from runoff
(NRCS, 2010i)

An area dominated by shrubs, trees, or a
combination of both located directly
adjacent to a sensitive body of water
(NRCS, 2010j)

CODE
391

*maintain or lower water
temperatures by creating
shade
*Reduce sediment and
other contaminants
associated with excess
runoff
*Reduce pesticide drift
into surface waters
*Increase carbon storage

*Areas adjacent
to intermittent
or permanent
streams, lakes
ponds and
wetlands
*Not applied to
stabilize
shorelines or
stream banks

15

*Minimum
capacity to
effectively
transport
runoff and
associated
sediment
from0 a 10
year, 24 hour
storm event
*Width less
than 100 feet
*Side slopes at
a value
unobtrusive to
current
agricultural
practices
*Stable outlet
with
appropriate
capacity
(NRCS, 2010i)
*Must have
sufficient
width, length,
height, and
plant density
to achieve
goals
*Vegetation
must be
dominated by

Table 2-1 (cont'd)
in biomass and soils
(NRCS, 2010j)

No Till

Limiting disturbing soil operations to
only those that are absolutely necessary
and management the amount and
distribution of plant residue
(NRCS, 2010k)

CODE
329

16

*Reduce sheet and rill
erosion of sediment and
other runoff associated
pollutants
*Reduce wind erosion of
soils
*Improve SOM content in
soils
*Reduce particulate
emissions from soil
*Increase plant available
moisture
*Provide food and habitat

*All
agricultural
land where
crops are
planted

planted trees
and shrubs that
are able
regenerate in
the given soil,
weather
conditions and
geographical
area
*Overland
flow through
the buffer is
kept to sheet
flow
*Native and
non invasive
vegetation
only (local
cultivars
acceptable)
*No residue is
burned
*Uniform
distribution of
residues
*includes all
operations
between and
including
harvesting the
previous crop
and harvesting
the current

Table 2-1 (cont'd)
for various wildlife

Stream Bank
Stabilization
(Streambank
Protection)

Stabilizing treatments used to protect
streams or constructed channels
(NRCS, 2010e)

CODE
580

17

*Prevent damage to
facilities, land uses, or
significant loss of land
near streams and
constructed channels
*Maintain flow capacities
of streams and constructed
channels
*Reduce downstream
effects of sediment erosion
*Enhance wildlife habitat
near stream or channel
banks
(NRCS, 2010e)

*Natural
streambanks or
constructed
channel banks
that are
susceptible to
erosion
(NRCS, 2010e)

crop in the
Soil Tillage
Intensity
Rating (STIR)
value
*STIR not
greater than 30
*Residue
amount needed
are determined
with current
approved
water erosion
prediction
technology
(NRCS,
2010k)
*All
treatments
must be in
agreement
with local,
state and
federal laws
*Avoid
detrimental
effects to
endangered or
threatened
species and
their habitat
*Proposed
treatments

Table 2-1 (cont'd)
interface
effectively
with existing
structure
Windbreak

Shelters made of one or more linear
rows of trees or shrubs for the purpose
of reducing wind erosion of soil
(NRCS, 2010d)

CODE
380

*Reduce soil erosion due
to wind
*Protect plants
*Improve air quality

*Where linear
stands of trees
or shrubs are
desirable
*Where
aesthetics hold
significant
value
*Wind
protection is
needed

Sediment Basin

A basin with an outlet formed by and
excavation or embankment to collect
excess sediment runoff
(NRCS, 2010c)

CODE
350

*Capture and store excess
sediment runoff for a
sufficient period of time to
allow settling
(NRCS, 2010c)

*When
treatments that
target the
source of
sediment runoff
are deemed
inappropriate
*Where basin
failure will not
result in
extensive
damage to
structures or
loss of life
(NRCS, 2010c)

18

*20 year
function
period
*Species must
be adapted to
the region
*Orientation
as close to
perpendicular
to average
wind direction
as possible
*Hazard class
of constructed
basin is low
*Total dam
height is less
than 11 meters
*Minimum
sediment
storage of 60
m3 per ha of
treated area
Minimum
detention
storage of 250
m3 per ha of
drainage area

Table 2-1 (cont'd)
(NRCS,
2010c)
Conservation
Crop Rotation

Crops grown in a planned sequence on
the same field aimed at improving
environmental quality by reducing
sediment erosion and other pollution
(NRCS, 2010b)

CODE
328

19

*reducing sheet and rill
erosion
*reducing irrigationinduced erosion
*reducing wind erosion of
soil
*maintaining or improving
soil organic matter content
*managing nutrient
balances
*managing plant pests
including weeds
*insects and disease
*providing food for
domestic livestock
*providing food or habitat
for wildlife
*Reducing the use of
pesticides
(NRCS, 2010b)

*Anywhere
there are
agricultural
lands on which
crops are grown
*However,
pasture and
grazing lands,
or any land on
which crops are
grown
occasionally are
inappropriate
for this BMP
(NRCS, 2010b)

* Specific
crops and
practices
associated
with these
rotations
must agree
with those
outlined by
aproved
university
publications
* Crops
chosen in
the rotation
must provide
ample
biomass
when needed
to
effectively
reduce soil
erosion to
within
thresholds
defined by
current
approved
erosion
prediction

Table 2-1 (cont'd)
technology
* If selected
crops fall
short of
biomass
requirements
a
conservation
crop rotation
should
include
cover crops
(NRCS,
2010b)

20

2.5 Existing BMP (conservation practice) Public Programs
BMP programs as they are known today are considered to begin with the Clean Water
Act-CWA (Shortle et al., 2012). Under the act, point source (PS) pollution was defined as a
federal responsibility and NPS pollution decisions were placed under state control (Shortle et al.,
2012). These policies are described as a complicated system of varying initiatives (Shortle et al.,
2012). Because of this, NPS management programs vary from state to state. Although each state
may have differing policies, most use voluntary programs to address NPS pollution control.
(Shortle et al., 2012; NRCS, 2012a)
2.5.1 Main USDA BMP Programs
Currently, the main United States Department of Agriculture (USDA) nationwide
conservation programs that fund and promote BMPs are the Conservation Reserve Program
(CRP), The Environmental Quality Incentive Program, (EQIP), and The Conservation
Stewardship program (CSP) (NRCS, 2012b). These programs are run by the Natural Resource
Conservation Service (NRCS, 2012b) a subset of the USDA (NRCS, 2012b). The following is a
summary of each NRCS program currently in place that is directly related to water quality
improvement through the use of BMPs.
2.5.1.1 Environmental Quality Incentives Program (EQIP)
The EQIP program is designed to assist agricultural producers by helping to fund the
implementation and maintenance of BMPs for up to 10 years per contract (NRCS, 2012b). In
providing assistance through the EQIP program, the NRCS plans to help producers meet federal,
state and local regulations (NRCS, 2012b). EQIP attempts to work with producers to identify the
most suitable BMPs to use, and helps producers with a conservation plan in agreement with

21

locally adapted NRCS technical standards (NRCS, 2012b). This is the largest of the programs,
with an annual budget of about $1.3 billion, the majority of which was spent on projects with
water quality improvement goals (Shortle et al., 2012). Over $487 million worth of requests were
sent in by farmers applying to this program in 2008 (Shortle et al., 2012).
2.5.1.2 Conservation Reserve Program (CRP)
Although not specifically mentioned in the 2008 NRCS Farm Bill, the CRP program is
the largest program based on budget (NRCS, 2012b). CRP promotes a specific BMP, native
grass, by paying producers to leave qualified land fallow, thus reducing NPS agricultural
pollutants (NRCS, 2012b). The goal is to influence farmers to convert erosion and pollution
sensitive crop lands to more secure native grass stands (NRCS, 2012b). These native grass stands
come in the form of filter strips, riparian buffers, or entire fallow fields, among others, and
participating producers receive an annual payment for participation (NRCS, 2012b).
2.5.1.3 Conservation Stewardship Program (CSP)
The conservation stewardship program is a program promoting BMPs through payments
made to producers who are making operation-level modifications towards environmental
improvement (NRCS, 2012b). No size limit is imposed on operations, so a great number of
producers in the United States are eligible to apply for a contract (NRCS, 2008). Participants are
paid based on performance: an annual payment for adopting BMPs, and a special payment for
implementing an environmentally friendly crop rotation (NRCS, 2008).

22

2.5.2 Other USDA Programs
In addition to EQIP, CRP, and CSP, the NRCS runs other voluntary programs with water
quality improvement goals (NRCS, 2009). There are a total of 14 such programs included in the
2008 NRCS Farm Bill (NRCS, 2009).
2.5.2.1 Agricultural Management Assistance Program (AMA)
The AMA program funds producer’s voluntary efforts to improve water quality and
erosion control on their lands (DOA, 2008a). Specific actions suggested under this program
include structural BMPs, tree planting, soil erosion control, pest management , or replacement of
traditional farming with organic methods (DOA, 2008a). Total budgeting for this program
includes $15 million be distributed yearly from 2008 through 2012 (DOA, 2008a). AMA is not
nationwide, it is only ―...available in 16 states where participation in the Federal Crop Insurance
Program is historically low: Connecticut, Delaware, Hawaii, Maine, Maryland, Massachusetts,
Nevada, New Hampshire, New Jersey, New York, Pennsylvania, Rhode Island, Utah Vermont,
West Virginia, and Wyoming (DOA, 2008a).
2.5.2.2 Chesapeake Bay Watershed Initiative (CBWI)
The Chesapeake Bay Watershed Initiative Designed to assist producers in BMP
implementation on agricultural lands within the Chesapeake Bay watershed. The initiative aims
to control erosion and nutrient levels in ground and surface water (DOA, 2008b). Actions aimed
at prevention of further water quality degradation, and remedial efforts to restore original
environmental health can be funded under the CBWI (DOA, 2008b). Funding priority within the
CBWI is given to the Susquehanna, Shenandoah, Potomac, and Patuxent River basins, which

23

outlet into the Chesapeake Bay in Maryland (DOA, 2008b). If selected, participants agree to
―improve water quality and quantity, and restore, enhance, and preserve soil, air, and related
resources‖ (DOA, 2008b).
2.5.2.3 Cooperative Conservation Partnership Initiative (CCPI)
The CCPI ―provides targeted assistance to producers for enhancing conservation
outcomes on agricultural and nonindustrial private forest land‖ (DOA, 2008c). Addressing
conservation priorities, encouraging regulatory compliance and cooperation of producers, and
promotion and demonstration of novel, innovative BMPs comprise the listed goals of the CCPI.
State Conservationists are responsible for how 90% of CCPI funds are spent (DOA, 2008c). All
CBWI funds come from allocating 6 percent of each of the total funds from the EQIP, CSP, and
WHIP programs (DOA, 2008c). Local, State, and Tribal governments, producer cooperatives,
universities, and non government organizations are all eligible to apply to the CCPI for funding
of project (DOA, 2008c).
2.5.2.4 Conservation of Private Grazing Land Program (CPGL)
Addressing natural resource concerns and ―enhancing the economic and social stability‖
on grazing and pasturelands are the major goals of the CPGL (DOA, 2008h). A program
specifically for grazing land was deemed necessary by the NRCS because much of the
agricultural land in the United States is grazing land and many ―ecological and economic
benefits‖ can be traced back to conservation efforts on these lands (DOA, 2008h). Opportunities
for producers listed under this program include funding for improving the aesthetic level of
grazing land, implementing and maintaining BMPs, improving both water quality and water

24

quantity, and wildlife habitat (DOA, 2008h). Also included are funding opportunities for creating
recreational options and increasing diversification (DOA, 2008h).
2.5.2.5 Agricultural Water Enhancement Program (AWEP)
Financial and technical assistance for ranchers and farmers under the AWEP program is
directed at conserving ground and surface water on agricultural lands (DOA, 2008f). AWEP
funds projects that fit into the following categories: conservation plans that include modeling
and/or ―condition assessment,‖ projects that reduce the amount of water necessary for
agricultural practices, projects that focus on water quality remediation, irrigation efficiency
improvement, ―activities designed to mitigate the effects of drought,‖ and any other project
deemed to add to water conservation efforts (DOA, 2008f). Producers who are eligible for EQIP
can apply to the AWEP (DOA, 2008f). Decisions on AWEP fund dispersal are partial to
applications that include a large area, ―allow for monitoring or evaluation,‖ include irrigation
management changes, or help one or more producers comply with a regulation (DOA, 2008f).
2.5.2.6 Conservation Innovation Grants (CIG)
CIG are focused on supporting innovative BMPs that work to protect the environment
and natural recourses (DOA, 2008g). This is a completive grant program where non-Federal and
Tribal governments as well as non-government organizations can compete for award money
(DOA, 2008g). Projects that receive CIG funding are often focused on large areas, such as the
watershed or multi-state scale (DOA, 2008g). Every year, specific concerns are stated as eligible
grant awards, sub competitions are also allowed under CIG (DOA, 2008g). CIG can cover up to
50% of project proposed (DOA, 2008f). Disadvantaged producers, such as Tribes or community
organizations can receive preference in CIG decisions (DOA, 2008g).

25

2.5.2.7 Farm and Ranch Lands Protection Program (FRPP)
The FRPP program purchases conservation easements for entities with land enrolled in
other conservation programs (DOA, 2008i). This action attempts to keep agricultural land that is
deemed conservatively operated from changing land use (DOA, 2008i). These agreements may
last up to 5 years for producers that have a ―proven record of acquiring and monitoring
conservation easements‖ (DOA, 2008i). More than 216,000 ha were enrolled in this program in
2008, including over 400 contracts and spanning 49 states (DOA, 2008i).
2.5.2.8 Grassland Reserve Program (GRP)
Stated under the goals for GRP are restoration, protection, and enhancement of grassland,
with a focus on working grazing operations (DOA, 2008d). The GRP is designed to enhance
―…plant and animal biodiversity…‖ and to protect ―…grassland[s] and land[s] containing shrubs
and forbs under threat of conversion…‖ (DOA, 2008d). From 2003 through 2007, the GRP was
involved in 250 easements including 47,000 ha located in 38 states (DOA, 2008d). There is a
$50,000 limitation on annual payments through this program given out in up to 50% cost share
payments (DOA, 2008d). Priority is given to land with expiring CRP contracts; GRP contracts
can be 10, 15, 20 years, or the maximum length of time allowed by the state in which the
contract is signed (DOA, 2008d).
2.5.2.9 Wetlands Reserve Program (WRP)
Aimed at private landowners, the WRP ―…provides technical and financial assistance to
private landowners and Tribes to restore, protect, and enhance wetlands‖ (DOA, 2008e). In an
attempt to meet these goals , the WRP specifically funds BMPs that reduce sediment and

26

chemical runoff, help to increase ground water levels and decrease flooding, and ―…provide
opportunities for educational, scientific, and limited recreational activities (DOA, 2008e). Three
options are available for landowners to join the program, they are: ―Permanent Easement,‖ ―30
year Easement,‖ or a ―Recreational Cost Share Agreement‖ (DOA, 2008e). The USDA pays the
entire easement in 30 year and permanent easements, and up to 75% of restoration costs for the
Recreational Cost Share Agreement (DOA, 2008e).
2.6 Status of BMP’s in the United States and Michigan
Although many different BMPs are known to be effective, and multiple public programs
funding and promoting them exist, ―it has been well established that agricultural NPS policies
are not having the desired outcomes‖ (Shortle et al., 2012). Often, these policies and programs
fall short of their goals (Gilinsky et al., 2009). Further criticisms deem these programs inefficient
(Office of the Press Secretary, 2009; Shortle et al., 2012). The major public programs aimed at
improving water quality EQIP, CRP, and CSP are falling under increasing scrutiny for
inefficiency and ineffectiveness (Shortle et al., 2012; NRCS EQUIP; Gilinsky et al., 2009) .
Inability of government officials to monitor producer’s progress and establish
accountability in the program caused almost 50% of EQIP funds to be spent on unfulfilled
contracts (USDA, 2009). Land retirement, though in theory is a sound solution to pollution
reduction, may be an ineffective tool in real agricultural systems. The low efficiency of these
USDA and NRCS programs is due partly to the fact that while technical support for BMP
projects has increased 60% from 1985 to 2006, expenditures have increased roughly 500%
(USDA, 2009). Further criticism states all three of these programs have ineffective targeting
strategies or none at all, thus missing a vital component to efficient spending towards water

27

recourse protection (Shortle et al., 2012). The USDA has also been accused of poor allocation of
funds to the most important problems water quality problems in the nation (Shortle et al., 2012).
THE USDA and NRCS BMP programs have shown some results by reduced NPS
pollution in multiple systems throughout America including the Chesapeake Bay and the Great
Lakes (NRCS, 2011a; NRCS, 2011b; NRCS, 2010k; Shortle et al., 2012). This has, however,
cost billions of dollars and clear instances of waste can be pointed out (USDA., 2006; Shortle et
al., 2012).
2.7 Attitude of Producers
All USDA, NRCS programs promoting and funding BMPs discussed thus far have been
voluntary. Therefore, it is important to review the attitude of producers to BMPs. There is a
sizable body of literature dedicated to describing factors that lead to producers adopting BMPs,
and nationwide, specific conclusions can be difficult to draw from the mass of data (Reimer et
al., 2012)
However difficult, understanding producer’s attitudes towards BMP implementation is
very important to future water quality and natural resource protection, as billions of dollars are
spent on BMP related voluntary program funding (Baumgart-Getz et al., 2012; Greiner et al.,
2009). Baumgart-Getz et al. (2012) found the most important factors that influence producer’s
adoption of BMPs are access to quality of information, financial capacity, and a sound
connections with watershed and community groups, by surveying 46 BMP studies that spanned
26 years from 1982 to 2007 (Baumgart-Getz et al., 2012). Prokopy et al. (2008) reviewed 55
studies covering over 25 years and reported that education level, capital income, farm size,

28

access to information, positive environmental attitudes, environmental awareness, and utilization
of social networks all show positive correlations to BMP adoption (Prokopy et al., 2008).
Producers who consider themselves strongly economically motivated are less likely to
adopt BMPs, while those who listed themselves as predominantly lifestyle and conservation
driven are more likely (Greiner et al., 2009). Because each geographical area has a unique set of
social attitudes and influences, ―…a sound understanding of farmer’s motivations and risk
attitudes is required—in a regional, industrial, and environmental context…‖(Greiner et al.,
2009).
Smith et al. report that the main reasons producers do not adopt BMPs are too much
government control, few options for producers, heavy paperwork load, low payments, and
complicated program information, insufficient information available about programs, and high
penalties for failing to meet program requirements. This study is unique compared to the two
previous studies mentioned, because rather than profiling producers and finding descriptive
factors to correlate with BMP adoption, it lists specific reasons producers do not adopt BMPs
from a survey (Prokopy et al., 2008; Greiner et al., 2009). Since ―…farmer adoption rates can be
improved by focusing on the generally consistent determinates of agricultural BMP adoption,‖ a
BMP programs should take into account the regional social climate in the area (Prokopy et al.,
2008). Programs should also address the main reasons BMPs are not adopted, e.g., economics
(Greiner et al., 2009). Most of the current NRCS programs do not specifically address these
important points in an effective way.
2.8 Incorporation of Best Science and Understanding for BMP Implementation

29

Most BMP programs include a statement requiring prediction technology, and as stated
above, the social component of understanding producers attitudes is paramount to program
success (NRCS, 2010a; Greiner et al., 2009). Therefore, the most effective BMP programs
should combine the best scientific methods of prediction available with an integral understanding
of the producer’s main concerns about BMPs. A few recently implemented programs have
attempted to use this understanding to increase the efficiency of water quality programs (EPA,
2003).
2.8.1 Water Quality Trading
According to the United States Environmental Protection Agency (EPA), roughly 40% of
rivers and 50% of lakes have insufficient water quality to support their designated uses, in spite
of multiple government sponsored programs with water quality improvement goals (EPA, 2012b;
NRCS, 2010a). The EPA further states that ―…solutions to…complex water quality problems
requires innovative approaches that are aligned with core water programs‖ (EPA, 2003).
In an effort to improve the effectiveness of water quality programs in this way, the EPA
established Water Quality Trading (WQT), a market-based approach (EPA, 2003). WQT is
similar to a carbon trading system, where facilities whose pollution reduction costs are higher are
allowed to purchase equivalent pollution reductions from another facility or source at a lower
cost (EPA, 2012c). The basis of WQT is ―… the fact that sources in a watershed can face very
different costs to control the same pollutant‖ (EPA, 2012c). WQT is designed to provide
assistance and guidance to states, Tribes, and interstate agencies for developing their own water
quality trading programs (EPA, 2003). According to the EPA market-based approaches are more

30

flexible and are more likely to achieve water quality improvement goals than traditional
regulations (EPA, 2003).
As of 2005, there were 30 trading programs sponsored by WQT in various states, and
currently there are forty eight (Peterson et al., 2005; EPA, 2012d). WQT trading programs exist
in 26 states, seven of which have a statewide trading framework in place (EPA, 2012d).
Pollutants are not necessarily being reduced simply because a WQT program is in place
(Peterson et al., 2005). Although programs exist, trading may not actually be taking place
(Peterson et al., 2005). In fact as of September 2011, thirteen of the programs had not led to any
trading (EPA, 2012d). Many of the other programs have few trades, in fact, Peterson et al. state
that ―…the most commonly noted feature of existing [WQT] programs is low trading volume‖
(Peterson et al., 2005). Low trading volumes are due primarily to high trading ratios, and limited
information between trading agreements (Peterson et al., 2005). High trading ratios simply mean
that each calculated unit of pollution in the trade system must be offset by a similar reduction
(Peterson et al., 2005). Though this system helps to ensure loading levels within the WQT
program bounds, economists liken it to transaction costs, and in turn with low trading volumes
(Peterson et al., 2005). Information between parities is limited in the trade agreements, as
potential traders often do not know all the trade prices available within the WQT framework, and
so are less likely to make a trade (Peterson et al., 2005). Although WQT addresses some of the
problems with current BMP programs, it fails to gain enough participation to be effective and
efficient.

31

2.8.2 BMP Auction
The BMP auction was created in an effort to address low levels of participation in
voluntary water quality programs (Smith et al., 2009). The BMP auction is built to address
producers’ main concerns, listed by Smith et al. (2009), in order to increase participation. A
BMP auction is a program in which producers submit bids by defining their own prices for
management and implementation of BMPs. These bids are ranked by effectiveness and funded
accordingly (Smith et al., 2009). A BMP that is cost effective is one that shows a high
environmental benefit per dollar spent (such as highest sediment runoff reduction per dollar)
(Smith et al., 2009). The intent of the program was to ensure the most efficient bids will receive
funding (Smith et al., 2009).
In addition to assessing efficiency, the BMP auction allows the bidders to request their
own prices for implementing BMPs (Smith et al., 2009). This creates a competitive
environment, which will drive BMP prices down, allowing the government agency applying the
program to decrease water pollution using less funding (Smith et al., 2009). By given price
naming to the producers, the government control over the programs seems less intrusive, thus
addressing the producer concern of invasive government management of programs and low
payouts. Although producers can name their own price, a maximum is usually set; this is set to a
reasonable amount for both producer and agency (Smith et al., 2009). Using the BMP auction,
agency managers can create competition for funding, and pay for the most effective BMPs
(Smith et al., 2009).
The BMP auction relies heavily on stakeholder support (Smith et al., 2009). In fact Smith
et al. suggest using local personnel with rapport among producers to market a proposed auction

32

(Smith et al., 2009). This group of people is called the ―stakeholder leadership group‖ (Smith et
al., 2009). By using this leadership group, producers will be more likely to join the auction and
bid for BMP funding (Smith et al., 2009).
A specific list of potential BMPs is used to simplify the auction, and a universal, yet
simple signup sheet is created for producers to send in bids, thus reducing paperwork to a
minimum (Smith et al., 2009). Producers are encouraged to submit multiple bids, but not exceed
a set maximum payment per farm (Smith et al., 2009). This streamlined approach addresses the
producers’ concerns of complicated programs and too much paperwork.
In recent literature, the auction format for water quality programs has received praise
(Thurston et al., 2008; Rolfe and Windle, 2011; Smith et al., 2009). A BMP auction has received
funding and has been implemented successfully in the Pomona Lake Watershed in Kansas
(Smith et al., 2009). The BMP auction is a novel approach that addresses producers’ main
concerns about joining voluntary BMP programs and allows funding agencies to make effective
decisions on how funds should be allocated to reduce pollution most efficiently.
2.9 Available Modeling Tools
The BMP auction requires ranking bids comparatively by modeling their outcomes in
some way (Smith et al., 2009). Almost every NRCS program dedicated to improving water
quality states the need for predictive technologies in funding decisions (NRCS, 2012a).
Therefore, in order for the BMP auction to be as successful as possible, the best modeling tools
and strategies should be used to rank the bids. The following are descriptions and explanations of
advantages and disadvantages of the most common modeling tools used in water quality
prediction applications.

33

2.9.1 Spreadsheet Tool for Estimating Pollutant Load (STEPL)
STEPL is a watershed scale spreadsheet tool for estimating pollutant loads at stream
outlets (EPA, 2012e). Simple algorithms are used to estimate loads for sediment and nutrients
(EPA, 2012e). The output of the model is a spreadsheet quantifying surface runoff, nutrient
loads, 5-day biological oxygen demand (BOD5), and sediment delivery (EPA, 2012e). Also
included are estimated reductions in these loads based on the implementation of BMPs in the
watershed (EPA, 2012e). An online data access system exists for users to define the input data
(Tetra Tech, 2012).
2.9.1.1 STEPL Model Components
STEPL requires user input for land use, animals present in the watershed, precipitation
and irrigation, soil and USLE parameters, and presence of septic systems or point discharges
(Tetra Tech, 2005). Model components include runoff, groundwater, sheet and rill erosion, gully
and stream bank erosion, and pollutant transport (Tetra Tech, 2005). Annual sediment load
estimation is calculated with the universal soil loss equation and sediment delivery ratio
(Nejadhashemi and Mankin, 2007; Nejadhashemi et al., 2011). Pollutant concentrations in runoff
volume are used to calculate annual nutrient loading; these factors are influenced by
management practices and land (Nejadhashemi and Mankin, 2007; Nejadhashemi et al., 2011).
Load and sediment reductions are calculated with predefined BMP efficiencies (Nejadhashemi
and Mankin, 2007; Nejadhashemi et al., 2011).
2.9.1.2 STEPL: Published Work and Applications

34

There are not many articles published that employ STEPL. It was designed for the EPA
to make management decisions on a watershed scale (EPA, 2012e). Since the model only
calculates these loads on an annual basis is usually used for estimations over a long period of
time Nejadhashemi and Mankin, 2007). STEPL is appropriate for assessing current situations in
management practice strategies in preliminary planning stages. (Nejadhashemi and Mankin,
2007)
2.9.2 GIS Pollutant Load Application (PLOAD)
CH2M HILL designed the PLOAD model for calculating NPS pollutant loads in
watersheds on an annual and seasonal basis (EPA, 2001). Any best management practices
applied within the watershed the watershed boundary and land-use including event mean
concentrations or export coefficients for each land-use are required to run the model (POLAD1).
(Nejadhashemi and Mankin, 2007; Nejadhashemi et al., 2011) This model is appropriate to use
when there is a significant amount of uncertainty in the effectiveness of watershed management
practices or the reliability of total maximum daily load threshold (EPA, 2001; Nejadhashemi and
Mankin, 2007). Some modifications to this model include tools that show nonpoint source
pollution as tables and maps and can compare different model runs (Nejadhashemi and Mankin,
2007).
2.9.2.1 PLOAD Model Components
The EPA Simple Method and the curve number method are used to calculated mean
event concentrations and runoff (EPA, 2001). Event mean concentrations are calculated for each
land use are calculated and used as inputs in the EPA Simple Methods to calculate pollutant

35

loads (EPA, 2001). Other inputs included in the EPA Simple method are ratio of runoff
producing storms and a land use specified runoff coefficient (EPA, 2001).

2.9.2.2 PLOAD: Published Work and Applications
PLOAD is not very prevalent in published studies, though some water quality and NPS
pollution studies exist that discuss the PLOAD model (Nejadhashemi and Mankin, 2007;
Nejadhashemiet al., , 2011) (Bosch et al., 2004; Endreny and Wood, 2003; Renaud et al., 2006)
Although PLOAD is relatively easy to use, it is not recommended for extensive modeling
(Nejadhashemiet al., , 2011).
2.9.3 Long Term Hydrologic Impact Analysis (L-THIA)
L-THIA was designed by the College of Engineering at Purdue University to compare the
impacts of differing land-use scenarios (Bhaduri et al., 2000). The model uses daily runoff from
climate records the current number soil data and the event mean concentrations and land-use to
calculate average annual runoff for each land-use and management combination. L-THIA is a
model dedicated to long-term analysis reporting yearly averages of these long-term periods. The
model is focused on comparing the impacts of different land-use scenarios (Bhaduri et al., 2000).
L-THIA outputs runoff and nonpoint source pollution in graphs and tables (Bhaduri et. al.,
2001).
2.9.3.1 L-THIA Model Components
The structure for L-THIA is a lumped parameter model (Bhaduri et al., 2000). Annual
runoff is calculated with the curve number method, but does not consider snow melt in the

36

calculations (Bhaduri et al., 2000). Daily runoff from climate records, soil type, and the event
mean concentrations, and land-use are used to calculate average annual runoff for each land-use
and management combination (Bhaduri et al., 2000).
2.9.3.2 L-THIA: Published Work and Applications
A number of studies report using L-THIA for NPS pollution prediction in various
watersheds (Bhaduri et al., 2000; Tangi et al., 2005; Choi et al., 2005; Muthukrishnan, 2002;
Yang et al., 2006; Lu et al., 2009). L-THIA is easy to use and requires simple, available inputs;
however since it employs the current number method it is not as effective an area where snow
and permanently frozen soils significantly impact the watershed (Bhaduri et. al., 2000). Also, it
is not effective in areas with highly variable moisture conditions (Bhaduri et. al., 2000).
2.9.4 Hydrologic Simulation Program-FORTRAN (HSPF)
The EPA designed the HPSF model for predicting watershed effects to land use change,
as well as the effects of NPS and point source pollution (Bhaduri et al., 2000). In-stream
processes are combined with overland flow to estimate loads. The HSPF model uses hydrologic
response units to model overland flow (Bhaduri et. al., 2000). A hydrologic response unit (HRU)
is a unit of the watershed with uniform characteristics in each descriptive category: land use
type, soil type and slope (USGS, 2010).
2.9.4.1 HSPF Model Components
For each HRU, surface flow and interflow and ground flow are calculated based on
infiltration and meteorological estimations (Johnson et. al., 2003; EPA, 2000). These results are
averages and represent each hydrological response unit's response to existing conditions. This

37

model allows the processes within the watershed that are most important to be exemplified by
the user. Actual physical data from watersheds concerning surface and ground water flow can be
hard or impossible to measure. Therefore, for this model, like many other large-scale watershed
hydrological models, the parameter values are often found through calibration. Calibration
involves adjusting the necessary parameters in order to make the predicted values satisfactorily
similar to the measured values (Johnson et. al., 2003).
2.9.4.2 HSPF: Published Work and Applications
This model has been widely used in documented in literature (USGS, 2010; Donigian,
2002; Chen et al., 1998; Laroche et al., 1996; Johnson et al., 2003; Van Liew et al., 2003; Albek
et al., 2004; Ackerman et al., 2005). The EPA has used this model for total maximum daily load
(TMDL) analysis through the watershed data management software BASINS. Watersheds of
small and large scale have been modeled successfully with HPSF. It is appropriate for predicting
flow and sediment for yearly and monthly intervals, however severe weather significantly
reduces the model's effectiveness (Borah et. al., 2004). Point and nonpoint sources can be
calculated, however as the time interval is reduced the model becomes less and less accurate.
(Borah et. al., 2004). HPSF can be helpful to study land-use change and pollution management
scenarios (Borah et. al., 2004).
2.9.5 Annualized Agriculture Non-Point Source Model (AnnAGNPS)
AnnAGNPS estimates NPS pollution through continuous simulation daily time step
calculations (Bosch et al., 2001). This model is based on the idea of flow between discrete cells;
these cells effectively form a raster set dividing the entire watershed with uniform square areas
(Polyakov et. al., 2007). By dividing the watershed into discrete cells of similar land use, soil

38

type, and slope, surface runoff, sediment runoff, nutrients and pesticides can be calculated at
multiple places throughout the watershed (Bosch et al., 2001). AnnAGNPS models conservation
practices, including some BMPs, and there effects on pollution in the watershed (Bosch et al.,
2001). The AnnAGNPS model has many input parameter categories, 34 in all (Polyakov et. al.,
2007). Model outputs can be obtained from any cell at a user-specified time (Bosch et al., 2001).
The model also includes an input editor to easily manipulate parameters after model set up, and
can be integrated with GIS (Polyakov et. al., 2007).
2.9.5.1 AnnAGNPS Model Components
AnnAGNPS includes components for hydrology, sediment nutrient and pesticide
transport, irrigation, precipitation, and snowmelt (Bosch et al., 2001). Chemical oxygen demand
is used as an indicator the degree of pollution (Polyakov et. al., 2007). The curve number method
is used to calculate surface runoff, and Manning’s equation for channel runoff (Bosch et al.,
2001). The weather data used in the model can be either observed or simulated (Bosch et al.,
2001). Darcy’s equation is used for lateral subsurface flow, and Hooghoudt’s equation for flow
through tile drains (Bosch et al., 2001). AnnAGPS incorporates the Revised Universal Soil Loss
(RUSLE) equation to estimate sediment loads (Das et al., 2007). The transport of sediment
through channels is modeled with the revised Einstein equation (Bosch et al., 2001).
2.9.5.2 AnnAGPS: Published Work and Applications
AnnAGPS is widely used and documented to estimate NPS pollution loads in watersheds
(Bosch et al., 2001; Das et al., 2007; Zema et al., 2012; Pease et al., 2010; Polyakov et al., 2007;
Borah, 2011; Parajuli et al., 2009; Kliment et al., 2008). These studies focus on the evaluation of
conservation practices, landuse change and BMPs. Different watersheds can be compared in

39

therefore problem areas can be identified. The effect of management operations can then be
assessed using the model as well. This is easily done by adjusting the input parameters to be
consistent with the proposed management practices (Polyakov et. al. 2007).
2.9.6 Soil and Water Assessment Tool (SWAT)
The Soil and Water Assessment Tool (SWAT) is a semi-distributed, continuous
simulation model with a daily time step (Arnold et al. 1998). Many management and land use
situations can be model with SWAT in gauged and unguaged watersheds (Karamouz et al. 2010).
The SWAT model can simulate sediment, nutrient, and pesticide movement in multiple forms
through all bodies of water in a watershed (Arnold et al. 1998). Point and non-point sources of
many types of pollutants can be evaluated using the SWAT Model (Arnold et al. 1998). The
SWAT model is a widely used and accepted tool for whole watershed scale modeling
applications for tracking sediment, nutrient, and other pollutant transport (Arabi et al. 2006).
2.9.6.1 SWAT Model Components
Included in the SWAT model are components for hydrology, soil and vegetative
processes (Kemanian et al., 2011). The model splits whole watersheds into sub watersheds and
further yet into HRUs (Kemanian et al., 2011). Water flow and pollutants are routed to watershed
outlets through streams (Kemanian et al., 2011). Inputs include data (measured or simulated) for
temperature and precipitation, soil and land use maps, slope classification, and management
practices (Nejadhashemi and Mankin, 2011). Surface runoff is calculated using the modified
SCS curve number method, and the Modified Universal Soil Loss Equation (MUSLE) estimates
sediment yields for each HRU (Saleh and Du, 2004; Neitsch et al., 2005). Flow is modeled with
Mannings equation (Neitsch et al., 2005).

40

2.9.6.2 SWAT: Published Work and Applications
The SWAT model is used often for variety of watershed modeling applications
(Kemanian et al., 2011; Saleh and Du, 2004; Neitsch et al., 2005; Gassman et al., 2010; Yanga
et al., 2009; Vieth et al., 2010) (Parajuli et al., 2009; Bosch et al., 2010; Luo et al., 2011; Boscha
et al., 2011; Schuola et al., 2008; Cibin et al., 2010; Zhang et al., 2011; Gassman et al., 2007). By
2007, 250 peer reviewed publications used or discussed the SWAT model (Gassman et al.,
2007). The number of published works that employ this model have grown significantly since.
Although SWAT has some problems with accurate digital elevation model (DEM) subbasin
delineation and does not identify braided streams, it is a very useful and effective watershed
modeling tool (Luo et al., 2011; Gassman et al., 2007).
2.9.7 Revised Soil Loss Equation 2 (RUSLE2)
RUSLE2 is a process and empirically based model that predicts rill and inter-rill erosion
caused by runoff and rainfall (Muthukrishnan, 2002). It is used to predict annual average rates of
erosion per unit area over long periods of time (NRCS, 2003b). RUSLE2 is indented to inform
conservation planners and estimate sediment deliveries on a field scale, It is not a whole
watershed scale model (NRCS, 2003b).
2.9.7.1 RUSLE2 Model Components
RUSLE2 requires database inputs for soil type, climate, and crop management zones, and
inputs of slope length, percent slope, and management operations per field (NRCS, 2003b).
RUSLE2 is based on the Universal Soil Loss Equation (USLE), which includes aggregate factor
values for rainfall erodibility, soil erodibility, slope length and percent, and the impacts of

41

cropping management systems (RSULE4). RUSLE2 is designed to be used with available
databases constructed for the USLE equation (Yoder et al., 2003).
2.9.7.2 RUSLE2: Published Work and Applications
Some publications employ RUSLE2 for sediment load calculations (Foster et al., 2001;
Pal and Al-Tabbaa, 2009). It is mainly used by the NRCS to catalogue soil erosion and evaluate
and predict results from conservation efforts (NRCS, 2003b). RUSLE2 is effective for one-fieldat-a-time conservation evaluation and planning, where fields have homogenous management
practices, soil types, and slope properties. Since the output, average annual soil loss per acre, is
normalized over area, a large area such as a watershed cannot be modeled with RUSLE2. The
reason for this is as area increases, there are more and more heterogeneous inputs, and the
prediction becomes impossible.
2.9.8 High Impact Targeting (HIT)
HIT is a combination of two models, the RUSLE model and the Spatially Explicit
Delivery Model (SEDMOD) (IWR, 2010). RUSLE estimations are used for soil load estimation;
these estimations are then used as inputs to the SEDMOD model in order to simulate stream
routing (IWR, 2010). The model outputs field scale maps highlighting problem areas and
sediment load estimations at the watershed scale (IWR, 2010).
2.9.8.1 HIT Model Components
SEDMOD is cell based, and uses flow path slope gradient, flow path slope shape, flow
path hydraulic roughness, stream proximity, soil texture, and overland flow to model steam
processes (Ouyang et al., 2005). The SEDMOD outputs are delivery ratio estimates for a

42

spatially explicit areas in a watershed given clay content, elevation, and land use inputs (Ouyang
et al., 2005). Sediment delivery ratio is calculated using the drainage area of the specified field
and extra adjustment made based on nearby conditions (Ouyang et al., 2005). The RUSLE model
(very similar to RUSLE2) calculates an annual sediment erosion value normalized over the area
defining the input boundaries (O’Neil, 2011). The targeting maps are found by combining these
results (O’Neil, 2011).
2.9.8.2 HIT: Published Work and Applications
This hybrid RUSLE/SEDMOD model approach was first presented in 2005, though it
was not yet deemed ―high impact targeting,‖ and the data frame work was not yet built (Ouyang
et al., 2005). However, no peer reviewed studies were found that use the HIT model for sediment
erosion and delivery studies since its introduction. Although SEDMOD calculates delivery ratios
for sediment in a watershed, it does not fully account for in-stream processes that vary
throughout the year.
2.9.9 Agriculture Policy/Environmental eXtender (APEX)
The purpose of the APEX model is to evaluate the effects of management practices on
agricultural land (Williams and Izaurralde, 2005). The model runs on a daily time step, and can
run simulations for very long periods of time (Williams and Izaurralde, 2005). APEX can
simulate a number of management practices including irrigation, drainage, furrow diking, buffer
strips, terraces, waterways, fertilization, manure management, lagoons, reservoirs, crops
rotations, pesticide applications, grazing, and tillage (Williams and Izaurralde, 2005). APEX was
designed to model on a farm scale, which can be sub-divided into fields (Williams and
Izaurralde, 2005).

43

2.9.9.1 APEX Model Components
APEX includes components for weather, hydrology, soil erosion, manure erosion,
pesticide fate, soil temperature, crop growth, tillage, plant environmental control, and economics
(Williams et al., 2008). Surface runoff is predicted with the SCS curve number, and the Green
and Ampt Method is used for infiltration (Williams et al., 2008). The TR-55 Method estimates
peak runoff and both vertical and horizontal flow patterns are considered in subsurface flow
(Williams et al., 2008). Multiple options are available to the user for evapotranspiration
calculations: the Hargreaves and Samani, Penman, Priestley-Taylor, Penman-Monteith, and
Baier-Robertson methods (Williams et al., 2008). Soil erosion is estimated with seven different
methods, including modifications of the RUSLE, USLE, and Modified Universal Soil Loss
Equation (MSULE) (Williams et al., 2008).
2.9.9.2 APEX: Published Work and Applications
Many studies have used the APEX model to assess management practices, and estimate
NPS pollution loads on the farm and small watershed scale (Santhi et al., 2008; Gassman et al.,
2002; Gassman et al., 2004; Mudgal et al., 2008; Tuppad et al., 2009; Seleh et al., 2003; Wang et
al., 2008). APEX is a sophisticated tool with many user options (Williams and Izaurralde, 2005;
Williams et al., 2008; Santhi et al., 2008).
2.10 Model Comparison Summary
The following table compares the previously described models over six categories: model
scale, time step, major inputs, major outputs, BMP simulation capability, and whether or not the

44

model has significant support for its stated application in literature. These categories help
compare the models’ operation, BMP evaluation potential, and effectiveness.

45

Table 2-2. Comparison of Currently Used Watershed and Field Scale Models
MODEL

Scale

Time
Step

Major
Inputs

Major
Outputs

BMP
Simulation
Capability

STEPL

Watershed

Single
Event

Annual
sediment
load
estimation

Some BMPs

PLOAD

Watershed

Annual

Watershed

Daily

Annual
Nutrient
and
sediment
loads
Annual
pollutant
loads

Management Very Low
BMPs

L-THIA

*Landuse
*Herd
Presence
*USLE
Parameters
*Point Source
Presence
*Management
practices
applied in
watershed
*Soil Data
*Daily
weather
*Soil Type
*land use data

HSPF

Watershed

Single
Event

*Weather
*Soil
*Elevation
*Land use
*Land Use
*Soil type
*Elevation

Single
Event
Pollutant
loads
Sediment,
nutrient
and
pesticide
loads

Management High
BMPs

AnnAGNPS Watershed

Daily

46

Relative
Prevalence
in
Literature
Low

Management High
BMPs

Some BMPs

High

Table 2-2 (cont'd)

SWAT

Watershed

Daily

RUSLE2

Field

Annual

HIT

Watershed

Annual

APEX

Farm
Daily
(small
Watershed)

*Daily
Weather
*Land use
*Elevation
*Soil Type
*Management
Practices

Sediment,
nutrient,
pesticide
loads on
an annual,
monthly,
or daily
basis
*Weather
Average
*Soil type
Annual
*Management soil loss
practices
per unit
area
*Weather
Average
*Soil Type
annual
*Management sediment
Practices
delivered
*Clay
to stream
percentage

Many BMPs

Some BMPs

None

*Daily
Weather
*Soil type
*Management
practices
*Elevation
*Land use

Many BMPs

High

47

Annual,
monthly,
or daily
pollutant
loads in
steam or
from
subbasin

High

Comprehensive Medium
list of BMPs

3. INTRODUCTION TO METHODOLOGY AND RESULTS

The content of this thesis is in the form of two research papers that were submitted to
scientific journals. The first paper is titled ―Evaluating the Impact of Field-scale Management
Strategies at the Watershed Outlet‖ and had a goal of quantifying the reliability of some widely
used environmental models to make field-scale sediment runoff estimations. The Raisin River
watershed was chosen as the study area, which is located in southeast Michigan roughly 100 km
southwest of Detroit, MI. Three watershed scale models capable of providing field-scale
sediment yield estimations (SWAT, Field_SWAT, and HIT) were tested against RUSLE2, a
calibrated, detailed field scale model. Forty-one fields with site-specific management practice
information were randomly selected to test the reliability of the three watershed models in
quantifying sediment yield for every field. Information gained from this exercise was used in the
next study to advise the design of the best method aimed at quantifying the environmental and
economic impacts of field-scale management changes at the watershed outlet.
The second paper is entitled ―Evaluating the Impact of Field-scale Management
Strategies at the Watershed Outlet." In this study, four methods were proposed and tested with
the goal of providing realistic sediment yield savings estimations at the watershed outlet based
on field-scale management operations. Watershed-scale sediment reduction loads from 80 fieldscale BMP scenarios were evaluated at both field and watershed outlets in the River Raisin
watershed. The applicability, advantages, and disadvantages of these approaches are discussed in
this study. Finally, an economic analysis was performed to compare producer requested prices
versus the prices defined by the USDA’s Environmental Quality Incentives Program (EQIP) for
BMP implementation.

48

4. EVALUATING THE CAPABILITIES OF WATERSHED-SCALE MODELS IN
ESTIMATING SEDIMENT YIELD AT FIELD-SCALE

Andrew R. Sommerlot, A. Pouyan Nejadhashemi, Sean A. Woznicki, Subhasis Giri, Michael D.
Prohaska
4.1 ABSTRACT
Many watershed model interfaces have been developed in recent years for predicting
field-scale sediment loads. They share the goal of providing data for decisions aimed at
improving watershed health and the effectiveness of water quality conservation efforts. The
objectives of this study were to: 1) compare three watershed-scale models (Soil and Water
Assessment Tool (SWAT), Field_SWAT, and the High Impact Targeting (HIT) model) against
calibrated field-scale model (RUSLE2) in estimating sediment yield from 41 randomly selected
agricultural fields within the River Raisin watershed; 2) evaluate the statistical significance
among models; 3) assess the watershed models’ capabilities in identifying areas of concern at the
field level; 4) evaluate the reliability of the watershed-scale models for field-scale analysis. The
SWAT model produced the most similar estimates to RUSLE2 by providing the closest median
and the lowest absolute error in sediment yield predictions, while the HIT model estimates were
the worst. Concerning statistically significant differences between models, SWAT was the only
model found to be not significantly different from the calibrated RUSLE2 at

= 0.05.

Meanwhile, all models were incapable of identifying priorities areas similar to the RUSLE2
model. Overall, SWAT provided the most correct estimates (51%) within the uncertainty bounds
of RUSLE2 and is the most reliable among the studied models, while HIT is the least reliable.

49

The results of this study suggest caution should be exercised when using watershed-scale models
for field level decision-making, while field specific data is of paramount importance.
4.2 INTRODUCTION
Agricultural non-point source (NPS) pollution significantly threatens the water quality of
lakes and streams in the United States (EPA 2005). Excess sediment loads are an important link
between agricultural land use and water quality (Bossio et al., 2010). Therefore, in the United
States and worldwide, several publicly sponsored programs have been established with the goal
of reducing NPS from agricultural lands (Shortle, 2012). These programs have had mixed
success, and have met obstacles in effectively reducing NPS primarily due to a lack of data
concerning the sources and amounts of excess pollution loads (Thomas and Froemke, 2012). By
collecting and synthesizing watershed data, water quality improvement projects can be made
more effective, as more information allows for better decision-making. Meanwhile, conducting
monitoring to quantifying pollution sources throughout a watershed is usually infeasible due to
economic and time constraints. Under these conditions, watershed models can be employed as an
alternative method capable of providing data on NPS pollution quantities and sources (Daggupati
et al., 2011).
Many government sponsored water quality improvement projects rely on watershed scale
analysis. For this reason, watershed models such as the Soil and Water Assessment Tool
(SWAT), the water Erosion Prediction Project (WEPP), the Annualized Agricultural Non-Point
Source (AnnAGNPS) model, the Long-Term Hydrologic Impact Assessment model (L-THIA),
the PLOAD model, and the Hydrological Simulation Program - FORTRAN (HSPF) model have
been used extensively to quantify NPS and the effects of BMP implementation on water quality

50

(Shen et al., 2009; Parajuli et al., 2009; Im et al., 2009; Nejadhashemi et al., 2011; Giri et al.,
2012). However, there are scaling concerns with these modeling approaches. Downscaling the
watershed-scale results to individual fields, which are the target of conservation practices, is
difficult because the majority of watershed model outputs are generated at the subbasin level or
smaller scales but lumped together based on physiographic characteristics. For example, in the
SWAT model, areas with homogeneous land use, soil type, and slope are lumped into hydrologic
response units (HRUs), which can vary in size from less than one hectare to hundreds of
hectares. At the same time, these lumped units do not follow actual field boundaries and
therefore, it is difficult and likely inaccurate to provide field specific recommendations regarding
conservation practice implementation strategies. In addition, the most common agricultural
practice information is incorporated at subbasin or HRU levels in watershed models, which is not
site or field specific.
In the past few years, a series of watershed model interfaces were developed claiming to
provide field-scale information from watershed models, such as the High Impact Targeting (HIT)
model (O'Neil, 2010; Bartholic, 2009) and Field_SWAT (Pai et al., 2011). However, the results
of these new downscaling techniques have not been tested against calibrated field-scale models
such as the Revised Universal Soil Loss Equation, Version 2 (RUSLE2), which is the goal of this
study. The primary objective of this study is to compare sediment yield estimations from
Field_SWAT, HIT, and a detailed SWAT model that was delineated according to agricultural
field boundary maps in the River Raisin Watershed in Michigan. The specific objectives of this
study are to: 1) compare watershed-scale modeling approaches against a calibrated field-scale
model (RUSLE2) in estimating sediment yield from 41 agricultural fields randomly selected
within the River Raisin Watershed; 2) evaluate the statistical differences among modeling

51

estimations; 3) assess the models’ capabilities in identifying areas of concern; and 4) evaluate the
reliability of the watershed models in estimating field-scale sediment loads.
4.3 METHODOLOGY
4.3.1 Study Area
This study was performed using data from the River Raisin Watershed, located in
southeast Michigan and northern Ohio (Figure 1). The main reach in this watershed is the River
Raisin, which flows into Lake Erie near Monroe, Michigan. The watershed is comprised of
268,000 ha of land located in Hillsdale, Jackson, Lenawee, Monroe, Washtenaw, and Fulton
counties. Most of the land in the River Raisin watershed is used for agriculture (60%), while the
remaining land is categorized as forest (13%), urban (12%), wetlands (7%), and range grass and
shrubs (1%), according to the National Land Cover Database (NLCD, 2009). Corn, soybeans,
wheat, and pasture are the most common agricultural uses, with corn and soybeans produced on
over half of the agricultural area. The watershed elevation ranges from 121 m to 391 m above sea
level, with an average elevation of 300 m.

52

Figure 4-1: Saginaw River watershed. For Interpretation of the references to color in this
and all other figures, the reader is referred to the electronic version of this thesis.
53

4.3.2 Data Inventory
Various datasets were collected, including streamflow, stream sediment concentration,
physiographical characteristics of the watershed, climatological data, and both specific and
common agricultural operations and rotations.
Two sets of data (streamflow and sediment concentration) were collected for watershed
model calibration. The mean daily streamflow data was collected from USGS station 04176500
near Monroe, Michigan, for the time period of January 1, 1983 through December 31, 2011. In
addition, total suspended solids (TSS) measurements were obtained from the US Environmental
Protection Agency (USEPA) Storage and Retrieval (STORET) database. The STORET database
contains a wide range of water quality monitoring data from many sampling locations across the
United States. There were 129 measurements available at sampling location 580046 located near
the River Raisin watershed outlet. Daily TSS data were collected from July 6, 1998 to October 3,
2005. The data were not continuous, averaging 1.5 observations per month.
Physiographic characteristics that were used include elevation, land use, agricultural field
boundaries, and soil datasets. Elevation data was obtained from the US Geological Survey
(USGS) National Elevation Dataset (NED). The USGS provides a 30 m resolution elevation map
that is available for the continental United States. For land use, the Cropland Data Layer (CDL
2007) is available in a 56 m resolution raster format. Aerial maps and Common Land Unit
(CLU) boundaries created by the NRCS were used to catalogue agricultural fields. Soil data was
obtained from the from the Natural Resources Conservation Service (NRCS) National Geospatial
Management Center in the form of the Soil Survey Geographic Database (SSURGO). The soil

54

map is available in raster format at 1:24,000 scale and consists of 433 unique soils in the study
area.
Daily precipitation and maximum and minimum temperatures were collected from the
National Climatic Data Center (NCDC) Cooperative Data Network (COOP). Data from two
stations (209218 and 200032) located within the boundaries of the watershed were used.
Climatological data was available for these two stations from January 1, 1970 through December
31, 2011.
Field specific data was collected from 41 fields in the River Raisin watershed. Crop
rotations and management operations were obtained for each field through a one-by one
producer survey with the help of local NRCS staff. In addition, common land use practices were
obtained from US Department of Agriculture (USDA) 2007 Censes of Agriculture (USDA,
2007).
4.3.3 Model Descriptions
Four models (RUSLE2, SWAT, Field_SWAT, and HIT) were used to calculate sediment
yield for 41 fields randomly selected within the River Raisin Watershed. These four models are
described in the following sections.
4.3.3.1 RUSLE2
The USDA Agricultural Research Service (USDA-ARS) developed the RUSLE2 model
in 2003 for the NRCS to assist with field-scale sediment erosion estimation. RUSLE2 was
designed to guide conservation planning by estimating rill and interrill erosion caused by rainfall
and runoff. The model is based on the Revised Universal Soil Loss Equation (RUSLE), which

55

estimates average annual sediment yield per unit area based on soil properties, land use, and
daily precipitation and temperature data.
A

R K L S P

(4-1)

The average annual soil loss (A) per unit area is calculated using the following
parameters: climate erodibility (R) that is based on rainfall and temperature, soil erodibility (K)
measured under standard conditions, the slope length (L), the slope steepness (S), the land cover
management factor (C), and the support practice factor (P) (Foster, 2003). User inputs include
weather data, soil database, location database, and management operations. RUSLE2 includes
many agricultural practices for creating management operations. Validation of the RUSLE2
model was performed on simulated and natural plots using 2,000 plot-years and 10,000 plotyears of data, respectively.
4.3.3.2 SWAT
The SWAT model is a spatially distributed, physically based watershed scale model
developed by the USDA-ARS (Arnold et al., 1998; Gassman et al., 2007). The modeling
procedure begins with watershed delineation, which consists dividing the watershed into
subbasins based on stream network and topography, and further into HRUs based on
homogeneous land use, soil type, and slope. SWAT outputs are calculated at the HRU level,
aggregated to the subbasin level, and routed through stream network to the watershed outlet. The
SWAT model is capable of simulating a broad list of components including hydrology, land
management, weather, plant growth, chemical transport, and nutrient transport (Gassman et al.,
2007).

56

Detachment, transport, and degradation/deposition are the three stages of sediment
erosion modeled in SWAT. The MUSLE equation (4-2) is used to estimate the detachment of
sediment particles from the land during a rainfall event.

sed 11.8 (Qsurf q peak areahru ) 0.56 Kusle Cusle Pusle LS usle CFRG

(4-2)

The variable sed is the daily sediment yield (metric tons), Qsurf represents the surface
runoff volume in (mm/ha), qpeak is peak runoff rate (m3/s), areahru is the area of HRU (ha), the
USLE soil erodibility factor, Kusle , assumed as 0.013 m2 hr/(m3-metric ton cm), the remaining
variables (unitless) Cusle, Pusle, LSusle, CFRG , represent the USLE land cover factor, the USLE
support practice factor, the USLE topography factor, and the coarse soil fragment factor,
respectively (Neitsch et al., 2005).
The transport stage, or the amount of detached sediment that enters the reach, is modeled
using equation 4-3.

sed

( sed ' sedstor,i 1) 1 exp

surlag
tconc

(4-3)

The variable sed is equal to the sediment discharged into the main channel on a given day
(metric tons), sed´ represents sediment yield from the HRU on a given day (metric tons), the
mass of sediment stored from the preceding day is represented by sedstor,i-i (metric tons), the
surface runoff lag coefficient is surlag, and tconc is the time of concentration for a subbasin (hr)
(Neitsch et al., 2005).
57

This final stage of sediment transport in a watershed is dynamic, as sediment may be
deposited or move about in a stream network depending on environmental factors. In order to
model this phenomenon, SWAT employs a maximum sediment concentration value, which acts
as a threshold. This value determines whether deposition or degradation will occur as a function
of the peak runoff rate (Neitsch et al., 2005).
4.3.3.3 Field_SWAT
Field_SWAT is a graphical user interface tool for mapping SWAT HRU outputs to field
boundaries. The MATLAB program environment was used to develop the interface but is not
required for operation of Field_SWAT. Field_SWAT aggregates water runoff and sediment yield
based on HRUs from an existing SWAT project. The main purpose of Field_SWAT is to aid
users in visualization of SWAT outputs at the field-scale and is listed as a potential tool for fieldscale targeting of conservation programs (Pai et al., 2011). The field level HRU aggregation
output is described by equations 4-4 and 4-5 below.

m
i 1 xi dt v

(4-4)

In the above equation, xi represents a value in a vector X(t) = (x1, x2, ..., xi) of a particular
SWAT response variable for a subbasin (the value of water or sediment runoff from HRU i in the
subbasin at time t). The variable v represents the daily, monthly or annual SWAT output from a
subbasin. The number of HRU is the subbasin of interest, represented by m. Therefore, equation
4 computes the summation of all HRU outputs at time step t in the subbasin.
In order to capture the model output for individual fields for the above subbasin, equation
4-5 is introduced:

58

n y dt
j 1 j

w

(4-5)

In the above equation, yj represents the jth value of a vector V(t)= (y1, y2, ..., yj)
containing instantaneous values of the SWAT model outputs for a subbasin. The number of
fields (from the field boundary layer created by the user) within the subbasin is n, and w is the
daily, monthly, or annual SWAT output from a subbasin. Equations 4 and 5 comprise the
mapping algorithm in Field_SWAT, and are subject to the constraint that the value of v be equal
to the value of w.
4.3.3.4 The High Impact Targeting (HIT) Model
The High Impact Targeting (HIT) model was designed by the NRCS, the Michigan
Department of Agriculture (MDA), the Huron Conservation District, and the Michigan State
University Institute of Water Research to identify highly erosive areas in a watershed (Koches,
2010). It is an online tool that uses geographic information systems (GIS) to display high impact
area maps. The purpose of the HIT model is to give conservation districts, watershed groups, or
other watershed stakeholders the necessary information to make decisions and create plans to
reduce sediment erosion within a watershed (Koches, 2010).
The HIT model inputs include land use, soil clay content, and elevation, soil erodibility,
rainfall, and support practice factors to calculate the mass of soil erosion (VanderMolen, 2010).
The main outputs of the model are field-scale maps identifying high risk areas for sediment
erosion and loading, and total erosion and sediment estimations at a watershed scale.
(vanderMolen, 2010).

59

The HIT model is a combination of the Revised USLE model and the Spatially Explicit
Delivery Model (SEDMOD), and estimates annual erosion and sediment loads entering streams
(Koches, 2010). The SEDMOD model is used to estimate the percentage of eroded soils that
enter the reach system in a given area, and the Revised USLE model is used to estimate the
amount of soil erosion (Vandermolen, 2010).
4.3.4 Model Calibration/Validation
In this study, the calibration procedure involved calibrating two SWAT projects. In the
first SWAT project, predefined subbasin and stream network maps were used, while in the
second SWAT project (Field_SWAT), the subbasin and river network maps were delineated by
the SWAT automatic watershed delineation tool. Both the SWAT and the Field_SWAT models
were calibrated according to guidelines described by Moriasi et al. (2007). Based on the average
monthly values, for flow and sediment the Nash-Sutcliff coefficient of efficiency (NSE) should
be greater than or equal to 0.5, while the ratio of the root mean square error to the standard
deviation of measured data (RSR) should be less than or equal to 0.70. In addition, the percent
bias (PBIAS) should be within ± 25% for flow, and within ± 50% for sediment. The calibration
period for flow and sediment for both SWAT models was January 1, 1998 through December 31,
2001, and the validation period was January 1, 2002 through December 31, 2005. Additional
calibration is unnecessary for the RUSLE2 model because it has been calibrated using over
10,000 plot-years of data (NRCS, 2003a), while the HIT model cannot be calibrated because
model results are preprocessed and incorporated in the model at 10 m resolution.

60

4.3.5 Field-Scale Sediment Yield Estimations
Among the studied models, RUSLE2 is the only model that was calibrated and validated
at the field scale; therefore, the results obtained from the three watershed models will be tested
against the RUSLE2 model in order to quantify their effectiveness at predicting sediment yields
at the field-scale. According to NRCS guidelines, at least 15 years of climate data and preferably
20 to 30 years should be used to estimate average annual sediment yield at the field-scale
(NRCS, 2003b). In addition, this data should be obtained as close as possible to the period of
study. Therefore, the most recent climatological data was obtained from two weather stations
within the region of study from 1983 to 2011. Among the studied models, only the HIT model
does not allow the user to adjust the climate database. However, because the Revised USLE
model is an integrated component of the HIT model, it is fair to assume that 30-years of daily
climatological data were used. Below, the procedure to obtain long-term average sediment yields
from 41 fields within the River Raisin watershed is described for each model.
4.3.5.1 RUSLE2
The most detailed, field-scale data was used in RULSE2 model in order to create a point
of comparison with other models. All slope length, slope percentage, and soil types were
identified within the boundary of each field. As described earlier, crop rotations and management
operations were also obtained for all 41 fields through the one-by-one producer survey. The
results of one field survey are presented in Table 4-1.
In the next step, the RUSLE2 model was run for all combinations of slope, slop length,
and soil type within the field of interest. In order to accelerate the process, a spreadsheet toolbox

61

was designed using Microsoft Excel to assist with RUSLE2 estimations. RUSLE2 outputs were
averaged for each field and the results were presented in tons/ha/yr.

62

Table 4-1. Comparison of SWAT and RULSE2 Management Operations for field number 1
Date
21-Apr
5-May
5-May
1-Nov
13-May
13-May
13-May
14-May
7-Jun
1-Oct
21-Apr
5-May
5-May
1-Nov
13-May
13-May
13-May
14-May
7-Jun
1-Oct

RUSLE2 Operation
Fertilizer application anhydrous knife 1
Planter, double disk opener w/fluted coulter,
Sprayer, post emergence
Harvest, killing crop 50pct standing stubble
Fertilizer application surface broadcast
Disk, tandem heavy primary op.
Cultivator, field 6-12 in sweeps
Planter, double disk opener w/fluted coulter
Sprayer, post emergence
Harvest, killing crop 20pct standing stubble
Fertilizer application shank low disturbance,
Planter, double disk opener w/fluted coulter
Sprayer, post emergence
Harvest, killing crop 50pct standing stubble
Fertilizer application surface broadcast
Disk, tandem heavy primary op.
Cultivator, field 6-12 in sweeps
Planter, double disk opener w/fluted coulter
Sprayer, post emergence
Harvest, killing crop 30pct standing stubble

RUSLE2 Crop
Corn, grain

Soybean

Corn, grain

Soybean

63

SWAT Operation
Fertilizer application
Plant/begin growing season
Pesticide application
Harvest and kill operation
Fertilizer application
Tillage operation
Tillage operation
Plant/begin growing season
Pesticide application
Harvest and kill operation
Fertilizer application
Plant/begin growing season
Pesticide application
Harvest and kill operation
Fertilizer application
Tillage operation
Tillage operation
Plant/begin growing season
Pesticide application
Harvest and kill operation

SWAT Crop

Year
1

CORN

2

SOYB

3
CORN

4

SOYB

4.3.5.2 SWAT Model
In general, SWAT is incapable of estimating sediment yield for individual fields within
the watershed; the results of several fields (HRUs) are aggregated at the subbasin scale. At the
same time, the automatic delineation tool in SWAT defines subbasins according to a topography
map (digital elevation model) that is not detailed enough to capture field boundaries. To solve
this problem and obtain SWAT model results for the individual fields, the automatic delineation
tool in SWAT was used to create the most detailed subbasin map possible by selecting the
minimum number of cells required for stream definition. In the next step, the field soil map was
superimposed on the subbasin map to create a new subbasin map that includes the fields of
interest. The total number of subbasins created at this stage was 163. In the next step, the
predefined stream network was created by using the NHDPLUS map and the 10 m resolution
digital elevation model. The process of field-scale delineation for the SWAT model is unique, as
no similar work was found in the literature. The main challenge in this process is the need for
varying degrees of detailed delineation, which is time consuming.
In the next step, unique management operations and crop rotations corresponding to
RUSLE2 were created in SWAT for all fields of interest. An example of a management
operation/rotation is provided in Table 1. This is unique, as it is a common modeling exercise to
apply non-unique management operations for each crop to all subbasins in a SWAT project.
Applying field-scale data to the SWAT project ensured close similarity to the calibrated
RUSLE2 model, although not all management operations created in SWAT were identical to the
management operations created with RUSLE2, due to SWAT model limitations. After the model
set up, calibration, and validation, the SWAT model was run for the 30-year study period.

64

Finally, the subbasin/field level outputs were averaged to estimate long-term sediment yields per
area for the fields of interest.
4.3.5.3 Field_SWAT
As described earlier, Field_SWAT uses a field boundary map to aggregate the sediment
yield estimation obtained at HRU level. Therefore, Field_SWAT requires a calibrated SWAT
model, and a shapefile containing all field boundaries with a unique record corresponding to
each field. Like the previous SWAT modeling exercise, the automatic delineation tool in SWAT
was used to create the most detailed subbasin map by choosing the minimum number of cells
required for stream definition. In the next step, SWAT generates a stream network map. The
difference between Field_SWAT and SWAT in this case was that only the most common
management practices and rotation operations can be introduced for each crop in each subbasin
because the introduction of field specific management operations is not feasible. After
calibration and validation, the SWAT model was run for the 30-year period. Long-term average
sediment yields were obtained for all fields of interest through the Field_SWAT user interface.
4.3.5.4 The High Impact Targeting (HIT)
Within the HIT model no data inventory or model run is required because all modeling
was preprocessed. The only required procedure is to download the raster-based sediment yield
map from the HIT model website (http://35.9.116.206/hit2/home.htm). In the next step, the
predefined field map is superimposed on the sediment yield raster map and the average sediment
yields are calculated for each field using the Zonal Statistics command in a GIS platform such as
ArcGIS.

65

4.3.6 Data Analysis
Four techniques were used to evaluate SWAT, Field_SWAT and HIT models’
effectiveness in estimating sediment yield and the applicability of these models in identifying
areas of concern for excessive sediment generation at the field-scale. First, the range, median,
and absolute error values for each model were compared to RUSLE2. Second, Fisher’s least
significant difference (LSD) test was preformed to compare sediment yield estimation among the
models. Third, the models’ capabilities in identifying areas of concern were assessed. Finally, the
reliability of the watershed models in estimating sediment yield was evaluated within the bounds
of uncertainty of the RUSLE2 model results.
4.3.6.1 Evaluating Field-Scale Sediment Yield
A box and whisker plot was created based on the long-term average sediment yield to
compare the overall range and median values for all four studied models (Figure 2). This graph
was used for simple visual comparison. However, in order to make more informed decisions, the
absolute error in estimating sediment yield was also calculated for all fields. Both procedures
were performed using the R software version 2.15.1.
Statistical analysis was performed using the SAS 9.2 statistical package. The experiment
was a completely randomized design with one factor (model), with four treatments: SWAT,
Field_SWAT, HIT, and RUSLE2. Each model treatment was used to determine soil erosion on
41 fields within the watershed. The data were analyzed using one-way ANOVA. First, the
statistical assumptions of normality of residuals and homogeneity of residual variances were
assessed. Normality of the residuals was assessed using the Shapiro-Wilk test for normality and
normal probability plots. Residuals were found to deviate from normality at α=0.05. Therefore,

66

the response variable (sediment yield) was transformed using both log and square root
transformations, where the log transformed was selected for use in the final model. Homogeneity
of residual variances was assessed using Levene’s test, which determined that the variances were
significantly different between treatments at α = 0.05. Therefore, the analysis was completed
with heterogeneous variances using the GROUP option of the REPEATED statement in the
mixed procedure of SAS 9.2. The Akaike Information Criterion (AIC) and Bayesian Information
Criterion (BIC) values were found to be lower with heterogeneous variances than the analysis
with homogeneous variance, thus the analysis considering unequal variances was used to draw
conclusions regarding differences in the models. After the final statistical model was selected
(log transformed, heterogeneous variances), Fisher’s LSD was used to determine statistically
significant differences in sediment yield calculations between models. More detailed information
on the statistical analysis method can be found in the appendix.
4.3.6.2 Comparing Identification of Areas of Concern
Areas of concern are defined as areas that relative pollution yields generation per unit
area are significantly higher than the surrounding areas (Giri et al., 2012). These areas are
usually divided to three priority levels: high, medium, and low. In this study, the Jenks natural
breaks method of classification was used on the results of all four models in order to classify the
41 randomly selected fields into priority levels. The Jenks natural breaks method ensures that the
data from any two priority levels are significantly different (Jenks, 1967).
4.3.6.3 Uncertainty Analysis
Each of the three watershed models (Field_SWAT, SWAT, and HIT) were evaluated
against RUSLE2 estimates with uncertainty bounds. The calibrated RUSLE2 model has

67

uncertainty bounds dependent on the value of the model output prediction. For predictions
between 9.0 and 67.3 tons/ha/yr, RUSLE2 has the lowest uncertainty, ±25 percent. For
prediction values between 2.2 and 8.9 tons/ha/yr, or between 67.4 and 112.1 tons per ha per year,
there is an uncertainty level of ±50 percent; RUSLE2 predictions have the highest uncertainty
level (>±100) when they are less than 2.2 tons/ha/yr. Finally, for model predictions greater than
112.1 tons/ha/yr, the uncertainty bound is less than ±50 (NRCSb, 2003). These uncertainty
bounds were applied to the RUSLE2 data set, and upper and lower uncertainty bounds were
calculated, creating a confidence region around the RUSLE2 predictions that the watershed
models were evaluated against. The entire data set (data for every field from all four models) was
organized in increasing order by RUSLE2 prediction value, and graphed (Figure 4).
In addition to the uncertainty bounds, the performance of the three models was evaluated
using P-factor and R-factor analysis. The P-factor is defined as the percentage of estimated
values that fall within the uncertainty bounds. The R-factor is defined as the average width of the
uncertainty bounds divided by the standard deviation of the corresponding measured variable. In
an ideal case for a set of model predictions, the P-factor would be equal to 1, and the R-factor
equal to 0, meaning that all estimations from the model fall within a narrow uncertainty bracket.
The P-factors and R-factors were calculated for each of the three watershed models with respect
to RUSLE2 uncertainty bounds.
4.4 RESULTS AND DISCUSSION
4.4.1 Model Calibration/Validation
As described earlier, the calibration procedure involved calibrating two SWAT projects.
The calibration and validation was satisfactory for both models according to criteria described by

68

Moriasi et al. (2007). The statistical results for the overall calibration and validation period are
displayed in Table 4-2.
Table 4-2. Combined calibration and validation for SWAT and Field_SWAT

Models
SWAT
Field_SWAT

NSE
0.54
0.50

Streamflow
RSR
0.67
0.70

PBIAS
2.68%
5.02%

NSE
0.53
0.52

Sediment
RSR
0.69
0.69

PBIAS
34.9%
-1.94%

4.4.2 Evaluating Field-Scale Sediment Yield
The long-term average sediment yields obtained from the 41 field-scale analysis are
presented in Figure 2. The box and whisker plot provides the range and median values for each
model. This allows comparison between three watershed-scale modeling approaches that claim
to provide field-scale results versus the calibrated RULSE2 model.
Among the studied models, SWAT produced the greatest maximum value (60 t/ha/yr)
and largest range (59.7 t/ha/yr). Field_SWAT had the second highest range and maximum values
(20.65 and 20.66 t/ha/yr, respectively). The difference in range between SWAT and
Field_SWAT is that in Field_SWAT, the most common management practices and rotation
operations were introduced in each subbasin. However, in the SWAT model the use of the
detailed predefined subbasin map allowed the introduction of field specific management
operations. Therefore, it is expected that the results of the SWAT model are closer to the
RUSLE2 model. Comparing the median sediment yield values of 41 fields confirmed the above
hypothesis, the median SWAT value of 2.32 t/ha/yr is the closest to the median RUSLE2 model
of 3.36 t/ha/yr and the Field_SWAT is the second best with a median value of 1.36 t/ha/yr.

69

Meanwhile, the very large estimation of sediment yield in the SWAT model (60 t/ha/yr) resulted
in distortion of the mean value (6.15 t/ha/yr), which is almost twice as large as the RUSLE2
model average (3.84 t/ha/yr).
The HIT model produced surprisingly low estimates compared to the other models, with
a range almost 16 times lower than the RUSLE2 estimation. The RUSLE2 maximum value of 13
t/ha/yr was 12.18 t/ha/yr greater than the maximum HIT prediction of 0.82 t/ha/yr. Meanwhile,
these maximums were obtained from two different fields between models, which creates more
doubt about the reliability of the HIT model. This discrepancy is likely due to input data.
RUSLE2 requires very detailed and site specific information while the HIT model only requires
county-level agricultural census data for crop operations/rotations.
Regarding the absolute error between the three watershed models and RUSLE2, SWAT
had the lowest median value of 1.28 t/ha/yr and the highest maximum value of 53.5 t/ha/yr. The
second best model is Field_SWAT with a median error of 2.63 t/ha/yr while for the HIT model
this value is 3.00 t/h/yr. However, both of these models have lower maximum error (19.21 and
12.54 t/ha/yr, respectively) than SWAT. On average, SWAT and Field_SWAT over-predicted
sediment yield by 154% and 173%, while the HIT model under predicting by 2084%. Among all
studied models the worst estimation for a single field was produced by the HIT model, which
under-predicted the sediment yield by 75 times. The worst-case scenarios for SWAT and
Field_SWAT were 9 and 14 times higher than the RUSLE2 perditions, respectively.

70

Figure 4-2. Comparison of four modeling strategies for estimating sediment runoff from 41
randomly selected fields in the River Raisin Watershed. In each box plot, the bold line is
drawn at the median of the sample set, the top and bottom box ends are drawn at the 75
and 25 percentile of the sample set, respectively. RUSLE2 estimations were considered
benchmark values.

71

In the Type 3 test of fixed effects, the models were found to be highly statistically
significant (P<0.0001) at α = 0.05. Therefore, Fischer’s Least Significant Difference (LSD) test
was performed for all pairwise comparisons of the treatments (Table 4-3). All comparisons were
found to be significantly different at α = 0.05 except for the SWAT versus RUSLE2 comparison.
More detailed information on the results from the statistical analysis can be found in the
appendix.
Table 4-3. P-values for differences of least squares means between models

Model
Field_SWAT
HIT
RUSLE2
SWAT

Field_SWAT
<0.0001
0.0118
0.0344

HIT
<0.0001
<0.0001
<0.0001

RUSLE2
0.0118
<0.0001

SWAT
0.0344
<0.0001
0.9938

0.9938

4.4.3 Comparing Identification of Areas of Concern
Identifying areas of concern is an important step toward developing an implementation
plan to control and mitigate non-point source pollution, because pollution yields from different
fields within a watershed are disproportionate. Meanwhile, measuring pollution yields from all
fields through monitoring is expensive and impractical. Therefore, watershed managers rely on
watershed models to guide them in the decision making procedure. In this section of study, our
goal was to examine the reliability of three watershed models in identifying areas of concern,
which can be later used for targeting best management practices. Based on the sediment yield
results obtained from 41 randomly selected fields and the Jenks natural breaks method, the areas

72

of concern were divided into high, medium, and low priority. However, different ranges were
identified for the three propriety areas because different models provided different results that
varied in range (Figure 4-3).

73

Figure 4-3.Spatial distribution of fields identified as “high,” “medium,” and “low” priority within the 41 fields randomly
selected for analysis in the River Raisin watershed for (a) Field_SWAT, (b) SWAT, (c) HIT, and (d) RUSLE2

74

The fields within each priority area were compared to those of RUSLE2 in order to assess
the effectiveness of each model at correctly identifying areas of concern (Table 4-4). The first
value in each cell is the number of fields that the given model categorized the same as RUSLE2.
For example, in the first row and column, the ―0‖ indicates that Field_SWAT did did not have
any high priority fields in common with RUSLE2. The ―0%‖ in the same cell is calculated by
dividing the number of fields with identical priority assignments to RUSLE2 (zero in this case)
in the given category (for this example, high) to the total number of fields that labeled as a
―high‖ priority from the RUSLE2 model. Overall, RUSLE2 identified seven fields as high
priority, 17 as medium priority, and 17 as low priority. However, none of the models were
similar to the RUSLE2 priority classifications. The SWAT model was able to assign one high
priority field in common with RUSLE2. In addition, the SWAT model has the most similar
medium and low priority predictions with RUSLE2. Meanwhile, the low priority areas had the
best estimates comparing to RUSLE2 for all watershed models. The HIT model had the worst
performance of any model, failing to identify any correct priorities for the high or medium
categories. Overall, the results show that all three of the watershed models are ineffective at
field-scale identification of areas of concern.
Table 4-4. Total number of the fields and the percentage of correctly identified fields within
each category compared to RUSLE2.
Priority
Level
High
Medium
Low

Field_SWAT

SWAT

HIT

0 (0%)
1 (6%)
9 (53%)

1 (14%)
4 (24%)
15 (88%)

0 (0%)
0 (0%)
7 (41%)

75

4.4.4 Uncertainty Analysis
Since the results of the calibrated RUSLE2 model were not exact, uncertainty bounds can
provide a better understanding of the true value of sediment yield for each field. With this
information, a more realistic comparison can be made regarding the reliability of the
Field_SWAT, SWAT, and HIT models. This information is presented in Figure 4-4.

76

Average Annual Sediment Load
(t/ha)

20
18
16
14
12
10
8
6
4
2
0

Fields
SWAT

Field SWAT

HIT

RUSLE2

Uncertainty Bound

High UB

Figure 4-4. Estimations from all four models organized by increasing RUSLE2 sediment runoff predictions. Uncertainty
bounds are provided for the RUSLE2 estimations (three outliers from the SWAT model not shown)

77

Overall, the SWAT model performed consistently across all levels of sediment yield; however,
four of the estimations were significantly above the uncertainty bound and could be classified as
outliers. The HIT model results are only acceptable within the uncertainty bounds for low
predictions. Meanwhile, Field_SWAT performed better in the low and mid-range sediment yield
estimations. In addition, because the data is organized by increasing RUSLE2 estimations,
models that show a similar trend will likely make better estimates of priority areas than those that
do not. The only model to demonstrate a prominent, increasing trend was SWAT. None of the
models was able to provide a sufficient estimation of sediment yield on more erosive fields.
Therefore, even though it may be acceptable to use watershed models in estimating sediment
yields from low and medium priority areas, for the highly erosive areas, the results of more
detailed field-scale model such as RUSLE2 is recommended.
In addition to the uncertainty bounds, the performance of the three models was evaluated
using the P-factor and R-factor (Table 4-5).
Table 4-5. P-factors and R-factors for Field_SWAT, SWAT and the HIT Model
Uncertainty Factor
P-factor
R-factor

Field_SWAT
0.27
0.91

SWAT
0.51
0.31

HIT
0.39
22.06

According to Abbaspour (2009) P-factors close to 1 are desired and R-factors less than 1 are
acceptable. None of the models performed exceptionally well, however SWAT and Field_SWAT
had satisfactory R-values while the HIT model had an R-factor significantly greater than 1. The
greatest P-value came from the SWAT and HIT models. However, Field_SWAT had fewer
predictions within the uncertainty bounds than the two other models and therefore the value of

78

the P-factor was the worst. Overall, the SWAT model performed the best among the studied
models by having the closest P-factor to 1 and a satisfactory R-factor less than 1. In general, the
uncertainty analysis shows that SWAT predictions are most similar to those of RUSLE2, with
Field_SWAT performing worse than SWAT but slightly better than the HIT model.

4.5 CONCLUSION

In recent years, various watershed models and interfaces have been developed with the goal of
providing results that can be used for decision making at the small-scale and for individual
farms. SWAT, Field_SWAT, and HIT are examples of these models. However, the results of
these models have not been compared against detailed and calibrated field-scale models such as
RUSLE2. Therefore, evaluating these models against reliable field-scale data is an important step
towards using these models responsibly. The specific objectives of this project are to: 1) compare
three watershed-scale models (SWAT, Field_SWAT, and HIT) against a calibrated field-scale
model (RUSLE2) in estimating sediment yield from 41 randomly selected agricultural fields
within the River Raisin watershed; 2) evaluate statistically significant differences among the
models; 3) assess the watershed models’ capabilities in identifying areas of concern at the field
level; 4) evaluate the reliability of the watershed-scale models for field-scale studies.
The SWAT model produced the most similar estimates in comparison to RUSLE2 by
providing the closest median and the lowest absolute error in sediment yield predictions. The
worst predictions were associated with the HIT model that overall under-predicts sediment yield
by about 1600%.

79

Concerning statistically significant differences between models, the results confirmed the
superiority of SWAT as the only watershed model found to be statistically similar to RUSLE2 at
= 0.05, with a p value equal to 0.9938.
The watershed models were evaluated regarding their capabilities in identifying the areas
of concern at the field level. Results showed that almost all of the watershed models are
incapable of identifying low, medium, and high priorities similar to RUSLE2 model
classification. However, the SWAT model performed slightly better than the two other watershed
models, while HIT performed the worst by completely missing all fields that were categorized as
high and medium priority areas by the RUSLE2 model.
Finally, the reliability of the watershed-scale models for the field-scale study was
evaluated using uncertainty bounds generated for the RUSLE2 model. SWAT provided the most
correct estimates within the uncertainty bounds, at 51%, while Field_SWAT provided 27%, and
HIT 39%. For highly erosive fields, all models performed poorly. The HIT model preformed
better in minimally erosive fields and SWAT performed reasonably well for low and medium
erosive areas. The SWAT model predictions within RULSE2 uncertainty bounds were somewhat
consistent for the entire range of sediment yield estimations, with a few outliers. The HIT model
was only able to estimate low sediment yields within the RULSE2 uncertainty bounds, and
Field_SWAT provided correct estimations within the bounds for mainly the low and mid-range
sediment yield. Uncertainty analysis using the P and R-factors sugggests that SWAT model is
the most reliable among the studied models, while HIT is the least reliable.
Although SWAT is a watershed scale model, it provided estimations similar to RUSLE2
when delineated at the field-scale, likely due to incorporation of detailed management operations

80

similar to those used in RUSLE2. For Field_SWAT, field specific inputs are difficult/impossible
to incorporate due to the large size of subbasins in comparison to average field size. Even though
the HIT model generated the results at the highest resolutions among the studied models (10 m),
usage of county level agricultural management and rotations information for each crop likely
resulted in poor model performance. These results suggest that field specific data is of paramount
importance when making field-scale sediment yield estimations. Therefore, caution should be
exercised when using the watershed models for field level decision-making. This study
demonstrated that none of the studied watershed models provided satisfactory field-scale
targeting results, while the overall performance of the SWAT model is not significantly different
from the RUSLE2 model in estimating the field-scale sediment yield. Meanwhile, SWAT
generated several outliers that are significantly larger than the RUSLE2 model results. Therefore,
it is recommended that field-scale models such as RUSLE2 should be used to evaluate the
impact of management operations at the small scale. Future work should involve the integration
of field-scale agricultural operations at the watershed level, which provide valuable information
concerning the true costs of best management practice implementation scenarios.

81

5. EVALUATING THE IMPACT OF FIELD-SCALE MANAGEMENT STRATEGIES
AT THE WATERSHED OUTLET

Andrew R. Sommerlot , A. Pouyan Nejadhashemi, Sean A. Woznicki, Michael D. Prohaska

5.1 ABSTRACT
Nonpoint source pollution from agricultural lands is a significant contributor of sediment
pollution in lakes and streams in the United States. Therefore, quantifying the impact of
individual field management strategies at watershed-scale provides valuable information to
watershed managers and conservation agencies. In this study, four methods employing some of
the most cited models in field and watershed analysis were compared in order to find a practical
yet accurate method. The models used in this study include field-scale model (the Revised
Universal Soil Loss Equation 2 -RUSLE2), the Spatially Explicit Delivery Model (SEDMOD),
and a watershed-scale model (Soil and Water Assessment Tool - SWAT). These models were
used to develop four modeling strategies (methods) as follows: Method 1: predefined field-scale
subbasin and reach layers were used in the SWAT model; Method 2: subbasin-scale sediment
delivery ratio was employed; Method 3: results obtained from the field-scale RUSLE2 model
were incorporated as point source inputs to the SWAT watershed model, and; Method 4: a hybrid
solution combining analysis from the RUSLE2, SEDMOD, and SWAT models. The analysis
was performed on the River Raisin watershed in southeast Michigan. Method 4 was selected as
the most accurate among the studied methods since it accounts for three stages of sediment
transport from a field to the watershed outlet. In addition, the effectiveness of six types of best
management practices (BMPs) both in terms of the amount of water quality improvement and
associated cost was assessed. Economic analysis was performed using Method 4, and producer

82

requested prices for BMPs were compared with prices defined by the Environmental Quality
Incentives Program (EQIP). The results show that on a per unit area basis, producers requested
higher prices than EQIP in four out of six BMP categories. Meanwhile, the true cost of sediment
reduction at the field and watershed scales were greater than EQIP in five out of six BMP
categories according to producer requested prices.

5.2 INTRODUCTION
Non-point source (NPS) pollution from agricultural lands poses a significant threat to
water quality in the United States. Runoff from agricultural lands is the main cause of water
quality problems in rivers and lakes; a major component of this pollution is excess sediment
runoff driven by rainfall events (EPA, 2005). Many publicly sponsored programs are aimed at
reducing sediment runoff in an effort to protect and preserve water resources (Shortle, 2012).
However, efforts to reduce water pollution have been mainly aimed at point sources, while NPS
pollution remains predominantly uncontrolled (Thomas and Froemke, 2012). The lack of success
in NPS pollution control is due largely to the difficulty of identifying specific problem areas that
are significant sources of pollution (White et al., 2009). Lack of regulation and enforcement also
cause NPS pollution to remain uncontrolled (EPA, 2005).
Monitoring projects aimed at quantifying water quality usually involve high
implementation and operational costs and require long periods of time and a great amount of data
to form conclusions. To address these difficulties, models can be employed to gain valuable
knowledge faster than monitoring and at lower costs. Watershed models provide a way to
quantify NPS, identify critical source areas of pollution, and compare management strategies

83

(Daggupati et al., 2011). Therefore, these models are useful and often necessary tools in the
planning and evaluation stages of water quality improvement projects.
Several studies have been completed regarding the application of watershed models to
quantifying NPS. For example, Shen et al. (2009) evaluated the performances of Water Erosion
Prediction Project (WEPP) and the Soil and Water Assessment Tool (SWAT) for soil erosion
prediction in the Zhangjiachong watershed. The WEPP model provided slightly better
predictions then the SWAT model; however, both produced satisfactory results. Parajuli et al.
(2009) used the Annualized AGricultural Non-Point Source (AnnAGNPS) model and the SWAT
model to predict sediment yields (among other outputs) in the Cheney Lake watershed located in
Kansas. SWAT preformed better than AnnAGNPS for sediment yield prediction over the 45month evaluation period. Im et al. (2007) compared predictions of sediment yield from the HSPF
and SWAT models in the Polecat Creek watershed in Virginia. Both HSPF and SWAT produced
satisfactory results, and HSPF preformed slightly better for time steps greater than a month.
However, all of the above models were found effective in NPS quantification. In addition,
watershed models are widely used to identify critical source areas. For example, recently,
Nejadhashemi et al. (2011) compared the applicability of Spreadsheet Tool for Estimating
Pollutant Load (STEPL), the Long-Term Hydrologic Impact Assessment model (L-THIA), the
PLOAD model, and the SWAT model to identify critical source areas. They concluded that
SWAT was the only model capable of identifying critical source areas, and results from the other
models were not satisfactory. In addition, Giri et al. (2012) performed a comprehensive study to
compare different targeting techniques (based on various factors such as pollutant concentration,
load, and yield) to identify the critical source areas using the SWAT model. They concluded that
concentration based targeting is the most effective in reducing nutrients, while load based

84

targeting techniques are more effective in reducing sediment at the watershed outlet. Finally,
watershed-scale impact assessment of best management practice (BMPs) implementation
scenarios have been extensively studied (Gitau et al., 2008; Ullrich and Volk, 2009; Lee et al.,
2010; Tuppad et al., 2010; Gassman et al., 2010; Betrie et al., 2011; Giri et al., 2012),
demonstrating that a watershed-scale model is a powerful tool for use in management plan
development.
The modeling exercises mentioned above are important and necessary to make informed
watershed management decisions; however, execution of a large-scale BMP implementation plan
is infeasible due to lack of rigorously enforced NPS regulations. In reality, BMPs are
implemented on individual fields, and due to the voluntary nature of most BMP installation
programs, many BMPs covering a significant portion of a watershed is unlikely. Under this
condition, understanding the true cost and effectiveness of individual BMPs both at the field and
watershed-scales is important to guide informed decision making for conservation programs such
as the BMP Auction (Smith et al., 2009).
Many field-scale models are available for evaluation of BMP effectiveness, such as the
Revised Universal Soil Loss Equation 2 (RUSLE2) and Agricultural Policy Environmental
Extender (APEX). Although very useful for field-scale analysis, watershed-scale impacts cannot
be quantified. Also, field-scale results obtained from watershed scale models such as SWAT may
be unreliable due to the limitations of land use, topography, and soil input data resolutions for
field-scale study (Daggupati et al., 2011). Therefore, there is need for an integrated modeling
framework capable of assessing the impact of field-scale management strategies at the watershed
scale, which is the main objective of this study. The study area is the River Raisin watershed,
located in southeast Michigan. Four techniques were proposed and tested to evaluate watershed85

scale sediment reduction loads from 80 field-scale BMP scenarios. The methods tested were
using: (1) predefined field-scale subbasin and reach layers in the SWAT model; (2) subbasinscale sediment delivery ratio; (3) results obtained from the field-scale RUSLE2 model as point
source inputs to the SWAT watershed model; (4) a hybrid solution combining analysis from the
RUSLE2, the Spatially Explicit Delivery Model (SEDMOD), and SWAT models. The
applicability, advantages, and disadvantages of these approaches are discussed in this study.
Finally, an economic analysis was performed to compare producer requested prices versus the
prices defined by the USDA’s Environmental Quality Incentives Program (EQIP) for BMP
implementation.
5.3 MATERIALS AND METHODS
5.3.1 Study Area
The River Raisin watershed (Hydrologic Unit Code 04100002) is located approximately
100 km south west of Detroit, Michigan. Almost the entire watershed is within Michigan
boundaries, excluding a small portion in Ohio. The watershed is located partially in five
counties: Hillsdale, Jackson, Lenawee, Monroe, Washtenaw, and Fulton, with most of the area in
Lenawee County. The River Raisin flows east into Lake Eire near Monroe, Michigan. Sixty-six
percent of the total 268,100 ha watershed area is used for crops and pastureland, according to the
Nation Land Cover Database (NLCD, 2009). The remaining land cover is 13% forest, 12%
urban, 7% wetlands, 1% range grass and brush, and 1% water. Major agricultural land use
includes corn, soybeans, wheat, and pastureland. Mean elevation is 300 m above sea level with a
maximum elevation of 391 m, and a minimum of 12 m, according to the United States
Geological Survey.

86

Figure 5-1. Study Area - River Raisin Watershed

87

5.3.2 Data Inventory
In this study, a wide range of data was required for the modeling practices. The following
is a summary of all data collected.
Mean daily streamflow data was available from January, 1990 through December, 2009
from USGS station number 04176500 located on the River Raisin near Monroe, Michigan. A
total of 7,305 records were collected. Ninety-nine percent of the data were accepted for
publication, and 1% were labeled as preliminary. The United States Environmental Protection
Agency (EPA) Storage and Retrieval (STORET) database contained 129 total suspended solid
(TSS) measurements spanning irregularly from July 6, 1998 to October 3, 2005. The station
number 580046 was located near Monroe, Michigan at latitude 42.07°, longitude -84.13°. For
the months that had observations, there were on average 1.5 observations per month with a
maximum of 5 and a minimum of 1. Meanwhile, weather data consisting of daily maximum and
minimum temperatures and total daily precipitation were obtained from the National Climatic
Data Center from two weather stations. Stations 209218 and 200032 are located within the
watershed boundary. Data from station 209218 are available from January, 1990 through April,
2008, and data from station 200032 are available from January, 1990 through December, 2009.
Elevation data used in this study was obtained from the USGS National Elevation Dataset
(NED) website. The USGS has digital elevation maps available for the entire continental United
States at a resolution of 1 arc-second (30 m) and 1/3 arc-second (10 m). The 10 m resolution
elevation map (NED10) was used in this study. Land cover data was obtained from the USDA
National Agricultural Statistics Service (NASS) Crop Data Layer (CDL), 2007 edition. The land
cover raster had a resolution of 56 m.

88

All fields boundaries were catalogued by common land use unit (CLU) created by the
Natural Resources Conservation Service (NRCS) and aerial maps. Two sets of soil maps were
used in this study. For the field-scale modeling, field coordinates were used to locate the field
boundaries on the NRCS Web Soil Survey (WSS) online tool. WSS provides detailed
descriptions of all soil map units within the area of interest. However, for the watershed-scale
study, soil data in raster format were obtained from The Soil Survey Geographic Database
(SSURGO). The SSURGO database provides high-resolution soil data maps. The soil raster map
for the Raisin Watershed listed 433 unique soil types.
5.3.3 Models Used
Three types of models were used in this study: a field-scale model (RUSLE2), an
overland sediment delivery model (SEDMOD), and a watershed-scale model (SWAT). Brief
descriptions of these models are provided below.
5.3.3.1 RUSLE2
RUSLE2 is a sediment erosion estimation tool developed by the USDA Agricultural
Research Service (ARS) for the NRCS in 2003. The main purpose of RUSLE2 is to guide in
conservation planning, and to estimate rill and inter-rill erosion by rainfall and runoff using the
Revised Universal Soil Loss Equation (RUSLE). The RUSLE (Equation 1) estimates average
annual sediment yield per unit area based on disaggregated daily precipitation and temperature
values.
A R K L S P

(5-1)

89

where A is the average annual soil loss from rill and inter-rill erosion caused by rainfall,
and overland flow, measured in U.S. tons per acre per year, R is the rainfall-runoff erosivity
factor, K is the soil erodibility measured under a standard condition, L is the slope length, S is the
slope steepness, C is the cover management factor, and P is the support practice factor (Foster et
al., 2003). The S and L factors are defined directly by the user and RUSLE2 uses algorithms to
calculate all of the remaining parameters based on the inputs. Inputs to RUSLE2 include userbuilt management operations from built-in events, weather data, soil database, and county
location database. RUSLE2 has been validated using 10,000 plot-years of data from natural
runoff sites, and 2,000 plot-years from simulated plots (Foster et al., 2003).
5.3.3.2 SEDMOD
The Spatially Explicit Delivery Model (SEDMOD) sediment delivery framework was
first introduced by Fraser et al. (1999) to estimate sediment deposition from surface runoff
before reaching a stream. Sediment delivery ratio (SDR) is estimated using six parameters: flow
path slope gradient, flow path slope shape, flow path hydraulic roughness, stream proximity, soil
texture and overland flow (Fraser, 1999). These parameters are provided by input raster files
including those for elevation, soil texture, soil transmissivity, stream network, roughness, and
soil loss (Kandel, 2010). SEDMOD uses a raster file for each of the six inputs parameters to
create intermediate grids including streamline, flow direction, flow accumulation, gradient, path
shape, profile curvature, moisture, cell length, path slope and proximity estimations. The SDR is
then calculated using a linear weighting model, and a composite raster is created. Input
parameters are used to calculate a delivery potential using Equation 2 (Kandel, 2010).

90

DP = SGrSGw + SSrSSw + SRrSRw + SPrSPw + STrSTw + OFrOFw

(5-2)

where DP is delivery potential, SG is flow path slope gradient, SS is flow path slope
shape, SR is flow path surface roughness, SP is stream proximity, ST is soil texture, OF is the
overland flow index. Subscript r represents the rating of each parameter (ranging from 0-100)
and subscript w is the relative weight of each parameter.
An empirical equation for delivery ratio-watershed area used by (Fraser, 1999) assumes
that an ―average‖ plot would have a delivery ratio of 100%. The average plot has an area of
2

0.00049 km similar to the plot size used for the USLE study. Therefore, an intercept parameter
used in Equation 3, (C) can be calculated (Kandel, 2010).

DRa C A

1
8

(5-3)

where DRa is the delivery ratio-watershed area (%) and A is the watershed area measured in
2

km . Finally, the spatially distributed delivery ratio, which is equal to SDR, is calculated using
Equation 4 (Kandel, 2010).
SDR = DP + [DRa –μ (DP)]

(5-4)

where SDR is the delivery ratio, and μ (DP) is the mean composite layer of DP.
5.3.3.3 SWAT
SWAT was developed by the USDA-ARS and is a physically based, spatially distributed
watershed model (Arnold et al., 1998; Gassman et al., 2007). The SWAT model divides the
watershed into subbasins, and further into HRUs, or hydrologic response units. These HRUs are

91

the basic land area unit that SWAT uses to calculate model outputs. The HRUs are represented
as land areas containing homogeneous land cover, soil type, and slope. Model components
include hydrology, land management strategy, weather, plant growth, chemical transport,
nutrient transport (Gassman et al., 2007). The SCS curve number method is used to calculate
runoff in SWAT, and sediment erosion is calculated using the modified universal soil loss
equation (MUSLE) for each HRU. Because this study focuses on sediment, an in-depth
explanation of sediment runoff and transport estimations is discussed.
Sediment runoff is calculated as an average annual erosion as a function of runoff, peak
runoff rate, HRU area, soil characteristics, land cover, and topography (Neitsch et al., 2005).
There are three main stages of sediment erosion: detachment, transport, and degradation and
deposition. Detachment of sediment particles from land is calculated using the MUSLE Equation
(5-5).
sed 11.8 (Qsurf q peak areahru )0.56 Kusle Cusle Pusle LSusle CFRG

(5-5)

In the above equation sed is daily sediment yield in metric tons, Qsurf is the volume of
surface runoff measured in mm per ha, qpeak is the peak runoff rate measured in cubic m per s,
areahru is the area of the HRU measured in ha, Kusle is the USLE soil erodibility factor with
units of 0.013 metric tons square meters hours per cubic meter metric ton centimeters, Cusle is
the USLE land cover factor, Pusle is the USLE support practice factor, LSusle is the USLE
topography factor, and CFRG is the course soil fragment factor (Neitsch et al., 2005).

92

After the detachment stage of erosion, the amount of sediment transported and released
into the main channel is calculated using Equation 6 (Neitsch et al., 2005).

sed

( sed ' sedstor,i 1) 1 exp

surlag
tconc

(5-6)

where sed is the sediment discharged into the main channel on a given day measured in metric
tons, sed´ is the amount of sediment load from the HRU on given day measured in metric tons,
sedstor,i-i is the mass of stored sediment from the preceding day measured in metric tons, surlag
is the surface runoff lag coefficient, which is a user defined, and tconc is the concentration time
measured in hours (Neitsch et al., 2005). By varying the surlag coefficient, watersheds with high
and low amounts of storage can be modeled. Lateral and base flow can also contribute to
sediment in the main channel.
The third stage of erosion is degradation and deposition. Deposition occurs when
sediment leaves the stream flow and settles on the streambed, due to high concentration.
Degradation occurs when sediment concentration is low, allowing sediment on the streambed to
become suspended and travel with stream flow. The threshold value for these processes,
maximum sediment concentration, is a function of the peak runoff rate of the stream (Neitsch et
al., 2005).
5.4 SWAT Model Calibration
Among the models described above, only SWAT model can be calibrated due to lack of
observed values at the edges of the fields and streams. Calibrating the SWAT model consists of

93

iterative model runs and parameter adjustments until model predictions are deemed satisfactory
compared with observed data. Calibrations for flow and sediment were performed using manual
calibration. The calibration/validation period was approximately eight years spanning from
January 1, 1998 through September 30, 2005. Calibration was performed on a daily basis from
January 1, 1999 to December 31, 2001 and the validation was performed from January 1, 2002 to
September 30, 2005. These dates correspond to the limiting data source, in this case, the daily
TSS values. Three statistical parameters were used to compare daily-observed data against daily
model simulations: Nash-Sutcliffe efficiency (NSE), percent bias (PBIAS), and the ratio of the
root mean square error to the standard deviation of measured data (RSR). A model predicting
two or more process can be considered accurate if the NSE > 0.5, PBIAS < ±25%, and RSR ≤
0.70 for streamflow, and the NSE > 0.5, PBIAS < ± 55%, and the RSR ≤ 70 for sediment
(Moriasi et al., 2007).
The NSE compares the magnitude of residual variance with data variance (Moriasi et al.,
2007). NSE was computed using Equation 7:
n
NSE 1

Yiobs Yisim

2

i 1
2
n
Yiobs Y mean
i 1

The variable Yi

obs

(5-7)

represents the ith value from the observed data set and Yi

sim

represents the ith value from the simulated data set. The numerator in the expression is the sum
of squares of the residuals. The denominator is the sum of squares of the observed values
subtracted from the mean of all observed values. This portion of the expression consists of Yi

94

obs

,

defined above, and Ymean , the mean of the observed values. The NSE ranges from negative
infinity to one, an optimal value of one.
PBIAS quantifies the tendency of simulated values to be greater or less than observed
values (Moriasi et al., 2007). Zero is the optimal value of PBIAS, indicating simulated values are
equally distributed above and below the observed values. Equation 8 is used to calculate PBIAS.
In this equation the numerator in the expression sums the residuals, each multiplied by 100; the
denominator sums the observed values.
n
PBIAS

i 1

Yiobs Yisim

100

(5-8)

n

Yiobs
i 1

The RSME-observations standard deviation ratio (RSR) parameter measures the degree of
collinearity between the observed and simulated data sets (Moriasi e al., 2007). RSR ranges from
0 to 1. Equation 9 was used to calculate RSR.

n
RSR

i 1
n
i 1

(Yiobs Yisim ) 2
(5-9)

(Yiobs Y mean ) 2

RSR combines error index analysis with a scaling factor (Moriasi et al., 2007). A value of
0 for RSR indicates zero RMSE, or a perfect simulation. Therefore, RSR close to zero are
satisfactory (Moriasi, et al., 2007).

95

5.5 Evaluating the Impact of Field-Scale Management Strategies at the Watershed Outlet
As described earlier, four techniques were proposed to evaluate the impacts of field-scale
management strategies at the watershed outlet for a 10-year period: January 1, 2000 to December
31, 2009. The methods tested were using: (1) predefined field-scale subbasin and reach layers in
SWAT model; (2) subbasin-scale sediment delivery ratio; (3) results obtained from the fieldscale RUSLE2 model as point source inputs to the SWAT watershed model; (4) a hybrid solution
combining analysis from the RUSLE2, SEDMOD, and SWAT models.
5.5.1 Predefined Field-Scale Subbasin and Reach Layers in SWAT Model (Method 1)
The goal of this section was to create a stand-alone SWAT model that could effectively
quantify sediment loads at the field-scale and then track sediment through in-stream processes to
the watershed outlet. The field-scale SWAT project required custom watershed delineation in
order to ensure each field had quantifiable outputs and could be uniquely edited to guarantee
correct field-scale data, since it is unadvisable to rely on the accuracy of NLCD cover data at
such a small scale (Daggupati et al., 2011). For this reason, the subbasin map was made to
include all the field boundaries. The first step in the process was to load the elevation map
(NED) into the SWAT delineation framework and perform automatic delineation using
minimum area for flow accumulation, thus rendering the most detailed subbasin map and reach
network that the automatic delineation module could create. Then, the common land use unit
(CLU) field boundary shapefile was used to draw the individual field in the subbasin map (figure
2). The newly created subbasins included all field boundaries, so that each field from the field
map had a corresponding subbasin in the subbasin map. In the SWAT model, every unique
subbasin/field must have a unique reach segment. Meanwhile, the NHDPlus stream network is

96

not detailed enough and does not pass through all individual fields (figure 3); therefore, for every
extra subbasin created, a corresponding reach segment was added. The splitting of subbasins and
adding of reach segments was done manually using the high-resolution elevation map (10 m
resolution), NHDPlus stream network, and NHDPlus catchment map simultaneously as guides.
The goal was to create a subbasin and stream network as close to reality as possible. After the
shapefiles (subbasin and reach) were created, all records in the subbasin and reach files were
edited manually to ArcSWAT predefined network guidelines, which include giving each
subbasin and the reach a corresponding unique identification number. In addition, the reach flow
direction had to be defined by listing the identification numbers of the reach or reaches each
subbasin flowed into. The challenging nature of this task arises from the need for selective detail,
which is unattainable using automatic delineation in ArcSWAT. The subbasin and reach network
were designed to be very detailed where fields of interest existed and less detailed in all other
locations. By delineating the watershed in this way, relatively small fields could still be included
as unique subbasins and the calculation time would still be reasonable by keeping the total
number of subbasins to a workable number (163 subbasins in this study). The remaining model
set up included hydrologic response unit (HRU) definition. The HRU classification was set to
20% land cover, 10% soil cover, and 20% slope class by area. All agricultural land uses included
in the field data (corn, soybeans, winter wheat, tomatoes, and cabbage) were exempted from the
HRU classification. Finally, the model was calibrated and validated for the flow and sediment
loads at the watershed outlet. The creation of the field-level subbasin in SWAT model is a
unique aspect of this method, and to the best of our knowledge, has not been described in
published work.

97

Figure 5-2: (a) Automatically delineated subbasin and (b) predefined subbasin map

98

Figure 5-3: Comparing the NHDPlus stream network (a) versus the manual stream
network delineation (b)

99

5.5.1.1 RUSLE2 Field-scale Sediment Loads
In order to provide consistency between the four methods comparison and estimate
sediment load at higher accuracy level than SWAT model, RUSLE2 was used to calculate
sediment loads from each field. Detailed information on crop rotations, management practices,
and soil types are needed to attain satisfactory results. Producer-specific information can be
difficult to acquire, but is necessary for best results. In this study, all management and crop
rotation data were obtained in collaboration with the Lenawee County NRCS and Conservation
District. Every field was assigned specific crop rotations and managements based on information
provided by the farmers themselves. Although time consuming, this is a critical step in accurate
quantification of sediment yields at the field-scale. For the RUSLE2 model, detailed soil
descriptions were obtained from WSS. For a given field, between one to thirteen soil types were
identified. The total sediment load for each field was estimated by summing sediment load from
each soil type multiplied by the area coverage. For this reason, a spreadsheet tool was designed
in Microsoft Excel to assist in total load calculations. The spreadsheet tool was designed around
a matrix that included all possible RUSLE2 answers given the management operations and soil
types. This matrix was created based on one soil type at a time, rather than one management at a
time, which greatly reduced the amount of user time required for calculating sediment runoff
values with RUSLE2. Then, the total sediment load for each field was calculated by summing
the multiplications of sediment load (obtained from RUSLE2) by the number of hectares covered
by the specific management and soil type combination occupied in the field. In this way,
accurate total sediment load for each field could be calculated while reducing the amount of user
effort and time spent in the RUSLE2 interface.
100

5.5.1.2 Replacement of SWAT Field-Scale Sediment Runoff Estimation with RUSLE2
In order to incorporate the sediment loads estimated from the RUSLE2 model into the
calibrated SWAT model, the support practice factor was set to zero for each subbasin
representing a field within the SWAT model to eliminate sediment erosion. In the next step, a
point source input was added to each field equal to the amount of sediment runoff calculated by
RUSLE2 one at the time. After running the current scenario, BMPs were applied by changing the
point source value for each field from the current situation to the proposed BMP situation, also
calculated with RUSLE2. After each model run (before or after BMP implementation), the
sediment yield at the watershed outlet was obtained. Next, the difference between the average
annual sediment loads from the current and BMP scenarios was calculated. This difference was
defined as sediment savings at the watershed outlet due to the implementation of the BMP.
5.5.2 Subbasin-Scale Sediment Delivery Ratio (Method 2)
The second method was a combination of the RUSLE2 model and subbasin level SDR
calculated using SWAT. The average SDR was calculated for each subbasin defined by the
automatic delineation procedure. The method described by Woznicki and Nejadhashemi (2012)
was employed in this study. They defined SDR for a subbasin as the ratio of long-term sediment
yield at the watershed outlet to the total sediment load at the subbasin. This method involves
forcing sediment load to zero for one subbasin at a time and comparing the total sediment yield
at the watershed outlet to the amount of sediment at the watershed outlet with no subbasins
sediment yield set to zero. Sediment load is forced to zero for a subbasin by changing the USLE
support practice factor (USLE_P) to zero for every HRU in the subbasin of interest. The SDR
was calculated using Equation 5-9.

101

SDRi

YO,P 1 YO,P 1
Li,P 1 Li,P 1

(5-9)

The variable YO,P=1 represents the sediment yield at the watershed outlet under the base
scenario, when all subbasins are contributing sediment runoff. The sediment yield at the
watershed outlet when USLE_P is less than 1 is represented by YO,P<1 . The sediment load from
subbasin i contributed to reach i is Li,P=1 and the sediment load contributed to reach i from
subbasin i when USLE_P is less than 1 is Li,P<1. This process was completed for all subbasins
and SDR by subbasin map was created as shown in Figure 5-4.

102

Low

Medium

High

Figure 5-4. Subbasin-scale delivery ratio map

103

A SDR value was assigned to each field by layering the field map over the SRD map
created in the previous step. The overlay command used in the ArcGIS software and fields
contained within a certain subbasin were assigned that subbasin’s SDR. This SDR was
multiplied by the average annual sediment calculated for each BMP with the RUSLE2 model.
Sediment output at the watershed outlet was obtained for the period of study (2000-2009) before
and after BMP implementation on each field. The difference between these two values represents
average annual sediment savings at the watershed outlet from each BMP.
5.5.3 SWAT without Field-scale Delineation (Method 3)
The third method employed a SWAT model for in-stream process calculation and
RUSLE2 model to estimate sediment load at the edge of the field before and after the BMP
implementation. Similar to Method 1, the process begins by performing automatic watershed
delineation using minimum area for flow accumulation, thus rendering the most detailed
subbasin map and reach network that the automatic delineation module can create. Already
knowing the coordinate systems of all fields, the closest reach to each field was identified and
point source discharge to the reach was added to the SWAT project. Therefore, this method is a
simplified version of Method 1, in which the needs for manual extension of the NHDPlus reach
to each field within the watershed is eliminated. The model was run for each field twice (before
and after BMP implementation) and the 10 years (2000-2009) average sediment yield at the
watershed outlet was calculated. The difference between the average annual sediment loads was
defined as sediment savings at the watershed outlet.

104

5.5.4 Hybrid Solution Combining Analysis from the RUSLE2, SEDMOD, and SWAT
(Method 4)
The fourth method is the most accurate among the studied methods and designed to
account for three stages of sediment transport from a field to the watershed outlet: 1) estimate
sediment runoff from the field using the RUSLE2 model, 2) account for deposition that occurs in
overland flow between the field of interest and the closest NHDPlus reach using the SEDMOD
model, and 3) calculate sediment transportation from stream entry point to watershed outlet using
the SWAT model. This method is similar to Method 3, but employs SEDMOD to calculate the
percent of sediment runoff from a field that actually enters a nearby stream, while Methods 2 and
3 assume all sediment leaving a field enters the stream network in the watershed. The field-scale
estimate of sediment load was performed using the RUSLE2 model and spreadsheet tool similar
to Methods 1 through 3. In the next step, the SEDMOD model was used to estimate SDR from
each field to the closest stream. The SDR was calculated at 10 m resolution for the entire
watershed (figure 5). In the next step, the common land use unit map containing the fields of
interest was overlapped with the SEDMOD SDR map. Then, the average field level SDR was
calculated for all fields within the watershed using the Zonal Statistic command in the ArcGIS
software. The point sources were set in the SWAT model for the corresponding fields as a
product of RUSLE2 model times the SDR from SEDMOD. A high-resolution delineation was
needed for the point source placement. Since SWAT model only allows one point source per
subbasin; therefore, an ideal SWAT model for this application would have one point source for
every field. A Python program was developed that ran SWAT once for every unique point source
before and after BMP implementation for each field. Once the runs were completed, the total
sediment at the watershed outlet was compared

105

Figure 5-5. Grid-based delivery ratio map at 10 m resolution using the SEDMOD model

106

5.5.5 Best Management Practices (BMPs) Implementation Scenarios
The BMPs used in this study were proposed by the producers as part of the BMP Auction
program (Smith et al., 2009). In the BMP Auction program, producers submit bids to a buyer,
which would usually be a governmental agency involved in water resource management. The
bids consist of a proposed BMP and requested price for installation. Producer defined costs
create a competitive market and the costs are likely close to the true value of the BMP
(Johansson, 2006). This is the reverse of the conventional process, in which the governmental
agency defines the prices for a wide range of BMPs. Allowing producer to name their own prices
gives watershed planners a unique opportunity to see a more realistic value of a BMP from the
producers’ standpoint. After performing water quality analysis for each field participated in the
BMP Auction program, the bids are ranked based on dollar requested per ton of sediment
reduction at the watershed outlet. In this study, the BMPs were proposed by the producers
consist of a single or a group of BMPs. Six different BMP categories were included in this study:
cover crop, cover crops and filter strips, filter strips, residue management, residue management
and cover crops, and residue management and filter strips. All four methods were used to
evaluate the sediment reduction at the watershed outlet for all 80 BMPs within the six BMP
categories. This was performed to compare the inconsistency and variability in different BMP
efficiency caused by using different simulation techniques (methods).
5.5.6 Economic Analysis
Currently, the primary USDA nationwide conservation programs that fund and promote
BMPs are the Conservation Reserve Program (CRP), the Environmental Quality Incentive
Program, (EQIP), and the Conservation Stewardship program (CSP). These programs are

107

administrated by the NRCS a subset of the USDA (NRCS, 2012). However, in this study we are
focusing on the EQIP program because it is the largest program, with an annual budget of about
$1.3 billion (Shortle et al., 2012). The EQIP program is designed to assist agricultural producers
by helping to fund the implementation and maintenance of BMPs for up to 10 years per contract
(NRCS, 2012).
In the River Raisin watershed, we identified a group of the producers who were willing to
adopt the BMPs supported by EQIP program if they were given a chance to set their own price
and select their own BMP through the BMP Auction program. EQIP prices are generally set
based on dollars per unit area. This is an easy way to make payments to producers interested in
BMPs implementation; however, it does not take into account actual sediment reduction cost at
the field or the watershed outlet. In this analysis, prices defined by producers are compared to
prices from the EQUP program for the six BMP categories.
In this economic analysis, no external costs were considered. The monetary values used
represent the implementation and maintenance costs of the best management practices. In the
case of producer requested prices, the monetary values were those asked for by the producers in
the survey. In the case of the EQIP defined prices, the monetary values were payments defined
under the EQIP program based on the type of best management practice. The reason externalities
were not considered in this analysis was the goal of this economic exercise: to increase the cost
effectiveness of BMP programs. When evaluating cost effectiveness, the monetary values of
implementation and maintenance–the equivalent of what a government sponsored BMP program
would pay to implement a practice–are the costs of interest. Introducing externalities would
cause the results of the economic analysis to have prices that do not reflect actual money being
paid to producers, and the producer and EQIP price definitions could not be directly compared.
108

5.6 RESULTS AND DISCUSSION
5.6.1 SWAT Model Calibration/Validation
The SWAT model was the only model used in this study that requires calibration. Since
each method required a slightly different SWAT model setup, four separate calibration/validation
procedures were performed. Table 1 summarizes the overall calibration of validation results for
each of the four methods. Based on the criteria described in section 2.3.3.1, the model
performance during calibration and validation periods are satisfactory.
Table 5-1. Overall SWAT model calibration and validation for all four Methods

Method
1
2
3
4

NSE
0.58
0.69
0.55
0.50

Streamflow
RSR
0.64
0.56
0.67
0.70

PBIAS
9.25%
3.25%
5.05%
5.02%

NSE
0.58
0.53
0.58
0.52

Sediment
RSR
0.64
0.68
0.65
0.69

PBIAS
19.9%
26.9%
-7.28%
-1.94%

5.6.2 Predefined Field-Scale Subbasin and Reach Layers in SWAT Model (Method 1)
The first method only produced a few results, all on fields near the watershed outlet.
Among these fields, the farthest one located about 72 km upstream of the outlet. Implementation
of BMPs on all other fields did not affect sediment load at the watershed outlet. This might be
due to the fact that in SWAT model the small field sized subbasins did not produce enough
surface runoff to carry sediment off-site. The point source loads (calculated independently from
SWAT), resulted in extremely high sediment concentrations in the reach for each field. These
concentration values were above the maximum sediment capacity for the flow rate, and thus,
most of the sediment was settled within the subbasin reach and was not routed to the watershed
109

outlet. Thus, the limiting factor in this method is very low estimations of flow for very small
subbasins. Meanwhile, for some fields (6 out of 80), a small sediment reduction at the watershed
outlet was detected. A possible reason for these results that is the small amount of sediment that
was leaving the subbasin was close enough to the outlet to not settle within the river network
before reaching the watershed outlet.
The processes described above highlight the disadvantage of field-scale study in the
SWAT model. This stems from the requirement of the SWAT model that all subbasins must
contain a reach. Since field-sized subbasins were made, reaches that do not actually exist must be
added to the river network, although they were created with scrutiny and followed likely patterns
based on a high-resolution digital elevation model, the high-resolution NHDPlus stream network,
and were combined with the subbasin drainage areas to make a robust SWAT project. In reality,
the overland flow portion of the sediments’ movement to the streams is much longer, and not all
fields drain directly into streams. This method would perhaps be appropriate in a watershed with
many open drainage systems adjacent to the fields. Although this network might be realistic, the
same problem of insufficient flow for sediment transport would likely occur.
5.6.3 Subbasin-Scale Sediment Delivery Ratio (Method 2)
Method 2 was designed for ease of use and took a relatively shorter time to complete than
the other methods. This low time requirement is because no field is directly modeled within the
SWAT model. However, initial SWAT runs are still required to establish SDR values for each
subbasin, which according to Woznicki et al. (2012) is equal to the number of subbasins in the
SWAT project. Since in this study 80 BMPs scenarios were tested, at least 160 SWAT runs are
required for Methods 1, 3, and 4 while just over 100 runs are required for Method 2 because

110

some fields are located within the same subbasin. In addition, the added time for defining,
calculating, and applying point sources is avoided in Method 2. Overall, the second method was
successful in quantifying sediment load at the watershed outlet for all fields. There are, however,
limitations with this model. In Method 2, all fields within the same subbasin are all given the
same delivery ratio, regardless of proximity to the stream, soil type, and a host of other factors
that determine SDR. Therefore, this method ignores the net sediment deposition that occurs in
overland flow process. On the other hand, 100% of sediment load at the edge of a field reaches
the nearest stream, which is not a realistic representation.
5.6.4 SWAT without Field-scale Delineation (Method 3)
The third method built on the idea of Method 2, but rather than estimating SDR as a
single value (long-term annual average of SDR per subbasin) this method included point sources,
thus accounting for daily in-stream processes throughout the year. However, it was again
assumed that 100% of the sediment leaving the fields entered the reach, therefore ignoring
deposition during the overland processes. Similar to Method 2, automatic delineation was used in
Method 3 and; therefore, in the case that multiple fields were contained within one subbasin, the
fields were forced to share a point source. This method is also capable of capturing field-scale
management practices at the watershed outlet.
5.6.5 Hybrid Solution Combining Analysis from the RUSLE2, SEDMOD, and SWAT
(Method 4)
The final method accounts for three stages of sediment delivery from the field to the
watershed outlet. Some of the most used and trusted models are employed for their respective
applications, and the results are compiled. RUSLE2 effectively quantifies average field sediment

111

runoff, while SWAT tracks the sediment transport from its introduction into the stream to the
watershed outlet. An estimation of the amount of sediment entering the nearest stream from the
field is made using SEDMOD model. The SWAT model is used to quantify in-stream processes
from the stream inlet to the watershed outlet. Overall, this hybrid solution had the most logically
complete strategy to estimate the impacts of field-scale managements at the watershed outlet.
The main difference in this method compared to the others is the inclusion of a sediment
deposition after leaving the field and before reaching to the stream system affected by overland
flow. The SEDMOD model is based on soil characteristics, elevation, and proximity to reach.
Although this is not a comprehensive understanding of overland flow, it provides an
approximation that is otherwise lacking in these procedures. Estimated percentage of sediment
leaving the fields and entering the reach ranged from 13% to 29%, much lower than the previous
assumption of 100%. Thus, the sediment point source values for Method 4 were much lower than
the other methods.
5.6.6 Overall Method Comparison
In order to summarize our findings, a comparison of the advantages and disadvantages of
each method described above is presented in table 5-2.

112

Table 5-2. Overview of the four methods
Method
Advantages
1
*Field-scale subbasins allow for direct
estimation of sediment load for each
individual field

Disadvantages
*Due to low runoff rate for small
subbasins (fields) most sediment loads
are deposited within the subbasins
*Data preparation period is timeconsuming
*Unable to estimate the impact of fieldscale management practices at the
watershed outlet

2

*Simplest method tested
*Fastest calculation time

* No field is directly modeled within the
SWAT model
*Assumes 100% of sediment leaving field
enters the reach
*SDR resolution is limited to the subbasin level
*By using the long-term average SDR value,
ignores the temporal SDR variation

3

*Uses the full potential of the
SWAT model to estimate in-stream
processes during sediment delivery
*Requires fewer point sources than
Method 2

*Ignores overland sediment deposition
*Multiple fields can be represented by the same
point source
*Field location within subbasins is ignored

4

* Accounts for three stages of
sediment delivery: field-scale,
overland sediment deposition, and
in-stream process
*Considers field location

*Requires extra step for estimating overland
sediment deposition using the SEDMOD model
*SWAT snaps the close point sources and
(correspondence to different fields) force them
to share a similar point discharge to a reach

113

5.6.7 Overall Method Comparison in Evaluating BMP Implementation Scenarios
Each method was used to evaluate the sediment reduction at the watershed outlet for all
80 BMPs (six BMP categories). This was performed to compare the inconsistency and variability
in BMPs efficiencies caused by using different simulation techniques (methods). As it was
discussed earlier, the fourth method was considered as the most accurate among the studied
methods and designed to account for three stages of sediment transport from a field to the
watershed outlet. Therefore, the performances of different methods were compared to Method 4.
The results are presented in Figure 6.
In the cover crop category, method 3 had the widest range of predicted sediment
reductions at the watershed outlet; the minimum value was 1.74 tons and the maximum value
was 64.3 tons. The median value for Method 3 was 17.7 tons. Method 2 predicted the next
highest range (51 tons) and median (6.08 tons) for cover crop values. Therefore, comparing to
Method 4 (median is 3.03 tons), Methods 2 and 3 overestimating the sediment reduction 101%
and 484%, respectively. Method 1 predicted a minimum sediment reduction equal to 0 tons, a
maximum of 0.86 tons, and a median of 0 tons. However, the range and median values for all
other BMP categories are 0.
Method 2 had the highest range in cover crop and filter strip category and predicted a
minimum of 0.536 tons, a maximum of 33.9 tons, and a median of 10.6 tons. Method 3 predicted
the highest median value at 14.6 tons, with a range of 4.05 tons. Methods 2 and 3 over predicted
sediment reduction at 303% and 455%, respectively, when compared to Method 4.
Similar trends were observed for filter strip, residue management, and residue
management and filter strip categories. However, within the residue management and cover crop

114

category, smaller variation was observed between Method 2 through 4. The median values for
Methods 2, 3 and 4 are 9.79, 8.80, and 4.21 tons, respectively. Meanwhile, the variations in
BMPs’ effectiveness are quite large and varies from 152.2 tons for Method 3 to 16.2 tons for
Method 4.
Finally, in the residue management and filter strip category, the largest sediment
reduction was observed for all methods except Method 1. In this BMP category, Method 3 had
the highest median value at 38.7 tons, follow by 25.2 tons for Method 2 and 13.8 tons for
Method 4. At the same time, the ranges for BMP efficiency are the smallest between all BMP
categories.
Overall, evaluation of field-scale impact of management practices at the watershed outlet
using Method 1 resulted in the lowest median values for all BMP categories and in many cases
showed no impact at the watershed outlet. This highlights the disadvantage of field-scale study in
the SWAT model that originates from a low estimation of flow for each field (subbasin)
incapable of transporting the majority of eroded sediments to the reach and the watershed outlet.
Meanwhile, Methods 2 and 3 consistently produced higher median sediment loads than method
4. This is likely due to ignoring the sediment delivery component from the field to the stream in
these methods. Method 4 employs the SEDMOD model to account for this stage of sediment
transport, and sediment loading values at the stream were reduced by over 70% in most cases.
The lower median value for Method 4 is a trend across all BMP categories and proves that field
overland flow and deposition between the field outlet and stream inlet is an important step to
include in field-scale BMP evaluation.

115

Figure 5-6. Comparison of four modeling strategies: a is cover crop, b is cover crop and
filter strip, c is filter strip d is residue management, e is residue management and cover
crop, and f is residue management and filter strip
116

5.6.8 Economic Analysis
The focus of an economic analysis of a watershed project is defined by the project goals.
Three main goals addressed by this economic analysis were identified as: 1 increase producer
participation, then develop an analysis focused on the price of BMP implementation per unit of
application area to provide insight into how much producers believe certain BMPs to be worth;
2) sustain field productivity by keeping soil on site – a field-level analysis of BMP
implementation will allow watershed stakeholders to find the BMPs most effective at erosion
reduction; and 3) protection of lakes, downstream structures, and aquatic ecosystems –
evaluating the price of BMPs on a per ton sediment reduced at the watershed outlet allows
watershed planners to identify BMPs that will result in greatest water quality improvement.
5.6.8.1 Goal 1: Improve Producer Participation in Conservation Programs
The goal of improving producer participation is simple: increase interest and therefore the
number and quality of agricultural practices aimed at improving water quality. The BMP Auction
addresses the top reasons producers list that keep them out of government funded projects (Smith
et al., 2009). Economically, this means allowing producers to define their own prices for BMP
implementation. These prices are most often defined on a per unit area basis. In this study,
producers defined their own prices, and, in the Table 3, they are compared to EQIP prices that
would be paid for the same practices through the NRCS. The EQIP and producer prices are
weighted based on area used for each BMP. For example, the filter strip EQIP price is $1,032.90
per ha, but the cover crop and filter strip combination it is only $61.57, which is the weighted
average for the area used for cover crop and filter strip. Therefore, the $/ha for cover crop and

117

filter strip is much less than just filter strip because the majority of the area is devoted to cover
crops rather than filter strip, while the entire area for filter strip is under one BMP.
Table 5-3: Median average $/ha for producer requested and EQIP defined prices

BMP Category
Cover Crop
Cover Crop and Filter Strip
Filter Strip
Residue Management
Residue Management and Cover
Crop
Residue Management and Filter Strip

Median Producer Requested
Price
$74.13
$128.60
$97.41
$24.71

Median EQIP
price
$48.16
$61.57
$1,032.90
$22.61

$27.18

$35.39

$45.20

$36.76

In four out of six BMP categories, median producer requested prices were greater than
that of EQIP. The residue management and cover crop combination and filter strip categories
stand out as having lower prices than the EQIP. Meanwhile, EQIP is over-valuing most of the
BMP categories compared to producers on per unit area basis. It is important to note that initially
we were expecting that the producers’ requested price for BMP implementation would higher
than the EQIP program because they are currently participating in the EQIP program. However,
the large discrepancy in price requested for the filter strip results from the fact that he filter strip
program under EQIP is rigid (in terms of width, harvesting, and the length of the contract).
However, the length of the contract under the BMP Auction can be one to several years. These
differences suggest that producers may be willing to implement the filter strip under significantly
lower cost if the regulations become less rigid.

118

5.6.8.2 Goal 2: Field Soil Conservation and Agricultural Sustainability
Evaluating prices based on sediment runoff reduction at the field-scale allows watershed
planners to measure the effectiveness each BMP to save soil on site. This analysis is appropriate
when project goals are related to soil conservation and agricultural sustainability. Both
producers’ requested and EQIP prices per unit sediment runoff reduction at the field-scale are
presented in the Table 5-4. This comparison is important, because it illustrates the difference in
true price of sediment reduction at the field-scale.
Table 5-4: Producer Requested and EQIP Median Prices per Ton Sediment Reduction at
the Field Outlet

BMP Category
Cover Crop
Cover Crop and Filter Strip
Filter Strip
Residue Management
Residue Management and
Cover Crop
Residue Management and
Filter Strip

Median Producer Requested Price
per Ton of Sediment Reduction
($/ton)
$73.59
$199.29
$107.94
$22.05

Median EQIP Price per
Ton of Sediment
Reduction ($/ton)
$47.22
$98.10
$1,151.92
$20.17

$23.10

$18.90

$12.09

$9.83

In five out of six BMP categories, the producer requested median cost per ton of sediment
reduction was more than those defined by EQIP. The filter strip category was the exception,
suggesting that, water quality can be significantly improved if the EQIP program becomes less
rigid in its requirements. Meanwhile, the combined residue management and filter strip program
is the most effective BMP category among the studied BMPs with the price of $12.09 and $9.83
for the producer requested and EQIP, respectively.

119

5.6.8.3 Goal 3: Watershed-Scale Sediment Reduction
In the previous sections of this paper, a comparison between four methods illustrated how
different modeling techniques can have widely varying results, even when similar input data are
used. For this reason, it is extremely important to select the simplest approach that not only
accurate but also effectively address the goal of the study (Nejadhashemi et al., 2009).
Ultimately, Method 4 was selected for economic analysis because the comparison and evaluation
of results deemed this method as the most accurate that can answer the question in hand, which is
the true cost of sediment reduction at the watershed outlet. In addition, Method 4 was the only
method that directly accounted for all three stages of sediment transport and produced
measurable results. Measuring the effectiveness of BMPs based on sediment reduction at the
outlet gives watershed planners a way to measure BMP effectiveness based on overall watershed
water quality improvement.
In figure 7, the ―producer requested‖ prices are total prices submitted by the each
producer divided by the calculated sediment reduction at the watershed outlet. The EQIP prices
are the total cost for each BMP defined under EQIP guidelines divided by the reduction at the
watershed outlet. Overall similar trends were observed at watershed-scale and field-scale.
However, the box and whisker plots also allow examining the range in addition to the median
values. Among the studied BMP categories, only filter strip has the higher range in EQIP
program than the producer requested costs. Meanwhile, the median cost of sediment reduction at
watershed outlet using filter strip is $10,914 under the EQIP, significantly larger than producers
requested price of $811. However, the median price for the rest of the BMPs under the EQIP

120

program is varies from $155 (residue management and filter strip) to $1140 (cover crop and filter
strip). These values are $193 (residue management) to $2041(cover crop and filter strip) per ton
of sediment reduction under the BMP Auction program (producer requested price). This can
provide valuable information to watershed manager and stakeholders to estimate the true cost of
conservation programs.

121

Figure 5-7. Comparison of dollar spent per ton sediment reduction at the watershed outlet
under farmer defined and EQIP prices: a is cover crop, b is cover crop and filter strip, c is
filter strip d is residue management, e is residue management and cover crop, and f is
residue management and filter strip

122

5.7 CONCLUSION
In this study, four methods were compared in order to evaluate a simple but effective
technique for quantifying the impact of field-scale management practices at the watershed outlet
in the River Raisin watershed. The methods tested were using: (1) predefined field-scale
subbasin and reach layers in SWAT model; (2) subbasin-scale SDR; (3) results obtained from the
field-scale RUSLE2 model as point source inputs to the SWAT watershed model; (4) a hybrid
solution combining analysis from the RUSLE2, the Spatially Explicit Delivery Model
(SEDMOD), and SWAT models.
Method 1 proved to be an ineffective way to evaluate the field-scale BMP
implementation strategy based on sediment reduction at the watershed outlet. The limiting factor
in this method is very low estimations of flow for very small subbasins. Meanwhile, Methods 2
and 3 were able to quantify the field-scale management practices at the watershed outlet, but
ignore the net sediment deposition occurs in overland flow process. Method 4 in the most
accurate among the studied methods and designed to account for three stages of sediment
transport from a field to the watershed outlet: 1) estimate sediment runoff from the field, 2)
account for deposition that occurs in overland flow between the field of interest and the closet
reach, and 3) calculate sediment transportation from stream entry point to watershed outlet. In
this method, RUSLE2, SEDMOD, and SWAT hybrid model effectively quantified all the fieldscale management actions at the watershed outlet. Overall, the results of this study showed that
the SWAT model is not capable of capturing field-scale activities and delineation should be
limited to reasonably sized subbasins suggested by automatic delineation or predefined NHD
Plus catchments. In addition, the second stage of sediment transport, sediment deposition on

123

overland, should be included in future studies dealing with development of field-scale BMP
implementation strategies.
The above-described methods were used to evaluate the impact of field-scale BMP
implementation scenarios at the watershed outlet. Six different BMP categories were included in
this study: cover crop, cover crops and filter strips, filter strips, residue management, residue
management and cover crop, and residue management and filter strips. Overall, analysis using
Method 1 provided the lowest median value for all BMP categories, and failed to measure any
difference at for over 90% of the fields in this study. Using the SWAT model at the field-scale
has the main disadvantage of the low flow estimations for small subbasins (fields), causing
estimated sediment erosion to stay on-site, never reaching the stream network. Methods 2 and 3
produced higher median sediment loads than method 4. The lack of a sediment delivery
component is the most likely reason for the higher estimates. Method 4 accounts for all stages of
sediment transport to the watershed outlet. The SEDMOD model is used to estimate overland
flow and deposition between the field outlet and stream inlet. This stage of sediment transport
reduced loading values at the stream by over 70% in some cases. Method 4 predicted lower
median values across all BMP categories, proving that overland flow and deposition between the
field outlet and stream inlet is important to consider when evaluating the effects field-scale
BMPs have at the watershed outlet.
An economic analysis using producer requested prices and EQIP program for BMP
implementation has identified how producers value BMP on a per unit area basis compared to
current water quality programs. The true costs of sediment reduction on both the field-scale and
watershed scale from producer requested prices and current water quality improvement projects
were quantified. On a per unit area basis, producers requested higher prices than EQIP in four
124

out of six BMP categories. Meanwhile, the true cost of sediment reduction at the field and
watershed scales were greater than EQIP in five out of six BMP categories according to the
producer requested prices. These results indicate that the true values of BMPs through the eyes
of the producers are not being adequately addressed with the EQIP program. More producer
input is needed when government agencies design water quality improvement programs. Or
alternatively, true cost can be more directly addressed by implementing non-traditional programs
such as the BMP Auction. In either case, it is clear that producer input and participation is
paramount to the success of water quality programs.
Future work focusing on development of an all-in-one model capable of providing
information at three stages of sediment transport from a field to the watershed outlet would give
watershed planners an effective tool for decision making. In addition, it is important to perform
similar economic analysis in different watersheds to better understand the true costs of
conservation programs.

125

6. CONCLUSIONS

Recently, different methods for estimating field-scale sediment loads have been
developed. These methods were designed to provide information to decision makers with the
intention of improving water quality at the watershed level. SWAT, Field_SWAT, and the HIT
model are a few of these methods. No information was provided in the literature about the
reliability of the field-scale estimations from these models. Quantifying the effectiveness of these
methods to produce accurate field-scale sediment yield estimations is an important step towards
using the models responsibly. In this study, the reliability of each of the three models was tested
against a detailed, calibrated field-scale model, RUSLE2, resulting in the following conclusions:
the SWAT model produced the most accurate estimates in comparison to RUSLE2, by
providing the closet median and the lowest absolute error in sediment yield predictions
a statistical analysis found SWAT to be the only watershed that provided a data set
insignificantly different from the calibrated RUSLE2 model, showing that SWAT
performed best at field-scale analysis among the watershed-scale models
all watershed-scale models were found to be incapable of identifying the three areas of
priorities similar to RUSLE2 model
the SWAT model provided the most correct estimates within the uncertainty bounds, at
51%, while Field_SWAT provided 27%, and HIT 39%, showing that none of the
watershed models are satisfactory for field-scale analysis
Based on these conclusions, four methods were compared to evaluate techniques designed
for quantifying the impact field-scale management changes have at the watershed outlet in the
River Raisin watershed. Since none of the watershed-scale models alone were capable of field126

scale analysis, the field-scale component of each of the following methods was calculated using
RUSLE2. The SWAT model was used for the in-stream dynamics portion of the analysis. The
methods tested were: (Method 1) predefined field-scale subbasin and reach layers in SWAT
model; (Method 2) subbasin-scale sediment delivery ratio; (Method 3) results obtained from the
field-scale RUSLE2 model as point source inputs to the SWAT watershed model; (Method 4) a
hybrid solution combining analysis from the RUSLE2, SEDMOD, and SWAT models. Each of
these methods was used to evaluate impacts of BMPs at the watershed outlet. Six different BMP
categories were included in this study: cover crop, cover crops and filter strips, filter strips,
residue management, residue management and cover crop, and residue management and filter
strips. The following conclusions can be drawn from the analysis:
Method 4, the hybrid RUSEL2-SEDMOD-SWAT model was identified as the only
reliable method of the four tested to estimate the effects of individual BMPs at the
watershed outlet
the true values of BMPs defined by producers are not being adequately addressed with
the EQIP program; therefore, more producers inputs are needed when government
agencies design water quality conservation programs

127

7. RECOMMENDATIONS FOR FUTURE RESEARCH

In this study, a detailed analysis of field-scale sediment yield estimations and their effects at
the watershed outlet was performed using multiple models and methods. In addition, an
economic analysis quantifying the true costs of BMPs both at field and watershed scale was
performed. However, there is a clear need for research in order to integrate effective field-scale
estimations and true costs of BMPs into watershed-scale analysis, decision making, and
conservation planning. Suggestions for future research based on the findings of this study
include:
Investment in collecting and digitizing filed-scale management operations, which found
to be a key in reliable estimation of sediment load both at the field and watershed scales.
The development of an all-in-one model capable of providing information at three stages
of sediment transport from a field to the watershed outlet. This would be an effective tool
for watershed planners. Depending of the goals of a water quality program, BMPs could
be ranked and funded accordingly.
Expanding the economic analysis outlined in this paper to different watersheds around
the country. These actions would increase the understanding of the true costs of
conservation programs in different localities. Studies such as this can help with the
development of more effective conservation programs and shed light on the true costs of
sediment reduction, building a knowledge base that can guide conservation planners in
future water quality programs.

128

APPENDIX

129

APPENDIX

Figure A-1. Schematic of logical flow path for statistical analysis

130

Figure A-2. Histogram of residuals (data not transformed)
Table A-1. Residuals (data not transformed), P-values
Test
Shapiro-Wilk
Kolmogorov-Smirnov

--Statistic--W
0.502371
D
0.322231

-----p Value-----Pr < W
<0.0001
Pr > D
<0.0100

131

Figure A-3. Histogram of residuals (log transformed raw data)
Table A-2. Residuals (log transformed raw data), P-values
Test
Shapiro-Wilk
Kolmogorov-Smirnov

--Statistic--W
0.887311
D
0.125953

-----p Value-----Pr < W
<0.0001
Pr > D
<0.0100

*1.4 check normality of residuals sqrt;

132

Figure A-4. Residuals (square root transformed raw data)
Table A-3. Residuals (square root transformed raw data), P-values
Test
Shapiro-Wilk
Kolmogorov-Smirnov

--Statistic--W
0.753258
D
0.198728

-----p Value-----Pr < W
<0.0001
Pr > D
<0.0100

*Based on W-statistic, select log transform;

133

Figure A-5. Residual Variances for each model

134

Table A-4. Levene’s test for equality of variances
Type 3 Tests of Fixed Effects

Effect
scen

Num Den
DF DF F Value Pr > F
3

160

4.49

0.0047

Table A-5. Reject null hypothesis (alpha=0.05). The residuals are not homogeneous
Model
2.1
2.2

AIC
330.9
235.1

135

BIC
330.9
235.3

Table A-6. ANOVA Table
Type 3 Tests of Fixed Effects

Effect

Num
Den
DF DF F Value Pr > F

scen

3

73

82.44

<.0001

Least Squares Means

Effect

scen

scen
scen
scen
scen

FSWAT
HIT
RUSLE
SWAT

Standard
Estimate Error
1.0097
0.2101
1.3849
1.3862

DF

0.1071
40
0.01985 40
0.09859 40
0.1382
40

t Value

Pr > |t|

9.43
10.59
14.05
10.03

<.0001
<.0001
<.0001
<.0001

Differences of Least Squares Means

Effect

scen

_scen

scen
scen

FSWAT HIT
FSWAT RUSLE

scen
scen
scen
scen

FSWAT
HIT
HIT
RUSLE

SWAT
RUSLE
SWAT
SWAT

Standard
Estimate
Error

DF

t Value

Pr > |t|

0.7996
-0.3752

0.1089 42.7
0.1456 79.5

7.34
-2.58

<.0001
0.0118

-0.3765
-1.1748
-1.1761
-0.00133

0.1748 75.3
0.1006 43.2
0.1396 41.7
0.1697 72.3

-2.15
-11.68
-8.43
-0.01

0.0344
<.0001
<.0001
0.9938

136

Table A-7. Input data for statistical analysis
SWAT
SWAT
SWAT
SWAT
SWAT
SWAT
SWAT
SWAT
SWAT

1
2
3
4
6
7
8
9
10

HIT
HIT
HIT
HIT
HIT
HIT
HIT
HIT
HIT

1
2
3
4
6
7
8
9
10

RUSLE
RUSLE
RUSLE
RUSLE
RUSLE
RUSLE
RUSLE
RUSLE
RUSLE

1
2
3
4
6
7
8
9
10

FSWAT
FSWAT
FSWAT
FSWAT
FSWAT
FSWAT
FSWAT
FSWAT
FSWAT

2.424
1.831
3.381
0.666
6.919
1.421
0.815
1.594
1.841

1
2
3
4
6
7
8
9
10

…
0.101
0.134
0.096
0.137
0.379
0.437
0.166
0.226
0.309
…
1.457
1.457
1.457
3.138
5.156
2.242
2.690
0.650
0.650
…
14.717
20.668
12.129
8.535
0.799
2.466
1.201
3.050
3.275
…

137

List A-1. Code used to execute statistical analysis
data andy;
input scen$ sub sed;
cards;
;run;
proc print data=andy;
run;
*checking assumptions;
proc mixed data=andy;
class scen;
model sed=scen/ outp=mr;
proc print data=mr; run;
*1.1 checking normality of residuals;
proc univariate data=mr normal plot;
var resid; histogram resid;
run;
*1.2 data transform;
data andy2; set andy;
sedadd=sed+1;
sedlog = log(sedadd);
sedsqrt = sqrt(sed);
run;
proc print data=andy2;
run;
*1.3 check normality of residuals log;
proc mixed data=andy2;
class scen;
model sedlog=scen/ outp=mrlog;
run;
proc univariate data=mrlog normal plot;
var resid; histogram resid;
run;
*1.4 check normality of residuals sqrt;
proc mixed data=andy2;
class scen;
model sedsqrt=scen/ outp=mrsqrt;
run;
proc univariate data=mrsqrt normal plot;
var resid; histogram resid;
run;
*Based on W-statistic, select log transform;

138

List A-1 (cont'd)
*1.5 Checking equality of variances;
proc sort data=mrlog; by scen; run;
proc univariate data=mrlog normal plot;
by scen;
var resid;
run;
*1.6 Levene's test;
data mrlog; set mrlog;
res2=resid*resid;
absres=abs(resid);
proc mixed data=mrlog;
class scen;
model res2=scen;
run;
*2.1 Model assuming equal variances;
proc mixed data=andy2;
class scen;
model sedlog=scen;
run;
*2.2 Model assuming unequal variances - this model is selected based on lower
AIC and BIC;
proc mixed data=andy2;
class scen;
model sedlog=scen/ddfm=satterthwaite;
repeated /group=scen;
run;
*3.1 Final model;
proc mixed data=andy2;
class scen;
model sedlog=scen/ddfm=satterthwaite;
repeated /group=scen;
lsmeans scen/pdiff;
run;

139

REFERENCES

140

REFERENCES

ACKERMAN, D., SCHIFF, K. C. & WEISBERG, S. B. 2005. Evaluating HSPF in an Arid,
Urbanized Watershed. Journal of the American Water Resources Association, 41, 477486.
ALBEK, M., ÖĞÜTVEREN, Ü. B. & ALBEK, E. 2004. Hydrological modeling of Seydi Suyu
watershed (Turkey) with HSPF. Journal of Hydrology, 285, 260-271.
ALSTON, J. M., ANDERSEN, M. A., JAMES, J. S. & PARDEY, P. G. 2010. Persistence Pays,
233 Spring Street, New York, NY, 10013, Springer Science and Business Media, LLC.
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, 73-89.
BARTHOLIC, J., O'NEIL, G. & SHI, Y. A web-accessible watershed-based system target's land
areas at highest risk for sediment loss to streams. The NABS 57th Annual Meeting, 2009
Michigan State University,1405 S. Harrison Rd.,101 Manly Miles Bldg.,East Lansing,MI
48823.
BAUMGART-GETZA, A., PROKOPYB, L. S. & FLORESS, K. 2012. Why farmers adopt best
management practice in the United States: A meta-analysis of the adoption literature.
Journal of Environmental Management, 96, 17-25.
BHADURI, B., HARBOR, J., BENGEL, B. & GROVE, M. 2000. Assessing Watershed-Scale,
Long-Term Hydrologic Impacts of Land-Use Change Using a GIS-NPS Model.
Environmental Management, 26, 643-658.
BORAH, D. K. 2011. Hydrologic procedures of storm event watershed models: a comprehensive
review and comparison. Hydrological Processes, 25, 3472-3489.
BOSCH, D., CHO, J., LOWRANCE, R., VELLIDIS, G. & STRICKLAND, T. 2010.
Assessment of Riparian Buffer Impacts within the Little River Watershed in Georgia
USA with the SWAT Model. Watershed Technology: Improving Water Quality and
Environment. Universidad EARTH, Costa Rica: American Society of Agricultural
Engineering.
BOSCH, D., THEURER, F., BINGNER, R., FELTON, G. & CHAUBEY, I. 2001. Evaluation of
the AnnAGNPS Water Quality Model. Agricultural Non-point Source Water Quality
Models: Their Use and Application, Southern Cooperative Series Bulletin #398, 45-62.

141

BOSCH, D. D., SHERIDAN, J. M., BATTEN, H. L. & ARNOLD, J. G. 2004. Evaluation of the
SWAT Model on a Coastal Plain Agricultural Watershed. Transactions of the ASAE, 47,
1493-1506.
BOSCHA, N. S., ALLANB, J. D., DOLANC, D. M., HAND, H. & RICHARDSE, R. P. 2011.
Application of the Soil and Water Assessment Tool for six watersheds of Lake Erie:
Model parameterization and calibration. Journal of Great Lakes Research, 37, 263-271.
BOSSIO, D., GEHEB, K. & CRITCHLEY, W. 2010. Managing water by managing land:
Addressing land degradation to improve water productivity and rural livelihoods.
Agricultural Water Management, 97, 536-542.
BROWN, T. C. & FROEMKE, P. 2012. Nationwide Assessment of Nonpoint Source Threats to
Water Quality. Bioscience, 62, 136-146.
CHEN, C.-F., LIN, J.-Y., KANG, S.-F., LEE, Y.-J. & YANG, C.-H. 2010. Predicting the LongTerm Performance of a Structural Best Management Practice with the BMP ToolBox
Model. Environmental Engineering Science, 27, 55-64.
CHEN, D. Y., CARSEL, R. F., MCCUTCHEON, S. C. & NUTTER, W. L. 1998. Stream
Temperature Simulation of Forested Riparian Areas: I. Watershed-Scale Model
Development. Journal of Environmental Engineering, 124, 304-315.
CHOI, J.-Y., ENGEL, B. A., THELLER, L. & HARBOR, J. 2005. Utilizing Web-Based GIS and
SDSS for Hydrological Land Use Change Impact Assessment. Transactions of the
ASABE, 48, 815-822.
CIA 2012. The World Factbook. In: AGENCY, C. I. (ed.).
CIBIN, R., SUDHEER, K. P. & CHAUBEY, I. 2010. Sensitivity and identifiability of stream
flow generation parameters of the SWAT model. Hydrological Processes, 24, 1133-1148.
DABROWSKI, J. M. & SCHULZ, R. 2003. Predicted and measured levels of azinphosmethyl in
the Lourens River, South Africa: Comparison of runoff and spray drift. Environmental
Toxicology and Chemistry, 22, 494-500.
DAGGUPAT, P., DOUGLAS‐MANKIN, K. R., SHESHUKOV, A. Y., BARNES, P. L. &
DEVLIN, D. L. 2011. Field Level Targeting Using SWAT: Mapping Output From HRUs
to Fields and Assessing Limitations of GIS Input Data. Transactions of the ASABE, 54,
501-514.
DAS, S., RUDRA, R. P., GHARABAGHI, B., K. GOEL, P., SINGH, A. & AHMED, I. 2007.
Evaluation of Nutrient Component of AnnAGNPS Model in a Watershed in Ontario.
Watershed Management to Meet Water Quality Standards and TMDLS (Total Maximum
Daily Load) Fourth Conference. San Antonio, Texas USA: American Society of
Agricultural and Biological Engineers.

142

DE LANGE, H. J., DE JONGE, J., DEN BESTEN, P. J., OOSTERBAAN, J. & PEETERS, E. T.
H. M. 2004. Sediment pollution and predation affect structure and production of benthic
macroinvertebrate communities in the Rhine–Meuse delta, The Netherlands. Journal of
the North American Benthological Society, 23, 557-579.
DEASY, C., BRAZIER, R. E., HEATHWAITE, A. L. & HODGKINSON, R. 2009. Pathways of
runoff and sediment transfer in small
agricultural catchments. Hydrological Processes, 23, 1349-1358.
DIEBEL, M. W., MAXTED, J. T., NOWAK, P. J. & VANDER ZANDEN, M. J. 2008.
Landscape Planning for Agricultural Nonpoint Source Pollution Reduction I: A
Geographical Allocation Framework. Environmental Management, 42, 789-802.
DOA 2008a. Farm Bill 2008 At a Glance: Agricultural Management Assistance. In: SERVICE,
D. O. A. N. R. C. (ed.).
DOA 2008b. Farm Bill 2008 At a Glance: Agricultural Water Enhancement Program. In:
SERVICE, U. S. D. O. A. N. R. C. (ed.).
DOA 2008c. Farm Bill 2008 At a Glance: Chesapeake Bay Watershed. In: AGRICULTURE, U.
S. D. O. (ed.).
DOA 2008d. Farm Bill 2008 At a Glance: Conservation Innovation Grants. In: SERVICE, U. S.
D. O. A. N. R. C. (ed.).
DOA 2008e. Farm Bill 2008 At a Glance: Conservation of Private Grazing Land Program. In:
SERVICE, U. S. D. O. A. N. R. C. (ed.).
DOA 2008f. Farm Bill 2008 At a Glance: Cooperative Conservation Partnership Initiative. In:
SERVICE, U. S. D. O. A. N. R. C. (ed.).
DOA 2008g. Farm Bill 2008 At a Glance: Farm and Ranch Lands Protection Program. In:
SERVICE, U. S. D. O. A. N. R. C. (ed.).
DOA 2008h. Farm Bill 2008 At a Glance: Grassland Reserve Program. In: SERVICE, U. S. D.
O. A. N. R. C. (ed.).
DOA 2008i. Farm Bill 2008 At a Glance: Wetlands Reserve Program. In: SERVICE, U. S. D. O.
A. N. R. C. (ed.).
DONIGIAN, A. S. 2002. Watershed Model Calibration and Validation: The HSPF Experience.
Water Environmental Federation Proceedings. Water Environment Federation.
DOWD, B. M., PRESS, D. & LOS HUERTOS, M. 2008. Agricultural nonpoint source water
pollution policy: The case of California's Central Coast. Agriculture, Ecosystems &
Environment, 128, 151-161.

143

ENDRENY, T. A. & WOOD, E. F. 2003. Watershed Weighting of Export Coefficients to Map
Critical Phosphorous Loading Areas. Journal of the American Water Resources
Association, 36, 1053-1062.
EPA 2000. BASINS Technical Note 6: Estimating Hydrology and Hydraulic Parameters for
HSPF. In: AGENCY, U. S. E. P. (ed.).
EPA 2001. PLOAD version 3.0 An ArcView GIS Tool to Calculate Nonpoint Sources of
Pollution in Watershed and Stormwater Projects. United States Environmental Protection
Agency.
EPA. 2003. Final Water Quality Trading Policy [Online]. Available:
http://water.epa.gov/type/watersheds/trading/finalpolicy2003.cfm [Accessed 1 April 2012
2012].
EPA 2005. Protecting Water Quality from Agricultural Runoff. In: AGENCY, U. S. E. P. (ed.).
Nonpoint Source Control Branch (4503T), 1200 Pennsylvania Avenue, NW Washington,
DC, 20460: United States Environmental Protection Agency.
EPA. 2012a. 2000 National Water Quality Inventory [Online]. Available:
http://water.epa.gov/lawsregs/guidance/cwa/305b/2000report_index.cfm [Accessed 1
April 2012].
EPA. 2012b. Contaminated Sediments Program [Online]. Available:
http://www.epa.gov/glnpo/sediments.html [Accessed 1 April 2012].
EPA 2012c. Managing Nonpoint Source Pollution from Agriculture. In: AGENCY, U. S. E. P.
(ed.).
EPA. 2012d. State and Individual Trading Programs [Online]. Available:
http://water.epa.gov/type/watersheds/trading/tradingmap.cfm [Accessed 1 April 2012
2012].
EPA. 2012e. Water Quality Trading [Online]. Available:
http://water.epa.gov/type/watersheds/trading.cfm [Accessed 1 April 2012 2012].
EPA. 2012f. Welcome to STEPL and Region 5 Model [Online]. Available: http://it.tetratechffx.com/steplweb/ [Accessed 1 June 2012 2012].
FOSTER, G. R., YODER, D. C., WEESIES, G. A., MCCOOL, D. K., MCGREGOR, K. C. &
BINGNER, R. L. 2003. User’s Guide Revised Universal Soil Loss Equation Version 2
RUSLE2. In: SERVICE, U.-A. R. (ed.). Washington, D.C.
FOSTER, G. R., YODER, D. C., WEESIES, G. A. & TOY, T. J. 2001. The Design Philosophy
Behind RUSLE2: Evolution of an Empirical Model. Soil Erosion Research for the 21st

144

Century. Honolulu, Hawaii, USA: American Society of Agricultural and Biological
Engineers.
GASSMAN, P. W., ARNOLD, J. J., SRINIVASAN, R. & REYES, M. 2010. The Worldwide
Use of the SWAT Model: Technological Drivers, Networking Impacts, and Simulation
Trends. Watershed Technology: Improving Water Quality and Environment. Universidad
EARTH, Costa Rica: American Society of Agricultural Engineering.
GASSMAN, P. W., OSEI, E., SALEH, A. & HAUCK, L. M. 2002. Application of an
Environmental and Economic Modeling System for Watershed Assessments. Journal of
the American Water Resources Association, 38, 423-438.
GASSMAN, P. W., REYES, M. R., GREEN, C. H. & ARNOLD, J. G. 2007a. The Soil and
Water Assessment Tool: Historical Development, Applications, and Future Research
Directions. Transactions of the ASABE, 50, 1211-1250.
GASSMAN, P. W., REYES, M. R., GREEN, C. H. & ARNOLD, J. G. 2007b. The Soil and
Water Assessment Tool: Historical Development, Applications, and Future Research
Directions. Transactions of the ASABE, 50, 1211-1250.
GASSMAN, P. W., WILLIAMS, J. R., BENSON, V. W., IZAURRALDE, R. C., HAUCK, L.
M., JONES, A., ATWOOD, J. D., KINIRY, J. R. & FLOWERS, J. D. 2004. Historical
Development and Applications of the EPIC and APEX models. 2004 ASAE/CSAE Annual
International Meeting Sponsored by ASAE/CSAE. Fairmont Chateau Laurier, The Westin,
Government Centre Ottawa, Ontario, Canada: American Society of Agricultural
Engineers.
GEORGAS, N., RANGARAJAN, S., FARLEY, K. J. & JAGUPILLA, S. C. K. 2009.
AVGWLF-Based Estimation of Nonpoint Source Nitrogen Loads Generated Within
Long Island Sound Subwatersheds. Journal of the American Water Resources
Association, 45, 715-733.
GILINSKY, E., CAPACASA, J. M., BAKER, M. G. & KING, E. S. 2009. An Urgent Call to
Action. State-EPA Nutrient Innovations Task Group.
GIRI, S., NEJADHASHEMI, A. P. & WOZNICKI, S. A. 2012. Evaluation of targeting methods
for implementation of best management practices in the Saginaw River Watershed.
Journal of Environmental Management, 103, 24-40.
GORDON, L. J., FINLAYSON, M. C. & FALKENMANRK, M. 2010. Managing water in
agriculture for food production and other ecosystem services. Agricultural Water
Management, 97, 512-519.
GREGOIRE, C., ELSAESSER, D., HUGUENOT, D., LANGE, J., LEBEAU, T., MERLI, A.,
MOSE, R., PASSEPORT, E., PAYRAUDEAU, S., SCHUETZ, T., SCHULZ, R.,
TAPIA-PADILLA, G., TOURNEBIZE, J., TREVISAN, M. & WANKO, A. 2009.

145

Mitigation of Agricultural Nonpoint-Source Pesticide Pollution in Artificial Wetland
Ecosystems – A Review. Climate Change, Intercropping, Pest Control and Beneficial
Microorganisms, 2, 293-338.
GREINER, R., PATTERSON, L. & MILLER, O. 2009. Motivations, risk perceptions and
adoption of conservation practices by farmers. Agricultural Systems, 99, 86-104.
HERENDEEN, N. & GLAZIER, N. 2009. Agricultural Best Management Practices for Conesus
Lake: The Role of Extension and Soil/Water Conservation Districts. Journal of Great
Lakes Research, 35, 15-22.
HOWARTH, R. W., SHARPLEY, A. & WALKER, D. 2002. Sources of Nutrient Pollution to
Coastal Waters in the United States: Implications for Achieving Coastal Water Quality
Goals. Estuaries, 25, 656-676.
HSIEH, P.-H., KUO, J.-T., WU, E. M.-Y., CIOU, S.-K. & LIU, W.-C. 2010. Optimal Best
Management Practice Placement Strategies for Nonpoint Source Pollution Management
in the Fei-Tsui Reservoir Watershed. Environmental Engineering Science, 27, 441-449.
IM, S., BRANNAN, K. M., MOSTAGHIMI, S. & KIM, S. M. 2007. Comparison of HSPF and
SWAT models performance for runoff and sediment yield prediction. Journal of
Environmental Science and Health, Part A: Toxic/ Hazardous Substances and
Environmental Engineering, 42, 1561-1570.
INC., T. T. 2005. User’s Guide Spreadsheet Tool for the Estimation of Pollutant Load (STEPL)
Version 3.1. 10306 Eaton Place, Suite 340, Fairfax, VA 22003.
IWR. 2010. HIT (High Impact Targeting) A tool for optimizing sedimentation reduction efforts
in the Great Lakes Basin [Online]. Institute of Water Research Available:
http://35.9.116.206/hit2/about.htm [Accessed 1 April 2012].
JENKS, G. F. 1967. The Data Model Concept in Statistical Mapping. 7, 186-190.
JOHNSONA, M. S., COONB, W. F., MEHTAA, V. K., STEENHUISA, T. S., BROOKSC, E. S.
& BOLLC, J. 2003. Application of two hydrologic models with different runoff
mechanisms to a hillslope dominated watershed in the northeastern US: a comparison of
HSPF and SMR. Journal of Hydrology, 284, 57-76.
KAINI, P., ARTITA, K. & NICKLOW, J. W. 2012. Optimizing Structural Best Management
Practices Using SWAT and Genetic Algorithm to Improve Water Quality Goals. Water
Resources Management, 26, 1827-1845.
KAPLOWITZ, M. D. & LUPI, F. 2012. Stakeholder preferences for best management practices
for non-point source pollution and stormwater control. Landscape and Urban Planning,
104, 364-372.

146

KAY, P., EDWARDS, A. C. & FOULGER, M. 2009. A review of the efficacy of contemporary
agricultural stewardship measures for ameliorating water pollution problems of key
concern to the UK water industry. Agricultural Systems, 99, 67-75.
KEMANIANA, A. R., JULICHB, S., MANORANJANC, V. S. & ARNOLDD, J. R. 2011.
Integrating soil carbon cycling with that of nitrogen and phosphorus in the watershed
model SWAT: Theory and model testing. Ecological Modelling, 22, 1913-1921.
KLIMENT, Z., KADLEC, J. & LANGHAMMER, J. 2008. Evaluation of suspended load
changes using AnnAGNPS and SWAT semi-empirical erosion models. CATENA, 73,
286-299.
KNOWLER, D. & BRADSHAW, B. 2007. Farmers’ adoption of conservation agriculture:
A review and synthesis of recent research. Food Policy, 32, 25-48.
KOCHES, J. 2010. HIT Model Home Page [Online]. 1 Campus Drive,
Allendale, MI
49401-9403, USA: Grand Valley State University. [Accessed April 1 2012].
LAROCHE, A., GALLICHAND, J. & PESANT, A. 1996. Simulating Atrazine Transport with
HSPF in an Agricultural Watershed. Journal of Environmental Engineering, 122, 622630.
LIN, L., DENG, Z.-Q. & GANG, D. D. 2009. Nonpoint Source Pollution. Water Environment
Research, 81, 1996-2018.
LINDAU, C., BOLLICH, P. & BOND, J. 2010. Soybean Best Management Practices for
Louisiana, USA, Agricultural Nonpoint Source Water Pollution Control.
Communications in Soil Science and Plant Analysis, 41, 1615-1626.
LUO, Y., SU, B., YUAN, J., LI, H. & ZHANG, Q. 2011. GIS Techniques for Watershed
Delineation of SWAT Model in Plain Polders. Procedia Environmental Sciences, 10,
2050-2057.
MAKAREWICZ, J. C. 2009. Nonpoint Source Reduction to the Nearshore Zone Via Watershed
Management Practices: Nutrient Fluxes, Fate, Transport and Biotic Responses —
Background and Objectives. Journal of Great Lakes Research, 35, 3-9.
MARINGANTI, C., CHAUBEY, I., ARABI, M. & ENGEL, B. 2011. Application of a MultiObjective Optimization Method to Provide Least Cost Alternatives for NPS Pollution
Control. Environmental Management, 48, 448-461.
MARINGANTI, C., CHAUBEY, I. & POPP, J. 2009. Development of a multiobjective
optimization tool for the selection and placement of best management practices for
nonpoint source pollution control. Water Resources Research, 45.

147

MAXTED, J. T., DIEBEL, M. W. & VANDER ZANDEN, M. J. 2009. Landscape Planning for
Agricultural Non–Point Source Pollution Reduction. II. Balancing Watershed Size,
Number of Watersheds, and Implementation Effort. Environmental Management, 43, 6068.
MILLER, C. 2012. Sediment Yield and Dam Capacity in the Great Lakes Watershed [Online].
Available: http://engineering.wayne.edu/cee/research/sediment-yield.php [Accessed 1
April 2012].
MORIASI, D. N., ARNOLD, J. G., VAN LIEW, M. W., BINGNER, R. L., HARMEL, R. D. &
VEITH, T. L. 2007. Model Evaluation Guidelines for Systematic
Quantification of Accuracy in Watershed Simulations. Transactions of the ASABE, 50,
885-900.
MOSS, B. 2008. Water pollution by agriculture. Philosophical Transactions of The Royal
Society of Biological Sciences, 363, 659-666.
MUDGAL, A., BAFFAUT, C., ANDERSON, S. H., SADLER, E. J. & THOMPSON, A. 2008.
APEX Model Assessment of Variable Landscapes on Runoff and Dissolved Herbicides.
ASABE Annual International Meeting. Rhode Island Convention Center, Providence,
Rhode Island: Agricultural Society of Agricultural and Biological Engineers.
MUTHUKRISHNAN, S. 2002. Development and testing of an integrated water pollution
assessment tool for analyzing impacts of urban sprawl. Ph.D., Purdue University.
NATIONS, E. O. T. 2012. Encyclopedia of the Nations [Online]. Advameg Inc. [Accessed 1
April 2012].
NEITSCH, S. L., ARNOLD, J. G., KINIRY, J. R. & WILLIAMS, J. R. 2005. Soil and Water
Assessment Toll Theoretical Documentation. Blackland Research Center, Texas
Agricultural Experiment Station, 720 East Blackland Road, Temple, Texas.
NEJADHASHEMI, A. P., WOZNICKI, S. A. & DOUGLAS-MANKIN, K. R. 2011.
Comparison of four models (STEPL, PLOAD, L-THIA, and SWAT) in simulating
sediment, nitrogen, and phosphorus loads and pollutant source areas. Transactions of the
ASABE, 54, 875-890.
NORD, M., COLEMAN-JENSEN, A., ANDREWS, M. & CARLSON, S. 2009. Household
Food Security in the United States, 2009. Measuring Food Security in the United States.
108 ed.: United States Department of Agriculture.
NRCS 2003a. Rusle2 Revised Universal Soil Loss Equation-Version 2 Predicting Soil Erosion
By Water: A Guide to Conservation Planning.

148

NRCS. 2003b. RUSLE2 Revised Universal Soil Loss Equation 2 [Online]. Available:
http://www.ia.nrcs.usda.gov/news/factsheets/RUSLE2FactSheet.html [Accessed April 1
2012].
NRCS 2008a. Conservation Stewardship Program Fact Sheet. In: SERVICE, N. R. C. (ed.).
NRCS 2008b. Farm Bill 2008. In: SERVICE, N. R. C. (ed.).
NRCS. 2009. 2008 Farm Bill [Online]. Available:
http://www.nrcs.usda.gov/wps/portal/nrcs/detail/national/programs/farmbill/?&cid=nrcs1
43_008208 [Accessed 1 April 2012].
NRCS 2010a. Assessment of the Effects of Conservation Practices on Cultivated Cropland in the
Upper Mississippi River Basin. Conservation Effects Assessment Project (CEAP). United
States Department of Agriculture Natural Resources Conservation Service.
NRCS 2010b. Natural Resources Conservation Service Conservation Practice Standard.
Conservation Crop Rotation.
NRCS 2010c. Natural Resources Conservation Service Conservation Practice Standard.
Sediment Basin.
NRCS 2010d. Natural Resources Conservation Service Conservation Practice Standard.
Windbreak.
NRCS 2010e. Natural Resources Conservation Service Conservation Practice Standard.
Streambank and Shoreline Protection.
NRCS 2010f. Natural Resources Conservation Service Conservation Practice Standard. No Till.
NRCS 2010g. Natural Resources Conservation Service Conservation Practice Standard. Riparian
Forest Buffer.
NRCS 2010h. Natural Resources Conservation Service Conservation Practice Standard.
Irrigation Water Management.
NRCS 2010i. Natural Resources Conservation Service Conservation Practice Standard. Grassed
Waterway.
NRCS 2010j. Natural Resources Conservation Service Conservation Practice Standard. Filter
Strip.
NRCS 2010k. Natural Resources Conservation Service Conservation Practice Standard. Terrace.
NRCS 2011a. Summary of Findings: Assessment of the Effects of Conservation Practices on
Cultivated Cropland in the Chesapeake Bay Region.

149

NRCS 2011b. Summary of Findings: Assessment of the Effects of Conservation Practices on
Cultivated Cropland in the Great Lakes Region. United States Department of Agriculture
Natural Resource Conservation Service.
NRCS. 2012a. Conservation Stewardship Program [Online]. Available:
http://www.nrcs.usda.gov/wps/portal/nrcs/main/national/programs/financial/csp.
NRCS 2012b. Field Office Technical Guide. In: SERVICE, N. R. C. (ed.).
O'NEIL, G. 2011. High Impact targeting (HIT) A web-Accessible System to Target Areas at
Highest Risk for Sediment Loading. 1405 S. Harrison Road, 101 Manly Miles Bldg., East
Lansing, MI 48823-5243: Institute of Water Research.
O'NEILL, D. 2007. The Total External Environmental Costs
and Benefits of Agriculture in the UK. In: AGENCY, E. (ed.).
OLMSTEAD, S. M. 2010. The Economics of Water Quality. Review of Environmental
Economics and Policy, 6, 44-62.
OUYANG, D., BARTHOLIC, J. & SELEGEA, J. 2005. Assessing Sediment Loading from
Agricultural Croplands in the Great Lakes Basin. The Journal of American Science, 1,
14-21.
O’NEIL, G., SHI, Y. & SEEDANG, S. Using the Spatially Explicit Delivery Model and the
Revised Universal Soil-Loss Equation to Target High-Risk Sediment Loading Areas at
Multiple Scales. Managing Agricultural Landscapes for Environmental Quality –
Strengthening the Science Base, 2010. Soil and Water Conservation Society’s.
PAI, N., SARASWAT, D. & SRINIVASAN, R. 2012. Field_SWAT: A tool for mapping SWAT
output to field boundaries. Computers & Geosciences, 40, 175-184.
PAL, I. & AL-TABBAA, A. 2009. Suitability of different erosivity models used in RUSLE2 for
the South West Indian region. Environmentalist, 29, 405-410.
PARAJULI, P. B., NELSON, N. O., FREES, L. D. & MANKIN, K. R. 2009. Comparison of
AnnAGNPS and SWAT model simulation results in USDA-CEAP agricultural
watersheds in south-central Kansas. Hydrological Processes, 23, 748-763.
PEASEA, L. M., ODUORB, P. & PADMANABHANA, G. 2010. Estimating sediment, nitrogen,
and phosphorous loads from the Pipestem Creek watershed, North Dakota, using
AnnAGNPS. Computers & Geosciences, 36, 282-291.
PETERSON, J. M., SMITH, C. & VALENTIN, L. 2005. A Water Quality Trading Simulation
for Northeast Kansas. American Agricultural Economics Association Annual Meeting.
Providence, Rhode Island.

150

PIMENTEL, D. 2009. Environmental and Economic Costs of the Application of Pesticides
Primarily in the United States, Springer Science+Business Media B.V.
POLYAKOV, V., FARES, A., KUBO, D., JACOBI, J. & SMITH, C. 2007. Evaluation of a nonpoint source pollution model, AnnAGNPS, in a tropical watershed. Environmental
Modelling & Software, 22, 1617-1627.
PORTER, J., COSTANZA, R., SANDHU, H., SIGSGAARD, L. & WRATTEN, S. 2009. The
Value of Producing Food, Energy, and Ecosystem Services within an Agro-Ecosystem.
AMBIO: A Journal of the Human Environment, 38, 186-193.
PROKOPY, L. S., FLORESS, K., KLOTTHOR-WEINKAUF, D. & BAUMGART-GETZ, A.
2008. Determinants of agricultural best management practice adoption: Evidence from
the literature. Journal of Soil and Water Conservation, 63, 300-311.
QUILBÉ, R., ROUSSEAU, A. N., LAFRANCE, P., LECLERC, J. & AMRANI, M. 2006.
Selecting a Pesticide Fate Model at the Watershed Scale Using a Multi-criteria Analysis.
Water Quality Research Journal of Canada, 41, 283-295.
REIMER, A. P., WEINKAUF, D. K. & PROKOPY, L. S. 2012. The influence of perceptions of
practice characteristics: An examination of agricultural best management practice
adoption in two Indiana watersheds. Journal of Rural Studies, 28, 188-128.
ROLFE, J. & WINDLE, J. 2011. Comparing a best management practice scorecard with an
auction metric to select proposals in a water quality tender. Land Use Policy, 28, 175184.
SALEH, A. & DU, B. 2004. Evaluation of SWAT and HSPF Within Basins Program for the
Upper North Bosque River Watershed in Central Texas. Transactions of the ASAE, 47,
1039-1049.
SALEH, A., GASSMAN, P. W., ABRAHAM, J. & RODECAP, J. 2003. Application of SWAT
and APEX models for Upper Maquoketa
River Watershed in Northeast Iowa. ASAE Annual International Meeting. Riviera Hotel
and Convention Center, Las Vegas, Nevada, USA: Agricultural Society of Agricultural
Engineers.
SANTHI, C., ARNOLD, J. G., WILLIAMS, J. R., HAUCK, L. M. & DUGAS, W. A. 2001.
Application of a Watershed Model to Evaluate Management Effects on Point and
Nonpoint Source Pollution. Transactions of the ASAE, 44, 1559-1570.
SCHULZ, R. 2004. Field Studies on Exposure, Effects, and Risk Mitigation of Aquatic
Nonpoint-Source Insecticide Pollution. Journal of Environmental Quality, 33, 419-448.

151

SCHUOLA, J., ABBASPOURA, K. C., SRINIVASANB, R. & YANGA, H. 2008. Estimation of
freshwater availability in the West African sub-continent using the SWAT hydrologic
model. Journal of Hydrology, 352, 30-49.
SECRETARY, O. O. T. P. 2009. Chesapeake Bay Protection and Restoration. In: HOUSE, T.
W. (ed.).
SHARPLEY, A. N., KLEINMAN, P. J. A., JORDAN, P., SWEDISH, L. B. & ALLEN, A. L.
2009. Evaluating the Success of Phosphorus Management from Field to Watershed.
Journal of Environmental Quality, 38, 1981-1988.
SHEN, Z. Y., GONG, Y. W., LI, Y. H., HONG, Q., XU, L. & LIU, R. M. 2009. A comparison
of WEPP and SWAT for modeling soil erosion of the Zhangjiachong Watershed in the
Three Gorges Reservoir Area. Agricultural Water Management, 96, 1435-1442.
SHORT, A. G. 2011. Governing Change: Land-Use Change and the Prevention of Nonpoint
Source Pollution in the North Coastal Basin of California. Environmental Management.
SHORTLE, J. S., RIBAUDO, M., HORAN, R. D. & BLANDFORD, D. 2012. Reforming
Agricultural Nonpoint Pollution Policy in an Increasingly Budget-Constrained
Environment. Environmental Science and Technology, 46, 1316-1325.
SMITH, C. M., NEJADHASHEMI, A. P. & LEATHERMAN, J. C. 2009. Using a BMP Auction
as a Tool for the Implementation of Conservation Practices. Journal of Extention, 47.
STEINMAN, A. D., OGDAHL, M. E. & RUETZ, C. R. I. 2009. An environmental assessment
of a small shallow lake
(Little Black Lake, MI) threatened by urbanization. Environmental Monitoring and
Assessment, 173, 193-209.
TANG, Z., ENGEL, B. A., PIJANOWSKI, B. C. & LIM, K. J. 2005. Forecasting land use
change and its environmental impact at a watershed scale. Journal of Environmental
Management, 76, 35-45.
TEGTMEIER E.M., DUFFY M. D. 2004. External costs of Agricultural Production in the
United States. International Journal of Agricultural Sustainability, 2, 1-20.
TECH, T. 2012. STEPL Model Input Data Server Version 1.0 [Online]. Available:
http://it.tetratech-ffx.com/steplweb/steplweb.html [Accessed 1 April 2012].
THURSTON, H. W., AYLOR, M. A. T., ROY, A., MORRISON, M., SHUSTER, W. D.,
TEMPLETON, J., CLAGETT, M. & CABEZAS, H. 2008. Applying a Reverse Auction
to Reduce Stormwater Runoff. AMBIO: A Journal of the Human Environment, 37, 326327.

152

TUPPAD, P., KANNAN, N., SRINIVASAN, R., ROSSI, C. G. & ARNOLD, J. G. 2010.
Simulation of Agricultural Management Alternatives for Watershed Protection. Water
Resources Management, 24.
TUPPAD, P., WINCHELL, M. F., WANG, X., SRINIVASAN, R. & WILLIAMS, J. R. 2009.
ARCAPEX: ARCGIS Interface for Agricultural Policy Environmental Extender (APEX)
Hydrology/Water Quality Model. International Agricultural Engineering Journal, 18,
59-71.
TWEETEN, L. & THOMPSON, S. R. 2009. Long-term Global Agricultural Output SupplyDemand Balance and Real Farm and Food Prices.
USDA 2006. Economic Information Bulletin 16. In: AGRICULTURE, U. S. D. O. (ed.).
USDA. 2009. Financial Assistance [Online]. Available:
http://www.nrcs.usda.gov/wps/portal/nrcs/main/national/programs/financial [Accessed 1
April 2012].
USGS. 2010. HSPF Hydrological Simulation Program--Fortran [Online]. 5285 Port Royal
Road, Springfield, VA 22161: National Technical Information Service (NTIS),.
Available: http://water.usgs.gov/software/HSPF/ [Accessed 1 April 2012].
VAN LIEW, M. W., ARNOLD, J. G. & GARBRECHT, J. D. 2003. Horologic Simulation on
Agricultural Watersheds: Choosing Between Two Models. Transactions of the ASABE,
46, 1539-1551.
VANDERMOLEN, J. The Hit Model: Better information leads to better decisions. HIT Model
Workshop, 2010 Ferris State University
Big Rapids, MI.
VEITH, T. L. V. L., M.W. BOSCH, D.D.ARNOLD, J.G. 2010. Parameter Sensitivity and
Uncertainty in SWAT: A Comparison Across Five USDA-ARS Watersheds.
Transactions of the ASABE, 53, 1477-1486.
VITOUSEK, P. M., NAYLOR, R., CREWS, T., DAVID, M. B., DRINKWATER, L. E.,
HOLLAND, E., JOHNES, P. J., KATZENBERGER, J., MARTINELLI, L. A.,
MATSON, P. A., NZIGUHEBA, G., OJIMA, D., PALM, C. A., ROBERTSON, G. P.,
SANCHEZ, P. A., TOWNSEND, A. R. & ZHANG, F. S. 2009. Nutrient Imbalances in
Agricultural Development. Science, 324, 1519-1520.
WANG, X., GASSMANB, P. W., WILLIAMS, J. R., POTTERA, S. & KEMANIANA, A. R.
2008. Modeling the impacts of soil management practices on runoff, sediment yield,
maize productivity, and soil organic carbon using APEX. Soil and Tillage Research, 101,
78-88.

153

WILLIAMS, J. R. & IZAURRALDE, R. C. 2005. The APEX Model. In: JOINT GLOBAL
CHANGE RESEARCH INSTITUTE, B. A., SUITE 201, COLLEGE PARK, MD 207402496 (ed.). 720 East Blackland Rd., Temple, TX 76502-9622: Texas Agricultural
Experiment Station.
WILLIAMS, J. W., IZAURRALDE, R. C. & STEGLICH, E. M. 2008. Agricultural
Policy/Environmental eXtender Model Theoretical Documentation Version 0604. 720
East Blackland Road, Temple, Texas, 76502: Blackland Research and Extension Center
and the Joint Global Change Research Institute Pacific Northwest National Lab and
University of Maryland.
WOZNICKI, S. A. & NEJADHASHEMI, A. P. 2012. Sensitivity Analysis of Best Management
Practices Under Climate Change Scenarios. Journal of the American Water Resources
Association, 48, 90-112.
YANG, L., MA, K., GUO, Q., ZHAO, J. & LUO, Y. 2006. Zoning planning in non-point source
pollution control in Hanyang district. Huan Jing Ke Xue, 27, 31-6.
YANGA, Q., MENGA, F.-R., ZHAOA, Z., CHOWB, T. L., BENOYC, G., REESB, H. W. &
BOURQUEA, C. P.-A. 2009. Assessing the impacts of flow diversion terraces on stream
water and sediment yields at a watershed level using SWAT model. Agriculture,
Ecosystems & Environment, 132, 23-31.
YODER, D. C., FOSTER, G. R., WEESIES, G. A., RENARD, K. G., MCCOOL, D. K. &
LOWN, J. B. 2003. Evaluation of the RUSLE Soil Erosion Model [Online]. Available:
http://www.bae.ncsu.edu/www3/acad/Regional-Bulletins/Modeling-Bulletin/rusle-yoder001016.html [Accessed 1 April 2012 2012].
ZEMA, D. A., BINGNER, R. L., DENISI, P., GOVERS, G., LICCIARDELLO, F. &
ZIMBONE, S. M. 2012. Evaluation of runoff, peak flow and sediment yield for events
simulated by the AnnAGNPS model in a belgian agricultural watershed. Land
Degradation & Development, 23, 205-215.
ZHANG, X., SRINIVASAN, R., ARNOLD, J., IZAURRALDE, R. C. & BOSCH, D. 2011.
Simultaneous calibration of surface flow and baseflow simulations: a revisit of the
SWAT model calibration framework. Hydrological Processes, 25, 2313-2320.

154