 

LIBRARY
Michigan State
University

 

 

 

PLACE IN RETURN BOX to remove mis checkout from your record.
TO AVOID FINES return on or before date due.
MAY BE RECALLED with earlier due date if requested.

 

DATE DUE DATE DUE DATE DUE

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

2/05 c:/CIRC/DateDue.Indd-p.15

EVALUATING THE AGNPS MODEL FOR
PRIORITIZING VEGETATIVE FILTER STRIPS
WITHIN AGRICULTURAL WATERSHEDS

BY

Sharon Anne Vennix

A THESIS
Submitted to
Michigan State University

in partial fulﬁllment of the requirements
for the degree of

MASTER OF SCIENCE
Department of Agricultural Engineering

2003

ABSTRACT

Evaluating the AGNPS Model to Prioritize Vegetative Filter Strips
within Agricultural Watersheds

By: Sharon A. Vennix

Recent developments in GIS interfaces have greatly improved hydrologic
water quality modeling; larger watersheds can accurately be assessed at higher
spatial resolutions for prioritizing site-speciﬁc areas where best management
practices, such as vegetative ﬁlter strips, would be most effective. In this study,
the AGNPS model in conjunction with the AGNPS GIS interface was used to
prioritize locations of vegetative ﬁlter strip effectiveness within areas of a 45,000
ha (111,000 ac) agricultural watershed located in mid Michigan. Vegetative ﬁlter
strips alone were found to be relatively inefﬁcient (<25%) especially in areas
where concentrated ﬂow occurs (drainage area > 4 ha or 10 ac), which are areas
of primary concern in terms of sediment load. Therefore conservation efforts
should focus on reducing sediment load in areas of concentrated ﬂow.

In addition, an unconditional stochastic simulation was preformed to
identify the uncertainty in the AGNPS model due to errors in 30 m USGS DEM.
Comparing the AGNPS sediment yield at the outlet of a 107 ha (264 ac)
watershed when using 30 m USGS DEM and 10 m simulated elevation data
resulted in a 37% decrease. However, the analysis also identiﬁed that change in
cell size (from 30 m to 10 m) reduces the sediment load by 41%. These
uncertainties are profoundly affecting the AGNPS estimates of sediment yield,
which may dramatically inﬂuence the analysis of prioritizing conservation

practices within watersheds.

AGKNOWLEDGEMENTS

To my Advisors...
You have shaped my life signiﬁcantly.

To my Family...
You have loved me unconditionally.

To my Friends...
You have brought me lifelong happiness and hope.

and especially to Frank...
You are my inspiration.

My accomplishments belong to you.

Thank you everyone for all that you
have brought to my life.

~Sharon

TABLE OF CONTENTS

LIST OF FIGURES ....v..ii

CHAPTER1:INTRODUCTION..............................................................1
1.1 Background

A

1.2 Objectives

CHAPTER 2: LITERATURE REVIEW .. .

6
8
2.1 Vegetative Filter Strips 8

2.11 Overview... 8
21.2 Vegetative Filter Strip EffectIveness 9
2.13 Design Considerations... 10
2.1.4 MaIntenance14

2.2 Conservation Programs 15
2.2.1 Overview . 15
2.2.2 Conservation Reserve Program (CRP) ........16
2.2.3 Conservation Reserve Enhancement Program (CREP) ........ 17
2.2.4 Environmental Quality Incentives Program (EQIP) .............. 18
2.2.5 WIldlife Habitat Incentives Program (WHIP) 18
2.2.6 Wetlands Reserve Program (WRP)19
2.2.7 Emergency Watershed Protection (EWP)... ......19

2.3 Current HydrologicWater Quality Models ...20
2.3.1 Overview... . .............20
2.3.2 Areal Nonpoint Source Watershed EnVIronmental .... ...22

Response Simulator (ANSWERS)
2.3.3 KlNematic runoff and EROSion (KINEROS) model .23
2.3.4 Dynamic Watershed Simulation Model (DWSM) ......24

2. 4 Agriculture Non-Point Source Pollution Model (AGNPS) .. ..

2.41 Overview... . .
2.4.2 Model Structure... ..
2.4.3 Input and Output Database
2.4.4 Validation...

CHAPTER 3: MATERIALS AND METHODS
3.1 StudyArea

3.2 AGNPS Modeling Scenarios

3.2.1 Overview”

3. 2. 2 Filter strip and No Filter Strip Input Database

3. 3 AGNPS Uncertainty Due to DEM Error.

3 3 1 Method for Evaluating Error Propagahon. .. ... ... ... ... ...

3. 3. 2 Unconditional Stochastic Simulation“
3. 3. 3 Methodology Using the Study Area’ 3 Watershed

CHAPTER 4: RESULTS AND DISCUSSION
4.1 Stony Creek Secondary Sub-watershed Results...

4.2 Prioritizing Vegetative Filter Strips.....”..

4.2.1 Overview...
4. 2. 2 Filter Strip Segment Results

.24

......24
......25
......29
2.4.5 Sensmvuty

.37
..37

.41

.......41
....... 45
......48
...48

..I...55
...55

4.3 Uncertainty in the AGNPS Model Due to Errors in the DEM...
4.3.1 OverVIew

4.3.2 Local Effects.

4.3.3 Global Effects..

4.3.4 Resampled USGS DEM Results
4.3.5 Cell Size Effects In the AGNPS Model

CHAPTER 5: CONCLUSIONS.
5.1 General Conclusions.

5.1.1 Prion‘tizing VegetatIve Filter Strips
5.1.2 AGNPS Uncertainty Due to DEM Error

5.2 FutureConsiderations.............................................................

ages

(0
N

.105

Table 2.1:
Table 2.2:

Table 2.3:

Table 3.1:

Table 3.2:

Table 4.1:

Table 4.2:

Table 4.3:

Table 4.4:

Table 4.5:

Table 4.6:

Table 4.7:

Table 4.8:

Table 4.9:

Table 4.10:

LIST OF TABLES

NRCS suggestions for ﬁlter strip length (NRCS, 1999a) ............. 1 1
Vegetation species and seeding rates for trapping 14
sediment within a ﬁlter strip (NRCS, 1999a).

AGNPS cell input parameters. ...29
AGNPS input parameters and the corresponding... ...46
data sources.

Land use parameters for the Stony Creek watershed... ...48
Filter strip effectiveness at the outlet of the 27 secondary .......... 65
sub-watersheds.

ANOVA results for Bad Creek, ﬁlter efﬁciency verses ..70
drainage area.

ANOVA results for Lost Creek, ﬁlter efﬁciency verses ............... 71

drainage area.

ANOVA results for Muskrat Creek, ﬁlter efﬁciency verses ........... 72
drainage area.

ANOVA results for Spaulding Drain, percent ﬁltered sediment ...73
verses drainage area.

Thirty-one ﬁlter strip segments ranked by the amount of... ...76
sediment ﬁltered within the EBC watershed.

ANOVA results for the EBC watershed, ﬁlter strip efﬁciency ........ 78
verses drainage area.

Ranking order from 1-31 of vegetative ﬁlter strip effectiveness ....82
without the concentrated ﬂow areas.

Sedimentation results at the outlet from the nine ...87
simulations.

Sedimentation results at the outlet from the resampled DEM ....... 88

vi

Figure 2.1:

Figure 3.1:
Figure 3.2:
Figure 3.3:
Figure 3.4:
Figure 3.5:
Figure 3.6a-d:

Figure 3.7:

Figure 3.8:

Figure 3.9:

Figure 3.10:

Figure 3.1 1a-i:

Figure 4.1:

Figure 4.2:

Figure 4.3:

LIST OF FIGURES

Young et al. (1987) sensitivity analysis indicating ................. 35

slope-associated parameters (LS) are the most sensitive
in the AGNPS model when predicting sediment yield.

Location ofthe Stony Creekwatershed..........
Land use in the Stony Creek watershed.
Soil types ofthe Stony Creekwatershed..........

Topography ofthe Stony Creek watershed.......

Thirty-one secondary sub-watersheds of Stony Creek .....

Four GIS layers for the AVNPSM interface...

Monte Carlo framework for evaluating uncertainty

in the spatial modeling application (AGNPS model)
due to error in the input data (elevation dataset).

Stochastic simulation approach for evaluating uncertainty

.38
.39
4O
41
..... 42
43
.50

in the AGNPS model due to error in coarse elevation data.

Location of study area and MSU F arm watershed site ......

The variogram model of the error between the Survey...

DEM and the USGS resampled DEM of the MSU Farm
watershed.

DEM Realizations 1-9 of study area watershed

100% sediment ﬁltered for the 27 secondary

sub-watersheds of Stony Creek.

The Bad Creek sub-watershed —cells within the ﬁlter strip
grouped into categories of ﬁlter efﬁciency and drainage
area.

The Lost Creek sub-watershed —cells within the ﬁlter strip

grouped into categories of ﬁlter efﬁciency and drainage
area.

vii

..... 56

.58

59

..... 67

..... 68

Figure 4.4:

Figure 4.5:

Figure 4.6:

Figure 4.7:

Figure 4.8:

Figure 4.9:

Figure 4.10:

Figure 4.1 1:

Figure 4.12:

Figure 4.13:

Figure 4.14:

Figure 4.15:

Figure 4.16:

The Muskrat Creek sub-watershed -cells within the ﬁlter... .

strip grouped into categories of ﬁlter efﬁciency and
drainage area.

The Spaulding Drain sub-watershed -cells within the

ﬁlter strip grouped into categories of ﬁlter efﬁciency
and drainage area.

Bad Creek average ﬁlter strip efﬁciency verses drainage. . ..

area for the cells within the ﬁlter strip.

Lost Creek average ﬁlter strip efﬁciency verses drainage

area for the cells within the ﬁlter strip.

Muskrat Creek average ﬁlter strip efﬁciency verses ...........

drainage area for the cells within the ﬁlter strip.

Spaulding Drain average ﬁlter strip efﬁciency verses .........

drainage area for the cells within the ﬁlter strip.

The EBC watershed ﬁlter strip segments of equal length

(500 m), numbered from 1-31 corresponding to a ranking
of ﬁlter strip effectiveness (ﬁltered sediment in tons) with
1 the most effective and 31 the least effective.

The EBC watershed average ﬁlter strip efﬁciency verses...

Drainage area of the cells within the ﬁlter strip.

Topography and drainage areas greater than 4 ha ............

(~10 acres) within the ﬁlter strip of the EBC watershed.

30 m DEM (Left) and 10 m Survey DEM (Right) of ............

MSU Farm watershed.

Error Map -difference of USGS DEM and Survey

DEM of MSU Farm watershed (white = no error,
red=positive error, blue=negative error).

The stream derived from the nine DEM realizations... ..

are showing locations of local uncertainty in the
30 m USGS DEM.

30 m grid cell layout (Left) and 10 m grid cell layout

(Right) for the study area watershed.

viii

...68

.69

...70

...71

...72

...73

...74

...78

...79

...83

.84

85

89

NOMENCLATURE

USDA, Soil Conservation Service Cuge Number Method, 1972 (Section 2.4)

CN = Curve number (dimensionless)

P = Precipitation rate (in/hr)
Q = Runoff volume (in)
S = Retention parameter (dimensionless)

§mith and WilliamsI 1980 (Section 2.4)

A =Area (ac)

CS = Channel slope (ftIft)

LW = Watershed length width ratio (dimensionless)
Qp = Peak Runoff Rate (cfs)

R0 = Runoff volume (in)

Wischmeier and SimthI 1978 (Section 2.4)

C = Cover management factor (dimensionless)

El = Rainfall energy-intensity (100ft-ton inch/acre hour)

K = Soil erodibility factor (ton-acre hourl100-acre foot-ton inches)
LS = Slope length factor (dimensionless)

P = Practice factor (dimensionless)

SL = Soil loss (tons/acre)

SSF = Factor to adjust for slope shape within cell (dimensionless)

Foster et al. 1981 and Lane 1982 ction 2.4

D(x) = Deposition rate (Iblsec-ftz)

g'.(x) = Effective transport capacity per unit width (lb/sec-ft)

Lr = Reach length (ft)

q(x) = Discharge per unit width (cfs)

03(0) = Sediment discharge into the upstream end of the channel reach (lb/sec)
O,(x) = Sediment discharge at the downstream end of the channel reach (lb/sec)
q.(x) = Sediment load per unit width (lb/ftz)

0.. = Lateral sediment inﬂow rate (lb/sec)

VI“; = Particle fall velocity (fps)

w = Channel width (ft)

x = Downstream distance (ft)

BagnoldI 1990 (Section 2.4)
9’; = Effective transport capacity per unit width (lb/sec-ft)

g. = Transport capacity (dimensionless)

k = Transport capacity factor (dimensionless)
n = Effective transport factor (dimensionless)
v = Average channel ﬂow velocity (fps)

Vs. = Particle fall velocity (fps)

c = Shear stress (lbetZ)

Variggram Eguation (Section 3.3)

h = Lag distance (units described by data)
Var = Variance of the argument
y(h) Semivariance

20:) Value of the regionalized variable of interest at location It
Z(x+h) = Value at the location x+h

Error Eguation (Section 3.3)
Coarse DEM (units described by data)

Error (units described by data)
Higher Accuracy DEM (units described by data)
A set of locations (possibly gridded) in a spatial dataset

CII'I'IO
II II II II

Chapter 1

1 Introduction

1.1 Background

Non-point source (NPS) pollution is the nation’s number one source of
water quality problems (EPA, 2002a). Thirty-three percent of US. surface waters
were surveyed in 2000; the survey determined that 40% of streams, 45% of
lakes, and 50% of estuaries are contaminated to the degree that they cannot
meet their standard uses (EPA, 2002b). NPS pollution dramatically impacts the
quality of aquatic habitats and the species that they sustain; therefore wildlife
populations and human uses, such as drinking water and recreation, are also
being affected.

NPS pollutants are deﬁned as pollutants carried by rainfall or snowmelt
moving over and through the ground (Novotny and Olem, 1994). This runoff
carries natural and human made pollutants such as fertilizers; pesticides; toxic
chemicals; acid drainage from abandoned mines; bacteria and nutrients from
livestock; and sediment from cropland, forestland, and eroding stream banks
(Novotny and Olem, 1994). The pollutants are eventually deposited into lakes,
rivers, estuaries, wetlands, and also reach sources of underground drinking
water.

The Environmental Protection Agency (EPA) issued a 2000 National
Water Quality Inventory Report stating that agriculture is the leading source of
NPS pollution in surveyed rivers and lakes. Approximately 60% of surface

waters are polluted by agricultural activities where sediment is one of the major

pollutants (EPA, 20020). Sediments and sediment-bound nutrients increase
turbidity and eutrophication, which decreases dissolved oxygen in surface water.
Erosion and sedimentation also reduces the life expectancy of roads, ditches,
culverts, and bridges. The damage due to deposited sediment is estimated to be
between 2.2 and 7 billion dollars a year (Lovejoy et al., 1997).

Conservation practices, also known as Best Management Practices (BMP)
such as conservation tillage, reconstructed wetlands, and vegetative ﬁlter strips
have been vital in reducing surface water pollution from agricultural runoff.
Vegetative ﬁlter strips in particular are crucial to water quality because they have
a wide range of applications (Lawrence et al., 2002). Deﬁned as an area of
herbaceous vegetation situated in-ﬁeld, at the edge of ﬁelds, or adjacent to
streams, rivers, or small lakes, vegetative ﬁlters slow runoff and ﬁlter sediments
and nutrients from agricultural ﬁelds and animal production systems (NRCS,
1999). Depending on the type of vegetative ﬁlter strip, research has indicated
that ﬁlters must be between 1 and 10 m (3.3 and 33 ft) wide to ﬁlter 30-80% of
sediments and nutrients that could potentially reach surface waters (Ghadiri,
2001 ).

Federal or state funded cost-sharing programs developed to encourage
landowners to implement conservation practices have existed since the 1930’s.
Approximately 1.3 million acres of vegetative ﬁlter strips are currently covered
under various cost-sharing programs (NRCS, 2002a). Each program has
different incentives but the same objective: taking farmland out of production to

ﬁlter runoff. Unfortunately, ﬁnancial incentives are limited, the eligibility

guidelines are broad, and most conservation programs have little or no
evaluation procedure. Existing technology does not allow for inexpensive large-
scale analysis of non-point source pollutants, therefore under current guidelines
75% of cropland that lie adjacent to streams or rivers is eligible for conservation
funding.

Evaluating non-point source pollution is a complex process. Changes
throughout agricultural watersheds are mainly occurring on three scales: in the
ﬁeld, on the farm, and throughout the watershed. Environmentally critical land
along with conservation practices can accurately be assessed on the three levels
with the help of hydrologic water quality models (Lawrence et al., 2002). These
models estimate the hydrology and transport of sediments and nutrients
deposited into streams or rivers throughout a watershed. As a result,
environmentally critical land can be assessed at high resolutions with or without
the use of conservation practices. Therefore, conservation efforts can be
analyzed both economically and on an environmental level, locally or throughout
the watershed.

Hydrologic water quality models such as those of Beasley et al. (1980),
Knisel (1980), Skaggs (1980), Leavesley et al. (1983), WIIIiams et al. (1984),
Abbott et al (1986), Lenord et al. (1987), Young et al. (1989), Lane and Nearing
(1989), Woolhiser et al. (1990), Chung et al. (1992), Bicknell et al. (1993), Arnold
et al. (1998), Ahuja et al. (1999), Bingner and Theurer (2001), Borah et al.
(2002), and Ogden and Julien (2002) have been developed with different

objectives in mind; thus simulated sediment results have varying degrees of

accuracy. WIth the capability of higher computing systems and the recent
integration of geographic information systems (GIS), these models have gained
widespread acceptance as accurate and cost-effective tools for evaluating
agricultural BMPs such as vegetative ﬁlter strips (Tim and Jolly, 1994, Mitchell et
al., 1993, Srinivasan and Arnold, 1994).

In this study, the Agricultural Non-point Source Pollution (AGNPS) model
developed by Young et al. (1989) is used as a tool to evaluate and prioritize
areas of vegetative ﬁlter strip effectiveness throughout an agricultural watershed.
AGNPS version 5.0, in conjunction with its recently developed geographic
information systems (GIS) interface (AVNPSM) is a user-friendly, single storm
event based, model developed by the United States Department of Agriculture-
Agricultural Research Service (USDA-ARS) in cooperation with the Minnesota
Pollution Control Agency and Soil Conservation Service (Young et al., 1989).
Unlike many models, AGNPS has been validated on watersheds throughout the
US. and in various countries around the world. The research has stated that the
AGNPS model compares well with ﬁeld data (Young et al., 1989; Mitchell et al.,
1993; Perrone and Madramootoo, 1999).

In addition, AGNPS is a distributed parameter model that subdivides
watersheds into uniform cells to capture the spatial variability of the physical
characteristics of a watershed. The input database, consisting of 21 input
parameters for every homogeneous cell, is compiled with the help of a GIS
interface called ArcWew Non-Point Source Pollution Model (AVNPSM) (He et al.,
2001).

The AVNPSM interface was developed to easily compile AGNPS 21 input
parameters from four GIS layers; watershed boundary, soil, land use/cover, and
a digital elevation model (DEM), so multiple higher resolution scenarios can
efﬁciently be evaluated (He et al., 2001). The grid layout for the AGNPS model is
usually based on the GIS layer with the lowest resolution. USGS DEMs have 30
m accuracy in some parts of the nation and are readily available on the lntemet
for free download. Therefore, because the DEM resolution is usually larger than
the soil and land use/cover GIS layers, the DEM resolution deﬁnes the size of the
grid cell for the AGNPS model.

The main objectives of the AGNPS model are to simulate runoff, sediment
and nutrient yields of agricultural watersheds ranging from a few hectares to
20,000 ha (49,421 ac) (Young et al., 1989). The cell-by-cell analysis not only
pinpoints areas of excessive sedimentation, but also identiﬁes areas where
BMPs such as vegetative ﬁlter strips are effective in reducing NPS pollutants (He
et al., 1993).

The study focuses on using the AGNPS model with the AVNPSM interface
as a tool to prioritize vegetative ﬁlter strips within the Stony Creek watershed.
The watershed, located in Clinton County, Michigan, drains 45,452 ha (112,314
acres) and is a subbasin of the Grand River, a major tributary of Lake Michigan
(NRCS, 2001). Recently, water quality concems prompted an environmental
assessment of Stony Creek which identiﬁed sediment as the primary pollutant in
the watershed (NRCS, 2001). The reoccurring sedimentation problem is

contributing to the degradation of habitat not only in Stony Creek, but also the

Grand River and ultimately Lake Michigan. Erosion and sediment deposition are
also reducing the longevity of roads, ditches, culverts, and bridges throughout the
watershed. An estimated 90,500 tons of sediment enters the stream each year,
where damages are estimated at approximately $15,500 per km ($25,000 per
mile) (NRCS, 2001).

Government subsidized cost-sharing programs are aiding in the rapid
installation of BMPs throughout the watershed to increase water quality.
Vegetative ﬁlter strips are one of the main BMPs that are implemented
throughout Stony Creek to stabilize streambanks and decrease sediment delivery
carried by agricultural runoff (NRCS, 2001). Unfortunately there are limited funds
for the cost sharing programs and therefore it’s necessary to evaluate areas
where vegetative ﬁlter strips would decrease the largest amount of sediment

throughout the watershed.

1.2 Objectives

The objective of this study is to prioritize areas along a stream segment
where a vegetative ﬁlter stn'p would signiﬁcantly reduce sediment delivery into
Stony Creek, by using the AGNPS model with the AVNPSM GIS interface.
Prioritizing critical areas of ﬁlter effectiveness will help watershed managers
install ﬁlter strips with greater certainty. This analysis will help determine where
funds from cost-sharing programs should be allocated, so future water quality

goals can be met.

In addition, a digital elevation model (DEM) uncertainty analysis was also
preformed to identify the ambiguity of the AGNPS sediment output when using a
30 m USGS DEM. The parameters derived from elevation data, such as aspect,
slope, slope length, and slope shape, are sensitive parameters when estimating
sediment yield throughout a watershed. Higher accuracy elevation data was used
to identify the errors in a 30 m USGS DEM. Using higher accuracy “true”
elevation data when evaluating sediment yield within the AGNPS model will not
only improve the accuracy of the output, but will explain the uncertainty in the
AGNPS model due to the error in the original elevation dataset (USGS 30 m
DEM).

It is important to note that the study only focuses on AGNPS modeling
results, there was no in-ﬁeld data measured to calibrate or validate the data
presented in this study. The AGNPS model has been calibrated and validated on
many watersheds worldwide; therefore, it is assumed that the model results are

representative of realistic scenarios.

Chapter 2

2 Literature Review

2. 1 Vegetative Filter Strips
2.1.1 Overview

Sediment, nutrients, pesticides, and other NPS pollutants carried by
agricultural runoff are major water quality concerns. As fanning operations
intensify and the landscape changes, implementing conservation practices or
BMPs in agricultural watersheds is becoming the main focus toward alleviating
NPS pollution. Vegetative ﬁlter strips for example, are used to reduce a wide
range of NPS pollutants from various sources, such as herbicide runoff from
cropland (Misra et al., 1996; Arora et al., 1996) and sediment and nutrient
delivery from feedlots (Young et al., 1980, 1980; Edwards et al., 1983; Dickey
and Vanderholm, 1981 ), dairy facilities (Livingston and Hegg, 1981), swine
operations (Sievers et al., 1975; Chaubey et al., 1994), and poultry production
areas (Srivastava et al., 1996; Edwards, et al., 1997).

In particular, vegetative ﬁlter strips have been studied extensively in small
agricultural research plots for removal of sediment delivery from cropland to
improve water quality (Meyer et al., 1995; Magette et al., 1989; Raffaelle et al.,
1997). Experiments have tested various grasses, ﬁlter widths, and slopes to
determine the inﬂuence of these variables on certain pollutant loads. The results
have been consistent in showing that under optimal conditions 75% or more of

total sediments are ﬁltered (Gharabaghi, et al., 2001). The following section will

focus on issues that involve the reduction of sediments by incorporating a

vegetative ﬁlter strip on cropland adjacent to streams to improve water quality.

2.1.2 Vegetative Filter Strip Effectiveness

Signiﬁcant reductions in sediment load take place when ﬂow passes
through a ﬁlter strip. Dillaha et al. (1989) found that 84% and 70% of sediment
was ﬁltered by incorporating a 9.1 m and 4.6 m (30 and 15 ft) wide ﬁlter strip.
Parsons et al. (1994) noted that ﬁlter strips of varying widths resulted in sediment
reductions of greater than 50% from ﬁeld runoff. Daniels and Gilliam (1996)
studied total sediment removal through a ﬁlter strip and found that approximately
80% was ﬁltered.

The effectiveness of ﬁlter strips to reduce sediment load from runoff has
been attributed to their ﬁltering capacity (Ghadiri et al., 2001). For the ﬁltration
process to work efﬁciently, ﬂow should be shallow and uniform (Dillaha et al.,
1989). Dillaha et al. (1989) observed in ﬂatter regions of a watershed a ﬁlter strip
can retain greater than 50% of sediment. Jin et al. (2001) similarly noted that the
type of vegetation, width, slope, ﬂow rate, and sediment type of the contributing
area determine sediment entrapment in vegetative ﬁlter strips. Magette et al.

(1 989) also indicated that when the ratio of ﬁlter width to pollutant-contributing
area decreases, the effectiveness of the ﬁlter strips also decrease. The
effectiveness of a ﬁlter strip, i.e., ﬁltering capacity, thus is a complex function of
the characteristics of the delivery area and the streamside area where the ﬁlter is

installed.

2.1.3 Design Considerations

Not all stream segments within a watershed are candidates for installing a
ﬁlter strip. The design of a ﬁlter strip depends on many factors including the
purpose of the ﬁlter (nutrient, sediment, or pesticide removal); seasonal weather
conditions; type of vegetation; and soil, slope, size, and land use of the
contributing area (NRCS, 1999a). Vegetative ﬁlter strips are also only one part
of an overall system of conservation practices that control the source and

transport of contaminants throughout the watershed (NRCS, 1999a).

Slope. Filter strips are appropriate down-slope of cropland where pollutants are
more likely carried. Steep slopes increase ﬂow velocity so that concentrated ﬂow
will rapidly exit through the ﬁlter, decreasing the interaction time between
pollutants and the vegetation and soil. Therefore, ﬁlters are most efﬁcient where
no slope or gradual slopes exist so that shallow, uniform ﬂow can slowly pass
through. Dillaha et al. (1989) found that slope and ﬁlter width affect sediment
yield and sediment concentration. In addition, Robinson et al. (1996) identiﬁed
that ﬁlter strips were less effective for ﬁltering sediment from a 12% slope than a
7% slope. Jin and Romkens (2001) also studied optimal slopes for sediment
removal and noted that, as the slope of the contributing area increased to 4% or
6%, the ﬁlter strip failed. The Natural Resource Conservation Service (NRCS)
has suggested ﬁlter strip locations on slopes greater than 1% but less than 10%

for optimal ﬁlter strip effectiveness (NRCS, 1999b).

10

Width. Dense, narrow ﬁlters may be the most cost-effective way to reduce
sediment delivery in most areas throughout a watershed, but wider ﬁlters may be
needed elsewhere depending on the characteristics and size of the contributing
area. For the purpose of sediment removal, the slope, land use, and soil type of
the contributing area determine the ﬁlter width. NRCS suggestions for ﬁlter width
are identiﬁed in Table 2.1 (NRCS, 1999a). These suggestions are based on
hydrologic class of the soil and percent slope of the contributing area. Other
research, has identiﬁed that ﬁlter widths of 5 to 10 m (16 to 33 ft) are sufﬁcient in
ﬁltering up to 80% of sediment (Dillaha et al., 1989; Dabney et al., 1993;
Srivastava et al., 1996). Line (1991) studied trapping efﬁciencies with respect to
width and found that ﬁlter width greater than 6.1 m (20 ft) produces a small

amount of change.

Table 2.1: NRCS suggestions for ﬁlter strip length (NRCS, 1999a).

% slope of the
ﬁlter strips Length of Flow (Feet)
contributing Hydrologic Soil Group of Filter

area Area
A B C D
0-1 20 20 22 24
1-3 20 25 28 30
3-5 24 30 33 36
5—8 28 35 40 42
8-12 32 40 44 48
12—15 40 50 55 60
15-20 48 60 66 72
>20 May need additional guidance

11

Filter strips are more effective at removing coarse aggregate sediment
than clay-sized sediment or ﬁne organic particles. Neibling and Alberts (1979)
found that ﬁlter strips removed over 90% of total sediment at widths ranging from
0.6 - 4.9 m (2 -16 ft). Thirty-seven, 78, 82, and 83% of the clay-sized fraction
was removed by the 0.6, 1.2, 2.4, and 4.9 m (2, 4, 8, and 16 ft) ﬁlter strips
respectively. erson (1967) found that a 3.1 m (10 ft) wide ﬁlter was sufﬁcient to
remove the maximum percentage of sand that went through the ﬁlter, 15.2 m (50
ft) for silt, and 122 m (400 ft) for clay. Gharabaghi et al. (2001) indicated that
aggregates larger than 40 mm (1.6 in) in diameter were captured in the upper
half of a ﬁlter strip. In addition, approximately 90% or more of aggregates
smaller than 40 mm (1.6 in) were removed when low to moderate ﬂow rates
existed (Gharabaghi et al., 2001). Therefore, the soil type is an important factor
when determining the width of the ﬁlter strip, is, if the soil type of the
contributing area is primarily clay the ﬁlter strip should be wider than if the soil
type was primarily sand.

Studies have also shown that approximately 70% of the sediment is
deposited between the contributing area and the up-gradient area of the ﬁlter
(NRCS, 1999b). Robinson et al. (1996) found that the most effective part of the
ﬁlter strip was in the ﬁrst 3 to 4 m (10 to 13 ft), suggesting that ﬁlter strips act as
grass barriers or buffers instead of ﬁlters. Dillaha et al. (1989) also indicated that
20 to 50% of the trapped sediment, in a 9.1 and 4.6 In (29 and 15 ft) ﬁlter strip,
was deposited above the ﬁlter strip. In addition, Ghadiri et al. (2001) determined

that the deposited sediment in the backwater of the ﬁlter was independent of strip

12

width no matter how high the slope of the contributing area. This analysis
suggests that vegetation characteristics and planting rates are important aspects

when designing an effective ﬁlter strip.

Vegetation. Identifying the correct vegetative species for the ﬁlter depends on
the characteristics of the contributing area as well as the planned purpose of the
ﬁlter. Grasses or legumes (alfalfa) are usually used as the vegetation for the
ﬁlter strip (Table 2.2). The NRCS suggests that vegetation species should have
stiff stems and high stem density near the ground surface (NRCS, 1999a).
Overall the planted vegetation should slow runoff, increase inﬁltration, reduce
erosion, and trap contaminants (NRCS, 1999b).

The NRCS CORE 4 Manual suggests to plant stem densities ranging from
1,500 to 2,500 bunches/m2 (139 to 231 bunches/f?) for optimal conditions
(NRCS, 1999b). In contrast, Jin and Romkens (2001) found that 2,500 to 10,000
bunches/m2 (231 to 926 bunches/f?) increase trapping efﬁciency by 45%. Table
2.2 shows the planting rates of different grasses and their function when planted

for vegetative ﬁlter strip installation (NRCS, 1999a).

13

Table 2.2: Vegetation species and seeding rates for trapping sediment
within a ﬁlter strip (NRCS, 1999a).

 

 

 

 

 

 

 

Seeding Seeding Established Established
Species or Seeding Cool/Warm Rate Rate Density (Stems Density (Stems
Mixture Season (lb/Acre) (kg/ha) per n2) per m2)
Smooth Bromegrass Cool 15-30 17-34 50 540
Orchardgrass Cool 10-15 1 1-17 70 758
Reed Canarygrass Cool 10 1 1 50 540
Tall Fescue Cool 1 5-25 1 7-28 60 648
Tall Wheatgrass Cool 8-12 9-13 60 648
Prar’n'e gasses
lnterrnediate Cool 8-12 9-13 60 648
Wheatgrass
Eastern Gamagrass Warm 8 9 40 432
Switchgrass Warm 5-10 6-11 50 540
Timothy Cool 5-10 6-1 1 60 648
Alfalfa 6-10 7-1 1
Bromegrass Cool 6-12 7-13 60 648
Alfalfa 6-10 7-11
Orchardgrass Cool 2—5 2-6 60 648
Alfalfa 6—10 7-11

 

 

2.1.4 Maintenance

Vegetative ﬁlter strip effectiveness decreases with time. Dillaha et al.
(1989) found that the efﬁciency of a 4.6 m and 9.1 m (29 and 15 ft) ﬁlter strip that
removed 81% and 91% of total sediments respectively, dropped by an average

of 9% between the ﬁrst and second rainfall simulations. Maintenance of the ﬁlter

14

strip throughout the year is an important aspect when achieving ﬁlter
effectiveness. Optimal conditions for sediment removal is shallow uniform ﬂow.
Shallow, uniform ﬂow across the ﬁlter must be encouraged to avoid the formation
of channels or rills (NRCS, 1999a). When channels or rills do develop the areas
must be repaired immediately. Filter strips with cross-slopes, indicating
channelized or concentrated ﬂow, are 40-95% less effective at removing
sediment (Dillaha et al., 1989). Inﬁltration can be enhanced by ﬁlling the
channels or rills with porous or absorbent material (crushed limestone or wood
products) to reduce runoff and absorb pollutants contained in the runoff.
Installing a grassed waterway or some other conservation practice can aid in
reducing sediment in areas where concentrated ﬂow cannot be redirected.
Sediment often accumulates within a ﬁlter strip or along the upper part of a
ﬁlter. If accumulation becomes a problem the sediment should be removed
before it reaches a height where ﬂow is diverted around the ﬁlter. Machinery
may be needed to re-establish the interface between the contributing area and

the ﬁlter strip so concentrated ﬂow does not occur.

2.2 Conservation Programs
2.2.1 Overview

Financial incentives for vegetative ﬁlter strips are available through six
USDA conservation programs: Conservation Reserve Program (CRP),
Conservation Reserve Enhancement Program (CREP), Environmental Quality

Incentives Program (EQIP), Wildlife Habitat Incentives Program (WHIP),

15

Wetlands Reserve Program (WRP), and Stewardship Incentives Program (SIP).
These programs were developed by the USDA to encourage the use of
conservation practices. Cost-share assistance is provided to any landowner that

establishes approved cover on eligible cropland.

2.2.2 Conservation Reserve Program (CRP)

CRP is the nation’s largest private lands conservation program and is
implemented through the USDA Farm Service Agency (FSA). The current
version of CRP was enacted in 1985. It reached 13.7 million hectors (33.9
million acres) through September 2002, at an approximate cost of $1.6 billion
(FSA, 2002). USDA economists estimate that it generates far more savings than
it costs. CRP is particularly attractive to farmers because in addition to paying for
50% of the cost of installation, it pays 100% of the annual soil rental rate and an
extra 20% of the soil rental rate if grassed waterways, ﬁeld windbreaks, ﬁlter
strips, and riparian buffers are installed (FSA, 1999a). CRP contracts require
landowners to enroll into the program for 10—15 years.

Land that is accepted for enrollment has to meet certain eligibility
requirements. Cropland that is in production, pastureland that is suitable for use
as riparian buffers, and EPA designated wellhead protection areas are all eligible
areas (FSA, 1999a). In addition, highly erodible or environmentally sensitive land
is eligible for installation of the following practices: riparian buffers, ﬁlter stn'ps,
grass waterways, shelter belts, ﬁeld windbreaks, living snow fences, contour

grass strips, salt tolerant vegetation, or shallow water areas for wildlife. The

16

Environmental Beneﬁts Index (EBI) is used to prioritize land offered for
enrollment (FSA, 1999b). Scores are based on cost, six environmental factors:
wildlife, water quality, erosion, enduring beneﬁts, air quality beneﬁts from

reduced wind erosion, and state or national conservation priority areas (CPAs).

2.2.3 Conservation Reserve Enhancement Program (CREP)

CREP is a federal-state conservation partnership program that targets
signiﬁcant environmental effects related to agriculture and is based on state
participation (F SA, 2000). CREP is a spin-off from the original CRP program.
The USDA’s objective is to share costs and resources to address speciﬁc local
environmental problems. For example, the Michigan CREP has been designed
to reduce the amount of sediment by over 784,000 metric tons (864,000 tons),
nitrogen by 726,000 kg (1.6 million pounds), and phosphorus by 363,000 kg (0.8
million pounds) (FSA, 2000). In addition, the Michigan CREP is intended to
protect water supplies used by over one million people, protect over 8,100 linear
km (5,000 miles) of streams from sedimentation, and improve wildlife habitat in
the project areas over the next 15 to 20 years (F SA, 2000). Participating states
receive funding for 40,500 ha (100,000 acres). Incentives for CREP participants
are very high. Landowners are eligible for ﬁve types of payments: base annual
rental payments (soil rental rate and consolidated annual CRP rental payment),
incentive payments (consolidated annual CRP rental payment), maintenance
payments (consolidated annual CRP rental payment), state cost-share

assistance payment, and state lump sum one-time payment (FSA, 2000). In

17

addition to the normal rental payment, an extra incentive of $12 per ha (SS/acre)

is also given to the landowner for any lands that are highly erodible.

2.2.4 Environmental Quality Incentives Program (EQIP)

Under the EQIP program farmers and ranchers may receive ﬁnancial and
technical help to install a conservation practice or implement structures to
manage conservation practices. EQIP does not involve land retirement but
rather conservation farming on working farms (NRCS, 2002b). Landowners that
are interested in the program are subject to a minimum of a one-year contract
and up to a 10-year contract involving ﬁnancial and technical assistance and
education. EQIP may pay up to 75% of the cost of certain conservation
practices. Limited resource farmers or beginning farmers may be eligible for up
to 90% of the cost of conservation. The total cost-share and incentive payments

are limited to $450,000 per individual (NRCS, 2002b).

2.2.5 Wildlife Habitat Incentives Program (WHIP)

The WHIP program is a voluntary program for people who want to develop
and improve wildlife habitat on their own land (NRCS, 2002d). NRCS provides
technical and ﬁnancial assistance to improve ﬁsh and wildlife habitat. Up to 75%
of the cost is paid to the landowner to establish the habitat (NRCS, 2002d).

Agreements generally span 5 to 10 years.

18

2.2.6 Wetlands Reserve Program (WRP)

WRP offers landowners the opportunity to protect, restore, and enhance
wetlands on their property. NRCS provides technical and ﬁnancial support to
help landowners. The program offers three enrollment options: permanent
easement, 30 year easement, and restoration. Each enrollment option has
different cost-sharing advantages to the landowner, up to 100% of the soil rental
rate and also 75—100% of the cost to restore the land (NRCS, 2002c). The goal
of the program is to achieve the natural wetland functions combined with

optimum wildlife habitat on every acre enrolled in the program.

2.2.7 Emergency Watershed Protection (EWP)

The EWP protects the lives and property threatened by natural disasters
such as ﬂoods, hurricanes, tornadoes, and wildﬁres (NRCS, 2002e). The NRCS
provides technical and ﬁnancial assistance to preserve life and property
threatened by excessive erosion and ﬂooding. EWP provides funding for
clearing debris from clogged waterways, restoring vegetation, and stabilizing
riverbanks, for example by installing vegetative ﬁlter strips. The measures that
are taken must comply with environmental and economic regulations. The
beneﬁts are payable only too individual property owners. Seventy-ﬁve percent of
funds needed to restore the property are provided. The community or local

sponsor pays the remaining twenty-ﬁve percent.

19

2.3 Cunent Hydrologic Water Quality Models
2.3.1 Overview

Modeling the performance of vegetative ﬁlter strips, or any other BMP,
involves simulating multiple complex mechanisms that take place within a
watershed. Hydrologic/water quality models are used as tools to analyze the
hydrology, sediment, and nutrient transport throughout watersheds and to
evaluation these processes with or without the use of BMPs. There are three
types of models that are used for NPS pollution modeling, continuous watershed
scale, single event based watershed scale, and ﬁeld based models. Most of
these models that exist today were developed in the 19703 and 19803 (Borah,
2002a).

The most commonly used continuous simulation models include:
Chemical, Runoff, and Erosion from Agricultural Management or CREAMS
(Knisel, 1980), Hydrological Simulation Program or HSPF (Bicknell et al., 1993),
Areal Nonpoint Source Watershed Environmental Response Simulator or
ANSWERS-continuous (Beasley et al., 1980), Soil and Water Assessment Tool
or SWAT (Arnold et al., 1998), and Annualized Agricultural NonPoint Source
Pollution Model or ANNAGNPS (Bingner and Theurer, 2001). CASC2D (Ogden
and Julien, 2002), Precipitation-Runoff Modeling System or PRMS (Leavesley et
al., 1983), and European Hydrologic System or SHE (Abbott et al., 1986a,b)

have long-term and single-event based modes.

2O

Continuous models are useful for analyzing long-term effects of
hydrological changes and watershed management practices. Due to their
complexity these models require large amounts of data for simulations which also
require higher computing systems. Continuous models take a long time to
calibrate and validate, therefore, the results from these models are not yet
robust, which makes it harder to rely on the output as an accurate interpretation
of the processes that take place within the watershed.

Commonly used ﬁeld-based models include Drainage Model or
DRAINMOD (Skaggs, 1980), Groundwater Loading Effects on Agricultural
Management Systems or GLEAMS (Leonard et al., 1987), Root Zone Water
Quality Model or RZWQM (Ahuja et al., 1999), Erosion-Productivity Impact
Calculator or EPIC (VVIIliams et al., 1984), Water Erosion Prediction Project or
WEPP (Lane and Nearing, 1989), and Agricultural Drainage and Pesticide
Transport or ADAPT (Chung et al., 1992). Field based models only analyze the
sediment and nutrient mechanisms within an agricultural ﬁeld boundary. This
simplistic ﬁeld scale assessment ignores the soil and land use/cover of the
upland area beyond the boundary.

Many of the equations and concepts from the ﬁeld-based models have
evolved into watershed scale models. For example, the SWAT model emerged
mainly from SWRRB, which is a single storm event based model developed by
Arnold et al. (1990), but features from CREAMS, GLEAMS, and EPIC are
important elements of the model as well. The ﬁeld-based models have often

been used as the basic framework for the watershed scale models. Each ﬁeld-

21

based model has varying degrees of accuracy where model validations from in-
ﬁeld analysis have been analyzed extensively.

Single event models, such as AGNPS (Young et al., 1989), ANSWERS
(Beasley et al., 1980), KINEROS (Woolhiser et al., 1990), and DWSM (Borah
2002a) can be used to evaluate watershed NPS pollutants and design BMPs for
severe or actual storm events. The single storm event based models are user-
friendlier than continuous models but have the luxury of simulating the effects of
pollutants throughout a watershed unlike the ﬁeld-based models. The main
strengths and weaknesses of the single event-based models are described in the

sections below.

2.3.2 Areal Nonpoint Source Watershed Environmental Response
Simulator (ANSWERS)

The ANSWERS model was developed to evaluate the effects of BMPs on
surface runoff and sediment loss from agriculture watersheds. The original
ANSWERS model was developed in the late 19703 (Beasley and Huggins,
1982). The model is based on one of the ﬁrst true distributed parameter
hydrologic models (Huggins and Monke, 1966). The model components and
capabilities are runoff, inﬁltration, subsurface drainage, soil erosion, and overland
sediment transport. ANSWERS principle application is watershed planning for
erosion and sediment yield control on complex watersheds and water quality
analysis associated with sediment bound chemicals (Beasley et al., 1980).

The watershed representation consists of square grids with uniform

hydrologic characteristics, some having companioned channel elements (Boroh

22

et al., 2001). Rainfall excess is calculated by surface detention with empirical
relations. Inﬁltration and runoff is calculated using the Mannings equation and
continuity equations. The main weakness of ANSWERS is its erosion model.
The erosion component is largely empirical and simulates only gross sediment
distribution of eroded sediment using Yalin’s method (Dillaha and Beasley,

1983).

2.3.3 KlNematic runoff and EROSIon (KINEROS) model

KINEROS is an event oriented, physically based model developed by the
USDA-ARS. The model describes the processes of interception, inﬁltration,
surface runoff, and erosion from small agricultural and urban watersheds. The
partial differential equations describing overland ﬂow, channel ﬂow, erosion, and
sediment transport are solved by ﬁnite difference techniques (USDA-ARS, 2002).
The spatial variation of rainfall, inﬁltration runoff, and erosion parameters can be
accommodated.

Applications of this model are limited to small watersheds and speciﬁc
combinations of space and time increments. KINEROS does a relatively good
job simulating runoff and sediment yield at watershed scales of up to
approximately 1000 ha (4 sq. miles) (USDA-ARS, 2002). BMP evaluation is
based on detention basins and alterations to hydrologic and hydraulic conditions

(Borah, 2002b).

23

2.3.4 Dynamic Watershed Simulation Model (DWSM)

DWSM was developed to simulate surface and subsurface storm water
runoff, propagation of ﬂood waves, upland soil erosion, sediment transport, and
nutrient and pesticide transport in agricultural watersheds (Borah, 2000a).
Watershed representation is overland, channel, and reservoir segments deﬁned
by topographic-based natural boundaries (Borah, 2002a). The model has routing
schemes developed using approximate analytical solutions of physically based
governing equations preserving the dynamic behaviors of water, sediment, and
the accompanying chemical movements within a watershed (Borah, 2002a). The
DWSM is a newly developed model therefore validations are in process. The
model is still in the development stages but has a potential for simulating and

predicting accurate results.

2.4 Agriculture Non-Point Source Pollution (AGNPS) Model
2.4.1 Overview

The Agricultural Non-Point Source pollution (AGNPS) model version 5.0 is a
single event empirically based distributed parameter model (Young et al., 1989).
The model uses one time step (storm duration) to generate one value for each of
the output variables: runoff volume, peak ﬂow, sediment yield, and average
concentrations of nutrients (Young et al., 1989). The AGNPS model’s distributed
parameter approach allows for every area within the watershed to be analyzed.

AGNPS is a suitable tool to use to study the overall response of BMPs from a

24

single severe or design storm. Therefore, BMPs such as vegetative ﬁlter strips
can accurately be prioritized throughout watersheds.

Recently, the development of a GIS interface called AVNPSM has
enhanced the capabilities of the model (He at al., 2001). AVNPSM has allowed
the AGNPS model to assess larger watersheds with higher resolutions, yielding
an accurate description of the physical characteristics of the watershed. The
interface decreases the time and labor involved in producing the input for the
model, thus allowing the model to become widely accepted by watershed

managers.

2.4.2 Model Structure

The AGNPS model components use equations and methodologies that
have been well established and are extensively used by agencies such as the
USDA-NRCS (Young et al., 1989). Runoff volume and peak ﬂow rates are
estimated using the SCS runoff curve number method. Peak runoff rate for each
cell is calculated by an empirical relationship proposed by Smith and Williams
(1980). The Universal Soil Loss Equation or USLE described by Wischmeier and
Smith (1978) estimates upland erosion and sediment transport. Sediment is
routed from cell to cell through the watershed to the outlet using a sediment
transport and depositional relationship described by Foster et al. (1981) which is
based on a steady-state continuity equation (Young et al., 1989). These

equations are described in the sections below.

25

Hydrology. Runoff volume and peak ﬂow rate is calculated based on the SCS

curve number method.

 

= (P — 025')2 (2_1)

Q P+O.8S

Where 0 is the runoff volume (in), P is the precipitation rate (in/hr), and S is the
retention parameter (dimensionless).
The retention parameter (S) is expressed in terms of a curve number (CN)

as follows:

1000
S =——1O 2.2
CN ( I

The curve number (CN) depends on the land use, soil type, and hydrologic soil
condition.
Peak runoff rate (Op) in units of cfs for each cell is estimated by using an

empirical relationship proposed by Smith and Williams (1980).

Qp = 3.79A0.7CSO.16(R0/25.4)(09032420017)Lw-OJ9 (2.3)

Where, A is the Area in Acres, CS is the channel slope in ft/ft, RO is the runoff

volume in in, and LW is the watershed length width ratio (dimensionless). The

26

coefﬁcient values for the peak runoff rate (Op) were determined from ﬁeld

measurements (Young et al., 1989).

Erosion and sediment transport. A modiﬁed form of the universal soil loss

equation (USLE) is used to estimate upland erosion for single storms.

SL = (EI)KLSCP(SSF) (2.4)

Where SL = soil loss (tons/acre), El = rainfall and runoff erosivity index (100 ft ton
inch/acre inch), K = the soil erodibility factor (dimensionless), LS = topographic
factor (dimensionless), C = the cover and management factor (dimensionless), P
= the supporting practice factor (dimensionless), and SSF = factor to adjust for
slope shape within the cell (dimensionless).

Soil loss is calculated for each cell in the watershed. Five particle size
classes -clay, silt, small aggregates, large aggregates, and sand for eroded soil
and sediment yield is estimated for the watershed (Young et al., 1989).

Detached sediment is routed from cell to cell through the watershed to the outlet.
The basic routing equation is derived from the steady-state continuity equation as

described by Foster et al. (1981) and Lane (1982):

Q. (x) = Q. (0) +9., (x/L.) — j (xiwdx (2.5)

27

Where Q.(x) is sediment discharge at the downstream end of the channel reach
(lb/sec), Q.(O) is the sediment discharge into the upstream end of the channel
reach (lb/sec), Q... is the lateral sediment inﬂow rate (lb/sec), x is the downstream
distance (ft), Lr is the reach length (ft), w is the channel width (ft), and D(x) is the

deposition rate (Iblsec-ft). Deposition rate (D(x)) is estimated as follows:

D(X) = W... / q(x)][q,(x) - g'. 06)] (2.6)

Where V.. is the particle fall velocity (fps), q(x) is the discharge per unit width
(cfs), q.(x) is the sediment load per unit width (Iblsec-ft), and g’,(x) is the effective
transport capacity per unit width (lb/sec—ft).

Effective transport capacity is computed using a modiﬁcation of the

Bagnold (1990) stream power equation, as follows:

2

, TV
g . = 778. = 77";- (2.7)

SS

Where 9,3 is the transport capacity (dimensionless), n is an effective transport
factor (dimensionless), k is the transport capacity factor (dimensionless), c is the
shear stress (lblftz), and v is the average channel ﬂow velocity (fps) determined

by Manning’s equation. Young et al. (1987) describes values for the effective

transport capacity.

28

Sediment load for each of the ﬁve-particle size classes leaving a cell is

calculated as follows:

 

 

 

206) =[ 2"“) IIQJOHQri'ﬂI V” iq.(o>—g'. (0)]- V”

2CI(J€)+AxV,, 2 (1(0) q(x)g'.(x)]] (2.8)

Symbols for equation 2.8 are deﬁned above. Equation 2.8 is the basic routing

equation that drives the sediment transport model.

2.4.3 Input and Output Database

The AGNPS model subdivides the watershed into uniform cells to capture
the spatial variability of its landscape. A database consisting of 21 input
parameters for every cell representing the watershed is required to run the model
(Table 2.3). This distributed parameter approach allows the AGNPS to take into
account the spatial variability of the landscape and also examine sedimentation,
runoff or nutrients, either for entire watershed or on a cell-by-cell basis. The cell-
by-cell analysis not only pinpoints excessively polluted areas, but also is

signiﬁcant for evaluating where best management practices are effective.

Table 2.3: AGNPS Cell Input Parameters

 

1. Cell Number 8. SCS curve number (CN) 15. Fertilization incorporation
2. Overland ﬂow direction 9. Mannings roughness coeff. (n) 16. Fertilization level

3. Receiving Cell number 10. USLE C factor 17. Pest Indicator

4. Average slope (%) 11. USLE P factor (P) 18. Point source indicator

5. Average slope length 12. Surface condition constant (SC) 19. Gully source indicator

8. Slope shape factor 13. Chemical oxygen demand factor 20. lmpoundment factor

7. USLE K factor (K) 14. Soil texture 21. Channel indicator

 

29

Developing the AGNPS input database is extremely time consuming and
labor intensive, especially for large watersheds simulated with small cell sizes.
Capturing the spatial variability of a watershed is difﬁcult when a large cell size is
used. Previous research has concentrated on identifying an optimal cell size to
adequately capture the landscape variability within a watershed so the model's
estimates for runoff, sediment yield, and nutrients are reasonably close to the
measured data (Bhuyan et al, 2001).

Brennan and Hemlet (1998) used a Jack-Kniﬁng procedure, a
geostatistical method, for selecting a base cell size for the entire watershed and
also locating areas of cell sub-divisions within the watershed. The base cell size
was predicted according to the layout of the SCS Curve Number data, which
represents the land use characteristics, within the watershed. The base cell size
of 2.14 ha (5.3 ac) was selected based on the mean square residual of the
krigged estimates, less than 2.14 ha (5.3 ac) was not used because the average
mean square residuals were not signiﬁcantly lower. Locating cell sub-divisions
were also used for indicating an enhanced representation of the land use/cover
characteristics within the watershed. This analysis found that when 0.03 ha (0.07
ac) cell sub—divisions were included in the 2.14 be (5.3 ac) grid layout the
variance increased by 10%, indicating a better representation of the watershed.

The labor and time constraints of subdividing watersheds into smaller grid
cell sizes for accurate representation of the physical characteristics is a limiting
factor when evaluating NPS pollutants in the AGNPS model. Fortunately, GIS

interfaces have been developed to aid in analyzing larger watersheds with higher

30

resolutions. Since the 19903, GIS interfaces have aided in the development of
the AGNPS input database allowing larger watersheds to be modeled at a higher
resolution (smaller cell size) leading towards a more accurate physical
representation of the watershed.

The GIS interfaces allow grid cell size to be comparable to the resolution
of the raster GIS data that is used to derive the input parameters for the model.
Therefore, The grid layout for the AGNPS model is usually based on the DEM
datasets because the grid resolution is larger (30 m or 98 ft) than the soil and
land use/cover GIS layers (1 m or 3.3 ft). Readily available soil and land
use/cover data is digitized from 1 m (3.3 ft) USGS digital ortho photo
quadrangles and readily available DEMs are have resolutions of 30 m (98 ft) in
various parts of the nation. Therefore, the coarse grid layout is limited to the
resolution of the elevation data and the accuracy of the spatial resolution of the
soil and land use/cover for the watershed is limited to the size of the grids as
well.

Over the years, there have been a number of AGNPS GIS interfaces
developed (Srinivasan and Engel, 1991; Cronshey et al., 1993; Engel et al.,
1993; He et al., 1993; Jankowski and Haddock, 1993; Klaghoffer et al., 1993;
Mitchell et al., 1993; Yoon et al., 1993). Tim and Jolly (1994) demonstrated the
use of an Arc/Info — AGNPS interface for assessing the effectiveness of best
management practices. In their study, they demonstrated that sediment load
was reduced by 41% by implementing a vegetative ﬁlter strip around all stream

segments within a 417 ha (1030 ac) agricultural watershed located in southern

31

Iowa. The grid cell size for the entire watershed was based on the manual labor
involved, time constraints, and computing capability, thus a 100 m by 100 m (328
ft by 328 ft) grid was used for the analysis. The 100 m (329 ft) grid cells located
adjacent to the stream were subdivided to create 20 m (66 ft) grids. The 20 m
(66 ft) grid cells that touch the stream were buffered, simulating the buffer strip
scenario. The analysis was a modeling scenario where no ln-ﬁeld data was
presented. The 100 m (329 ft) grid cell size may not have deﬁned the “true”
landscape characteristics of the watershed, therefore, the results at the outlet
may not have been accurate.

Most recently, He at al. (2001) developed AVNPSM, an ArcView GIS
interface for AGNPS. The AVNPSM, a Windows based ArcView (version 3.0a or
later) GIS interface, developed to easily collect and manipulate the 21 input
parameters needed for the AGNPS input database so multiple scenarios can be
evaluated. The input database, created by AVNPSM, is a database ﬁle that is
imported into the AGNPS model, version 5.0. The interface, which was written in
ArcView Avenue scripts, uses three GIS layers to develop the database. These
layers include soil, land use/cover, and a DEM.

AGNPS creates a tabular output that can easily be imported into ArcView
for evaluation. The outputs can be examined for each cell throughout the
watershed or at the watershed outlet. The tabular output provides estimates of
runoff volume (inches), peak runoff rate (cfs), sediment yield (tons), sediment
concentration (ppm), upland erosion (tons/acre), amount of deposition (%),

sediment generated within each cell (tons), mass of sediment attached and

32

multiple chemical outputs associated with nitrogen, phosphorus and chemical

oxygen demand (Young et al., 1989).

2.4.4 Validation

The AGNPS model has been validated on several watersheds all over the
world resulting in varying degrees of accuracy. Young et al. (1989) found a
coefﬁcient of determination (r2) to be 0.81 for sediment yield estimates on three
different watersheds ranging from 1,000 to 4,000 ha (2,500 to 10,000 ac) located
in north central United States. A 4 ha (10 ac) cell size was used for each
watershed. Perrone and Madramootoo (1999) tested AGNPS on a 2,700 ha
(6,700 ac) watershed in Quebec. A 9.25 ha (23 ac) cell size was used for the
evaluation. The storm events that were modeled resulted in an average error of
28.2%.

Mitchell et al. (1993) evaluated 50 sediment yield events from two
watersheds and ﬁve sub watersheds located in East Central Illinois. Half of the
events were used to calibrate the model and the other half were used for
validation. A 20 m by 20 m (66 ft by 66 fl) cell size was used to predict sediment
yield from small, mild-sloped watersheds ranging from 30 to 1.6 ha (74 to 4 ac).
Predicted vs. observed resulted in a mean error of 35% for the validated data
and 23% for the calibrated data.

The AGNPS model was used to assess runoff, soil erosion, and
associated NPS pollutants in a watershed located in Italy. Lenzi and Luzio

(1995) found that the predicted runoff volume was estimated precisely but peak

33

ﬂow rates were poorly predicted at high and low storm events. Predicted
sediment and nutrient loads were overestimated and underestimated in the same
events. The average error for predicted sediment and nutrient load resulted in
26% and 14% respectively. This analysis demonstrated that the AGNPS model
is capable of predicting sediment and nutrient loads for large storm events but

the authors suggest further investigation of the model.

2.4.5 Sensitivity Analysis

A sensitivity analysis is an essential process when understanding model
responses to parameter changes. Finding the parameters that have the largest
or no impact on the model is of great importance. Knowing the sensitivity of the
model parameters will aid in understanding errors that occur in the output.

Young et al. (1987) demonstrated a sensitivity analysis for all of the
parameters of the AGNPS model. The sensitivity analysis of varying all the
parameters by -50%, -25%, 25%, and 50% reported that the slope-associated
parameters (LS) were the most sensitive when evaluating sediment yield (Figure

2.1).

 

Sensitivity Analysis

 

0 Rain
I El

-. CN
x LS
X FSL
0 CS
+ CSS
- N
- K

LS Factors

    

Results

 

 

 

 

-50 -25 0 25 50
% varlatlon

 

 

 

Figure 2.1: Young et al. (1987) sensitivity analysis indicating slope-
associated parameters (LS) are the most sensitive in the AGN PS model
when predicting sediment yield.

Topographic attributes are important factors in predicting sediment loss
when using hydrologic models as shown by the sensitivity analysis in Figure 2.1
for the AGNPS model. Most watershed analyses, including evaluations using the
AGNPS model, use 30 m USGS DEMs to derive the slope associated
parameters (aspect, slope length, slope shape, percent slope). These elevation
datasets are readily available for free download off of the lntemet for various
parts of the US.

The 30 m USGS DEM data is in 7.5-minute units and corresponds to the
USGS 1224,000 scale topographic quadrangle map series for the US. with a
spatial resolution of 30 by 30 meters (USGS, 2002a). The uncertainty of USGS

35

DEMs have been evaluated in certain spatial modeling applications by using
various geostatistical methods (Fisher, 1999) but no data has been reported for

evaluating the 30 m USGS DEM uncertainty in the AGNPS model.

36

Chapter 3
3 Materials and Methods

3.1 Study Area

The Stony Creek watershed, located in Clinton County, Michigan drains
45,452 ha (112,314 acres) and is a subbasin of the Grand River, a major
tributary of Lake Michigan (Figure 3.1). Land use in the Stony Creek watershed
is 85% agricultural land consisting of corn, soybeans and wheat, as well as more
diverse crops such as mint, with the remaining 15% a mixture of urban areas,
forests, shrubland, and wetlands or water (Figure 3.2). Soil types vary
extensively throughout the watershed from well-drained sandy loam soils to
poorly drained clay soils (Figure 3.3). The topography is predominantly ﬂat but in
some areas gentle slopes exist, with the elevation ranging from 198 to 277 m
(650 to 909 ft) above sea level (Figure 3.4). Please note the following images

throughout this thesis are presented in color.

37

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Figure 3.1: Location of the Stony Creek watershed.

38

 

Agricultural Land
- Forest Land

Rangeland
E Urban and Built Up

- Water
:1 Wetlands

Figure 3.2: Land use in the Stony Creek watershed.

39

 

m Clay Loam
C] Fine Sand

Fine Sandy Loam
Gravel Pit

Loam
E] Loamy Sand
m Muck

l:l and

- Sandy Clay Loam
- Sandy Loam

- Sllt Loam

Slly Clay Loam
Sloan Loam

(:1 Stony Sandy Loam
- Water

 

Figure 3.3: Soil textures of the Stony Creek watershed.

40

   

Elevation Above
Sea Level (m)

:1 197 - 206
E] 206 - 215
[:l 215 - 224
I! 224 - 233
I 233 - 242
[j 242 - 250
250 - 259
- 259 - 268
- 268 - 277

    

Figure 3.4: Topography of the Stony Creek watershed.

3.2 AGNPS Modeling Scenarios
3.2.1 Overview

The AGNPS distributed parameter approach allows analysis at any point
throughout the watershed at the resolution of ﬂte speciﬁed grid cell size. The
AGNPS model is potentially capable of analyzing watersheds consisting of
30,000 cells, but experience has Tshown that the model tends to crash when the
database exceeds 15,000 cells, therefore only allowing analysis of smaller
watersheds. Larger cell sizes can be used when analyzing extensive watersheds
greater than 4,047 ha (10,000 ac) but this analysis sacriﬁces detailed

characteristics of the watershed. For the purposes of this study four primary sub-

41

characteristics of the watershed. For the purposes of this study four primary sub-
watersheds of Stony Creek (Bad Creek, Lost Creek, Muskrat Creek, and
Spaulding Drain) were analyzed and then subdivided into 31 secondary sub-
watersheds ranging from 284 to 1,214 ha (702 to 3000 ac) (Figure 3.5). The

grid-cell resolution for each sub-watershed was 30 meters (0.22 ac).

Spauldin . Drain

 

Figure 3.5: Thirty-one secondary sub-watersheds of the Stony Creek
watershed.

The AGNPS GIS interface, AVNPSM, was used to develop the AGNPS
input database for the 31 sub-watersheds. The interface uses three raster GIS
layers (soil, land use/cover, and a digital elevation model (DEM)) and a boundary
layer to develop the AGNPS input database for each watershed. An example of

the four GIS layers needed to develop the input database is shown in Figure 3.6.

42

 

 

 

 

 

 

 

 

 

Figure 3.6c: Soil type layer Figure 3.6d: 30 m USGS DEM

Figures 3.6a-d: Four ers layers for the AVNPSM interface.

43

Modeling scenarios consisted of simulating the hydrology and sediment
transport throughout the 31 sub-watersheds of Stony Creek with a baseline
scenario (no ﬁlter strip, non-buffered cells) and with a 30-meter vegetative ﬁlter
strip placed around each stream segment (ﬁlter strip scenario, buffered cells).
Filters were only simulated in agricultural areas, i.e., forests or urban areas
located along the streamside were not buffered. The scenarios were evaluated
with AGNPS by simulating a 10yr-24hr storm event with precipitation of 87.1 mm
(3.43 inches) and corresponding energy intensity of 1859 Nlm2 (70.47-
ft’ton/acre-inch) (Huff et al., 1992).

AGNPS creates a tabular output that can easily be imported into ArcView
for evaluation. Although AGNPS creates several outputs, sediment yield was the
only output that was reported in this study. Filtered sediment was calculated as
the difference in the ﬁlter strip and baseline scenarios, which measured cell
effectiveness. The ability of the cells to ﬁlter the entire amount of entering

sediment (percent sediment ﬁltered) was also calculated,

[(Baseline scenario - Filter strip scenario)! Baseline scenario] * 100 (3.1)

thus measuring the buffered cell efﬁciency. These watershed evaluation

procedures led to identifying and prioritizing areas of ﬁlter strip effectiveness and

efﬁciency throughout the 31 sub-watersheds in Stony Creek.

44

3.2.2 Filter Strip and No Filter Strip Input Database

The data sets used for the study area were obtained from a variety of
different sources (Table 1). The NRCS county soil survey database (SSURGO)
for Clinton County was used to identify soil texture and soil erodibility factor (K).
The SSURGO soils database is generally the most detailed level of soil
geographic data developed by the National Cooperative Soil Survey. The digital
vector data is collected and archived in a 7.5—minute topographic quadrangle
unit, mapped on a 1:15,840 scale using aerial maps or remotely sensed images
(USGS, 2002b)

A 7.5-minute digital elevation model provided by USGS was used to
determine slope, slope length, slope shape, and flow direction for the AGNPS
database. The USGS DEMs are digital representations of cartographic
information in raster form (USGS, 2002a). The 7.5-minute units correspond to
the USGS 124,000 scale topographic quadrangle map series, which provide
elevation values sampled at 30 by 30 m intervals (USGS, 2002a).

The land use database, digitized and reprojected by the State of Michigan
Center for Geographic Information Systems consists of the 1992 National Land
Cover Data (NLCD). The vector NLCD is derived from the mid-19903 30 m
spatial resolution Landsat Thematic Mapper satellite data and aerial maps
(USGS, 2002c). The Stony Creek’s land use included ﬁve different land cover
classes: woodland, shrublandlwetlands, water, farrnsteads, and cropland. Each

land cover class was assigned a value for the SCS curve number (CN), crop

45

management factor (C), overland Manning’s value (n), and surface condition

constant (SC) based on the digitized land use database (Table 3.1).

Table 3.1: AGNPS input parameters and the corresponding data sources.

 

 

Parameter Data Type Data Source
1. Cell Number Topography uses DEM
2. Overland ﬂow direction (1-8) Topography USGS DEM
3. Receiving Cell number Topography USGS DEM
4. Average slope (%) Topography uses DEM
5. Average slope length Topography USGS DEM
6. Slope shape factor (1, 2, or 3) Topography USGS DEM
7. USLE Kfactor (K) Soil SSURGO county soil database
8. SCS curve number (CN) Land use 1992 Naional Land Cover
9- Mannings mughness °°°"- 0" Land cover 1992 National Land Cover
10- USLE C factor (C) Land use 1992 National Land Cover
11- USLE P fact“ (P) Land cover 1992 National Land Cover
12. Surface condition constant (SC) Land use 1992 Naional Land Cover
13. Chemical oxygen demand factor
(COD) Land use 1992 National Land Cover

14. Soil texture (1. 2. or 3) Son SSURGO county soil database
15. Fertilization incorporation (1 or 0) Assume -

none
16. Fertilization level Assume _

none
17. Pest Indiwor (1 or 0) Assume -

none
18. Point source indicator (1 or 0) Assume -

none
19. Gully source indicator Topography USGS DEM
20. lmpoundment factor (1 or 0) Assume -

none
21. Channel indicator (1 or 0) Hydrology 1992 Name" Land CW“

 

Many of the 21 parameters were assumed for both scenarios. The USLE
conservation practice factor (P) was assumed to be 1 to simulate worst-case
occurrences. The soil texture number (sand=1, silt=2, clay=3) was set at a 2, the
closest soil texture number simulating loam and silt loam soils. The fertilizer,
pesticide, point source, and impoundment factors were set at zero, since the
study did not focus on nutrient or pesticide pollution.

To simulate a ﬁlter strip within AGNPS, four input parameters were
manipulated for the streamside cells: the curve number, C—factor, overland
Manning’s value and surface condition constant. Tim and Jolly (1994) also
chose these parameters for their filter strip analysis. The curve number for the
ﬁlter strip was deﬁned as brush-weed-grass mixture with good hydrologic
condition (SCS, 1986). The C-factor for the ﬁlter strips was assigned a value of
0.003, representing a ﬁlter with 95% vegetative density and a vegetative canopy
of 75% grass or grass-like plants (Wischmeier and Smith, 1978). The overland
Manning’s value was set at 0.25, representing a grass pasture (NRCS, 1979)
and the surface condition constant was set to 1.0, an internal indicator in AGNPS
for simulating a ﬁlter strip (Young et al., 1994). These values are listed in Table

3.2.

47

Table 3.2: Land use parameters for the Stony Creek watershed.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Land Cover Class Hydro CN C P n SC
Class

Cropland (85.1%) A 64 0.22 1 0.04 0.05
B 75 0.22 1 0.04 0.05
C 82 0.22 1 0.04 0.05
D 85 0.22 1 0.04 0.05

Urban (2.3%) A 59 0.01 1 0.015 0.01
B 74 0.01 1 0.015 0.01
C 82 0.01 1 0.015 0.01
D 86 0.01 1 0.015 0.01

Shrubland -including

wetlands (3.81%) A 30 0.08 1 0.2 0.29
B 58 0.08 1 0.2 0.29
C 70 0.08 1 0.2 0.29
D 78 0.08 1 0.2 0.29

Water (0.09%) - 100 0 0 0.99 0

Woodland (8.7%) A 30 0.002 1 0.4 0.29
B 55 0.002 1 0.4 0.29
C 70 0.002 1 0.4 0.29
D 77 0.002 1 0.4 0.29

Buffer A 30 0.003 1 0.25 1
B 48 0.003 1 0.25 1
C 65 0.003 1 0.25 1
D 73 0.003 1 0.25 1

 

 

 

 

 

 

 

 

 

3.3 AGNPS Uncertainty Due to DEM Error
3.3.1 Method for Evaluating Error Propagation

In this section, elevation uncertainty is assessed to measure its effect on
estimates of sediment yield within the AGNPS model. The uncertainty of terrain

data is a factor of the distance that a particular spot on the landscape is from the

48

nearest data point, the variation of terrain between the data points, and the
accuracy of the elevation measure in the datasets (lsaaks and Srivastava, 1989).
Therefore, as the elevation data points become more generalized, 9.9., 30 m
(0.22 ac) elevation data as opposed to 10 m (0.025 ac) elevation data, the
uncertainty is greater. This generalization effect has repercussions when using
the data in spatial modeling applications.

Measuring the uncertainty in the spatial modeling application (AGNPS)
due to errors in the input data (elevation data) helps to understand the
conﬁdence limits associated with the result, i.e., determining the error in the
output, given the operation and the errors in the input attributes (Heuvelink,
1999). There are many methods used to evaluate this uncertainty, such as the
Monte Carlo method (Heuvelink, 1999).

The Monte Carlo method (Hammersley and Handscomb, 1979; Lewis and
Orav, 1989) used in this study, computes the result of a spatial modeling
application (AGNPS) repeatedly, with randomly generated input data (elevation
data) sampled from their joint distribution (Heuvelink, 1999). The distribution of
the results is then statistically assessed, which reﬂects the uncertainty in the
spatial modeling application due to the error in the input data. Figure 3.7

characterizes the framework for the Monte Carlo process.

49

 

 
     

     
     

 
 

Spatial

#4332335!)

(AGNPS)

«- Input I : :
(Elevation ,
Datasct)

  

MODEL
RESULTS
(Sedimen
Yield)

 

 

 

 

Figure 3.7: Monte Carlo framework for evaluating uncertainty in the spatial
modeling application (AGNPS model) due to the error in the input data
(elevation dataset).

To assess uncertainty in the elevation data, error must be measured. In
this study, a higher accuracy elevation dataset was compared to a coarse

elevation dataset:

Eu = Hu - Cu. 0 = {(X1. Y1). (X2. Y2). .--(Xn.Yn)} (3-2)

where E is the error, C is the coarse elevation dataset, H is the higher accuracy
elevation dataset, and u are a set of locations (possibly gridded) in a spatial
dataset. In addition, E, C, and H are shorthand for spatial datasets consisting of
grids of values. Stochastic simulation approach was used to develop a set of
error realizations. From the distributions of the error maps a mean and a
standard deviation can be obtained which relates to the “true” values of the

higher accuracy elevation data (Wechsler, 2000).

50

A sequential Gaussian simulation was used to produce a set of error
realizations based on E. This process is commonly used in geostatistics to
evaluate error in the input data for any spatial modeling application (Gooverts,
1997). The elevation error was assumed to have a normal distribution to employ
this process. The following steps describe how the sequential Gaussian

simulation constructs the error realizations.

Step 1. A variogram must be employed to the error data to account for its spatial
structure. Accounting for the spatial structure in the error data allows analysis of
the “true” spatial patterns of the error in the elevation for that speciﬁc geographic
region. A variogram summarizes the relationship between differences in pairs of
measurements and the distance of the corresponding points from each other

within the dataset. The variogram is calculated by equation 3.3:

y(h) = (1/2)Var[Z(x)-Z(x+h)] (3.3)

where, h is the lag distance separating pairs of data points, Var is the variance of
the argument, Z(x) is the value of the regionalized variable of interest at location
x, and Z(x+h) is the value at the location x+h (Lin and Tang, 2000). Generally,
as the lag distance increases the plot should rise simulating at larger distances
the variability is greater.

At large values of the lag-distance (h) the plot tends to level off becoming

horizontal. At this distance the variogram has reached a sill, which is

51

theoretically the sample variance. The distance to the sill is called the range
where data points are spatially autocorrelated (Bailey and Gatrell, 1995). The
nugget is the magnitude of discontinuity at the origin. A variogram model type is
decided by plotting the empirical variogram and examining the behavior of the

sill, range, and nugget.

Step 2. A randomly generated path is deﬁned for visiting each of the nodes (u) or

cells throughout the error data.

Step 3. Sequentially visit each node (u).

Step 3a. Simple kriging is performed for grid node (u) using the variogram model
developed in step 1. The result is both an expected value and a kriging variance.
The expected value is a best linear unbiased estimate (BLUE), in the least
squares sense. The kriging variance is the minimized value of the estimation
variance or the variance of the error of estimation. The kriging variance is a
measure of the certainty the model has in its estimate and is a function of the
variogram model and the distance to surrounding known points. The variance is
higher for locations remote from unknown nodes. Kriging is a geostatistical
estimation method by which optimal weights are assigned to unknown values
based on the variogram model, since the variogram changes with distance the

weights depend on the known sample distribution.

52

Step 3!). Next, a random number from a normal (Gaussian) distribution is drawn
for “u” that has a variance equivalent to the kriged variance and a mean
equivalent to the kriged value. This number will be the simulated number for that

grid node. When all nodes have been simulated this is the ﬁrst realization.

Step 4. Repeat steps 2 and 3 for all the other realizations using a different
random number sequence to generate multiple error realizations of the original

error map.

The error realizations match the statistical characteristics speciﬁed by the
error model. They are equiprobable in the sense that each is representation of
the true error in the dataset. The error realizations are then added to the coarse
elevation data of the study area. Each new elevation realization for the study
area is used for evaluating sediment load in the AGNPS model. The
sedimentation results from using the new elevation realizations in the AGNPS
model are separately compared to the sedimentation results from the coarse
elevation data. The results from the simulations construct a distribution of
possible outcomes. The width of this distribution reﬂects the uncertainty in the

AGNPS model due to coarse elevation error.

53

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

True Coarse
Elevation Elevation : Error
Data Data
1
EIL
E Variogram ‘1
8
6
4!
I
.2
g
3
Distance (m)
Error
Model
l l i l
l I l l l I Coarse
Error Realizations + 33:21:11 :

 

l l i lr <

 

 

 

 

 

 

 

 

l I l j l [ Sediment Yield
tons
New Elevation Realizations —> 33%;?- : L—L

 

 

 

 

 

 

 

 

 

 

 

 

Figure 3.8: Stochastic simulation approach for evaluating uncertainty in the
AGNPS model due to error in the coarse elevation data.

3.3.2 Unconditional Stochastic Simulation

If the resulting realizations honor the data values at the sampled locations
the process is said to be a conditional simulation. Since the data values in this
study do not represent the actual study area values the process is unconditional.
The unconditional approach, used in this study, is applied because no
information about true elevation is available within the study region. Therefore,
an error model is developed by comparing higher accuracy “true” elevation data
with a coarse elevation dataset from a geographic region that has similar
geomorphological characteristics to the study area. A statistical error model
characterizing error magnitudes and spatial patterns observed for the coarse
elevation data can be developed by assuming that the error characteristics for

the two geographic areas are the same.

3.3.3 Methodology Using the Study Area’s Watershed

The unconditional stochastic simulation approach was employed on a 107
ha (264 ac) sub-watershed of Stony Creek (Figure 3.9). Thirty-meter USGS
DEM data is available for the study area but higher accuracy elevation data is
not. However, a 30 m USGS DEM and higher accuracy elevation data,
characterizing the “true” elevation, is available for a watershed located at
Michigan State University Farms (MSU Farm watershed), which is 43 km (27
miles) south of the Stony Creek study area (Figure 3.9). The higher accuracy
digital elevation data for the MSU Farm watershed (Survey DEM) is generated

from ground truth data surveyed at various spacing and gridded to a 0.6 m (2 ft)

55

resolution. The 0.6 m (2 ft) DEM was resampled to 10 m because the 10 m grid
for this watershed consisted of 10,968 cells, which was the largest database that

operated efﬁciently in the AGNPS model.

 

Clinton County

WAMW

Q

 

 

 

Locational
IMSUFann
(~43kmsouthotStudyArea)

Ingham County

 

 

 

1. 0 1. KW
E

 

Figure 3.9: Location of study area and MSU Farm watershed site.

The MSU Farm watershed and the study area have similar
geomorphological characteristics. Therefore, it is assumed that the error
characteristics for the USGS DEM covering the MSU Farm watershed are similar
to the unknown characteristics of the USGS DEM overlying the study area in
Stony Creek. By assuming that the error characteristics are the same a

statistical error model characterizes error magnitudes and spatial patterns

observed for the MSU Farm DEM. The objective of this analysis is to employ an
error model for the derivation of the USGS DEM of the MSU Farm watershed
from higher accuracy elevation data characterizing the “true" elevation. The error
model can then be used to generate realizations of error for the 30 m USGS
DEM.

The USGS DEM of the MSU Farm watershed was interpolated or
resampled to 10 m by nearest neighbor interpolation so it could be compared to
its 10 m Survey DEM. The resampled DEM is termed the USGS resampled
DEM. The difference between the two data sets was calculated and termed the
error (Equation 3.4). Error is a spatially extensive variable; thus, an error

magnitude is present for every cell in the watershed.

Thus,

USGS resampled DEM — Survey DEM = Error (3.4)

Gstat, a geostatistics modeling program, was used to ﬁt a variogram
model to the error (Pebesma, 1999). A variogram model was ﬁt to the empirical
semivariogram by a weighted least squares method and yielded a Gaussian
model with a sill of 1.76 and a range of 53.8 plus an exponential model with a sill
of 6.8 and a range of 115, where each lag had thousands of point pairs (Figure
3.10).

57

 

 

A A l v v
M.” v 1 vfa f

01 02 V
k
x

semivariarce
.5

—— 1.7672 Gau(53.8146) + 6.84041 Exp(115.075)

 

 

 

O - - -
0 100 200 300 400 500

distance it meters

Figure 3.10: The variogram model of the error between the Survey DEM and
the USGS resampled DEM of the MSU Farm watershed.

The variogram model, a mean error of 0.6 m (2 ft), and a mask map (ascii
grid) identifying the watershed area were used to create a set of error realizations
for the study area. The USGS DEM of the study area was interpolated bilinearly
to 10 m so each random error realization could be added to it (Equation 3.5).

The resampled USGS DEM for the study area is termed the Study Area USGS
resampled DEM. The DEM realizations create a collection of alternative equally

probable models of spatially distributed variable uncertainty or error (Wechsler,

58

2000). For the scope of this study, nine new DEM error realizations were created

(Figure 3.11a—i).
Thus the study area DEM realizations yield:

Study Area USGS resampled DEM + Error realizations = DEM realizations (3.5)

Elevation
Above Sea
Level (m)

218 - 221
221 ~ 223

   

Figure 3.11a: DEM Realization 1 Figure 3.11b: DEM Realization 2

59

Elevation
Above Sea
Level (m)

218 - 221
221 - 223

   

Figure 3.11c: DEM Realization 3 Figure 3.11d: DEM Realization 4

Elevation
Above Sea
Level (m)

218-221
221-223
223-220
220-228
228-231
231-233
233-230
23-230

1: 239 - 243

   

Figure 3.11e: DEM Realization 5 Figure 3.11f: DEM Realization 6

60

 

Elevation
Above Sea
Level (m)

218 - 221
221 - 223

   

Figure 3.119: DEM Realization 7 Figure 3.11h: DEM Realization 8

Elevation
Above Sea
Level (m)

219 - 221
221 - 223

 

Figure 3.11i: DEM Realization 9

Figure 3.11a-i: DEM Realizations 1-9 of Study Area watershed.

61

 

The nine DEM realizations were used to derive the four slope associated
parameters (aspect, slope, slope length, slope shape) for the AGNPS model to
simulate nine sediment yield results at the outlet of the study area watershed for
a 10 yr-24 hr storm event. The other parameters within the AGNPS model were
kept constant as the slope-associated parameters changed for each DEM
realization. The distribution of the sedimentation results at the outlet of the
watershed was statistically assessed to calculate a mean and a standard
deviation.

To identify the uncertainty in changing grid cell size from 30 m to 10 m, the
USGS resampled DEM for the study area was also used in the AGNPS model to
simulate sediment load at the outlet of the watershed. The sediment yield from
using each of the nine DEM realizations and the USGS resampled DEM were
compared to the sediment yield from using the 30 m USGS DEM for the study
area.

There is a precision problem with characterizing cell size in the AGNPS
model, version 5.0: when deﬁning cell size the model only allows two decimal
places. The 30 m cell size (0.22 acres) matches the actual acreage well enough
with two decimal places. However, the 10 m cell size of 0.0247 acres does need
more decimal places to characterize its tme acreage. The rounding effect
increases the cell size to 0.03 acres, which would inﬂate the watershed size and
also increase the results at the outlet signiﬁcantly over a number of cells, i.e.,
10,968 cells. Therefore, to account for these effects each simulation including

the 30 m data was increased in cell size by a factor of 100. The 10 m and 30 m

62

cell size yielded 2.47 acres and 22.44 acres respectively. The sediment yield at
the outlet will not result in actual sedimentation estimates but can be compared
on the same scale to identify uncertainty in the AGNPS model when estimating

sediment yield due to error in the DEM.

63

Chapter 4

4 Results and Discussion

4. 1 Stony Creek Secondary Sub-watershed Results

The sediment yield within the 31 Stony Creek secondary sub-watersheds
were analyzed by using the AGNPS model to identify site-speciﬁc areas along
stream segments where ﬁlter strips were effective (tons of sediment ﬁltered) as
well as efﬁcient (% sediment ﬁltered). Four secondary sub-watersheds were not
completed due to errors in their AGNPS input databases, three located in
Muskrat Creek (sub-watershed number 10, 15, and 16) and one located in Bad
Creek (sub-watershed number 19). The data for the remaining 27 secondary
sub-watersheds are presented in this analysis.

There was not a signiﬁcant difference in the predicted sediment loads at
the outlet of each secondary sub-watershed between the baseline and the ﬁlter
strip scenarios. Filtered sediment at the sub-watershed outlets varied from 3 to
50 tons as shown in Table 4.1. This analysis presented the overall effectiveness
or lack of effectiveness of ﬁlter strips throughout the watershed. However,
throughout this study, the main focus is to prioritize vegetative ﬁlter strips by
comparing areas of ﬁlter strip effectiveness and efﬁciency along stream

segments within the watershed.

Table 4.1: Filter strip effectiveness at the outlet of the 27 secondary sub-
watersheds.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Secondary Sub- No Filter Filter Strip Filtered

Primary Sub- Sub- watershed Strip/Baseline Results Sedimentl %
watershed watershed Size (ha) Results (tons) (tons) @s) Reduction

Lost Creek 1 1,320 500 473 27 5.4

2 604 225 212 13 5.8

Muskrat Creek 3 468 104 101 3 3.0

4 302 422 373 50 1 1 .8

5 959 290 264 26 8.9

6 540 451 425 26 5.7

7 1 ,255 371 350 20 5.5

9 755 230 216 15 6.4

1 1 768 276 260 16 5.9

12 420 215 200 15 7.0

13 568 231 213 19 8.0

14 521 179 170 9 5.0

15 872 261 252 9 3.5

Bad Creek 17 624 222 199 23 10.2

18 1,107 256 244 13 5.0

20 694 280 251 29 10.2

21 522 183 171 12 6.6

22 918 340 316 24 7.0

Spaulding Drain 23 301 124 114 10 7.9

24 370 149 136 13 9.0

25 1,030 268 247 22 8.1

26 415 104 97 7 6.7

27 349 147 135 12 8.1

28 264 119 106 12 10.5

29 358 142 125 17 12.2

30 732 229 213 16 7.2

31 323 116 107 9 7.8

 

 

 

 

 

 

 

 

 

The cell-by-cell analysis did not show one or two distinct areas of ﬁlter
efﬁciency or effectiveness throughout the 27 sub-watersheds. Therefore, it was

important to assess the reasons for the scattered results. In ArcView, version

65

3.2, a brushing technique was used to examine the ﬁlter strips that were
efﬁciently ﬁltering sediment (>50% reduction) throughout the 27 secondary sub-
watersheds (Figure 4.1). In these areas, the tons of sediment produced was
minimal compared to other areas within the watersheds and in most cases the
upland contributing area entering the ﬁlter strip was less than 4 ha (10 ac). This
analysis is simulating the effect of ﬁlter strips only being able to ﬁlter shallow,
uniform ﬂow as described by Dillaha et al. (1989). In the areas where the ﬁlter
strips upland contributing area is greater than 4 ha (10 ac) the majority of the
cells have a percent ﬁltered sediment of less than 25%. Therefore, the effect of
concentrated ﬂow is deﬁning areas where the ﬁlter strips are less effective in
reducing sediment, which indicates that ﬁlter efﬁciency is dependant on drainage

area.

.100% Filtered
ediment

 

Figure 4.1: 100% ﬁlter efﬁciency for the 27 secondary sub-watersheds of
Stony Creek.

Drainage area was compared to percent ﬁltered sediment for the 27 sub-
watersheds. The data was compiled into separate bar graphs for the four
primary sub-watersheds of Stony Creek to analyze the buffered cell’s distribution
of ﬁlter efficiency with respect to drainage area. Filter strip efﬁciency was
grouped into four categories (0-25%, 25-50%, 50-75%, 75-100%) and drainage
area was classiﬁed by natural breaks in ArcView into ﬁve different categories as
shown in the Figures 4.2, 4.3, 4.4, and 4.5. The cells that did not accumulate

any sediment in the no ﬁlter strip/baseline scenario were excluded from the data.

 

 

 

 

 

 

 

0-4 420 20-40 40-200 >200
Drainage Area (ha)

 

 

 

Figure 4.2: The Bad Creek sub-watershed -cells within the ﬁlter strip
grouped into categories of ﬁlter efﬁciency and drainage area.

67

 

 

zoo- I75—100%
ISO-75%
1325—5091.

 

 

D 0-25%

 

 

 

0-4 4—20 20-40 40-200 >200
Drainage Area (ha)

 

 

 

Figure 4.3: The Lost Creek sub watershed -cells within the ﬁlter strip
grouped into categories of ﬁlter efﬁciency and drainage area.

 

 

 

 

 

 

 

100°— I75-100%
9°°‘ .50-75%
800- Eras-50%
roo— bro-25%
N beret 60°—
ug'errs 5°°‘
m-
an-
m—l
‘°°‘ n Fl 11
0‘ 04 420 ' 2040 ' 40200 I >200
DrainageArea(ha)

 

 

 

Figure 4.4: The Muskrat Creek sub watershed —cells within the ﬁlter strip
grouped into categories of ﬁlter efﬁciency and drainage area.

68

 

 

 

 

 

roo-
l75-100%

500‘ .50-75%
ass-50%

500‘ CID-25%

Number or 40°“
curs
300~
2004

 

 

 

 

 

 

Figure 4.5: The Spaulding Drain sub watershed -cells within the ﬁlter strip
grouped into categories of ﬁlter efﬁciency and drainage area.

As shown in the ﬁgures above the ﬁlter efﬁciency from the four primary
sub—watersheds are closely related to drainage area. The cells within the ﬁlter
strip are more efﬁcient at lower drainage areas (0-4 ha or 0-10 ac) but aren’t
necessarily ﬁltering much sediment because the draining areas are so small. An
analysis of variance (ANOVA) was conducted on the ﬁlter strip data of the four
primary sub-watersheds to test whether ﬁlter efﬁciency is directly related to
drainage area. In each of the data sets the ANOVA indicated that ﬁlter efﬁciency
is signiﬁcantly related to drainage area (p < 0.01) (Tables 4.2-4.5 and Figures
4.6- 4.9).

69

Table 4.2: ANOVA results for Bad Creek, ﬁlter strip efﬁciency verses
drainage area.

Anova: Single Factor

 

 

 

 

 

SUMMARY
Average
Groups (Drainage Filter strip
Area in h_a) Count Sum eﬁiciency Van‘ance
0-4 1755 107691.06 61.36 970.91
4-20 181 2662.18 14.71 56.25
2040 23 240.14 10.44 42.59
40-200 24 148.99 6.21 30.69
>gro 1 14.08 14.08-
ANOVA
Source of Variation SS df MS F P-value F M?
Between Groups 4680118 4 117002.95 135.03 4.2E-102 2.38
Wrthin Groups 1714743 1979 866.47
Total 2182755 1983

 

 

 

 

 

 

 

 

 

 

 

Figure 4.6: The Bad Creek average ﬁlter strip efﬁciency vs. drainage area of
the cells within the ﬁlter strip.

70

Table 4.3: ANOVA results for Lost Creek, ﬁlter strip efﬁciency verses
drainage area.

Anova: Single Factor

 

 

 

 

 

 

 

 

 

 

 

 

 

 

SUMMARY
AVG/399
Groups (Drainage ﬁlter strip
Area in ha) Count Sum efﬁciency Variance
0—4 536 31440.85 58.66 979.94
4-20 67 1122.88 16.76 155.67
20-40 5 39.93 7.99 19.37
40-200 10 22.93 2.29 0.95
>200 15 84.34 5.6_2 7.59
ANOVA
Source of Variation SS df MS F P-value F crit
Between Groups 1735597 4 43389.91 50.96 3.63E-37 2.39
Wrthin Groups 5347344 628 851.49
Total 708294 632
70.00 «
60.00 L—g
E 50.00 -
8.
11. 40.00 1
g 30.00 1
32 20.00 «
10m «
0.00 «
0.4 4-20 20-40 40-200 >200
Drainage Area (ha)

 

 

 

Figure 4.7: The Lost Creek average ﬁlter strip efﬁciency vs. drainage area
of the cells within the ﬁlter strip.

71

Table 4.4: ANOVA results for Muskrat Creek, ﬁlter strip efficiency verses
drainage area.

Anova: Single Factor

 

 

 

 

 

 

 

 

 

 

 

 

 

SUMMARY
Average
Groups (Drainage ﬁlter strip
Area in he) Count Sum efﬁciency Variance
0-4 2460 144923.21 58.91 1025.33
4-20 191 2440.84 12.78 101.41
20-40 95 887.72 9.34 31.05
40-200 66 507.48 7.69 32.65
>@ 2 4 2 0.67
ANOVA
Source of Variation SS df MS F P-value F crit
Between Groups 7166658 4 179166.46 197.71 1.5E-149 2.38
Wrthin Groups 2545592 2809 906.23
Total 3262257 2_813
70.00 -
60.00 -
g 50.00 -
E 40.00-
; .....-
at 20.00 —
10.00 —
0.00 <
0-4 4-20 2040 40200 >200
Drainage Area (ha)

 

 

 

Figure 4.8: The Muskrat Creek average ﬁlter strip efﬁciency vs. drainage
area of the cells within the ﬁlter strip.

72

Table 4.5: ANOVA results for Spaulding Drain, ﬁlter strip efﬁciency verses
drainage area.

Anova: Single Factor

 

 

 

 

 

SUMMARY
Average
Groups (Drainage ﬁlter strip
Area in h_a) Count Sum efﬁciency Variance
0-4 1621 97909.99 60.40 913.82
4-20 288 7410.46 25.73 422.38
2040 77 925.24 12.02 54.67
40-200 52 604.46 1 1 .62 34.65
>2_00 4 23.35 5.84 35.64
ANOVA
Source of Variation SS df MS F P-value F crit
Between Groups 5293079 4 132326.97 167.67 3.2E-124 2.38
Wrthin Groups 1607636 2037 789.22
M 2136943 2941

 

 

 

 

 

 

 

 

 

04 4-20 204) 402m >200
Dralnaga Area (ha)

 

 

 

Figure 4.9: The Spaulding Drain average ﬁlter strip efﬁciency vs. drainage
area of the cells within the ﬁlter strip.

73

4.2 Prioritizing Vegetative Fitter Strips
4.2.1 Overview

Vegetative ﬁlter strips along stream segments were analyzed throughout
the 722 ha (1,784 ac) secondary sub-watershed 20 or also known as the East
Bad Creek (EBC) watershed. The EBC watershed was divided into 8,107 cells at
a resolution of 30-meters. The ﬁlter strip throughout the EBC watershed
consisted of 680 cells, which was divided into 500 m (1640 ft) lengths on each
side of the stream (Figure 4.10). A detailed analysis of the tons of sediment
ﬁltered as well as average ﬁlter efﬁciency for each 500 m ﬁlter strip segment was

assessed to identify a method of ﬁlter strip placement.

 

Figure 4.10: The EBC watershed ﬁlter strip segments of equal length (500
m), numbered from 1-31 corresponding to a ranking of ﬁlter strip
effectiveness (ﬁltered sediment in tons) with 1 the most effective and 31 the
least effective.

74

4.2.2 Filter Strip Segment Results

The sediment reduction of each ﬁlter strip segments varied from 2 to 6.4
tons, with a mean of 3.7 tons (Table 4.6). The average efficiency of each ﬁlter
strip segment also ﬂuctuated throughout the watershed, from 25 to 78%, with a
mean of 56% (Table 4.6). The ﬁlter strip segments were ranked based on the
amount of total sediment each segment ﬁltered, where one is the most effective

and 31 the least (Figure 4.10 and Table 4.6).

75

Table 4.6: Thirty-one ﬁlter strip segments ranked by the amount of
sediment ﬁltered within the EBC watershed.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Average
Total Amount Percent of Total
Ranking of Sediment Sediment Drainage
Number Filtered (tons) Filtered (%) Area (ha)
1 6.4 62 8.5
2 5.5 57 32.0
3 4.9 65 36.0
4 4.9 69 21.5
5 4.8 29 34.4
6 4.5 63 19.8
7 4.3 78 7.3
8 4.3 60 9.7
9 4.2 70 6.5
10 4.2 38 29.1
11 4.1 49 14.2
12 4.0 69 6.9
13 3.8 40 37.7
14 3.8 58 44.5
15 3.6 69 4.5
16 3.5 69 3.6
17 3.5 60 7.7
18 3.2 40 13.0
19 3.1 59 6.9
20 3.1 58 7.3
21 3.1 51 46.2
22 3.1 73 10.5
23 3.0 25 14.2
24 3.0 73 10.5
25 3.0 71 4.5
26 2.9 40 14.2
27 2.9 52 22.7
28 2.4 51 8.5
29 2.2 63 4.5
30 2.2 29 5.7
31 2.0 36 6.1

 

 

 

 

The average percent of sediment ﬁltered in Table 4.6 identiﬁes the most
efﬁcient ﬁlter strip segments. However, the averaging is inaccurately
representing the actual efﬁciency because within most ﬁlter strip segments the

majority of the ﬁlter strip drains small areas (04 ha), and in most cases the total

76

amount of sediment entering the stream from the smaller drainage areas was
ﬁltered, i.e., >50% of sediment ﬁltered. Within certain stream segments there
may be one or two distinct areas within the ﬁlter strip segment where the ﬁlter
efﬁciency was very low, i.e., less than 25%, with a large corresponding drainage
area indicating concentrated ﬂow. By averaging the efficiency throughout the
ﬁlter strip segment, shallow, uniform ﬂow areas that make up the majority of the
segment was diluting its inefﬁciencies.

For example, ﬁlter strip segment number 14 ﬁltered 58% with a total
drainage area of 44.5 ha (110 ac). In this particular segment, one cell ﬁltered
39.2 ha (97 ac) and the rest of the drainage area was distributed evenly across
the remaining cells within the segment. The efﬁciency of this cell was 5%, which
did not signiﬁcantly lower the overall ﬁlter efﬁciency for the ﬁlter strip segment
because 19 other cell’s ﬁlter efﬁciency was higher than 30%. This analysis
indicates the importance of assessing ﬁlter strip efﬁciency in each cell within the
ﬁlter strip segment instead of using the segment’s average.

To determine why ﬁlter strips within the EBC watershed ﬁltered better than
others the drainage area and slope were analyzed. First, an ANOVA was
preformed to assess the inﬂuence of ﬁlter efﬁciency and drainage area in the
EBC watershed (Table 4.7 and Figure 4.11). This analysis was comparable to
the 27 sub-watershed ANOVA results throughout Stony Creek, where the ﬁlter
efﬁciency is signiﬁcantly inﬂuenced (p< 0.01) by drainage area. Therefore
indicating as the drainage area of the ﬁlter strip increased, suggesting

concentrated ﬂow, ﬁlter strip efﬁciency decreases.

77

Table 4.7: ANOVA results for the EBC watershed, ﬁlter strip efﬁciency
verses drainage area.

Anova: Single Factor

 

 

 

 

 

SUMMARY
Average of
Groups (Drainage ﬁlter strip
Area in he) Count Sum efﬁciency Variance
0-4 615 36895.73 59.99 951.58
4—20 58 807.80 13.93 50.47
20-40 6 40.49 6.75 33.20
40—200 2 35.81 17.91 41.68
>200 1 0 0 -
ANOVA
Source of Variation SS df MS F P-value F crit
Between Groups 1328147 4 33203.67 38.27 6.81 E-29 2.39
Wrthin Groups 5873536 677 867.58
Total 72_0168.3 681

 

 

 

50.00 .

 

  

S
8

 

 

16 Sedlment Flltered
8
8

N
.O
8

 

 

5
8

 

0-4 4-20 20-40 40-200 >200

 

 

Drainage Area (ha)

 

Figure 4.11: The EBC watershed average ﬁlter strip efﬁciency verses
drainage area of the cells within the ﬁlter strip.

78

The slope of the EBC watershed is an insigniﬁcant parameter to analyze
for ﬁlter strip efﬁciency because the watershed is very ﬂat (Figure 4.12). The
average slope across the watershed is 0.82% with an overall elevation change of

6 m or 20 ft (244 m to 230 m).

   
   

n
b 236
{:1

—Elevation In meters
above sea level.

a Drainage area is
reater than 4
(~10 acres)

Figure 4.12: Topography and drainage areas greater than 4 ha (~10 acres)
within the ﬁlter strip of the EBC watershed.

79

The ﬁlter strip segment results show the importance of identifying the
efﬁciency in site-speciﬁc areas within the ﬁlter strip, especially the areas where
concentrated ﬂow exists. The corresponding 18 ﬁlter strip segments that have at
least one or more cells that drain greater than 4 ha (~10 acres) are segments: 2,
3. 4, 5, 6, 7, 10, 11, 13, 14, 18, 21, 22, 23, 24, 26, 27, and 30. The effects of
concentrated ﬂow are detrimental to ﬁlter strip efﬁciency. Individual cells within
the ﬁlter strips that have drainage areas greater than 4 ha generally were
inefﬁcient. The 24 cells that drained greater than 4 ha, deﬁning concentrated
ﬂow areas, are highlighted in Figure 4.12. The total amount of sediment
produced in the 24 areas from the no ﬁlter strip/baseline scenario resulted in 209
tons, which is 47% of the total amount of sediment entering the stream. The total
amount of ﬁltered sediment from installing the ﬁlter strips in the concentrated ﬂow
areas resulted in a reduction of 25 tons. The average ﬁlter strip efﬁciency in
these areas was only 15%.

The ﬁlter strips are very efﬁcient in the remaining areas throughout the
watershed where uniform ﬂow occurs (drainage area is < 4 ha). The average
efﬁciency is 63%. However, the total sediment delivered to the stream in these
areas is 231 tons. By incorporating a ﬁlter strip in these areas the sediment load
was reduced by 93.4 tons.

This analysis indicates that the majority of the sediment delivered to the
stream originates from the areas of concentrated ﬂow. When the drainage area
is greater than 4 ha excessive sedimentation exists as show by the 47%

sediment load. Filter strips are not effective in areas of concentrated ﬂow, e.g.,

80

15% average efﬁciency, therefore it is imperative to identify areas of
concentrated ﬂow and model the effects of reducing sediment by installing other
conservation practices, such as sedimentation basins, reconstructed wetlands, or
grass waterways. Vegetative ﬁlter strips should then be analyzed in areas where
shallow, uniform ﬂow occurs.

The remaining ﬁlter strip segments that do not have any areas of
concentrated ﬂow should then be analyzed. For example, removing the 24
concentrated ﬂow areas from the 17 ﬁlter strip segments would create an entirely
different ranking of the 31 segments as shown in Table 4.8. Filter strip segment
1 did not contain any concentrated ﬂow areas and also ﬁltered the most amount
of sediment therefore its rank remains the same. This scenario recognizes that
ﬁlter strips should be prioritized based on the amount of sediment each segment

ﬁlters but the concentrated ﬂow areas must separately be evaluated.

81

Table 4.8 Ranking order from 1 to 31 of vegetative ﬁlter strip effectiveness
without the concentrated ﬂow areas.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Filter sediment Filtered Sediment
Rank without First Total Amount in Concentrated without the
Concentrated Ranking of Sediment Flow Areas Concentrated Flow
Flow Areas Order Filtered (tons) (tons) Areas (tons)
1 1 6.4 0.0 6.4
2 4 4.9 0.2 4.7
3 8 4.3 0.0 4.3
4 9 4.2 0.0 4.2
5 12 4.0 0.0 4.0
6 15 3.6 0.0 3.6
7 7 4.3 0.8 3.5
8 16 3.5 0.0 3.5
9 17 3.5 0.0 3.5
10 6 4.5 1.1 3.4
11 2 5.5 2.3 3.2
12 19 3.1 0.0 3.1
13 20 3.1 0.0 3.1
14 25 3.0 0.0 3.0
15 24 3.0 0.1 2.9
16 5 4.8 1.9 2.9
17 27 2.9 0.2 2.7
18 11 4.1 1.5 2.6
19 22 3.1 0.5 2.6
20 21 3.1 0.5 2.6
21 10 4.2 1.7 2.6
22 3 4.9 2.5 2.5
23 28 2.4 0.0 2.4
24 14 3.8 1.5 2.3
25 29 2.2 0.0 2.2
26 31 2.0 0.0 2.0
27 30 2.2 0.3 1.9
28 26 2.9 1.1 1.8
29 13 3.8 2.5 1.3
30 18 3.2 2.0 1.3
l 31 23 3.0 2.1 1.0

 

 

 

 

 

 

82

4.3 Uncertainty in the AGNPS Model Due to DEM Error
4.3.1 Overview

Wsually comparing the 30 m USGS DEM and the Survey DEM of the MSU
Farm watershed suggest that there is a substantial amount of terrain information
missing from the 30 m USGS DEM (Figure 4.13). For example, shown in the
Survey DEM the roads, railroads, and hills are obvious, whereas in the 30 to
USGS DEM these areas are generalized and do not emerge from the image
(Figure 4.13). The error map measures the errors associated to the terrain
variation from the 30 m USGS DEM and the Survey DEM. The negative error
values are the lower areas (roads or railroad tracks) and positive error values are
the higher areas or hills. The elevation variability between the two datasets has
a range from -8.5 to 3.9 m (-28 to 13 ft), with a mean difference of -0.6 m (-2 ft)
and a standard deviation of 0.93 m (3 ft). Therefore, on average the 30 m USGS

DEM underestimates the Survey DEM by 0.6 m (2 ft) (Figure 4.14).

Elevation
Above Sea
Level (m)

— 258 . 258
m 258-281

261 - 283
263 - 206
266 - 208

   

Figure 4.13: 30 m USGS DEM (Left) and 10 m Survey DEM (Right) of MSU
Farm watershed.

83

Error in
Meters

~73 - -O.3
-0.3 - -5

 

 

Figure 4.14: Error Map —difference of USGS DEM and Survey DEM of MSU
Farm watershed (white=no error, red=positive error, blue=negative error)

4.3.2 Local Effects

Figure 4.14 depicts areas in the 30 m USGS DEM where high elevations
(273-278 m or 896-912 ft) are underestimated, whereas lower elevations (261-
256 m or 856-840 ft) are overestimated. This bias is standard when comparing
coarse elevation data to higher accuracy sources (Gooverts, 1997).
Unfortunately, such bias becomes a serious problem when trying to detect
patterns of extreme elevation values.

An excellent example of local uncertainty is identifying stream location.
Stream location is an important aspect of hydrologic modeling especially for
identifying areas of excessive NPS pollution, i.e., major sediment delivery areas.

Each stream was derived from the nine DEM realizations, which was developed

84

to help explain local uncertainty in the 30 m USGS DEM for the study area. The
stream derivations show areas where the stream varies a lot in low or level areas
of the watershed especially at the watershed outlet, and in steeper more deﬁned
areas upstream where there isn’t as much variation (Figure 4.15). The local
uncertainty can become a problem when identifying critical sediment delivery
areas as well as designing/modeling conservation practices to aid in their

reduction.

Steeper
more
deﬁned
areas

    

Low F lat

Figure 4.15: The streams derived from the nine DEM realizations are
showing locations of local uncertainty in the 30 m USGS DEM.

In distributed parameter models there is a local uncertainty for every
output in every cell representing the watershed due to DEM error. The local

uncertainty is very difﬁcult to quantify for every cell. The effect of local

85

uncertainty propagates to the outlet; therefore, this effect can be measured at the

outlet, which produces a global uncertainty measurement.

4.3.3 Global Effects

The overall uncertainty in the coarse DEM causes problems when using
the AGNPS model or any NPS pollution model to assess sediment yield at the
outlet. Comparing the 30 m USGS DEM and the nine 10 m DEM realizations
used in the AGNPS model to estimate sediment load resulted in an average
reduction of 37% with a standard deviation of 118 (Table 4.9). The standard
deviation identiﬁed that randomly changing the slope-associated parameters had
a large effect in simulating sediment yield at the outlet when using the AGNPS
model. The 37% reduction suggests that the 30 m USGS DEM is profoundly
overestimating the sediment yield.

The decrease in sediment load was logically consistent with the ﬁndings of
Wolock and Price (1994). They reported that as grid cell size decreased, the.
depth to water table increased. The denser elevations produced detailed slope
attributes thus, identifying more areas where sediment is deposited upstream,

which accounts for the decrease in sediment load for each simulation.

86

Table 4.9: Sedimentation results at the outlet from the nine simulations.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

DEMs used in the Sediment Load at the % Difference From the
AGNPS Model Watershed Outlet Original 30 m Data
30 m 3358
DEM Realization 1 1926 43
DEM Realization 2 2059 39
DEM Realization 3 2046 39
DEM Realization 4 2151 36
DEM Realization 5 2122 37
DEM Realization 6 2125 37
DEM Realization 7 2078 38
DEM Realization 8 2211 34
DEM Realization 9 2351 30
Mean 21 19 37
SD 1 18

 

 

4.3.4 Resampled USGS DEM Results

The resampled USGS DEM was used to measure the effect of changing
cell size from 30 m to 10 m when predicting sediment load in the AGNPS model.
As the slope-associated parameters derived from the elevation data remain the
same the 10 m cell size redeﬁnes the soil and land use/cover characteristics and
better represents the actual size of the watershed. This analysis has shown that
the cell resolution greatly reduces the sediment load by 41% (Table 4.10). A z-
score test was used to compare the sediment yield from the resampled USGS
DEM and the DEM realizations, which identiﬁed that they were signiﬁcantly
different at an alpha of 0.01. The analysis suggests that incorporating the effects
of uncertainty into the resampled DEM, e.g., DEM realizations, increases
estimates of sediment yield. This indicates that 30 m DEM error is affecting the

actual estimates of sediment yield. However, the results for sediment yield are

87

not actual tons because of the cell size precision problems described in the

materials and methods section.

Table 4.10: Sedimentation results at the outlet from the resampled DEM.

 

DEMs used in the Sediment Load at the % Difference From the
AGNPS Model Watershed Outlet Original 30 m Data

30 m 3358
Resampled DEM 1993 41

 

 

 

 

 

 

 

4.3.5 Cell Size Effects in the AGNPS Model

The change in cell size, original 30 m DEM cell size to 10 m resampled or
interpolated DEM, not only affects the representation of the soil and land
use/cover characteristics within the watershed, but it redeﬁnes the actual size of
the watershed. When larger cell size is used the estimates of sediment yield
may be larger because not all the cells are entirely within the actual boundary of
the watershed (Figure 4.16). The cells that lie on the boundary overestimate the
size of the watershed, which increases the sediment load. When smaller cell
size is used, the watershed is closer to its actual size, which decreases the
sediment load at the outlet and represents a closer actual estimate of sediment

load for the watershed.

88

 

Figure 4.16: 30 m grid cell layout (Left) and 10 m grid cell layout (Right) for
the study area watershed.

The coarser grid layout (30 m cell size) estimates the size of the
watershed at 116 ha (287 acres), while the higher density 10 m cell size
estimates the area of the watershed at 110 ha (271 acres). This effect is a major
concern because in addition to all of the other cell size effects (redeﬁning the
land use and soils) it too enhances the uncertainty of the sediment load not only

at the outlet but also throughout the entire watershed.

89

Chapter 5

5 Conclusions

5. 1 General Conclusions
5.1.1 Prioritizing Vegetative Filter Strips

The AGNPS model combined with the AVNPSM GIS Interface allows
evaluation of multiple watershed scenarios of high resolution to be analyzed for
BMP placement. This research has shown that prioritizing vegetative ﬁlter strips
within agricultural watersheds primarily should be based on ﬁlter strip efﬁciency
and effectiveness. The size of the upland contributing area and the ﬂow path
through the ﬁlter strip is directly related to ﬁlter efﬁciency (% sediment ﬁltered)
and thus ﬁlter strip effectiveness (tons of ﬁltered sediment). The model results
identiﬁed vegetative ﬁlter strip inefﬁciencies (<25%) when drainage area was
greater than 4 ha (10 ac). Filter strips were inefﬁcient in these areas because the
increase in drainage area simulated where concentrated ﬂow paths would occur.

The results of this study were similar to vegetative ﬁlter strip ﬁeld research
conducted by Dillaha et al. (1989) and Robinson et al. (1996), in which the size of
the upland contributing area affected ﬁlter strip performance. Therefore, it is
imperative to assess the effects of vegetative ﬁlter strips in areas where shallow,
uniform ﬂow occurs. The main problem with this analysis is that areas of
shallow, uniform ﬂow are usually not critical delivery areas of concern, i.e.,
excessive NPS pollutant loading areas.

Vegetative ﬁlter strips alone were found to be relatively inefﬁcient

especially in areas where concentrated ﬂow occurs. For these potential risk

90

areas, other conservation practices, such as grassed waterways, reconstructed
wetlands, or sedimentation basins, should be considered. Evaluating other
conservation practices in the AGNPS model will help to reduce the pollutant load
in the most efﬁcient way possible. The results at the watershed outlet may then
show a higher reduction in sediment than if ﬁlter strips alone are the sole
conservation practice considered.

The AGNPS model in conjunction with the AVNPSM interface proved to
be an efﬁcient, user-friendly tool for evaluating conservation efforts on a
watershed scale. The model in conjunction with prioritizing potential risk areas
should be incorporated into the conservation planning process when allocating
federal funds from cost-sharing programs. This type of watershed scale analysis
establishes the foundation for conservation decisions in critical areas subjected

to NPS sediment delivery.

5.1.2 AGNPS Uncertainty Due to DEM Error

The unconditional stochastic simulation technique does not ensure an
actual “real” elevation map of the study area but identiﬁes many equal probable
elevation maps within which we can state where the true elevation map may lie
(Ehlschlaeger and Shortridge, 1996). As shown in this study, by using higher
accuracy elevation data, of a small watershed with little terrain variation, the
AGNPS sedimentation estimates are dramatically decreased (37%) from the
estimates when using the coarse elevation data. Cell size effects were also

evaluated to identify the difference in sediment estimates when only changing

91

cell size from 30 m to 10 m, all other parameters stayed constant. The difference
of sediment yield at the outlet of the watershed from changing cell size resulted
in a 41% reduction. The percent reduction from changing the cell size (41%) and
from the DEM uncertainty (37%) were compared, which identiﬁed that, they were
statistically signiﬁcant. Although, this data states that by using a smaller grid cell
will allow a closer estimate of the actual sedimentation results by better
representing the watershed’s size and the soil and land use/cover characteristics
it is equally important to state the signiﬁcance of the DEM uncertainty.

In the present analysis, the overall difference due to 30 m USGS DEM
error and cell size effects was signiﬁcant and could become a detrimental
problem when using the results from the AGNPS model for identifying areas of
BMP placement within watersheds. Therefore, pinpointing potential risk areas in
terms of sediment load do require increase accuracy in elevation data but in
particular further analysis needs to be explored on the effects of changing cell

size.

5.2 Future Considerations

Future work should include the use of water quality models to ﬁrst identify
the critical delivery areas within the watershed, prioritize them by the amount of
sediment that is delivered into the stream, and then evaluate which best
management practice would be most appropriate. The technique for prioritizing
critical sediment delivery areas can be incorporated into the AGNPS AVNPSM

GIS interface by writing an ArcView Avenue script. The conservation practice

92

costs and soil rental rates can also be incorporated into the AVNPSM interface to
evaluate the economics of installing effective conservation practices.

Accurate elevation data is an important variable when using water quality
models. For example, the DEM analysis suggested an increased uncertainty in
the results generated by the AGNPS model due to the error in the 30 m USGS
DEM. Although the error was small it was signiﬁcant. Using actual higher
accuracy or ground truth elevation data for the study area to simulate sediment
yield in the AGNPS model should test the unconditional stochastic simulation
approach that was used in this study. The sedimentation results from using the
two elevation datasets in AGNPS should be compared as well as the simulated
error and the actual errors in the 30 m USGS DEM.

The grid cell size effects contributed signiﬁcantly in the uncertainty of
estimating sediment at the outlet of a watershed. These effects require an in-
depth evaluation to characterize the reasons why estimates of sedimentation at
the outlet and throughout the watershed differ. This in-depth analysis will help
identify which of the model limitations (accuracy in watershed size or soil/land
use parameters) is the greatest source of error in the ﬁnal estimate.

In addition, the AGNPS model must be very precise when characterizing
the cell size. The model characterizes grid cell size in acres, but only allows for
two decimal places. This becomes a problem when using a small cell size such
as 10 m. Over a large number of cells the acreage of the watershed will be very
different than the true size of the watershed if the cell size is rounded down or up.

The increase or decrease in the “true” cell size will increase the error throughout

93

the watershed and propagate to the outlet results. This is one of the major
limitations of the AGNPS model when using a small cell size. This precision

problem must be analyzed further.

94

References

Abbott, M.B., J.C. Bathurst, J.A. Cunge, P.E. O’Connell, and J. Rasmussen.
1986a. An introduction to the European Hydrological System - Systems
Hydrologique Europeen “SHE” 1: History and philosophy of a physically
based distributed modeling system. Journal of Hydrology 87: 45—59.

Abbott, M.B., J.C. Bathurst, J.A. Cunge, P.E. O’Connell, and J. Rasmussen.
1986b. An introduction to the European Hydrological System — Systems
Hydrologique Europeen “SHE” 2: Structure of a physically based
distributed modeling system. Journal of Hydrology 87: 61 -77.

Ahuja, L.R., K.W. Rojas, J.D. Hanson, M.J. Shaffer, and L. Ma (editors), 1999.
Root Zone Water Quality Model: Modeling Management Effects on Water
Quality and Cmp Production. Water Resources Publications, LLC,
Highland Ranch, CO.

Arnold, J.G., J. R. Williams, R. Srinnivasan, and K W. King. 1996. SWAT: Soil
and Water Assessment Tool. User Manual. USDA and Texas A&M
University, Temple, TX.

Arora, K, SK, Mickelson, J.L. Baker, D.P. Tierney, and C.J. Perters, 1996.
Herbicide retention by vegetative buffer strips from runoff under natural
rainfall. Transactions of the ASAE 39(6): 2155-2162.

Bagnold, R. A. 1966. An approach to the sediment transport problem from
general physics. Prof. Paper 422-J. U.S. Geologic Survey, Reston, VA.

Bailey, T. and Gatrell, A. 1995. Interactive Spatial Data Analysis. England:
Pearson Education Limited.

Beasley, D. B., L. F. Huggins and E. H. Monke. 1980. ANSWERS: A model for
watershed planning. Transactions of the ASAE 23(4): 938-944.

Bhuyan, S. J., K R. Mankin, J. K Koeliker, L. Marzen and J. A. Harrington. 2001.
Effect of Cell Size on AGNPS Predictions. ASAE Paper No. 01-2002, St.
Joseph, MI: ASAE

Bicknell, B.R., J.C. lmhoff, J.L. Kittie, Jr., A.S. Donigian, Jr., and RC. Johanson,
1993. Hydrologic Simulation Program - FORTRAN (H SPF) User's Manual
for Release 10. Report No., EPA/600lR-93l174. US EPA, Washington
DC.

Bingner, R. L. 1990. Comparison of the components of used in several sediment
yield models. Transactions of the ASAE 33(4): 1990.

95

Bingner, R.L. and F .D. Theurer, 2001. AnnAGNPS Technical Processes:
Documentation Version 2. URL:
http:/Mww.sedlab.olemlss.edulAGNPS.html.

Borah, D.K 2002a. DWSM-A dynamic watershed simulation model.
Mathematical Models of Small Watershed Hydrology and Applications.
Water Resources Publications, LLC, Highlands Ranch, CO. pp 113-166.

Borah, D. K 2002b. Watershed scale nonpoint source pollution models:
Mathematical Bases. ASAE Paper No. 022091, St. Joseph, MI: ASAE.

Brannan, K and J. M. Hamlett. 1998. Using geostatistics to select grid-cell
layouts for the AGNPS model. Transactions of the ASAE 41 (4): 1011-
1018.

Chaubey, l., D. R. Edwards, T. C. Daniels, P. A. Moore Jr., and D. J. Nichols,
1994. Effectiveness of vegetative ﬁlter stips in retaining surface-applied
swine manure constituents. Transactions of the ASAE 37(3): 845-850.

Chung, S. 0., A. D. Ward, and C. W. Schaulk. 1992. Evaluation of the Hydrologic
Component of the ADAPT Water Table Management Model. Transactions
of the ASAE 35(2):571-579.

Corwin, D. L. and P. J. Vaughan. 1997. Modeling nonpoint source pollutants in
the vadose zone with GIS. Environmental Science and Technology
31 (8)22157-2175.

Cronshey, R. G., Theurer, F. D., and Glenn, R. L., 1993. GIS - Water Quality
Computer Model Interface: A Prototype. NCGIA Second International
Conference/lrI/orkshop on Integrating Geographic Information Systems
and Environmental Modeling. Breckenridge, Colorado USA.

Dabney, S. M., K C. McGregor, L. D. Meyer, E. H. Grissinger, and G. R. Foster.
1993. Vegetative barriers for runoff and sediment control. Transactions of
the ASAE 38: 809-815.

Daniels, and J. W. Gilliam. 1996. Sediment and chemical load reduction by grass
and riparian ﬁlters. Soil Science Society American Journal 60: 246-251.

Dickey, E. C., and D. H. Vanderholm. 1981. Vegetative ﬁlter strip treatment of
livestock feedlot runoff. Journal of Environmental Quality 10(3): 279-284.

Dillaha, T. A., R. B. Reneau, S. Mostaghimi, and D. Lee. 1989. Vegetative ﬁlter
strips for agricultural nonpoint source pollution control. Transactions of the
ASAE 32(2): 513-519.

Edwards, D. R., P. A. Moore, Jr., T.C. Daniel, P. Srivastava, and D. J., Nichols,
1997. Vegetative ﬁlter strips removal of metals in runoff from poultry litter-
amended fescue grass plots. Transactions of the ASAE 40(1): 121 -127.

Edwards, W. M., L. B. Owens, and R. K White, 1983. Managing runoff from a
small, paved beef feedlot. Journal of Environmental Quality 12(2): 281 -
286.

Ehlschlaeger, C. R. and A. Shortridge. 1996. Modeling Elevation Uncertainty in
Geographical Analyses. Proceedings of the International Symposium on
Spatial Data Handling, Delft, Netherlands. 98. 15-9B.25.

Engel, B. R., Srinivasan, R., and Reverts, C., 1993. A Spatial Decision Support
System for Modeling and Managing Agricultural Non-Point—Source
Pollution. in: Goodchild, M. F., Parks, B. 0., and Steyaert, L. T.
Geographic Information Systems and Environmental Modeling, New York:
Oxford. pp. 231-237.

Environmental Protection Agency, 2002b. Water Quality Conditions in the United
States A Proﬁle from the 2000 National Water Quality Inventory. EPA-
841-F-02-003. US. EPA, Washington DC.

Environmental Protection Agency, 2002a. Nonpoint Source Pollution: The
Nation’s Largest Water Quality Problem. EPA-841-F-96—004A. URL:
http:llwww.epa.gov/owowlnpslfacts/point1 .htm.

Farm Service Agency. 1999a. Conservation Reserve Program. USDA Farm
Service Agency Online Fact Sheet. URL:
http:llwww.fsa.usda.gov/paslpublicationslfactslhtml/crpmiOO.htm.

F arm Service Agency. 19990. Environmental Beneﬁts Index. USDA Farm
Service Agency Online Fact Sheet. URL:
wvwv.fsa.usda.govlpaslpublicationslfactslebiold.pdf

Farm Service Agency. 2000. Conservation Reserve Enhancement Program
(CREP) Questions and Answers. USDA Farm Service Agency Online.
URL: http:l/www.fsa.usda.govldafplcepdlcrep.htm.

Farm Service Agency, 2002. Veneman announces payments of nearly $1.6
billion for conservation reserve program. USDA Ofﬁce of Communications
News Room 460-A. URL:
http:llvwvwfsa.usda.gov/paleullStory.asp?Storle=905

97

Fisher, P. F. 1999. In Longley, P. A., M. F. Goodchild, D. J. Magure, and D. W.
Rhind. 1999. GIS Second Edition. Wiley, NY. pp. 191-205.

Foster, G. R., L. J. Lane, J. D. Novvlin, J. M. Laﬂen, and R. A. Young. 1981.
Estimating erosion and sediment yield on ﬁeld-sized areas. Transactions
of the ASAE 24(5): 1253-1262.

Ghadiri, H., C. W. Rose, and W. L. Hogarth. 2001. The inﬂuence of grass and
porous barrier strips on runoff hydrology and sediment transport.
Transactions of the ASAE 44(2):259-268.

Gharabaghi, B., R. P. Rudra, H. R., Whiteley, and W. T. Dickinson, 2001.
Sediment-removal efﬁciency of vegetative ﬁlter strips. ASAE Paper No.
01-2071, St. Joseph, MI: ASAE.

Gooverts, P. 1997. Applied Geostatistics Series: Geostatistics for Natural
Resources Evaluation. NY: Oxford. pg 369-393.

He, 0., J. F. Riggs, and Y. -T. Kang. 1993. Integration of geographic information
systems and a computer model to evaluate impacts of agricultural runoff
on water quality. Water Resources Bulletin 29(6): 891 -900.

He, 0., C. Shi, C. Yang and B. P. Agosti. 2001. A windows-based GIS-AGNPS
interface. Journal of the American Water Resources Association 37(2): 1-
11.

Heuvelink, G. B. M. 1999. In Longley, P. A., M. F. Goodchild, D. J. Magure, and
D. W. Rhind. 1999. GIS Second Edition. Wiley, NY. pp. 207-218.

Huff, F. A. and J. R. Angel. 1992. Rainfall Frequency Atlas of the Midwest. Illinois
State Water Survey, Bulletin 71. Champaign, IL.

Huggins, L. F. and E.J. Monke. 1966. The mathematical solution of the hydrology
of small watersheds. Technical Report No. 1. Water Resources Research
Center. Purdue University, West Lafayette, IN. pg 130.

lnamdar, S. P. and T. A. Dillaha. 2000. Relationships between drainage area,
slope length, and slope gradient for riparian slopes in Virginia.
Transactions of the ASAE 43(4): 861 -866.

lnamdar, S. P., S. Mostaghimi, P. W. McClellan and K M. Brannan. 2001. BMP

impacts on sediment and nutrient yields from an agricultural watershed in
the coastal plain region. Transactions of the ASAE 44(5): 1191 -1200.

lsaaks, E. H., R. M. Srivastava. 1989. An Introduction to Applied Geostatistics.
New York: Oxford.

98

Jankowski, P. and Haddock, G., 1993. Integrated Nonpoint Source Pollution
Modeling System. NCGIA Second International Conference/Workshop on

Integrating Geographic Information Systems and Environmental Modeling.
Brackenridge, Colorado USA.

Jin, C. -X. and M. J. M. Romkens. 2001. Experimental studies of factors in

determining sediment trapping in vegetative ﬁlter strips. Transactions of
the ASAE 44(2): 277-288.

Klaghofer, E., Bimbaum W., and Summer, W., 1993. Linking Sediment and
Nutrient Export Moodels with a Geographic Information System. In:
H ydroGIS: Application of Geographic Information System in Hydrology
and Water Resources, Free, IAHS no. 211, pp. 501 -506.

Knisel, W. G., ed. 1980. CREAMS: A ﬁeld-scale model for chemicals, runoff and
erosion from agricultural management systems. Conservation Research
Report No. 26. USDA-SEA, Washington, DC.

Lane, L. J. 1982. Development of a procedure to estimate runoff and sediment
transport in ephemeral streams. Recent Developments in the Explanation
and Prediction of Erosion and Sediment Yield. Publication No. 137.

Lane, L.J. and MA. Nearing (eds), 1989. USDA -water erosion prediction
project: Hillslope proﬁle model documentation. NSERL Report No. 2,
USDA-ARS National Soil Erosion Research Laboratory, West Lafayette,
IN.

Lenzi, M. A. and M. D. Luzio. 1997. Surface runoff, soil erosion and water quality
modeling in the alpone watershed using AGNPS integrated with a
geographic information system. European Journal of Agronomy 6: 1-14.

Leavesley, G.H., R. W. Lichty, B. M. Troutman, and L. G. Saindon. 1983.
Precipitation-runoff modeling system — User’s Manual. US. Geological
Survey Water Resources Investigative Report 83-4238.

Leonard, R.A., W.G. Knisel, and DA Still, 1987. GLEAMS: Groundwater loading

effects on agricultural management systems. Transactions of the ASAE
30(5): 1403-1428.

Lin, Y. and T. Teng. 2000. Comparing Effects of Different Sizes of Aggregation
on Spatial Structure of Remotely Sensed Data. Asian Conference on
Remote Sensing. URL: www.9isdevelopment.net.

99

Livingston, W. H. and R. O. Hegg, 1981. Terraced pasture for disposal of diary
yard runoff. Proceedings 4‘” International Livestock Waste Symposium,
270-273. St. Joseph, MI: ASAE.

Lovejoy, S. 8., JG. Lee, T. 0. Randhir, and B. A. Engel. 1997. Research Needs
for Water Quality Management in the 21“ Century. Journal of Soil and
Water Conservation. January-February 1997: 18-21

Lowrance, R., S. Dabney, and R. Schultz, 2002. Improving water and soil quality
with conservation buffers. Journal of Soil and Water Conservation 57(2):
37-43.

Magette, W. L., R. B. Brinsﬁeld, R. E. Palmer, and J. D. Wood. 1989. Nutrient
and sediment removal by vegetative ﬁlter strips. Transactions of the ASAE
32(2): 663-667.

Meyer, L. D., S. M. Dabney, and W. C. Harmon. 1995. Sediment-trapping
effectiveness of stiff-grass hedges. Transactions of the ASAE 38(3): 809-
81 5.

Misra, A.K, J.L., Baker, S.K Mickelson, and H. Shang, 1996. Contributing area
and concentration effects on herbicide removal by vegetative buffer strips.
Transactions of the ASAE 39(6):2105-2111.

Mitchell, J. K, B. A. Engel, R. Srinnivasan, and S. S. Y. Wang. 1993. Validation
of AGNPS for small watersheds using and integrated AGNPS/GIS system.
Water Resources Bulletin 29(5): 833-842.

Munoz-Carpena, R., J. E. Parsons and J. W. Gilliam. 1993. Numerical approach
to the overland ﬂow process in vegetative ﬁlter strips. Transactions of the
ASAE 36(3): 761 -770.

Natural Resource Conservation Service. 1999a. Natural Resource Conservation
Service Conservation Practice Standard for Filter Strips Code 393. USDA-
NRCS Michigan Technical Guide, Section IV. USDA-NRCS, Washington
DC.

Natural Resource Conservation Service. 19990. Core4 Manual Conservation
Practices. USDA-NRCS, Washington, DC.

Natural Resource Conservation Service and Conservation Partners in Michigan.
2001. Stony Creek watershed plan receives funding for water quality
project. USDA-NRCS quarterly newsletter. URL:
http:l/www.mi.nrcs.usda.gov/News_and_lnformationlindex/February_New
sletter/february_newsletter_index.htm.

100

Natural Resource Conservation Service. 2001. Watershed plan and
environmental assessment for Stony Creek watershed: Clinton and Ionia
Counties, Michigan. USDA-NRCS: Lansing, MI.

Natural Resource Conservation Service. 2002a. Buffer Strips: Achieving our two
million mile goal. USDA-NRCS. URL: www.nrcs.usda.govlfeaturelbuffers.

Natural Resource Conservation Service. 2002b. Environmental Quality
Incentives Program. USDA-NRCS Farm Bill 2002 Fact Sheet. URL:
www.nrcs.usda.govlprogramslfarmbilllZOOZ/products.html.

Natural Resource Conservation Service. 2002c. Wetlands Reserve Program.
USDA-NRCS Farm Bill 2002 Fact Sheet. URL:
www.nrcs.usda.govlprogramsffarmbilII2002lproducts.html.

Natural Resource Conservation Service. 20020. Wildlife Habitat Incentives
Program. USDA-NRCS Farm Bill 2002 Fact Sheet. URL:
www.nrcs.usda.gov/programslfarmbiIll2002/products.html.

Natural Resource Conservation Service. 20029. Emergency Watershed
Protection. USDA-NRCS URL:
http:llvvww.nrcs.usda.govlprogramslewplewp.html.

Neibling, W. H. and E. E. Alberts. 1979. Composition and yield of soil particles
transported through sod strips. ASAE Paper No., 79-2065. St. Joseph, MI:
ASAE

Novotny, V. and H. Olem. 1994. Water Quality: Prevention, Identiﬁcation and
Management of Diffuse Pollution. New York: Van Nostrand Reinhold.

Ogden, FL and FY. Julien, 2002. CASCZD: A tvvo-dimensional, physically-
based, hortonian hydrologic model. Mathematical Models of Small
Watershed Hydrology and Applications. Water Resources Publications,
LLC, Highlands Ranch, CO. pp 69-112.

Park, S. W., S. Mostaghimi, R. A. Cooke and P. W. McClellan. 1994. BMP
impacts on watershed runoff, sediment and nutrient yields. Water
Resources Bulletin 30(6): 1011-1023.

Parsons, J.E., J.W. Gilliam, R. Munoz-Carpena, R.B. Daniels and TA. Dillaha.
1994. Nutrient and sediment removal by grass and riparian buffers.
Environmentally Sound Agriculture. Proc. of Second Conference. Univ.
FL, Gainesville, FL. pg. 147-154

Perrone, J. and C. A. Madramootoo. 1999. Sediment yield prediction using
AGN PS. Journal of Soil and Water Conservation 54(1): 415-419.

101

Pebesma, E. 1999. Gstat Users Manual. Department of Physical Geography,
Utrecht University, The Netherlands.

Raffaelle, J.B., K C. McGregor, G. R. Foster, and R. F. Cullum, 1997. Effect of
narrow grass strips on conservation reserve land converted to cropland.
Transactions of the ASAE 40(6): 1581 -1 587.

Robinson, C. A., M. Ghaffarzadeh, and R. M. Cruse. 1996. Vegetative ﬁlter strip
effects on sediment concentration in cropland runoff. Journal of Soil and
Water Conservation 50(3): 227-230.

Sievers, D. M., G. B. Garner, and E. E. Pickett. 1975. A lagoon-grass terrace
system to treat waste. Proc. 3'" International Livestock Waste Symposium.
ASAE Publication Proc-273. St. Joseph, MI: ASAE.

Skaggs, R.W. 1980. A Water Management Model for Artiﬁcially Drained Soils.
North Carolina Agricultural Research Service Technical Bulletin, No. 267.
Raleigh, NC: North Carolina Agricultural Research Service.

Smith, R. E., and J. R. Williams. 1980. Simulation of surface water hydrology. In
CREAMS, A Field Scale Model for Chemicals, Runoff, and Erosion from
Agricultural Management Systems. Conservation Reserve Report 26: 15.
USDA-ARS, Washington, DC.

Soil and Water Conservation Society. 2001. Realizing the promise of
conservation buffer technology. Soil and Water Conservation Society,
Iowa.

Soil Conservation Service. 1975, 1986. Urban hydrology for small watershed,
Technical Release No. 55. Soil Conservation Service, US. Department of
Agriculture, Washington, DC.

Soil Conservation Service. 1979. Engineering Field Manual. Soil Conservation
Service, US. Department of Agriculture, Washington, DC.

Srivastava, P., D. R. Edwards, T. C. Daniels, P. A. Moore Jr., and T. A. Costello.
1996. Performance of vegetative ﬁlter strips with varying pollutant source
and ﬁlter strip lengths. Transactions of the ASAE 39(6): 2231 -2239.

Srinivasan, R. and J. G. Arnold. 1994. Integration of a basin-scale water quality
model with GIS. Water Resources Bulletin 30(3): 453-462.

Srinivasan, R. and B. A Engel. 1991. A knowledge based approach to extract
input data from GIS. ASAE Paper No., 91 -7045. St. Joseph, MI: ASAE.

102

Tim, U. S. and R. Jolly. 1994. Evaluating agricultural nonpoint-source pollution
using integrated geographic information systems and hydrologic/water
quality model. Journal of Environmental Quality 23: 25-35.

USDA-ARS. 2002. KINEROSZ --A Kinematic Runoff and Erosion Model. USDA-
ARS, Southwest Watershed Research Center. URL:
www.tucson.ars.ag.gov/kineroslhome.html.

U.S. Geologic Survey. 2002a. USGS Digital Elevation Model Data. US.
Geological Survey. URL: edcwww.cr.usgs.govlglislhyperlguidelusgs_dem.

U.S. Geologic Survey. 2002b. Soil Survey Geographic (SSURGO) Data Base.
US. Geological Survey. URL: www.ftw.nrces.usda.govlssurgo.html

U.S. Geologic Survey. 2002c. Michigan 1992 NLCD Shapeﬁle by County
Metadata. Michigan Center for Geographic lnforrnation. URL:
www.michigan.govlcgi.

Wechsler, S. 2000. Digital Elevation Model (DEM) Uncertainty: Evaluation and
Effect on Topographic Parameters. URL:
www.csulb.edul~wechslerldissertatioan262/p262.html

Williams, J. R., A. D. Nicks and J. G. Arnold. 1985. SWRRB, A simulator for
water resources in rural basins. ASCE Hydraulics J. 111(6): 970-986.

Williams, J. R., P.T. Dyke and C. A. Jones. 1982. EPIC -A model for assessing
the effects of erosion on soil productivity. Proc. Third Int. Conf. On State-
of-the-Art in Ecological Modeling, Int. Soc. for Ecological Modeling.

Wilson, LG. 1967. Sediment Removal From Flood Water by Grass Filtration,
Transactions of the ASAE 10(1 )135-37

Wischmeier, W. H. and D. D. Smith. 1978. Predicting rainfall erosion losses —A
guide to conservation planning, Agricultural Handbook No. 537. USDA,
Washington, DC.

Wolock, D. M. and C. V. Price. 1994. Effects of digital elevation model map scale
and data resolution on a topography-based watershed model. Water
Resources Research 30(11): 3041 -3052.

Woolhiser, D.A., R.E. Smith, and DC. Goodrich, 1990. KINEROS, A kinematic

runoff and erosion model: Documentation and User Manual. ARS-77,
USDA-ARS, Fort Collins, CO.

103

Yoon, J., Padmanabhan, G., and Woodbury, L. H., 1993. Linking Agricultural
Nonpoint Source Pollution Model (AGNPS) to a Geographic lnfon'nation
System (GIS). In J. M. Harlin and K J. Lanfear (eds) Geographic
Information Systems and Water Resources, AWRA Proc., pp. 79-87.

Young, R. A., C. A. Onstad, D. D. Bosch, and W. P. Anderson. 1987. AGNPS,
Agricultural Nonpoint Source Pollution Model: A large watershed analysis
tool. Conservation Reserve Report 35. USDA-ARS, Washington DC. 77.

Young, R. A., C. A. Onstad, D. D. Bosch, and W. P. Anderson. 1989. AGNPS: A

nonpoint-source pollution model for evaluating agricultural watersheds.
Journal of Soil and Water Conservation 44: 168-173.

Young, R. A., C. A. Onstad, D. D. Bosch, and W. P. Anderson. 1994. Agricultural
Non-point Source Pollution Model, Version 4.03 AGNPS User’s Guide. St.
Paul, Minn. Pollution Control Agency and USDA-ARS: Washington, DC.

Young, R. A., T. Huntrods, W. Anderson. 1980. Effectiveness of Vegetated Buffer

Strips in Controlling Pollution from Feedlot Runoff. Journal of
Environmental Quality. (9) 32483-487.

104

APPENDICES

Appendix A: Gstat program

105

APPENDIX A

Gstat Program

106

#

# Unconditional Gaussian simulation on a mask

# (local neighborhoods, simple kriging)

#

# deﬁnes empty variable:

data(mserr): dummy, sk_mean=0.6, max=30;
variogram(mserr): 1.7672 Gau(53.8146)+6.84041 Exp(115.075);
mask: 'lc1 mask. asc’;

method: gs; # Gaussian simulation instead of kriging
predictions(mserr): 'lc1sim';

set nsim=10;

107

   

l11111111111111le(1(1le