EXPLORING RELATIONSHIPS BETWEEN IN-STREAM CONDITIONS AND
ECOLOGICAL HEALTH WHILE ASSESSING LANDUSE AND CLIMATE SCENARIOS
By
Matt Einheuser

A THESIS
Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of
MASTER OF SCIENCE
Crop and Soil Sciences
2011

ABSTRACT
EXPLORING RELATIONSHIPS BETWEEN IN-STREAM CONDITIONS AND
ECOLOGICAL HEALTH WHILE ASSESSING LANDUSE AND CLIMATE SCENARIOS
By
Matt Einheuser
Human disturbances can have significant impacts on physicochemical and biological
conditions of streams. A good understanding of the relationships among these factors will help
decision makers in sustainable management of the ecosystems. To address these issues, the
following research objectives were developed: 1) bridge the gap between hydrologic models and
ecological conditions using the Soil and Water Assessment Tool, 2) identify influential in-stream
variables to explain fish and macroinvertebrate measures, 3) compare fuzzy logic techniques
with statistical approaches to describe and model ecological health, 4) use in-stream variables
obtained from SWAT to predict the impacts of different landuse and climate scenarios, and
evaluate the effectiveness of best management practices, in regards to aquatic health. A high
resolution SWAT model was built for the Saginaw River basin of Michigan, and flow and water
quality outputs were linked with measured biological data. Results indicate that SWAT models
2

can be an effective tool to produce in-stream variables, explaining 21% to 57% of variation (R )
in ecological measures. Fuzzy logic methods are effective approach to model ecological health
and outperformed other statistical methods tested here. Average annual flow rate had the
strongest correlation with IBI, whereas nutrient concentrations showed the largest influence on
all other ecological measures. Results suggest that efforts to model historic baseline conditions
and to provide context for stream health assessments should include both pre-settlement land use
and climate conditions. Meanwhile, the conservation practice, native grass, showed the most
improvement to stream health, followed by residue management and no-tillage.

This thesis is dedicated to my family and friends for their much needed love and support.

iii

ACKNOWLEDGMENTS

I would like to thank everyone who has supported me and helped throughout my graduate
studies. I want to specifically thank my soon to be wife, Mary, for putting up with me and for all
her love, support, and advice throughout completion of this degree. I would like to thank my
parents, Dave and Denise, for giving me the opportunity and drive to receive my education and
continue on to graduate school. I would also like to thank my sister, Jess, for paving the way in
going to college and graduate school.
I would like to thank my major advisor and friend Dr. Pouyan Nejadhashemi for giving
me the opportunity, encouragement, and motivation to complete this research and thesis. I really
appreciate all the guidance and good laughs along the way. In addition, I would like to thank my
committee, Dr. Scott Sowa, Dr. Lizhu Wang, and Dr. Sasha Kravchenko for their invaluable help
and support throughout my research.
I would like to also thank all my friends and lab mates, Brad, Sean, Andy, Giri, Yaseen,
and Mike. Although, in many cases they more than likely slowed down my progress throughout
my research, they also made the time much more enjoyable and fun.
Finally, I would like to thank the United States Department of Agriculture, Natural
Resources Conservation Services for funding this research under the Great Lakes Northern
Forest Cooperative Ecosystems Studies Unit (CESU) Umbrella Agreement Number 68-3A75-886.

iv

TABLE OF CONTENTS

LIST OF TABLES ................................................................................................................... VIII
LIST OF FIGURES ..................................................................................................................... X
LIST OF ABBREVIATIONS .................................................................................................. XII
INTRODUCTION......................................................................................................................... 1
LITERATURE REVIEW ............................................................................................................ 4
1.1 OVERVIEW ........................................................................................................................... 4
1.2 STREAM / ECOLOGICAL HEALTH AND INTEGRITY ................................................................ 4
1.2.1 Indicators..................................................................................................................... 5
1.3 AGRICULTURAL INFLUENCES ON STREAMS ........................................................................ 14
1.3.1 Agriculture and Flow ................................................................................................ 15
1.3.2 Agriculture and Sediments ........................................................................................ 15
1.3.3 Agriculture and Nutrients ......................................................................................... 16
1.3.4 Agriculture and Pesticides ........................................................................................ 18
1.3.5 Agriculture and Bacteria ........................................................................................... 19
1.3.6 Agriculture and Water Temperature ......................................................................... 20
1.3.7 Agriculture and Dissolved Oxygen ........................................................................... 21
1.3.8 Agriculture, Physical Habitat and Stream Morphology ............................................ 21
1.4 STREAM CONDITIONS AND AQUATIC HEALTH ................................................................... 22
1.4.1 Flow and Aquatic Health .......................................................................................... 22
1.4.2 Sediments and Aquatic Health .................................................................................. 23
1.4.3 Nutrients and Aquatic Health.................................................................................... 25
1.4.4 Pesticides and Aquatic Health .................................................................................. 26
1.4.5 Bacteria and Aquatic Health ..................................................................................... 27
1.4.6 Water Temperature and Aquatic Health ................................................................... 27
1.4.7 Dissolved Oxygen and Aquatic Health ..................................................................... 29
1.4.8 Physical Habitat, Channel Morphology and Aquatic Health .................................... 30
1.4.9 Influences and Different Scales ................................................................................ 31
1.5 METHODS LINKING ENVIRONMENTAL STRESSORS AND ECOLOGICAL VARIABLES ............ 32
1.5.1 Linear Methods ......................................................................................................... 32
1.5.2 Non-Linear Methods ................................................................................................. 33
1.5.3 Soft-Computing Methods.......................................................................................... 33
1.6 CONSERVATION PRACTICES ............................................................................................... 35
1.6.1 Filter Strips................................................................................................................ 36
1.6.2 Terraces ..................................................................................................................... 36
1.6.3 Grassed Waterways ................................................................................................... 36
1.6.4 Constructed Wetlands ............................................................................................... 36
1.6.5 Conservation Tillage ................................................................................................. 37
1.6.6 Rotational Grazing .................................................................................................... 37
1.6.7 Contour Farming ....................................................................................................... 37
v

1.6.8 Strip Cropping........................................................................................................... 38
1.6.9 Conservation Reserve Program (Native Grasses) ..................................................... 38
1.6.10 Conservation Effects Assessment Project (CEAP) ............................................... 38
1.7 MODELING ......................................................................................................................... 40
INTRODUCTION TO METHODOLOGY AND RESULTS ................................................. 46
EXPLORING RELATIONSHIPS BETWEEN IN-STREAM CONDITIONS AND
ECOLOGICAL HEALTH WHILE ASSESSING LAND-USE AND CLIMATE
SCENARIOS...................................................................................................................... 48
1.8 ABSTRACT ...................................................................................................................... 48
1.9 INTRODUCTION ............................................................................................................ 49
1.9.1 Ecological Modeling ................................................................................................. 54
1.9.2 Motivation and Objectives ........................................................................................ 56
1.10 METHODOLOGY ....................................................................................................... 57
1.10.1 Study Area ............................................................................................................ 57
1.10.2 Biophysical Model ................................................................................................ 58
1.10.3 Water Quality Elements of SWAT ....................................................................... 60
1.10.4 Model Setup .......................................................................................................... 61
1.10.5 Sensitivity Analysis, Calibration, and Validation ................................................. 65
1.10.6 Fish Data ............................................................................................................... 67
1.10.7 Data Analysis ........................................................................................................ 73
1.10.8 Predictions............................................................................................................. 77
1.11 RESULTS ..................................................................................................................... 78
1.11.1 Calibration and Validation .................................................................................... 78
1.11.2 Model Results ....................................................................................................... 79
1.11.3 Model Interpretations and Selection ..................................................................... 81
1.11.4 Evaluation of Landuse and Climate Changes Impacts on Stream Health ............ 84
1.12 CONCLUSION ........................................................................................................... 101
STUDY AND MODEL THE EFFECTS OF CONSERVATION PRACTICES ON
STREAM HEALTH ........................................................................................................ 105
1.13 ABSTRACT ................................................................................................................ 105
1.14 INTRODUCTION ...................................................................................................... 106
1.14.1 Approaches to Linking Environmental Variables and Biota .............................. 109
1.14.2 Challenges and Objectives .................................................................................. 110
1.15 METHODOLOGY ..................................................................................................... 112
1.15.1 Study Area .......................................................................................................... 112
1.15.2 Soil and Water Assessment Tool ........................................................................ 113
1.15.3 Macroinvertebrate Data ...................................................................................... 118
1.15.4 Data Analysis ...................................................................................................... 119
1.16 RESULTS ................................................................................................................... 127
1.16.1 Calibration and Validation .................................................................................. 127
1.16.2 Model Results ..................................................................................................... 128
1.16.3 Model Selection and Interpretation ..................................................................... 134
1.16.4 Evaluations of Best Management Practices’ Effectiveness on Stream Health ... 136
1.17 CONCLUSION ........................................................................................................... 153
vi

CONCLUSIONS ....................................................................................................................... 157
RECOMMENDATIONS FOR FUTURE RESEARCH........................................................ 160
APPENDIX ................................................................................................................................ 163
REFERENCES .......................................................................................................................... 172

vii

LIST OF TABLES
Table 1. Current and pre-settlement landuse distributions ........................................................... 62
Table 2. Summary of the fish and SWAT output data at the 193 sites ......................................... 70
Table 3. Calibration and validation results ................................................................................... 78
2
Table 4. Average MSE and R amongst models from 10-fold cross validation .......................... 83

Table 5. Best performing models for all methods......................................................................... 84
Table 6. Paired t-test and changes among scenarios based on 193 fish sampling locations......... 95
Table 7. Paired t-test and changes among scenarios considering all reaches within the Saginaw
River watershed .............................................................................................................. 95
Table 8. Summary of macroinvertebrate and SWAT output data at 262 sites ........................... 121
Table 9. SWAT model calibration and validation summary ...................................................... 128
2

Table 10. Average R and MSE obtained from 10-fold cross validation based on training data135
2
Table 11. Average R and MSE obtained from 10-fold cross validation based on test data ..... 136

Table 12. Predictor variables used for the best performing macroinvertebrate model ............... 136
Table 13. Wilcoxon rank sum test and percent changes variables before and after BMP
implementation scenarios for 262 sampling locations ................................................. 139
Table 14. Wilcoxon rank sum test and percent changes variables before and after BMP
implementation scenarios for the entire basin .............................................................. 140
Table 15. The impact of BMP implementation scenarios at watershed scale based on
macroinvertebrate measures ......................................................................................... 144
Table 16. Stepwise linear regression models for IBI .................................................................. 163
Table 17. Stepwise linear regression models for percent intolerant individuals ........................ 164
Table 18. PLSR models for IBI .................................................................................................. 164
Table 19. PLSR models for percent intolerant individuals ......................................................... 165

viii

Table 20. Fuzzy logic models for IBI ......................................................................................... 166
Table 21. Fuzzy logic models for percent intolerant individuals................................................ 167
Table 22. Spearman rank correlations with red indicating significance ..................................... 168

ix

LIST OF FIGURES
Figure 1. Saginaw River watershed .............................................................................................. 58
Figure 2. Average monthly (a) precipitation and (b) temperature ................................................ 64
Figure 3. Location of fish data and calibration site ...................................................................... 69
Figure 4. IBI scores and total reach lengths among 193 fish locations ........................................ 87
Figure 5. Percent intolerant individuals and total reach lengths among 193 fish locations .......... 88
Figure 6. IBI scores and total reach lengths within Saginaw River watershed............................. 90
Figure 7. Percent intolerant individuals and total reach lengths within Saginaw River watershed
........................................................................................................................................ 91
Figure 8. Improvements and declines in IBI (a) and percent intolerant individuals (b) from
Presettle1 to Current. ...................................................................................................... 98
Figure 9. Improvements and declines in IBI (a) and percent intolerant individuals (b) from
Presettle2 to Current ....................................................................................................... 99
Figure 10. Improvements and declines in IBI (a) and percent intolerant individuals (b) from
Presettle1 to Presettle2 ................................................................................................. 100
Figure 11. Saginaw River watershed (040802)........................................................................... 113
Figure 12. Location of macroinvertebrate data and calibration site ........................................... 119
Figure 13. Improvements and declines in Family IBI from current scenario after no-till
implementation ............................................................................................................. 145
Figure 14. Improvements and declines in EPT taxa from current scenario after no-till
implementation ............................................................................................................. 146
Figure 15. Improvements and declines in HBI from current scenario after no-till implementation
...................................................................................................................................... 147
Figure 16. Improvements and declines in Family IBI from current scenario after residue
management implementation........................................................................................ 148
Figure 17. Improvements and declines in EPT taxa from current scenario after residue
management implementation........................................................................................ 149
Figure 18. Improvements and declines in HBI from current scenario after residue management
implementation ............................................................................................................. 150
x

Figure 19. Improvements and declines in Family IBI from current scenario after native grass
implementation ............................................................................................................. 151
Figure 20. Improvements and declines in EPT taxa from current scenario after native grass
implementation ............................................................................................................. 152
Figure 21. Improvements and declines in HBI from current scenario after native grass
implementation ............................................................................................................. 153

xi

LIST OF ABBREVIATIONS
%INTOL

Percent Intolerant Individuals (Fish)

ACBOD

Average annual carbonaceous biochemical oxygen demand concentration (mg/L)

ACHLA

Average annual Algal biomass (chl - a) concentration (mg/L)

ADO

Average annual dissolved oxygen concentration (mg/L)

AFLOW

Average annual flow (cms)

AMinP

Average annual mineral phosphorus concentration (mg/L)

ANFIS

adaptive neural-fuzzy inference system

ANH4

Average annual ammonium concentration (mg/L)

ANo2

Average annual nitrite concentration (mg/L)

ANo3

Average annual nitrate concentration (mg/L)

AOrgN

Average annual organic nitrogen concentration (mg/L)

AOrgP

Average annual organic phosphorus concentration (mg/L)

APercBase

% Average annual water yield contributed by groundwater (%)

ASED

Average annual sediment concentration (mg/L)

ATN

Average annual total nitrogen concentration (mg/L)

ATP

Average annual total phosphorus concentration (mg/L)

BMP

Best Management Practice

CCSM

Community Climate System Model

CDL

Crop Data Layer

CEAP

Conservation Effects Assessment Project

CrossArea

Cross-sectional area of reach (m )

EPT

Ephemeroptera, Plecoptera, and Trichoptera

2

xii

EPTtaxa

Total number EPT tax

Family IBI

Family index of Biological Integrity (Macroinvertebrates)

GIS

Geographic Information Systems

HBI

Hilsenhoff Biotic Index (Macroinvertebrates)

HRU

Hydrologic Response Unit

HUC

Hydrologic Unit Code

IBI

Index of Biological Integrity (Fish)

MNFI

Michigan Natural Features Inventory

MSE

Mean Square Error

MUSLE

Modified Universal Soil Loss Equation

N

Nitrogen

NCAR

National Center for Atmospheric Research

NED

National Elevation Dataset

NHDPlus

National Hydrography Dataset plus

NSE

Nash-Sutcliffe Efficiency

P

Phosphorus

PBIAS

Percent Bias

PLSR

partial least squares regression

R

2

Coefficient of determination

RMSE

Root mean square error

SCBOD

Average seasonal carbonaceous biochemical oxygen demand concentration
(mg/L)

SCHLA

Average seasonal Algal biomass (chl - a) concentration (mg/L)

SDO

Average seasonal dissolved oxygen concentration (mg/L)
xiii

SFLOW

Average seasonal flow (cms)

SMinP

Average seasonal mineral phosphorus concentration (mg/L)

SNH4

Average seasonal ammonium concentration (mg/L)

SNo2

Average seasonal nitrite concentration (mg/L)

SNo3

Average seasonal nitrate concentration (mg/L)

SOrgN

Average seasonal organic nitrogen concentration (mg/L)

SOrgP

Average seasonal organic phosphorus concentration (mg/L)

SPercBase

% Average seasonal water yield contributed by groundwater (%)

SSED

Average seasonal sediment concentration (mg/L)

STATSGO

State Soil Geographic Database

STN

Average seasonal total nitrogen concentration (mg/L)

STP

Average seasonal total phosphorus concentration (mg/L)

StreamGrad

Stream segment gradient (m/m)

SWAT

Soil and Water Assessment Tool

USDA

United States Department of Agriculture

USEPA

United States Environmental Protection Agency

USGS

United States Geological Survey

xiv

INTRODUCTION
With over 3.5 million miles of rivers and streams within the United States (USEPA, 2011a) and
given the numerous benefits and services provided by them, these freshwater ecosystems are of
great importance to humans and much attention has been given to their protection. Land- use
changes, climatic changes, and other stressors can have a profound effect on water resources.
Among these stressors, agriculture practices, specifically, have a large influence on streams and
rivers, effecting water quality, water quantity, and geomorphology (Dale and Polasky, 2007;
Zimmerman et al., 2003; Webb et al., 2008). This in not only observed at field scale, but also
large scale assessments identify agriculture as a leading source of current water quality problems,
through practices such as fertilizer, manure, and pesticide applications and sediment inputs from
increased erosion and runoff (Heimlich, 2003).
These effects to our streams can be further reflected in the aquatic organisms and
ecological health of the system. Alterations to flow regimes (Poff et al., 1997), changes in water
temperature (Wehrly et al., 2003), increases in sediment (Wood and Armitage, 1997) and
nutrient (Miltner, 2010) concentrations, and changes to physical habitat and stream
geomorphology (Rowe et al., 2009; Sullivan et al., 2004), have all been shown to influence the
aquatic organisms within river systems. Being that the ecological health of the stream cannot be
directly measured, biological indicators, such as fish or macroinvertebrates, are often used to
represent different communities’ responses to stressors (Karr, 1991; MDEQ, 1997; Barbour et
al., 1999).
Overall, the relationships between environmental variables, such as nutrients, and
macroinvertebrates or fish measures need to be further investigated (Wang et al., 2007; Weigel
and Robertson, 2007). In addition, there is a need to forecast the impacts that anthropogenic
1

activities have on stream health and fish distributions (Diebel et al., 2010). The first challenge
when linking disturbance and stream condition data to biological data is the lack in availability
of high resolution and complete datasets (Wang et al., 2008; Sutela et al., 2010). The use of
biophysical models can play an important role in successfully estimating this data otherwise
nonexistence. Few studies, however, have examined the use of models to help fill this gap.
Meanwhile, efforts are currently under way, through the Conservation Effects Assessment
Project (CEAP), to quantify the ecological benefits of best management practices (BMPs) and
conservation practices through modeling applications and tools (Maresch et al., 2008; Shields et
al., 2006). The second challenge is finding a suitable modeling technique or method that can
capture the complex relationships between stream conditions and ecological components (Wang
et al., 2008; Sutela et al., 2010). Alternative soft computing methods, such as fuzzy logic, may be
a possible solution to address this challenge (Marchini, 2011).
The overall goal of this research is to provide foundation for linking watershed models to
ecological health to better understand and further document the relationships involved between
environmental variables and biological integrity within a large watershed in Michigan. In
addition, this study investigates the consequences of agricultural practices on ecological health
and aims at providing critical information to support decision making regarding landuse
management.
The specific objectives of this study are to:


Bridge the gap between hydrologic models and ecological conditions using the Soil and
Water Assessment Tool (SWAT) to generate high resolution flow and water quality data

2



Identify the influential in-stream variables and their relationship to macroinvertebrate and
fish indices and metrics, ultimately indicating the overall ecological health



Employ and compare soft computing techniques and statistical analysis to describe and
model the relationships among stream health, flow, and water quality variables.



Demonstrate the applications of SWAT to model beyond sampled locations and forecast
conditions by predicting historical reference conditions under pre-settlement land-use and
climate data as well as determine the effectiveness of large scale best management
practice implementation.

3

LITERATURE REVIEW
1.1 Overview
This review discusses stream health and integrity, along with the aquatic organisms that
can serve as indicators. Because fish and macroinvertebrates are commonly used indicators
(MDEQ, 1997; Flinders et al., 2008; Barbour et al., 1999; Infante et al., 2008; Karr, 1991), they
will be discussed specifically, including frequently used indices. The review then describes how
agriculture can impact streams and reviews its effects on flow, sediments, nutrients, pesticides,
water temperature, dissolved oxygen, and other channel characteristics. It continues by
connecting the first two sections and shows how water quality, quantity, and other
characteristics, influenced by agriculture, can affect aquatic organisms and ecological health. It
also looks at methodological approaches that have been taken to make this connection.
Furthermore, the use of agricultural conservation practices to reduce the impacts on streams and
aquatic species is reviewed along with the Conservation Effects Assessment Project (CEAP) and
its efforts to quantify the effects of these practices on ecological integrity. The review ends with
a look at potential models and how they can play a role in these efforts.
1.2 Stream / Ecological Health and Integrity
Ecological health is a term that is not so clear cut and has been defined in many different
ways, making it difficult to measure and quantify (Karr, 1999). In general, the health of the
stream refers to a condition of the stream when it is flourishing, resilient, sustainable, and
maintains its societal values (Meyer, 1997). Associated with stream health, although different, is
the term integrity (Karr, 1996). Integrity refers to a quality or condition that is compared to an
original condition (Karr, 1996). . Biological Integrity, as defined by Karr and Dudley (1981) is
the ability of a system to support "a balanced, integrated, adaptive community of organisms
4

having a species composition, diversity, and functional organization comparable to that of
natural habitat of the region” (Karr, 1991). The definition, as later explained by Karr (1999),
involves three main principles; temporal and spatial scale, elements of biodiversity and
processes, and the aspect that “living things are embedded in dynamic evolutionary and
biogeographic contexts”. Biological integrity, along with physical and chemical integrity, makes
up ecological integrity (Karr, 1996). Both “ecological health” and “integrity” are terms
frequently used by agencies and are included in legislation and laws, including the Clean Water
Act (Karr, 1991). ). These definitions make integrity a valuable approach at looking at how
human disturbances have affected streams and rivers ecologically. To determine the integrity of
a system, we review measurements of the condition of a system, indicators within the system,
and the indexes that involve them.
1.2.1 Indicators
There exist numerous biological monitoring methods to measure and quantify the
ecological condition of a stream system. Using biological indicators is a highly accepted
technique in seeing how different communities respond to water quality issues and can often be
efficient in situations where physical habitat (Flinders et al., 2008) and low levels of pollutants
may be otherwise hard to detect (Barbour et al., 1999; Flinders et al., 2008). Common biota used
as indicators in aquatic systems are fish and macroinvertebrates (MDEQ, 1997; Flinders et al.,
2008; Barbour et al., 1999; Infante et al., 2008; Karr, 1991), while others like periphyton,
macrophytes, and diatoms are not as frequently used. Periphyton, however, have been shown to
be informative indicators because of their short life, their location at the base of the food web,
and their connection to fish and macroinvertebrates (Griffith et al., 2009; Hill et al., 2010;
Barbour et al., 1999).
5

Ideally, both fish and macroinvertebrates should be measured together because they can
react differently, vary in sensitivity to different stressors, and mirror conditions at different scales
(Griffith et al., 2009; Lammert and Allan, 1999; Infante et al., 2008; Flinders et al., 2008; Karr,
1981). Previous research from Infante and others (2008) showed weak community concordance
between fish and macroinvertebrates within their study region, suggesting that certain landscape
factors may be more stressful for certain organisms and/or the scale at which influences may
differ between organisms. More specifically, it has been shown that fish tend to respond to
broader scale factors like flow and land use, due to their range of movement and lifespan
(Lammert and Allan, 1999; Barbour et al., 1999; Plafkin et al., 1989). Conversely,
macroinvertebrates seem to be influenced more by local habitat, including substrate (Lammert
and Allan, 1999). This can be attributed to invertebrates living much of their lives within the
hyporheic zone of streams and their connection with the stream bed with little migration (Power
et al., 1999; Barbour et al., 1999). Flinders and others (2008) further supported this when they
found that watershed land-use predicted a higher percentage of fish community variation more
consistently than that of macroinvertebrates.
1.2.1.1 Fish as Indicators
Fish have several benefits when it comes to biological monitoring and being used as
indicators. Karr (1981) explained that fish cover many trophic levels, including piscivores,
herbivores, omnivores, and insectivores, which allows them to be a good representation of the
system and its interactions. This is also achieved by being at the top of the aquatic food chain
(Karr, 1981). Fish are advantageous because of their long lives and mobility, allowing for
observation of long term effects, as well as influences from larger, broader scales (Karr, 1981;
Babour et al., 1999). Because of fishes’ ability to often persist and recover from environmental
6

strain (Rowe et al., 2009), stress and toxicity effects can be observed, through missing taxa and
growth and reproductive variations (Karr, 1981). From a sampling stand point, fish are easy to
collect and identify. Their life history information is often known, along with distributions and
environmental requirements (Barbour et al., 1999). Not only does the public relate to reports and
information about fish, but the information collected is directly applicable to fisheries (Karr,
1981; Barbour et al., 1999).
1.2.1.1.1 Index of Biological Integrity
One common index recognized and used by many (Lammert and Allan, 1999) to measure
biological integrity is the Index of Biotic Integrity (IBI) which was introduced by James Karr
(1981). The IBI uses a set of metrics that cover hierarchical aspects of a system, from the
individual through the whole ecosystem (Karr, 1991). The original 12 metrics, intended for the
Midwest United States, are broken up into three main components including species richness and
composition, trophic composition, and fish abundance and condition. Within species richness
and composition, six metrics focus on quantifying native, benthic, water-column, long-lived,
intolerant, and tolerant species. With the trophic composition component, three metrics look at
percentages of omnivores, insectivores, and piscivores. The last three metrics, within fish
abundance and condition, quantify number of individuals, hybrids, and individuals with diseases
or abnormalities (Karr, 1991). After metrics are measured, they are scored by a comparison to
how the site would be expected to be found if undisturbed; this is usually done with a 1, 3, or a 5,
5 being the best. Totaling the metric scores obtains a final IBI score, which can be represented by
a class that can range from very poor to excellent (Karr, 1981). Over the years, others have used
alternative metrics and made appropriate modifications to the IBI to properly assess specific sites
(Cooper et al., 2009; Roth et al., 1996; Lammert and Allan, 1999; Wang et al., 2006; MDEQ,
7

1997; Wehrly et al., 2003; Lyons et al., 1996; Lyons, 1992). Several IBIs have been created for
the various regions of North America and specific states (Barbour et al., 1999). Furthermore,
indexes that are modified for stream type based on temperature also exist; an example being the
coldwater and warmwater IBIs (Lyons et al., 1996; Lyons, 1992). Streams are often classified as
being a cold, cool, or warm water streams, leading to the species and faunal assemblages they
support (MDEQ, 1997; Lyons et al., 2009). With warmwater streams supporting more diversity
and species richness (Karr, 1981; Lyon et al., 2009), just as important, coldwater streams may
not be represented well, using the same metrics. By adjusting metrics to fit species and measures
that are realistic to the site, more accurate estimations and scores can be obtained. In past studies,
underestimations of biotic integrity were obtained where warmwater or coldwater indexes were
used for streams that were classified as coolwater (Lyons et al., 2009). The MDEQ (1997) survey
protocols also call for different metrics to be used, anticipating the absence of many species in
coldwater streams, and highlighting the presence of others, like salmonids.
1.2.1.1.2 Modified Index of Well Being
The Modified Index of Well Being (MIWB) was created to resolve the issues with the
first Index of Well Being that often gave misleading results and is considered to be a sensitive
and consistent index to environmental stressors (Yoder, 1987; Barbour et al., 1999). This index
however, is made up of the same computational steps, with the elimination of certain species
when computing. Problems with the first index were often skewed due to tolerant species being
dominant and abundant, masking lack of diversity and other structural problems (Yoder, 1987).
The MIWB is made up of four estimates, two measuring diversity and two measuring abundance,
including relative number of all species, relative weight of all species, Shannon Diversity Index
of relative number of species, and the Shannon Diversity Index of relative weight. Modifications
8

to the original index includes highly tolerant species, exotics, and hybrids being eliminated from
the numbers and biomass estimates, while still being acknowledged within the Shannon
Diversity Index (Yoder, 1987). The estimates are put into a basic formula, resulting in a final
score (Barbour et al., 1999; Yoder, 1987). Higher scores for the MIWB reflect healthier
communities and better environmental conditions (Covert, 1997). When reviewing the literature,
very rarely was the MIWB used compared to Karr’s (1981) Index of Biotic Integrity. When
comparing results across agencies in a particular study, Covert (1997) found that IBI values were
less varied and consistent compared to the MIWB. The MIWB is often a complimentary index to
the IBI and can react faster to severe impacts and recovery within a lotic system (Yoder and
Smith, 1999).
1.2.1.2 Macroinvertebrates and Indicators
Frequently, macroinvertebrates have also been used to evaluate stream conditions and
integrity. Because of their smaller migration habits and sessile lifestyle, macroinvertebrates can
be efficient at reflecting localized sites and habitats (Barbour et al., 1999; Flinders et al., 2008;
USEPA, 2009c). They often respond quickly to stressors due to sensitive life stages, complex life
cycles, and varying pollution tolerances. While this allows macroinvertebrates to be affected by
short-term environmental conditions (Barbour et al., 1999), they also tend to live for periods
longer than a year (USEPA, 2009c; Barbour et al., 1999), making it possible to observe affects
over extended periods of time. Macroinvertebrates represent a range of trophic levels and being
that they are an important food source for fish (Barbour et al., 1999); they act as a link in the
food web connecting multiple organisms (USEPA, 2009c). As for sampling, macroinvertebrates,
much like fish, are generally easy to collect and identify. There is a great deal of background
information available, as well (Barbour et al., 1999; USEPA, 2009c). Over the years there have
9

been numerous indexes and metrics used in identifying and measuring stream conditions and
integrity using macroinvertebrates. Macroinvertebrate indexes provide useful information about
the condition of the community (Ohio EPA, 1989) by evaluating elements and processes
(Barbour et al., 1999), along with reflecting the health and integrity of a stream. There are two
main informative types: multi-metric indexes, that have several metrics that measure and cover
specific attributes of the assemblage; and pollution-tolerance indexes, which are based on taxonspecific tolerance values (Fore et al., 1996). There are many variations of these indexes that have
been presented for different regions and conditions. Below, are some of the more common and
basic ones.
1.2.1.2.1 Invertebrate Community Index
The Invertebrate Community Index (ICI) is a commonly used multi-metric index, which
was created by the Ohio EPA. Being based off of Karr’s fish IBI, the ICI is made up of ten
metrics reflecting the structure and composition of the community. The ten original metrics
include total number of taxa, number of Mayfly taxa, number of Caddisfly taxa, number of
Dipteran taxa, percent Mayfly composition, percent Caddisfly composition, percent Tribe
Tanytarsini Midge composition, percent other Dipteran and non-insect composition, percent
tolerant organisms, and number of qualitative EPT taxa. These metrics are then given a score of
6, 4, 2, or 0, based on a comparison with undisturbed reference sites, and summed up to a final
ICI value (DeShon, 1995; Ohio EPA, 1989). However, before being summed up, metrics may be
scaled to the relative attributes of the sample and the drainage area of the stream (Lammert and
Allan, 1999; Ohio EPA, 1989). Like most indexes, the ICI may be modified to take into account
regional variation and the levels at which identification is being done. An example of this is
where Lammert and Allan (1999) used a modified ICI, by omitting midges and tolerant species
10

metrics, in determining the influence of land cover and structure on biotic integrity in Michigan’s
River Raisin watershed.
1.2.1.2.2 Benthic Index of Biological Integrity
The Benthic index of Biotic Integrity is another multi-metric index, which was based off
Karr’s IBI for fish that has been used commonly and was developed by Kerans and Karr (1994).
This index focuses on taxa richness, composition, and biological process. All metrics represent
attributes that are reactive to human disturbances (Fore et al., 1996). They are compared with
undisturbed sites and given a score of 1, 3, or 5. These scores are then added up to give a final BIBI value. There are thirteen original metrics involved in this index including, total taxa richness,
intolerant snail and mussel species richness, mayfly richness, caddisfly richness, stonefly
richness, relative abundance of Corbicula, oligochaetes, omnivores, filterers, grazers, and
predators, proportion of individuals in two most abundant taxa, and total abundance. With the
metrics, total taxa richness and total abundance, statistical comparisons can be made among sites
(Kerans and Karr, 1994). Modifications also are needed and have been made for this index,
depending on local conditions and sampling methods. Fore and others (1996) modified and
created a B-IBI for the Pacific Northwest that used metrics that were similar to those of Kerans
and Karr’s (1994) that responded to logging. In addition, Lammert and Allan (1999) modified
metrics and scoring criteria to work for the River Raisin watershed in Michigan, based on
taxonomic differences in the area.
1.2.1.2.3 Biotic Index
The Biotic Index (BI) that was presented by Hilsenhoff (1987), also known as the HBI, is
an index based on the tolerance values of organic pollution for species and genera (Hilsenhoff
,1988; Barbour et al., 1999). Biotic indexes can target different types of stressors and the
11

tolerance values derived, determine the accuracy of the index (Lenat, 1993). The tolerance values
are multiplied by the number of individuals in that taxa, summed up across taxa, and then a
weighted average pollution tolerance is derived (Lenat, 1993; Fore et al., 1996; Lammert and
Allan, 1999). The sites with higher BI scores, which are scaled from 0 to 10, are determined to
be more degraded by pollution (Lammert and Allan, 1999). As mentioned previously, different
modifications have been made to the index based on tolerances to the pollutions of concern,
differences in season, stream size, and geographic regions (Lenat, 1993; Davis, 1995). Not only
has this index continued to be used and modified individually, but it has also been incorporated
into multi-metric indexes as well (Davis, 1995).
1.2.1.3 Habitat Indexes
With physical habitat being the fundamental driver of fish communities and structure
(Lammert and Allan, 1999), often agencies will use habitat indexes as well. With this evaluation,
habitat restraints on the biological potential of a stream can be determined. Having habitat
conditions combined with water quality conditions can allow for a better understanding of the
limiting factors of a system (MDEQ, 1997). Habitat characteristics that have been shown to be of
importance in influencing faunal assemblages include velocity, depth, water temperature,
substrate, woody debris, in stream and stream side cover, and diversity of pools, riffles, runs, and
bends (Infante et al., 2008; Barbour et al., 1999; Rowe et al., 2009).
1.2.1.3.1 Qualitative Habitat Evaluation Index
The Qualitative Habitat Evaluation Index (QHEI) is a habitat index that was developed to
give a measure of macrohabitat based on physical and functional factors that can affect fish
communities (Rankin, 1989; Miltner and Rankin, 1998). There are six components that make up
the QHEI; type and amount of substrate; type and amount of in-stream cover; channel
12

morphology, including sinuosity, development, channelization, and stability; Riparian zone
width and bank erosion; pool, glide, riffle, and run quality; and Gradient. These metrics are
scored based on comparison with streams that have high biological diversity and integrity and
summed up for a QHEI value (Rankin, 1989).
1.2.1.4 Michigan DEQ GLEAS Procedures
In their survey protocols and procedures, the Michigan Department of Environmental
Quality, Great Lakes and Environmental Assessment Section (GLEAS) have outlined a
standardized evaluation of nonpoint source impacts, developed specifically for Michigan. These
protocols involve evaluations of the habitat quality of a stream along with the biological integrity
based on fish and macroinvertebrate community metrics (MDEQ, 1997).
The Macroinvertebrate score is similar to that of the ICI (DeShon, 1995) and is
composed of nine metrics that are scored based on a comparison of an excellent site and summed
up for a final score. The nine metrics include total number of taxa, representing diversity; total
number of mayfly taxa, caddisfly taxa, and stone fly taxa, all of which are in the EPT group and
sensitive to pollution and disturbance; Percent mayfly composition; percent caddisfly
composition; percent contribution of the dominant taxon, an indication of community balance;
percent isopods, snails, and leeches, which are highly tolerant; and percent surface dependent
taxa, which are indicators of dissolved oxygen levels (MDEQ, 1997). These metrics have been
used by many; including Cooper and others (2009) when looking at sediment contamination and
faunal communities in urbanized watersheds.
The fish index within these protocols works like most of the indexes with metrics that are
scored based on ideal conditions and totaled for a final fish score. If sites have fewer than 50 fish

13

or extensive amounts of fish have anomalies, the site will be considered below acceptable
quality. Cold water streams will be judged by macroinvertebrates, with salmonid numbers
determining designation. The ten metrics that compose the index are; number of darter species,
which are sensitive to habitat degradation; number of sunfish species, which are responsive to
habitat structure and cover; number of sucker species, an intolerant species to habitat and
chemical degradation; number of intolerant species; percent omnivores, who can dominate
degraded sites; percent insectivores, who can respond to degradation; percent piscivores, which
are indicators of a healthy and diverse community; percent tolerant species; and percent
lithophilic spawners, who rely on gravel for spawning and can be largely impacted by sediments
(MDEQ, 1997). This index is not only used by itself but also has been integrated into the IBI as
seen by Lammert and Allan (1999).
1.3 Agricultural Influences on Streams
Because of the importance of having healthy streams with healthy biological integrity, it is
essential to recognize and be able to assess the impacts of anthropogenic activities, such as
agricultural practices. Agriculture makes up a large portion of the land-use in the United States.
We use approximately 20% of total land for the production of crops and a quarter of private land
is used for grazing (USEPA, 2009a). Recently, due to urbanization and the conversion of
farmland through development, land area designated for cultivation has shown some decline
(USEPA, 2009a). This decline, however, does not relieve the risks to our natural resources.
Difficulties in reducing these risks will continue with future conditions in climate change, water
supply challenges, and production of biomass for alternative fuel (Maresch et al., 2008).
Agricultural practices can have a large influence on our water resources and often results in
alterations in flow (Dale et al., 2007; Shields et al., 2006; Rowe et al., 2009; Poff et al., 1997);
14

increases in sediments (Rowe et al., 2009; Dale and Polasky, 2007; Zimmerman et al., 2003;
Wohl and Carline, 1996; Diana, 2006); an excess of nutrients (Bernot et al., 2006) and pesticides
(Gillium, 2007); increases in water temperature (Poole, 2001; Caissie, 2006; Webb, 2008); and
changes in stream morphology (Sullivan et al., 2004; Rowe et al., 2009).
1.3.1 Agriculture and Flow
Flow is a primary function of lotic systems and has been shown to be altered by different
land covers. Agriculture, through topography, geology, and vegetative cover, or lack thereof, can
influence flow patterns (Poff et al., 1997) and critical components of flow (Shields et al., 2006).
For example, increases in agricultural land-use can often lead to increases in both magnitude and
frequency of storm flows (Shields et al., 2006). Diana and others (2006) also observed a decrease
in flow stability associated with agriculture in southeastern Michigan. With natural vegetation
often lacking in agricultural areas, there is little to act against overland flow and runoff. This
combined with artificial drainage, like tiles, that rapidly discharge water into channels, illustrates
processes in which agriculture can have an effect on flow (Rowe et al., 2009). Irrigation and
water withdrawal is another activity tied to agriculture that shapes flows and infiltration patterns
(Dale and Polasky, 2007).
1.3.2 Agriculture and Sediments
With variable flows, higher peak rates, and higher velocities, comes easier and increased
transport of sediments. Sediments can be transported from either the channel itself or from the
catchment through erosion (Wood and Armitage, 1997). Both of these are found to be amplified
in agricultural regions. Rowe and others (2009) illustrated this connection between land-use and
sediments when looking at fine substrates and different land covers. Row-crop agriculture
showed a positive relationship with fine substrates. Easily erodible sediments from bare soil and
15

cultivated lands contribute to these increases along with easier transport of sediments with
increases in run-off (Zimmerman et al., 2003). Agricultural regions often have riparian areas
with reduced vegetation (Bernot et al., 2006) and this can not only promote easier sediment
transport with higher runoff velocities and reduced trapping efficiencies (Liu et al., 2008), but it
can also lead to less stable banks (Lyons et al., 2000). This in turn leads to bank erosion and
added sediments. Similar outcomes can be seen in areas of grazing. Wohl and others (1996)
exhibited this when comparing un-grazed and grazed reaches. They found both increases in
stream bank erosion and the amount of streambed composed of fine sediments. A Yankey and
others (1991) study even estimated that grazed reaches, compared to croplands, can supply two
to five times’ greater amounts of sediments (Wohl and Carline, 1996).
1.3.3 Agriculture and Nutrients
Along with sedimentation, nutrients are also a significant factor originating from
agriculture that contributes to the alterations within a watershed. Increases in nitrogen and
phosphorus are well documented in their association with agricultural practices (Bernot et al.,
2006), especially the use of fertilizers (Robertson and Vitousek, 2009; Gentry et al., 2007).
Fertilizers, soil disturbances, and biological fixation from crops all contribute in making
agriculture the leading source for nutrients in streams (Bernot et al., 2006). The transport of
nutrients from agricultural lands to channels can be further driven by climatic factors, application
timing, soil characteristics, and different flows (Heathwaite and Johnes, 1996). Groundwater,
runoff, drainage, and irrigation all help to determine the loads and concentrations of nutrients
(Domagalski et al., 2008). Irrigation using local water sources can result in lower loads than the
use of imported water due to the possibility that local water withdrawals might break the
connection between groundwater and a stream (Domagalski et al., 2008). Once nutrients reach a
16

stream, the flow and biological conditions determined by agricultural practices can continue to
direct impacts by affecting stream flow, with low discharges creating conditions for retention,
while higher discharges process and transport nutrients downstream (Royer et al., 2006; Bernot
et al., 2006).
Nitrogen can be in many forms and is often an abundant nutrient found in agricultural
regions and the watersheds they influence. Johnson and others (1997) found that the highest
nitrate and nitrite concentrations were in catchments that were dominated by row-crops
agriculture. Along with nitrate and nitrite, other inorganic forms of nitrogen, including
ammonium and ammonia, are of the more common forms of nitrogen that are introduced to
fields through microbial activity, animal waste, and fertilizers (Robertson and Groffman, 2007).
Manure can be an input of nitrogen where animals graze and where it is used as a fertilizer. In
other areas, often intense cropping operations, synthetic fertilizer is another main nitrogen source
(Robertson and Vitousek, 2009). The nitrogen from these sources can be transported both by
runoff and subsurface flow (Domagalski et al., 2008), with subsurface flows often being a
dominant means of transport (Heathwaite and Johnes, 1996).
Phosphorus is another main nutrient in streams that can reach degrading levels in areas of
agriculture and like nitrogen, phosphorus in these regions have shown increase in many studies
(Bernot et al., 2006). Fertilizers are a large contributor and source of inorganic phosphorus to a
system (Heathwaite and Johnes, 1996). Phosphorous levels can be further raised with the input of
organic phosphorus through animal wastes and manures. Phosphorus’s tendency to be immobile
and connect with soils, make overland transport through runoff and erosion the primary mode
(Sims et al., 1998; Heathwaite and Johnes, 1996). This makes application timing, soils,
precipitation, and flow driving forces behind its transport. Greater concentrations of phosphorus
17

have been observed in areas where fertilizer was applied on frozen soils, right before a
precipitation event (Gentry et al., 2007) and also in areas of heavily grazed grasslands
(Heathwaite and Johnes, 1996). Sharpley and others (1999) further explain the significances of
location and that P application near stream has more of an impact on a watershed than areas of
application further from the channel. Besides runoff, Phosphorus, like nitrogen, can also reach a
channel through leaching, subsurface runoff (Sims et al., 1998), and tile drainage (Royer et al.,
2006).
1.3.4 Agriculture and Pesticides
Pesticides are commonly used in protecting crops and agricultural practices from insects
(insecticides), unwanted vegetation (herbicides), bacteria (Antimicrobials), fungi (Fungicides),
and other pests (USEPA, 2009b). Gillium (2007) showed that 97% of streams in agricultural
areas had detectible pesticides. These pesticides can have significant negative impacts to streams.
They can enter channels through transports similar to those of nutrients, including overland
runoff, erosion, groundwater flow, and leaching. Wind and spray drift is another means of
transport for pesticides within agricultural areas (USGS, 2000). Runoff, however, continues to be
one of the most important routes (Schulz and Liess, 1999; Domagalski et al., 2008). Transport of
pesticides can be affected by soil characteristics, field slope, climatic conditions, and application
timing and amount (MSU Weed, 2010). Pesticides can accumulate in sediment and biota based
on their solubility in water and their persistence in soils. Low solubility and longer soil half-lives
can lead to more accumulation and affects, much like that of earlier used pesticides, including
DDT (USGS, 2000). Herbicides are the most widely used pesticides with hundreds of kinds
available for use (Helfrich et al., 2009). Some herbicides that are commonly used and detected
include Atrazine, Deethylatrazine, Simazine, and Metolachlor, which show high yields in basins
18

correlated with high use (Domagalski et al., 2008; Gillium, 2007). Additional herbicides that are
commonly used include 2, 4-D, Glyphosate and Acetochlor, along with several others (MSU
Weed, 2010). Other pesticides that have originated from agricultural uses and have been
monitored in streams include insecticides like Diazinon, Chloropyrifos (Gillium, 2007),
Fenvalerate, and Parathionethyl, which was observed in river channels through edge of field
runoff by Schulz and Liess (1999).
1.3.5 Agriculture and Bacteria
Agricultural practices can continue to impact water quality by amplifying the amount of
bacteria and pathogens found in streams and channels. Although this data is often unavailable
and not within the scope of our project, we recognize that they can have significant impacts on
our water resources. Animal and livestock grazing, along with manure applications in
agricultural areas, can lead to these increases in bacteria (Mishra et al., 2008; USEPA, 2006a;
Jamieson, 2002). With bacteria and pathogens living in the digestive system of humans and
animals, fecal matter then continues to carry these organisms to the land. Two main bacteria,
Fecal Coliform and Escherichia coli (E.coli), are often tested due to efficiency and their
indication of other more harmful bacteria, pathogens, viruses, and protozoa (USEPA, 2006a;
Mishra et al., 2008). The survival of these bacteria in agricultural areas and soils can be
contributed to soil characteristics such as nutrients, moisture, and organic matter (Jamieson et al.,
2002). Mishra and others (2008) showed that manure applications lead to higher edge of field
concentrations of Fecal coliform and E. coli. Lyautey et al. (2010) and Vidon et al. (2007) further
showed the influence of animals and animal waste on bacteria numbers by observing an increase
in E. coli densities with proximity to livestock production and stream sections where cattle have
access. Grazing practices and lack of vegetated riparian areas also have been observed to show
19

increases in bacteria in streams (Roodsari et al., 2005; Jamieson et al., 2002; Sullivan et al.,
2007). Transport of bacteria is often related to precipitation, soil characteristics, land cover, and
bacteria survival. Precipitation has been shown to influence transport of bacteria through runoff,
splash transport (Boyer, 2008), and leaching (Gagliardi and Karns, 2000). Once these bacteria
have reached a stream, agricultural effects on other characteristics of the stream like sediments,
temperature, and flow, further influence their survival and levels (Minnesota Pollution Control
Agency, 2008).
1.3.6 Agriculture and Water Temperature
Water temperature is an additional characteristic of streams that can be manipulated by
external factors from catchment and land-use. One factor that affects stream temperature is
stream morphology (Poole and Berman, 2001). Agriculture can affect this morphology by
increased bank erosion and increased sediments. This can further lead to channel widening and
shallowing, which can also increase temperatures (Webb et al., 2008). Another factor influencing
temperature is solar radiation and canopy cover from riparian zones (Wehrly et al., 2003; Poole
and Berman, 2001; Caissie, 2006). As discussed previously, agricultural lands often have limited
vegetated riparian area, meaning little shading. These conditions also lead to increase lateral
inputs and overland flow, which can affect temperature. Wehrly and others (2003) showed that
agricultural areas with surface runoff-dominated reaches that had little shading were found to be
warmer then streams that were groundwater dominated and had forested landscapes. Vegetated
cover, often lacking in agricultural areas, can also influence temperature by decreasing heat
exchange through wind and air movement (Poole and Berman, 2001; Webb et al., 2008). In
addition, irrigation and water withdrawals are another activity frequently tied to these areas,
which can further manipulate temperatures (Poole and Berman, 2001; Cassie, 2006).
20

1.3.7 Agriculture and Dissolved Oxygen
Dissolved oxygen (DO) is used in both biotic and abiotic components of lotic systems
and can, too, be affected by agricultural practices (Garvey et al., 2007). The relationship and
responses of DO to agriculture however, is much more complex. This complexity is due to all the
interactions involved, including temperature, nutrients, flow, and further biotic interactions
(Garvey et al., 2007; Wang et al., 2002). The temperature of a stream, affects the amount of
dissolved oxygen in it, with colder waters being able to hold more dissolved oxygen (USEPA,
2006b). Because agricultural regions affecting stream temperatures, as explained earlier, they
too, indirectly affect DO. Additionally, this indirect relationship is seen with flow. Garvey and
others (2007) explains how flow can affect oxygen levels through the interaction of water and
the atmosphere, along with the effects of substrates and depth. Further agricultural effects on
dissolved oxygen include factors that influence eutrophication. Due to light, dissolved oxygen in
streams varies often from day to night, through photosynthesis and respiration from organisms
within, including algae and periphyton. This lack of oxygen at night can reach extremes in
impacted areas (Wang et al., 2002; Loperfido et al., 2009; Hill et al., 2009). Nutrients, sediments,
light, temperature, and flow continue to influence abundance, respiration, production and
decomposition of these organisms which alter and often limit dissolved oxygen within a system
(Loperfido et al., 2009).
1.3.8 Agriculture, Physical Habitat and Stream Morphology
As shown, previously, agriculture can change both water quantity and quality
characteristics. In addition to those, there are also more physical conditions that can be impacted
by land use. Although majority of these variables are not included within the scope of our
project, we do recognize their importance. Structure and debris within a stream is one of these
21

conditions. The amount of woody debris and organic material that is found in a stream is
influenced and primarily driven by the vegetation and forested composition within the riparian
zone (Roth et al., 1996; Poole and Berman, 2001). Agricultural areas, having less forested
riparian areas, have been shown to have lower numbers of large woody debris within a channel
(Shields et al., 2006). Channelization, homogenous habitat, and decreases in stream sinuosity
can also be conditions associated with agricultural areas, through the straightening of channels
and increased drainage. This can lead to a decrease in bed stability, peak stream flow and power,
and other further impacts (Sullivan et al., 2004; Rowe et al., 2009).
1.4 Stream Conditions and Aquatic Health
The above mentioned stream conditions and factors affected by agriculture can have a
large influence on the aquatic biota and overall aquatic health of a system. These factors include
flow, sediment, nutrients, pesticides, bacteria, water temperature, dissolved oxygen, and channel
morphology.
1.4.1 Flow and Aquatic Health
Flow within a lotic system has a great deal of influence on ecological processes and fish
assemblages. As earlier explained, flow can affect temperature, sediment, nutrient, and pesticide
input, dissolved oxygen levels, and other stream conditions, which in turn affect aquatic
organisms. Flow can also affect the biota in more direct ways, including habitat availability,
wash out, and other stressors. Macrophytes and aquatic vegetation that are a resource for food,
juvenile habitat and cover, among other things, can be impacted by altered flows (Poff et al.,
1997; Bunn and Arthington, 2002). In addition, flow is tied to many important life events for
some fish, including spawning and recruitment, and changes can have negative implications on
these species behavior and survival (Bunn and Arthington, 2002). All of these impacts can be
22

observed through five components of flow; including magnitude, frequency, timing, duration,
and rate of change (Poff et al., 1997). With changes in flow, potential areas for habitat and or
connections to areas of habitat can be created or reduced (Sparks, 1995).A common way of
describing how flow can have an effect on fish and macroinvertebrates in this way is the IFIM,
Instream Flow Incremental Methodology. This model shows the amount of available habitat with
varying discharges based on water depth, velocity, and substrate (Poff et al., 1997). Within
several studies, fish have consistently been shown to be negatively affected, including diversity
of species, to changes in natural flow regimes. This is true for both decreases and increases in
flows. Also, in their review of 165 papers, Poff and Zimmerman (2010) found that 92% of them
showed negative responses of ecological metrics to alterations in flow. Lammert and Allan
(1999) observed flow’s effect on fish, revealing that IBI score variation was strongly associated
with flow stability. Abundance and diversity of macroinvertebrates are also commonly reduced
in studies with flow alterations (Poff and Zimmerman, 2010). These reductions can occur due to
changes in substrate habitat from flow and, in a more direct affect, increased drift and
downstream transport (Bunn and Arthington, 2002; Borchardt, 1993).
1.4.2 Sediments and Aquatic Health
Sediments can also impact the ecology of stream in several direct and indirect ways.
Primary production within a stream can be affected due to sediments limiting light penetration,
transporting nutrients, and covering substrate (Wood and Armitage, 1997). Izagirre et al. (2009)
observed sediments impacting periphyton negatively soon after siltation. Although they saw a
recovery, it was partially contributed to a shift in community structure. These influences on
periphyton can be linked to and show indirect effects from sediments on fish and
macroinvertebrates through habitat structure alteration, food availability, and dissolved oxygen
23

levels (Griffith et al., 2009). Although sediments are shown to have negative effects toward
larger aquatic organisms, it is also possible that they can improve habitats by limiting periphyton
growth and eutrophication (Hill et al., 2009).
Sediments can further affect fish in other ways by degrading their habitat within a stream;
with, increased turbidity, their feeding efficiency and abundance of food is obstructed (Wood
and Armitage, 1997; Karr and Dudley, 1981). Sullivan et al. (2009) showed results that indicated
that sediment loads and their duration affected certain feeding guilds more than others, with
opportunistic feeders being the most resilient. Fish can also be hindered by sediments that cover
and embed substrates, which impede on spawning, nesting, and juvenile habitat (Sullivan et al.,
2009; Newcombe and Jensen, 1996). Sediments continue to affect fish in more direct ways like
physiological stress when sediments clog gills (Wood and Armitage, 1997) and increase
respiration rate (Zimmerman et al., 2003). All these stresses can further impact fish assemblages
by reducing growth rates, causing delayed hatching, mortality, and natural fish migration
(Zimmerman et al., 2003; Wood and Armitage, 1997). Overall community structure and diversity
can be influenced by suspended sediments and sediment deposition. Rowe et al. (2009) observed
significant declines in fish IBI’s from sites where substrates were dominated by sediments.
Along with loads and concentrations, effects of sediments have been observed to increase with
particles size and have more of an affect to fish in warmer waters (Newcombe and Jensen, 1996).
Macroinvertebrates have been shown to be directly impacted from sediments as well.
Their abundance, diversity, density, and community structure have been observed in studies to be
affected, due to covered and modified substrate habitat, affecting their respiration, impairing
feeding, and increasing drift (Wood and Armitage, 1997). Wohl and Carline (1996) observed
lower densities in streams with higher sediment loads. Griffith et al. (2009) also continued to see
24

effects of sediments on macroinvertebrates in direct ways, including increases in percent
burrower taxa, as well as indirect means through interactions with periphyton as explained
earlier.
1.4.3 Nutrients and Aquatic Health
Sediments often are accompanied by nutrients when they reach channels. These nutrients,
although essential to organisms, can exceed levels that are needed and have considerable
negative impacts on the individual, community, and the system as a whole. In some situations,
where nutrients are low, supplied nutrients from agriculture can show positive effects on a lotic
system and aquatic organisms. However, often this is not the case and there is an excess of
nutrients in the system (Miltner, 2010). One of the main effects of nitrogen and phosphorus
overload in a lotic system is the acceleration of eutrophication (Carpenter et al., 1998). Both
nitrogen and phosphorus, predominantly in areas of adequate light (Hill et al., 2010); have been
shown to increase algal, macrophyte, periphyton, and phytoplankton growth (Bernot et al., 2006;
Carpenter et al., 1998). Shifts in algal community structure have also been observed, leading to
more tolerant species like blue-green algae becoming dominant in streams with high nutrients
(Minnesota pollution Control Agency, 2003). These alterations and increases can reach levels
that create dangerous algal blooms and conditions, which lead to low dissolved oxygen levels
and areas of hypoxia, which are unable to support other organisms (USEPA, 2009d; Robertson
and Vitousek, 2009). Not only can these increases in periphyton and algae negatively affect
dissolved oxygen levels, but they can continue to affect consumers and secondary consumers by
influencing feeding and habitat structure (Griffith et al., 2009).
Some fish are more intolerant than others to high levels of nutrients, along with the
conditions they create within the habitat. These increased levels of nitrogen and phosphorus have
25

been observed many times to affect fish assemblages and communities. For example, Wang et al.
(2006) showed that total phosphorus and total nitrogen contributed to most variation in fish
variables, including negative relationships with IBI, number of carnivores, omnivores, and
intolerant species. This negative relationship between these nutrients and IBI was also observed
by the Minnesota Pollution Control Agency (2003) and Miltner (1998) as well.
In addition, these effects have been reflected in macroinvertebrate communities. Miltner
and Rankin (1998) observed a decrease in ICI associated with NH3. It was also shown that EPT
taxa decreased, while some other taxa increased, with higher nutrient levels. Justus et al. (2010)
observed that biotic indexes for both fish and macroinvertebrates were negatively correlated with
nutrients and were found to be lowest where nutrient concentrations were at their highest.
1.4.4 Pesticides and Aquatic Health
Pesticides can be introduced to streams through agricultural practices and throughout the
years have proven to be detrimental to aquatic systems. One of the most well-known and
damaging pesticides was DDT (dichlorodiphenyl-trichloroethane). This, along with all other
organochlorine insecticides, has now been banned from use in the United States due to harmful
effects on the environment and biota (Helfrich et al., 2009). Other types of pesticides, however,
are still commonly used and can affect aquatic organisms directly through absorption, respiration
through the gills, and orally through feeding and drinking. Pesticides, although toxic to aquatic
life in high doses and for long periods of time, seldom cause kills. This is due to pesticides being
often short lived and strongly absorbent to sediments and mud. Insecticides are generally the
most harmful to a lotic system, being highly toxic to macroinvertebrates and fish (Helfrich et al.,
2009). Schulz and Liess (1999) observed several macroinvertebrate species disappearing and
others’ population densities showing significant reductions in streams affected by insecticides.
26

Schafer and others (2007) also showed negative relationships between pesticides and
macroinvertebrate community structure. These eliminations and shifts in macroinvertebrates can
affect food resources for specific fish and ultimately impact them as well (Pimentel, 2005).
Herbicides, like Atrazine, Metolachlor, and Simazine, also can be detrimental to streams through
their impacts on macrophytes and vegetation, which can affect habitat, food, and refuge for many
aquatic organisms (Solomon et al., 2008).
1.4.5 Bacteria and Aquatic Health
Fecal bacteria that originate from manure and animal waste in agricultural areas can have
impact on the ecological health of a stream. Along with the cloudiness and turbidity that it can
create, the decomposition of the organic material can contribute to low dissolved oxygen
concentrations as well. The bacteria, pathogens, viruses, and protozoan that are present can all
present health risks within a lotic system, especially for humans (USEPA, 2006a). Geldreich and
Clarke (1966) observed levels of fecal coliform in fish’s intestinal tracts that were reflective of
the pollution levels within the stream and indicated that these bacterial microorganisms could in
fact survive and multiply in favorable temperatures. Similar results were obtained by Del RioRodriguez and others (1997) who found that E.coli could be established in trout through feeding
on infected feed in water at warmer temperatures. These pathogens cannot only be found in the
digestive tract of fish, but also in their skin, organs, and muscles, making them a potential risk
for human consumption (Fattal et al., 1992).
1.4.6 Water Temperature and Aquatic Health
Temperature and thermal regimes within streams is significant when it comes to aquatic
biota and their distributions and physiologies (Wehrly et al., 2003; Poole and Berman, 2001).
Not only can temperature affect aquatic organisms directly, but it can also affect them indirectly
27

through its influence on oxygen levels, with colder waters are able to hold more dissolved
oxygen (USEPA, 2006b; Garvey et al., 2007). Nutrient toxicity is also influenced, with many
contaminates showing an increase in toxicity at higher temperatures (Caissie, 2006).
Temperature has also influenced alterations in production and consumption (Loperfido et al.,
2009). Additionally, there influence on the system affects rates of nutrient cycling and
productivity (Poole and Berman, 2001). Although warmer waters are known to support more
aquatic species and diversity (Karr, 1981; Lyons et al., 2009), warmer waters can also lead to
detrimental conditions as well.
Fish, through community composition, variations in species richness, and amounts of
standing stock, have shown that temperature is significant in determining their distribution and
its use for classifying them and their habitats is very valuable in management (Wehrly et al.,
2003; Lyons et al., 2009). Water temperature has been observed to have direct effects on fish,
including warmer waters associated with increased metabolic rates (Pool and Berman, 2001;
MDEQ, 1997) and increased growth rate (Hinz and Wiley, 1998; Caissie, 2006). With both
increases in metabolic rates and lower levels of dissolved oxygen, these factors can sometimes
lead to unfavorable and harmful conditions (Cassie, 2006). Increases in temperature have also
been shown to exceed optimal levels for fish and in effect show declines in the above mentioned
rates (Power et al., 1999). Many life events can be triggered by temperatures as well, including
fish movement, spawning, and smolt runs (Caissie, 2006; Power et al., 1999). Along with this,
habitat and food availability can be influenced by water temperature and its fluctuations.
Groundwater supplies are often associated with colder waters with less fluctuation, allowing for
refuge from extreme temperatures in both summer and winter (Poole and Berman, 2001; Power
et al., 1999). When groundwater is the primary source of flow, creating constant colder thermal
28

conditions, different fish, that are intolerant to warmer waters, are supported in that system (Chu
et al., 2008; Lyons et al., 2009; Wehrly et al., 2003). This variation in distribution of fish at
varying temperatures has systems commonly classified as cold water, cool water, or warm water
streams, with cool water being the most common within Michigan and Wisconsin (Lyons et al.,
2009).
Macroinvertebrates are also influenced by temperature in much of the same way as fish.
It can affect their habitat around them and increases in temperature have shown increases in their
metabolic rates (MDEQ, 1997; Poole and Berman, 2001) and growth rates (Caissie, 2006).
Pockets of desired temperatures also create localized refuge for macroinvertebrates in extreme
temperatures (Poole and Berman, 2001). Hinz and Wiley (1998) found that several temperature
measures, including annual maximum and mean daily summer temperature, were positively
correlated with macroinvertebrate biomass and standing stock.
1.4.7 Dissolved Oxygen and Aquatic Health
With oxygen often being a limiting substance for aquatic organisms, like fish (Garvey et
al., 2007), dissolved oxygen can be a significant predictor of ecological conditions within a
stream (Loperido et al., 2009). The dissolved oxygen within an aquatic system can be influenced
by temperature, light, flow, channel morphology, and the organisms within (Loperfido et al.,
2009; Garvey et al., 2007; Wang et al., 2002). Although densities of larger organisms seldom
reach levels where they can deplete a system of dissolved oxygen, smaller organisms like zebra
mussels, macrophytes, and periphyton can be significant in reducing DO levels (Caraco and
Cole, 2002; Garvey et al., 2007). Although macrophytes, algae, and periphyton all contribute to
creating dissolved oxygen through photosynthesis, they also deplete the system through
respiration at times of limited light (Loperfido et al., 2009; Francis-Floyd, 1992). This is why
29

aquatic systems see diurnal fluxes in dissolved oxygen levels, with low levels being reached at
night (Francis-Floyd, 1992). During those times in which oxygen levels reach insufficient
amounts, fish and macroinvertebrates, especially at earlier life stages, can be impacted, leading
to migration or even lethal states of hypoxia (Francis-Floyd, 1992). Other, more specific effects
on fish include growth declines (Garvey et al., 2007), altered heart rate, changes in circulation,
changes in respiration, and increases in breathing rate (Seager et al., 2000).
Lethal impacts of low dissolved oxygen have been observed along with its correlation with fish
IBI and macroinvertebrates (Minnesota Pollution Control Agency, 2003; Garvey et al., 2007).
1.4.8 Physical Habitat, Channel Morphology and Aquatic Health
Woody debris and substrate have also been connected to fish and macroinvertebrate
assemblages. Woody debris can affect flows and heterogeneity of depth and habitat along with
providing structure and cover for aquatic biota (Roth et al., 1996). Degerman and others (2004)
showed that brown trout were more frequent and abundant in sites with more woody debris.
Similar results were observed by Schneider and Winemiller (2008) who saw abundance and
richness increase in areas with woody debris, as well as macroinvertebrate abundance and
community structure.
A diversity of pools, riffles, runs, and bends allows for a diversity of species and
organisms that need and rely on them (Rowe et al., 2009; Sullivan et al., 2004). This variation of
habitat units supports life cycle stages and special requirements for species including spawning
areas and juvenile habitat. These areas also provide refuge for organisms, whether it is for
thermal conditions, dissolved oxygen needs, cover, feeding, flow patterns, or extreme conditions

30

(Caissie, 2006). Areas of channelization and little diverse habitats have been shown to support
less diversity in fish assemblages (Sullivan et al., 2004).
1.4.9 Influences and Different Scales
Agriculture can influence streams and their ecological integrity from both the local and
catchment scale, requiring a hierarchical approach of thinking when looking at its affects (Rowe
et al., 2009; Roth et al., 1996). There have been several studies that have observed local riparian
habitat to be a better predictor of biotic integrity, including Lammert and Allan (1999). Rowe
and others (2009) also showed these results with a stronger relationship with physical habitat at a
small scale. However, these and several other studies recognize the importance of the landscape
and catchment area and its ultimate influence on the local area (Infante et al., 2008; Hutchens et
al., 2009). Roth and others (1996) especially saw this influence and reported a higher correlation
between IBI and the catchment scale with local riparian conditions being a secondary and often
an ineffective predictor. These results have also been observed by others as well, including
Lyons and others (2000), who observed fish communities to be more influenced by conditions at
the larger scale. Although Wang and others (2006) observed local factors explaining fish
communities better, these were in catchments of relatively low disturbance. As the local and
catchment areas became more disturbed, the more local factors had less of an impact and
influence on fish communities. The contradicting results, of the scale at which agriculture has
more of a weight on fish, have additionally been seen with macroinvertebrates. Several studies
have shown that macroinvertebrates are more commonly influenced by local riparian conditions,
yet, some have shown that the watershed scale can also be a significant predictor as well
(Flinders et al., 2008). It is apparent that factors at a variety of scales contribute to the conditions
of aquatic biota communities within lotic systems and can be influential (Stewart et al., 2001;
31

Hutchens et al., 2009). For predictive modeling purposes, it is very common that the off stream
environmental variables including the types and percentages of different land uses are included
in development of a predictor for stream health indicators. However, using this variable in a
watershed that mainly dominated with one landuse (e.g. agriculture) can lead to poor predictions
(Heitke et al., 2006; Stauffer et al., 2000; Johnson and Host, 2010). In addition, the relationship
built base on the percentages of different landuse is insensitive to climatological variables, which
has an important influence on stream health. There continues to be a challenge in capturing the
confounding interactions among landscape factors that help in forming aquatic ecosystems
(Johnson and Host, 2010).
1.5 Methods Linking Environmental Stressors and Ecological Variables
Several approaches have been proposed to study the complex relationships between
stream health and human disturbance. Efforts to link disturbances and stream variables to aquatic
macroinvertebrates and fish have been made through numerous methods ranging from simple
linear models to complex soft computing techniques such as fuzzy logic.
1.5.1 Linear Methods
Simple models such as regression models have been considered reliable where there is a
lack of knowledge or limited data (Van Sickle et al., 2004). These models are relatively easy to
develop and can help in understanding how changes in independent variables affect the
dependent variable. There have been numerous studies demonstrating the use of linear regression
models for ecological studies (Van Sickle et al., 2004; Maret et al., 2010). Waite and others
(2010) used linear regression when looking at the relationship between environmental variables
and macroinvertebrate metrics. Other studies employed multivariate and dimensional approaches
like canonical correspondence analysis (CCA) (Pool et al., 2010; Wang et al., 2008) and
32

nonmetric multidimensional scaling (NMS) (Sutela et al., 2010). In addition, other multivariate
approaches such as partial least squares regression (PLSR) has advantages compared to general
linear regression methods when dealing with ecological processes, due to their ability to handle
complex interaction and redundancy among a large array of variables (Carrascal et al., 2009).
PLSR is a multivariate method that linearly combines several predictor variables into latent
factors that maximize the explained variance in the response variable or variables (Carrascal et
al., 2009).
1.5.2 Non-Linear Methods
As it is often the case in ecological processes, relationships are complex and non-linear.
To better understand and represent these relationships for purposes including modeling, Nonlinear statistical methods can be used. These methods include piecewise linear regression,
regression tree analysis (Wang et al., 2007; Weigel and Robertson, 2007), and kernel regression.
Maret and others (2010) used piecewise regression to identify thresholds in biotic responses and
nutrient concentrations. Wang and others (2007) employed regression tree analysis to analyze
linkages between nutrients and aquatic organisms (fish and macroinvertebrates). Kernel
regression is another non-linear technique that can be used to create a non-linear and flexible
regression function (Hastie et al., 2009). Kernel methods are memory-based and use a localized
weighting function along with different local smoothing techniques (Hastie et al., 2009).
1.5.3 Soft-Computing Methods
Other methods that can handle complex problems and are based on “inexact” computing
techniques are known as soft computing (Huang et al., 2010). Methods, such as fuzzy logic
(Adriaenssens et al., 2006) can provide robust solutions with tolerances to imprecision,

33

uncertainty, and approximation (Huang et al., 2010) and has shown an increasing trend in
ecological research (Marchini, 2011).
Fuzzy logic is a computing approach introduced in 1960’s by Zadeh (1965) and is a
method that deals with approximate reasoning (Huang et al., 2010), which can be data driven or
executed through expert knowledge. Fuzzy logic approach can be especially effective when
dealing with highly variable, complex, non-linear, and uncertain relationship of variables,
commonly observed in ecological studies (Adriaenssens et al., 2004; Chen and Mynett, 2003). In
addition, the fuzzy models constructed can often be easily interpreted based on their linguistic
nature. Fuzzy logic techniques involve the creation and building of membership functions and
inference rules, which can often be the most difficult part involved in fuzzy modeling (Chen and
Mynett, 2003) and can often be subjective (Adriaenssens et al., 2004). This is especially true
when working with fuzzy logic models based on Mamdani-Assilian methods (Mamdani, 1977).
When building a Mamdani-Assilian type model, both the input and output variables are
incorporated into membership functions. Whereas in the other approach, Takagi-Sugeno (Takagi
and Sugeno, 1985), the input variables are expressed in membership functions, however, the
output is expressed through as a linear relationship with the input variables. In addition, fusion of
fuzzy logic with other soft computing methods, such as neural networks, can provide adaptive
learning techniques that can be very useful when data is already available and a predetermined
model structure is not (MathWorks, 2010). The adaptive neural-fuzzy inference system (ANFIS)
is an example of the fusion technique.
Applying to environment and ecological processes, fuzzy logic has proven to be a valid
method for developing predictive models and decision support tools (Adriaenssens et al., 2004).
Fuzzy logic has been used for water quality assessment and classification, such as construction
34

of an index model for surface water quality classification (Icaga, 2007) and the evaluation of
watershed conditions and creation of a decision support system (Jensen et al., 2000). Fuzzy logic
has also been used to develop predictive models for both abiotic and biotic processes. Using
fuzzy logic for flow (Zhang et al., 2009) and flood forecasting (Nayak et al., 2005) have been
shown to be reliable. Fuzzy logic techniques have been successfully used for modeling algae,
macroinvertebrate, and fish communities. Chen and Mynett (2003) illustrated the use of fuzzy
logic developed with the combination of data mining techniques and heuristic knowledge in
predicting algal biomass and eutrophication for a lake system. Adriaenssens and others (2006)
illustrated that fuzzy models were valuable and practical for predicting the abundance of
macroinvertebrate taxa in a river basin based on conductivity, dissolved oxygen, water velocity,
and stream width. Predictions of fish assemblages and their physical habitat have also been
conducted using fuzzy logic approaches. For example, Mouton and others (2008) demonstrated
fuzzy logic application in developing a physical habitat model for European grayling in a river
system. Jorde and others (2000) used fuzzy logic to evaluate relationships among floodplain
processes, in-stream habitat quality, and fish communities.
1.6 Conservation Practices
There are several conservation and best management practices that are implemented by
land owners and organizations that aim to reduce the negative environmental consequences that
come from agriculture procedures. These conservation actions, more specifically, help in
reducing erosion and sedimentation, nutrient transport, water runoff and surface transport, and
impacts to stream morphology and habitat (NRCS, 2006).

35

1.6.1 Filter Strips
Filter strips are a conservation practice that works similarly to herbaceous riparian
buffers. Filter strips are strips of vegetation that is often placed adjacent to streams in the act of
removing or reducing contaminates pesticides, and sediments from runoff produced by fields.
These strips can also reduce the quantity of overland flow through infiltration and the potential
benefits, like in riparian buffers, can increase with width and vegetation used (NRCS, 2006).
1.6.2 Terraces
When trying to reduce runoff, terraces are often used as well. This best management
practice consists of a series of earth embankments or ridges that lay across a farmed hillside,
which retain runoff and direct it to a desired location. They can hold runoff for moisture and also
reduce erosion, which leads to less sediment transport to stream and channels (NRCS, 2006).
1.6.3 Grassed Waterways
Another conservation practice that works by directing water to favorable outlets is
grassed waterways. These are vegetated channels through fields that direct runoff in low
velocities to stable outlets, while reducing erosion and keeping contaminates and sediments from
going directly into streams (NRCS, 2006). Not only can these waterways direct runoff and trap
sediments efficiently, but they can also reduce velocities by prolonging infiltration time (Fiener
and Auerswald, 2002).
1.6.4 Constructed Wetlands
Sometimes grassed waterways can lead to outlets like constructed wetlands. These low
lying ecosystems with hydrophytic vegetation are an additional best management practice used

36

to improve water quality (NRCS, 2006). These wetlands can not only help in improving water
quality and regulate flow, but they can also supply important wildlife habitat (USEPA, 2000).
1.6.5 Conservation Tillage
Conservation Tillage is another conservation practice that is known to reduce erosion,
improve soil matter content, increase plant available moisture, and reduce soil particulate
emissions. This is the practice of managing of plant residue on the soil surface, while limiting
disturbances to the soils when applying nutrients and planting crops (NRCS, 2006). There are
several types of conservation tillage, including strip till, ridge till, and no till, all being beneficial
to lessening the impact on water quality in nearby aquatic systems (USEPA, 2009e).
1.6.6 Rotational Grazing
The managing of grazing animals and their harvest of vegetation is a conservation
practice that can improve plant species composition, riparian and watershed function, subsurface
and surface water quality and quantity, and soil condition, all while reducing soil erosion
(NRCS, 2006). There have been several studies where areas of intense grazing have led to
increased sediments and runoff (Wohl and Carline, 1996).
1.6.7 Contour Farming
Contour farming is the practice of using farming operations, tillage ridges, and plantings
to alter the transport of runoff from going directly downhill. This technique can reduce sediment
and contaminate transport, along with increasing infiltration as well. This practice is especially
efficient on sloped land with increased residue cover and roughness, achieved through vegetation
(NRCS, 2006).

37

1.6.8 Strip Cropping
Strip cropping is the act of growing rotational crops in an arrangement of strips across a
field in effort to reduce soil erosion from both wind and overland flow. Strips should alternate
between erosion-resistant crops and crops that are subject to erosion. Several factors determine
the efficiency of this practice including, number of strips, alignment, width, and orientation
(NRCS, 2006).
1.6.9 Conservation Reserve Program (Native Grasses)
The Conservation Reserve Program is a voluntary program that contracts and pays farmers
to designate, usually cropped land, to be planted with tree, shrubs, grasses, and native prairie
species. These actions can reduce erosion, sediments inputs into streams, and improve other
water quality conditions (NRCS, 2006).
1.6.10 Conservation Effects Assessment Project (CEAP)
Following the 2002 farm bill and the boost of conservation program funding, the USDA
Natural Resources Conservation Service (NRCS) and other USDA agencies established the
Conservation Effects Assessment Project (CEAP) in an effort to show the environmental benefits
of conservation practices at the national, regional, and watershed level (Maresch et al., 2008;
Duriancik et al., 2008; Mausbach and Dedrick, 2004). With the U.S. Federal government
spending about 4 billion annually in agricultural conservation programs (Shields et al., 2006),
being able to quantify and support these programs is critical. Outcomes of this project are
intended to continue to provide insight, understanding, and recommendations for conservation
decisions to everyone in the conservation community, from policy makers to the farmers
themselves (Duriancik et al., 2008; NRCS, 2010; Maresch et al., 2008). Educating farmers and
others on conservation options and effectiveness, has proven to be influential in practice
38

adoption and the move toward more conservation actions. CEAP also can provide other insights
into human dimensions of conservation practices and the factors influencing adoption (Maresch
et al., 2008). Several programs and conservation techniques are assessed and covered by CEAP,
including the Conservation Reserve Program (CRP), Wildlife Habitat Incentives Program
(WHIP), Conservation buffers, nutrient management, pest management, and tillage management
(Duriancik et al., 2008). There are three main components of CEAP, one being bibliographies,
literature reviews, and a scientific workshop. This component was initiated to distinguish what is
known about the impacts of conservation practices and what still needs further research and data
collection. This includes not only the effects, but also, methods that are used to assess the effects
(Duriancik et al., 2008). The second component is watershed assessment studies, which is aimed
at providing the effects of conservation practices on water and soil quality at the local level. This
can allow for more efficient management and implementation of conservation methods within a
watershed (Dureiancik, 2008). Earlier research focused in on environmental impacts at the
individual farm or field level, however CEAP recognizes and looks at the off-site effects through
the watershed, which are regularly unseen by the people who apply the conservation practice
(Maresch et al., 2008). The third component in CEAP is national and regional assessments.
Much like the second component, this too looks at the benefits and impacts of conservation
practices, but on a larger scale (Duriancik et al., 2008). It can also help in applying or designing
new conservation programs for more efficiency at reaching desired outcomes (Maresch et al.,
2008; NRCS, 2010). Within the national assessment, CEAP focuses on the effects of
conservation practices on four elements including croplands, wetlands, grazing lands, and
wildlife. Due to diversity of fish and wildlife resources, agricultural landscapes, and conservation
activities from region to region, the wildlife component is split up into four regions, including

39

the Midwest, Southeastern, Northeastern, and Western (Duriancik et al., 2008). Because fish and
wildlife are connected to landscapes and the conservation practices, the wildlife element is linked
and influenced by the other three elements (NRCS, 2010). Because of this connection and the
complexity of biological outcomes, the impacts on fish and wildlife are often difficult to
understand and measure. After the compiling of literature reviews, Duriancik and others (2008)
identified areas with a lack of research pertaining to and understanding the effects on fish and
wildlife from conservation practices. The research that was available often used habitat quality
and suitability, which is seen as a reliable predictor of the impacts.
1.7 Modeling
Other environmental effects, from conservation practices in agricultural regions, have
been assessed within CEAP through model development and application (Mausbach and
Dedrick, 2004). These include AnnGNPS, a watershed model; CONCEPTS, a channel evolution
model; REMM, a riparian ecosystem management model; and SWAT, a watershed model
(Shields et al., 2006). Although these models can be used to see water quality and quantity
effects, limited connection to ecological effects has been shown (Shields et al., 2006). The
development of models is often very complex, which has led to gaps in research. The
improvement of application and uncertainty of models is a strong focus for CEAP (Maresch et
al., 2008). Shields and others (2006) look at the possibilities of using current models and
adapting their use for assessing fish or macroinvertebrate habitats. Simulated flow, water quality,
channel morphology, and other outcomes, which are all valuable habitat characteristics, could,
with further processes, be applied to modeling habitat and ecological conditions.

40

1.7.1.1 SWAT
The Soil and Water Assessment Tool (SWAT) is a commonly used watershed scale
model that can simulate runoff, soil erosion, chemical and sediment transport, along with several
other hydrologic processes, in an aim to predict effects from land use and agricultural
watersheds. This GIS based and spatially explicit model, developed by the USDA-ARS temple
Texas, uses input data, including but not limited to weather, topography, soil, and vegetation.
Simulation, along with calibration and validation, then provides outputs for sub-watersheds and
reaches on different time scales (Casper et al., 2011; Neitsch et al., 2009; Shields et al., 2006).
Several studies have shown the use of SWAT in evaluating hydrologic condition in agriculture
areas and its possible impacts on aquatic health. For example, Rossi and others (2008) looked at
SWAT’s ability to simulate hydrology, pollution discharge from point sources, runoff, and
average stream flow, and found it to be an effective tool. The SWAT model’s prediction of
stream flow was also demonstrated to be more efficient than that of other models in uncalibrated
conditions (Heathman et al., 2008). Other studies have shown SWAT’s ability to evaluate best
management practices and nutrient loading (Sood and Ritter, 2010), sediments, and bacteria
(Parajuli et al., 2008). Casper and others (2011) used the SWAT model in efforts to simulate
hydrograph data for streams that did not have monitored or gauged data. They found that
predicted model outputs such as hydrographs, although significantly sensitive to resolution,
could be a viable option in linking to the habitat model, PHABSIM.
1.7.1.2 AnnAGNPS
Another watershed scale model developed by the USDA, used for simulating the long
term allocation and load of pollutants is the Annualized Agricultural Non-point Source Model
(ANNAGNPS) (Bingner and Locke, 2009). This model, much like SWAT, can simulate runoff,
41

sediment and chemical yields, in-channel transport, and the effects agricultural practices have on
them (Shields et al., 2006). Some differences between the two models include; AnnAGNPS uses
various sized designated cells, while SWAT uses HRUs; AnnAGNPS has less components and
capabilities, like pathogen transport and crop growth; and there are different methods and
simulators within the model (Parajuli et al., 2009). Although SWAT is often found to be a more
appropriate and efficient model for watersheds (Heathman et al., 2008), Parajuli and others
(2009) observed that AnnAGNPS can be used in simulating surface runoff and sediment yield in
‘impaired waters’ within a CEAP watershed in Kansas. Other studies have observed similar
results with AnnAGNPS and it has been used to show how conservation practices can reduce
sediment load, including Yuan and others (2008) who observed a 77% and 64% reduction of
sediment load when converting crop land to no-till soybeans and no-till cotton respectively.
1.7.1.3 APEX
When trying to simulate environmental and hydrologic impacts from agricultural
practices on a whole farm or smaller watershed scale, the Agricultural Policy/Environmental
eXtender model, APEX, is a viable option (Williams et al., 2008; Harman et al., 2004). APEX
was developed from the EPIC model and can evaluate erosion, economics, water quantity and
quality, soil quality, plant competition, weather, and pests, along with routing water, sediments,
nutrients, and pesticides (Williams et al., 2008). The outputs from APEX can then be put into
SWAT and be further evaluated (Duriancik, 2008). Harmon and others (2004) used APEX to
look at Atrazine loss related to different conservation practices on a crop farm and found several
practices, including filter strips and constructed wetlands, which reduced the amount lost. Both
runoff and sediment yield have also been shown to be reduced through different tillage systems,
using a validated and calibrated APEX model (Wang et al., 2008).
42

1.7.1.4 CONCEPTS
The Conservation Channel Evolution and Pollutant Transport System (CONCEPTS) is a
model developed for simulating channel hydraulics, morphology, and sediment transport
(Langendoen, 2000; Shields et al., 2006). This model can predict in-channel processes along
with riparian processes, including simulating streamside riparian vegetation’s effects on stream
morphology and pollutant loading (Bingner and Locke, 2009), along with stream bank erosion
and channel widening. Outputs from this model can be used within CEAP to see the effects of
varying riparian area and stream bank activities. Shield and others (2006) used CONCEPTS to
show the effects of deforestation on bed sediment gradation and percent gravel/cobble content.
Other studies used CONCEPTS with AnnAGNPS to identify sediment sources and flows
through different management and climatic scenarios (Kuhnle et al., 2005).
1.7.1.5 REMM
The earlier mentioned models have all been used and integrated with the Riparian
Ecosystem Management Model (REMM) in looking at riparian buffers and there efficiency at
reducing sediments and nutrients (Liu et al., 2007; Yuan et al., 2007; Langendoen et al., 2009).
REMM, developed by USDA, helps to simulate riparian buffers based on three zones and their
effects on, surface and subsurface hydrology; sediment and nutrient transport, removal, and
cycling; and buffer conditions, like vegetation type, size, slope, and harvesting (Lowrance et al.,
2000; Shields et al., 2006). Yuan and others (2007) used REMM to simulate water and sediment
movement and reduction along a riparian buffer, given inputs of sediment and water loading
from AnnAGNPS model outputs. REMM was also used with CONCEPTS to show reduction of
bank erosion rates with woody vegetation and coarse rooting systems (Langendoen et al., 2009).
Other significant uses that can apply to CEAP include large woody debris and leaf litter inputs,
43

which are important habitat components for fish and macroinvertebrates. These too can be
obtained with the help of REMM modeling (Shields et al., 2006).
1.7.1.6 PHABSIM
The Physical Simulation Model (PHABSIM) is also a hydraulic model. However,
PHABSIM relates changes in stream flow, including surface elevation and velocities, to
quantities of physical habitat, based on weighted useable area (WUA). Although PHABSIM
cannot relate hydrology directly to biota measures, like abundance or diversity, it can be
predictor of available physical habitat through defined hydraulic parameters and habitat
suitability criteria (Casper et al., 2011; USGS, 2010). This model uses the same approach as
IFIM, Instream Flow Incremental Methodology, using a curve of WUA over discharge, and can
be used in developing standards for withdrawals. Applications of PHABSIM have also been used
in urban rivers to look at modifications to the stream channel and effluent discharges (Booker
and Dunbar, 2004). However, some studies have shown little relations between WUA and
population parameters of fish (Nuhfer and Baker, 2004) and that PHABSIM carries a great deal
of uncertainty (Williams, 2010).
1.7.1.7 AQUATOX
Although there is little modeling available to predict all the ecological processes within a
system, because of the complexity, the U.S. EPA’s AQUATOX model is one of the few that
comes close. AQUATOX is a simulation of aquatic systems, including the processes, direct and
indirect effects, and relationships between pollutants, nutrients, sediments, dissolved oxygen,
periphyton, macroinvertebrates, fish, and many other components. This is done with the use of
450 equations within the model, system linkages with SWAT and EPA-BASINS, and uncertainty
analysis (Park et al., 2008). Carleton and others (2009) used AQUATOX to develop nutrient and
44

sediment criteria and thresholds of impairment to lotic systems, using biological indexes. Food
web modeling, bio-magnification of PCBs, and the recovery for specific species have also been
characterized and simulated through AQUATOX (Rashleigh et al., 2008). The Dominant
pathway for PCB’s was also observed in this study showing its course from detritus to daphnia to
shad to largemouth bass. Other known AQUATOX applications include the assessments of
pesticides, dissolved oxygen, nutrients, sediments, periphyton biomass, and fish dynamics (Park
et al., 2008).

45

INTRODUCTION TO METHODOLOGY AND RESULTS
This thesis is in the form of two research papers that have been submitted to scientific
journals. The first paper, entitled “Exploring Relationships between In-Stream Conditions and
Ecological Health while Assessing Land-use and Climate Scenarios”, aims to employ watershed
model results to obtain in-stream flow and water quality data and fill a critical gap in data
collection. This data was then used to describe and estimate fish index of biological integrity
(IBI) and the percent of intolerant fish individuals through series of models representing
ecological health at un-sampled stream reaches within the Saginaw River watershed in Michigan.
The ecological health models were then used to predict historical reference conditions for
streams under pre-settlement landuse and climate scenarios. The process started by setting up a
Soil and Water Assessment tool (SWAT) model to generate high-resolution flow and water
quality data for all reaches within the study area. The model was calibrated and validated for
stream flow, sediment concentrations, total nitrogen concentrations, nitrate concentrations, and
total phosphorus concentrations. Relationships between estimated flow and water quality
variables obtained from the watershed model and stream health measures were described using
statistical methods and fuzzy logic techniques based on 10-fold cross validation. The best
performing model to predict the Index of Biological Integrity (IBI) and percent intolerant
individuals was then used to forecast the ecological health measures under pre-settlement
conditions. Next, pre-settlement and current scenarios were compared to evaluate the potential
impacts that landuse and climate change may have on IBI and the percent of intolerant
individuals, as well as demonstrate the applications of linking ecological health to a watershed
model.

46

The second paper, entitled “Study and Model the Effects of Conservation Practices on
Stream Health”, aims to complement the first paper by estimating additional ecological health
measures using in-stream flow and water quality variables and ultimately study the impacts of
agricultural conservation practices on overall stream health measures. This paper, specifically,
focuses on connecting in-stream variables to macroinvertebrate measures (Family index of
biological integrity, Hilsenhoff Biotic Index, and number of Ephemeroptera, Plecoptera, and
Trichoptera taxa). Model setup, calibration, and validation were performed in the same matter as
the first paper. Relationships between model outputs and macroinvertebrate measures were
described using statistical methods and adaptive neural fuzzy inference system. The best
performing model was selected based on the 10-fold cross validation for each macroinvertebrate
measure and then used to forecast the potential impacts of agricultural best management practice
(BMP) implementation in a large-scale watershed. Three BMPs (no-tillage, residue management,
and native grass) were physically represented within the SWAT model. The BMP scenarios were
compared through statistical significant analysis and spatial representation of stream segment
degradation or improvement in regards to current land use scenario.

47

EXPLORING RELATIONSHIPS BETWEEN IN-STREAM CONDITIONS AND
ECOLOGICAL HEALTH WHILE ASSESSING LAND-USE AND CLIMATE
SCENARIOS
Matt Einheuser, A. Pouyan Nejadhashemia, Lizhu Wang, Scott Sowa

1.8 ABSTRACT
Land use and other human disturbances have significant impacts on physicochemical and
biological conditions of stream systems. A good understanding of the relationships among those
factors will help aquatic resource managers to make wise decisions in protecting un-impacted
systems and rehabilitate degraded systems. The objectives of this study were to employ
watershed model results to obtain in-stream flow and water quality data and fill a critical gap in
data collection. This data was then used to describe and estimate fish index of biological integrity
(IBI) and the percent of intolerant fish individuals representing ecological health at un-sampled
stream reaches within the Saginaw River watershed. Three methods were used in connecting instream variables to fish measures including stepwise linear regression, partial least squares
regression, and fuzzy logic. The model developed using fuzzy logic showed the best
2

2

performance based on the highest R for IBI (R = 0.48) and for percent intolerant fish
2

individuals (R = 0.21) and the lowest mean square error for IBI (MSE=268) and for percent
intolerant fish (MSE=275). Overall, average annual flow rate had the strongest correlation with
IBI, whereas nutrient concentration showed the largest influence on the percentage of intolerant
individuals. Based on the best model identified from the previous section, predictions were made
for pre-settlement landuse and climate conditions. Results showed overall significantly higher
48

IBI and percent intolerant individuals under pre-settlement landuse scenario. This implies that
landuse change from pre-settlement to current has profound negative impacts on stream health.
Results also showed that including pre-settlement climate factors have strong influences on
stream flow and water quality measures that interactively affect stream health as indicated by
fish measures. These results suggest that efforts to model historic baseline habitat conditions and
to provide context for stream health assessments should include both pre-settlement land use and
climate conditions.
1.9 INTRODUCTION
Land use and other anthropogenic activities have profound effects on water resources
through complex interactions among specific land practices and lotic systems’ structure, process,
and function. Agriculture and urbanization have especially strong influence on water quality and
quantity of streams and rivers (Dale and Polasky, 2007; Zimmerman et al., 2003; Webb et al.,
2008; USGS, 2011). Several of the reported impacts include altered flow (Dale and Polasky,
2007; Shields et al., 2006; Rowe et al., 2009), temperatures (Poole and Berman, 2001; Caissie,
2006; Webb, 2008), and channel morphology. Additionally, those land uses have shown
increases in sediment loading (Rowe et al., 2009; Dale and Polasky, 2007; Zimmerman et al.,
2003), nutrients levels (Bernot et al., 2006), and pesticide concentrations (Gillium, 2007).
Consequently, those influences may impact biological communities, such as fish and
macroinvertebrate assemblages, and hence aquatic system health.
A healthy stream can be described in several ways. In general it refers to the condition of the
stream when it is flourishing, resilient, sustainable, and maintains its ecological and societal
values (Meyer, 1997). Numerous biological monitoring methods measure the ecological
condition of a stream and quantify its health. Biological indicators is a commonly accepted
49

technique for assessing how different communities respond to water quality issues and can often
be efficient when physical habitat (Flinders et al., 2008) and low levels of pollutants may be
otherwise hard to detect (Barbour et al., 1999; Flinders et al., 2008). The most common and
widely used biotic indicators in aquatic systems are fish and macroinvertebrate assemblages
(MDEQ, 1997; Flinders et al., 2008; Barbour et al., 1999; Infante et al., 2008; Karr, 1991; Wang
et al., 2007; Sutela et al., 2010). Fish and macroinvertebrates can react differently, vary in
sensitivity to different stressors, and mirror conditions at different scales (Griffith et al., 2009;
Lammert and Allan, 1999; Infante et al., 2008; Flinders et al., 2008; Karr, 1981). Fish have
several strengths for biological monitoring and being used as indicators. Karr (1981) explained
that fish cover many trophic levels, including piscivores, herbivores, omnivores, and
insectivores, which allows them to be a good representation of aquatic systems and its
interactions. Fishes’ long lives and mobility allow for observation of long-term effects at broad
scales (Karr, 1981; Babour et al., 1999). Often, when using fish as indicators, there is an aim to
capture the biological integrity of the system. Biological integrity, as defined by Karr and Dudley
(1981) is the ability of a system to support "a balanced, integrated, adaptive community of
organisms having a species composition, diversity, and functional organization comparable to
that of natural habitat of the region” (Karr, 1991). Biological integrity involves three main
principles; temporal and spatial scales, elements of biodiversity and processes, and that biotic
species are important components in a dynamic evolutionary and biogeographic context (Karr,
1999). In addition to biological integrity, ecological integrity also includes biochemical and
physical integrities, which refer to the condition of the physical and chemical environment (Karr,
1996).

50

In this research, we used the index of biological integrity (IBI) of fish as an indicator of
stream health, which is widely recognized and used (Lammert and Allan, 1999; Van Sickle et al.,
2004; Wang et al., 2008; Weigel and Robertson, 2007). IBI scores are calculated by summarizing
multiple metrics. The original 12 fish metrics, developed for the Midwest United States, are
broken up into three main components including species richness and composition, trophic
composition, and abundance and condition. Within species richness and composition, six metrics
focus on quantifying native, benthic, water-column, long-lived, intolerant, and tolerant species.
With the trophic composition component, three metrics look at percentages of omnivores,
insectivores, and piscivores. The last three metrics, within fish abundance and condition
component, quantify number of individuals, hybrids, and those with diseases or abnormalities
(Karr, 1991). However, over the years, some modifications to the IBI have been made to
properly assess stream conditions for specific regions (Lyons, 1992; Lyons et al., 1996; Roth et
al., 1996; MDEQ, 1997; Lammert and Allan, 1999; Wang et al., 2008).
Flow, sediments, and nutrients are among the many factors influencing biotic composition
and overall stream health. Numerous studies have linked environment variables of both inchannel and landscape of different scales to fish biological integrity. Van Sickle et al. (2004)
used regression model to evaluate interaction among land use and flow variables and fish
metrics, and a coefficient of determination of 0.37 was obtained for predicting IBI. They further
employed their models to predict IBI values based on future alternative land-use scenarios.
Wang et al. (2008) developed a disturbance index for all stream reaches in Michigan based on
environmental factors and their relationship with fish measures, including IBI. Relationships
among nutrient concentrations and stream fish and macroinvertebrate measures have also been

51

used for the development of nutrient criteria for water resource management (Wang et al., 2007;
Weigel and Robertson, 2007; Sutela et al., 2010).
Stream flow regime has strong influence on ecological processes and fish assemblages. Not
only can flow be a reflection of stream size (Lyons, 1992), but flow is a driving force that affects
temperature, sediment, nutrient, pesticide, dissolved oxygen, and other stream conditions, which
in turn affect aquatic organisms (Poff et al., 1997). Flow also affects biota in more direct ways
such as habitat availability, wash out, and diluting stressors. In addition, flow is tied to many
important life events for some fish, including spawning and recruitment, and flow regime
changes can have negative implications on these species behavior and survival (Bunn and
Arthington, 2002). For example, changes in flow can directly impact habitat areas and river
network connections (Sparks, 1995). In the review of 165 studies, Poff and Zimmerman (2010)
reported that 92% studies showed negative responses of ecological metrics to flow alterations.
Lammert and Allan (1999) also showed that variation in IBI scores was strongly associated with
flow stability.
Sediments impact the function and process of stream systems in several direct and indirect
ways. Primary production within a stream can be affected due to sediments limiting light
penetration, transporting nutrients, and covering substrate (Wood and Armitage, 1997). Izagirre
et al. (2009) observed sediment negative impacts on periphyton soon after siltation. Such an
indirect influence of sediments through periphyton can be realized by alteration of habitat
structure, reduction of food availability, and changes in dissolved oxygen levels (Griffith et al.,
2009). In addition, sediments can directly affect fish through physiological stress, such as
sediments clogging gills (Wood and Armitage, 1997) and increasing respiration rate
(Zimmerman et al., 2003). All these stresses can further impact fish assemblages by reducing
52

growth rates, causing delayed hatching, increasing in mortality, and affecting natural fish
migration (Zimmerman et al., 2003; Wood and Armitage, 1997). Overall community structure
and diversity can be influenced by suspended sediments and sediment deposition. Rowe et al.
(2009) observed significant declines in fish IBI scores from sites where substrates were
dominated by sediments. Meanwhile, effects of sediments have been observed to have more
impact on fish in warmer waters (Newcombe and Jensen, 1996).
Sediments often are accompanied by nutrients when they reach the channel. These nutrients,
although essential to organisms, can exceed levels that are needed and have negative impacts on
the individual, community, and the entire system. In fact, an excess amount of nutrients is one of
the top ranked causes for the degradation of US waters during the past decade (USEPA, 2011b).
In some situations, where nutrients are low, supplied nutrients from agriculture can have positive
effects on aquatic organisms. However, often this is not the case and there is an excess of
nutrients in the system (Miltner, 2010). One of the main effects of nitrogen and phosphorus
overload in a lotic system is the acceleration of eutrophication (Carpenter et al., 1998). Not only
can these increases in periphyton negatively affect dissolved oxygen levels, but also they can
affect primary and secondary consumers by influencing food sources and habitat structure
(Griffith et al. 2009). High levels of nitrogen and phosphorus have been observed to affect fish
assemblages. For example, Wang et al. (2006) showed that total phosphorus and total nitrogen
contributed to most of the variation in their observed fish variables, including negative
relationships with IBI, number of carnivores, omnivores, and intolerant species. This negative
relationship between nutrients and IBI was also reported by the Minnesota Pollution Control
Agency (2003) and Miltner (1998).

53

Along with the above mentioned variables, there are several other variables that have large
influences on fish communities and ecological health. These variables include stream
morphology, temperature, physical habitat variables, etc. Although these variables are
recognized as being very important, considering all of these variables are beyond the scope of
this research.
1.9.1 Ecological Modeling
Several approaches have been proposed to study the complex relationships between
stream health and human disturbance or environmental stresses. Simple models such as general
regression models have been considered reliable where there is a lack of knowledge or limited
data (Van Sickle et al., 2004). These models are relatively easy to develop and can help in
understanding how changes in independent variables affect the dependent variable. There have
been numerous studies demonstrating the use of linear regression models for ecological studies
(Van Sickle et al., 2004; Maret et al., 2010; Waite et al., 2010). Other studies have employed
multivariate and dimensional approaches like canonical correspondence analysis (Pool et al.,
2010; Wang et al., 2008) and nonmetric multidimensional scaling (Sutela et al. 2010). In
addition, the use of methods addressing the non-linear nature of the relationships have also been
increasing, including the use of regression tree analysis (Weigel and Robertson, 2007),
regression neural networks (Sutela et al., 2010), and other non-linear models like piecewise
regression (Maret et al., 2010).
One non-linear approach that is particularly suitable for modeling and exploring the
relationship of ecological processes is fuzzy logic (Marchini, 2011). Fuzzy logic is a computing
approach introduced in 1960’s by Zadeh (1965) and is a method that deals with approximate
reasoning (Huang et al., 2010), which can be data driven or executed through expert knowledge.
54

Fuzzy logic approach can be especially effective when dealing with highly variable, complex,
non-linear, and uncertain relationship of variables (Adriaenssens et al., 2004). In addition, the
fuzzy models constructed can often be easily interpreted based on their linguistic nature.
Applying to environment and ecological processes, fuzzy logic has proven to be a valid method
for developing predictive models and decision support tools (Adriaenssens et al., 2004). Fuzzy
logic has been used for water quality assessment and classification, such as construction of an
index model for surface water quality classification (Icaga, 2007) and the evaluation of
watershed conditions and creation of a decision support system (Jensen et al. 2000). Fuzzy logic
has also been used to develop predictive models for both abiotic and biotic processes. Using
fuzzy logic for flow (Zhang et al., 2009) and flood forecasting (Nayak et al., 2005) have been
shown to be reliable. Fuzzy logic techniques have been successfully used for modeling algae,
macroinvertebrate, and fish communities. Chen and Mynett (2003) illustrated the use of fuzzy
logic developed with the combination of data mining techniques and heuristic knowledge in
predicting algal biomass and eutrophication for a lake system. Adriaenssens et al. (2006)
illustrated that fuzzy models were valuable and practical for predicting the abundance of
macroinvertebrate taxa in a river basin based on conductivity, dissolved oxygen, water velocity,
and stream width. Predictions of fish assemblages and their physical habitat have also been
conducted using fuzzy logic approaches. For example, Mouton et al. (2008) demonstrated fuzzy
logic application in developing a physical habitat model for European grayling in a river system.
Jorde et al. (2000) used fuzzy logic to evaluate relationships among floodplain processes, instream habitat quality, and fish communities.

55

1.9.2 Motivation and Objectives
Linking anthropogenic disturbances in watersheds with in-stream physicochemical and
biological conditions is challenged by the lacking of high resolution and complete landscape
datasets and the selection of modeling techniques that can adequately deal with the complex and
non-linear relationships of model components (Wang et al., 2008; Sutela et al., 2010). The Great
Lakes regional river database and classification system (GLRRDACS), including all streams and
rivers in Michigan, meets such high resolution and complete landscape database needs. This
database divides stream networks into confluence-to-confluence stream reaches, with each reach
having delineated local and network catchments (Brenden et al., 2008). The local and network
catchments provide the basis for attributing landscape-scale information to the streams, which is
essential for comprehensive evaluation of watershed conditions affecting biotic or habitat
conditions of a reach.
Quantifying linkages among watershed anthropogenic factors and in-stream stressors and
biological assemblages to provide management information is focus of many studies. However,
most such studies are either simply linked coarse-scale watershed condition measures (e.g., %
watershed land use) directly with stream reach-scale biological indicators or modeled fine-scale
watershed condition measures (e.g., land use in 30x30 m cells) for only river mouth. Our study
incorporated a highly detailed SWAT model (13,831 stream reach watersheds) and connects
environmental and ecological variables at an individual homogenous reach scale.
The objectives of this research were to (1) employ the Soil and Water Assessment Tool
(SWAT) to generate high resolution flow and water quality data for all reaches within a large
watershed so that the SWAT outputs could be estimated for all un-sampled stream reaches; (2)
link stream reach scale SWAT estimates to fish measures to identify how much those estimates
56

could explain in overall stream health; (3) describe the relationships among stream health, flow,
and water quality using statistical methods and fuzzy logic techniques, and (4) show the
applications of SWAT to model beyond sampled locations and forecast conditions by predicting
historical reference conditions for streams in the Saginaw River watershed using pre-settlement
land-use and climate data.
1.10 METHODOLOGY
1.10.1 Study Area
The Saginaw River watershed, hydrologic unit code (HUC) 040802, is located in the
central eastern portion of the lower peninsula of Michigan (Figure 1). This 6-digit HUC
Watershed is approximately 1,612,266 ha and is made up of six 8-digit HUC watersheds
including the Tittabawassee (04080201), Pine (04080202), Shiawassee (04080203), Flint
(04080204), Cass (04080205), and Saginaw (04080206) watersheds that discharge into Lake
Huron. The Saginaw River watershed is the largest watershed within Michigan. This watershed
has a large amount agricultural land use along with the America's largest contiguous freshwater
coastal wetland system (USEPA, 2011c). The basin’s land cover is made up of 42.9%
agriculture, 3.7% Rangeland, 24% forest, 14.1% developed, 14.1% wetland, and 1.2% water.
Agricultural production within the region in mainly composed of corn, soybean, and
pasturelands. The study area is dominated by warm water streams that belong to three Level-3
Ecoregions (USEPA, 2011d): 21% within Northern Lakes and Forests Ecoregion (NLF), 39%
within Southern Michigan/Northern Indiana Drift Plains Ecoregion (SMNIDP), and 40% within
Huron/Erie Lake Plains Ecoregion (HELP). Segments of the Saginaw River within the Saginaw
River watershed, along with its outlet, have been designated as an area of concern with degraded

57

fisheries, fish consumption advisories, loss of recreational values, and contaminated sediments
(USEPA, 2011c).

Figure 1. Saginaw River watershed

1.10.2 Biophysical Model
Although field sampled data for quantifying the relationships between in-stream water
quality and biotic assemblages is regularly measured, this approach is expensive and cannot be
performed for a large-scale study area for a long period due to budget and logistical constraints.
58

Biophysical models play an important role to estimate water quality where it may otherwise be
missing or impractical to collect. For example, Wang et al. (2008) used nutrient estimates from
spatially referenced regressions on watershed attributes (SPARROW) model to assess stream
health. Casper et al. (2011) used the Soil and Water Assessment tool (SWAT) to simulate flows
and then predict fish habitat using Physical Habitat Simulation (PHABSIM) model. In this
present study, the SWAT model (ArcSWAT v.2.3.4) developed by the USDA-ARS Temple
Texas, was used. SWAT is a spatially explicit watershed scale model that can simulate several
hydrologic processes and is commonly used in water resources management (Neitsch et al.,
2005). The model can be used to simulate runoff, soil erosion, chemical components, and
sediment transport, along with hydrologic processes, in an aim to predict effects of land use in
watersheds. Input data to the model include, but are not limited to, weather, topography, soil, and
vegetation (Arnold et al., 1998). In addition, both scheduled fertilizer applications and
management operations can be included in the SWAT modeling process to allow for detailed
simulations of real world processes and conditions along a continuous time scale.
Within SWAT, the watershed being modeled is delineated into subbasins that contain one
stream reach and at least one hydrologic response unit (HRU). An HRU is a unit area with
unique combination of land use, soil, and slope conditions based on thresholds designated by the
user (Neitsch et al., 2005). Once hydrologic balance, sediment loads, and nutrient losses are
calculated for each HRU, each variable is then accumulated within subbasins and routed through
the stream network. The hydrologic balance is achieved within SWAT by simulating canopy
interception, evapotranspiration, precipitation partitioning, snowmelt, surface runoff, lateral
subsurface flow, infiltration, redistribution of water through soil profile, and return flow from
aquifers (Gassman et al., 2007). A complete and detailed description of the processes within
59

SWAT can be found in Neitsch et al. (2005). However, here we briefly describe the sediment
and nutrient components.
1.10.3 Water Quality Elements of SWAT
In this study, we used SWAT to calculate erosion and sediment yield using the Modified
Universal Soil Loss Equation (MUSLE). MUSLE generates erosive energy through precipitation
data, as well as simulating yields through surface runoff. In addition to these factors, runoff
volume, peak runoff rates, subbasin area, soil surface characteristics, slope characteristics, land
cover, and management conditions were also contributed to the calculations (Neitsch et al.,
2005). Once these sediments reach the channel for routing, they are controlled and transported
through the processes of deposition and degradation (Neitsch et al., 2005).
In general, different forms of nitrogen (N) and phosphorus (P) are simulated as a function
of nutrient cycles with losses occurring through crop/plant uptake, surface runoff in their solution
phases and by erodible sediments, percolation, lateral subsurface, degradation, and volatilization
(Gassman et al., 2007). The supply and demand approach is used in determining plant use of
nitrogen and phosphorus as explained by Williams et al. (1984) and Santhi et al. (2006). Both
nutrients can be added to soils through fertilization, manure, or residue applications.
More specifically, Nitrogen can be found in three major form including organic, mineral
form held by soils, and mineral form in solution. Nitrate (NO3) can be transported through
surface runoff, lateral flow, and percolation, as a function of the runoff volume and nitrates
concentration in the soil. Organic N, on the other hand, is more connected with sediment
particles and is a function of sediment loading. Once the different forms of nitrogen have entered
the stream reach, channel routing and transformations are simulated through routines developed

60

for QUAL2E stream water quality model, with organic forms susceptible to settling (Neitsch et
al., 2005; Gassman et al. 2007).
Phosphorus is modeled in three major forms, including organic, insoluble mineral and plantavailable forms. Dissimilar to nitrogen, phosphorus’ mobility is often limited and diffusion acts
as its mode for movement through soils. Both organic and mineral forms attach to soil particles
and are calculated based on sediment loadings (Neitsch et al., 2005). Once in-stream, phosphorus
is routed and transformed through the routines based on QUAL2E, also having organic forms
being lost to settling.
1.10.4 Model Setup
SWAT requires many input datasets, including topography, land use, soils type, and
climate. For topography data, a 30m resolution USGS National Elevation Dataset (NED) was
obtained from the Better Assessment Science Integrating point and Nonpoint Sources (BASINS)
software, version 4.0 (BASINS, 2007).
Both current and pre-settlement land-use layers were used for this research. The current
condition SWAT model scenario (Current) was built based on the 2009 Cropland Data Layer
(CDL) map at 56 m resolution. This was obtained from the National Agriculture Statistics
Service within the U.S. Department of Agriculture (USDA) (Johnson and Mueller, 2010). The
CDL has crop-specific land cover classifications making it ideal for the study area, which is
dominated by agricultural lands. The second SWAT model was developed using pre-settlement
land-use data, which was acquired from the Michigan Natural Features Inventory (MNFI) and
represents mid-1800’s land cover. The land uses and their percentages of total area within the
watershed our shown in Table 1.

61

Table 1. Current and pre-settlement landuse distributions
Landuse

Water
Agriculture
Forest
Urban
Rangeland
Wetland

Current
% of Area

Pre-settlement
% of Area

1.2
42.9
24
14.1
3.7
14.1

0.8
0
82
0
1.5
15.7

Soil data were obtained from the USDA State Soil Geographic dataset (STATSGO),
which are primarily designed for planning, management and monitoring in areas such as river
basins. The STATSGO map, at a scale of 1:250,000, is linked to both physical and chemical
properties of the soil that are relevant to hydrologic modeling (USDA, 1995).
High resolution stream network data were obtained from the 1:24k National Hydrography
Dataset plus (NHDPlus) that was acquired by Michigan Institute for Fisheries Research. In total,
the Saginaw River watershed was delineated into 13,831 individual subbasins, each containing
individual stream reaches that represented spatial units having homogenous physicochemical,
geomorphological, and biological features. Both the stream network and subbasins were
predefined layers added to the SWAT model. It is important to point out that this is one of the
most comprehensive SWAT models built and is the first to include such detailed subbasins and
reaches.
Climatic data for the period of 1990-2008 were obtained from the National Climatic Data
Center (NCDC). A total of 21 stations provided precipitation data while 16 stations provided
temperature data within and near the watershed. Supplementary meteorological data was
generated by SWAT weather generator (Neitsch et al., 2005).
62

A third SWAT model scenario (Presettle2) was built based on both climate and landuse
data from mid-1800s. Climate data used for the pre-settlement scenario was obtained from the
Intergovernmental Panel of the Climate Change (IPCC) 20C3M scenario, representing historic
conditions (1870-1889). The dataset is available in a GIS format by the Community Climate
System Model (CCSM) project (www.ccsm.ucar.edu) and the National Center for Atmospheric
Research (NCAR) GIS Initiative. The precipitation and temperature data were downscaled by
NCAR to a 4.5 km scale for higher resolution (Hoar and Nychka, 2008). Conversion of observed
precipitation and temperature data to historic values was performed based off of the delta method
(Fowler et al., 2007; Woznicki et al., 2011) at each weather station. Monthly 20C3M
precipitation averages are compared with observed monthly precipitation averages and delta
ratios are calculated and applied to daily observed precipitation to obtain daily pre-settlement
values. This can be seen in equation 1 below:

(

)

(1)

where, is the daily precipitation for 20C3M, is the daily precipitation of the current observed,
is the monthly average precipitation for 20C3M (1870-1889), and is the monthly average
precipitation of the current observed (1990- 2008). To calculate daily pre-settlement (20C3M)
temperatures, monthly 20C3M temperature averages were compared with observed monthly
temperature averages and additive deltas were calculated. These deltas were applied to daily
observed maximum and minimum temperature values. This is presented in Equation 2:

(

)

63

(2)

where, is the daily temperature for 20C3M, is the daily temperature of the current observed,
is the monthly average temperature for 20C3M (1870-1889), and is the monthly average
temperature of the current observed (1990- 2008). Mean monthly temperature and precipitation
over all the years from all sites representing current and pre-settlement conditions can be seen in
the Figure 2.

Figure 2. Average monthly (a) precipitation and (b) temperature
Once data were obtained, the model was set up for the Saginaw River watershed and HRUs
were defined based on dominant landuse, soil, and slope class (0 to 2%, 2 to 5%, 5 to 10%,
>10%) within each subbasin. In addition, localized management operations and crop rotations

64

were applied to the model in replacement of SWATs default values. Further detail on the
operations and rotations can be found in Love and Nejadhashemi (2011).
1.10.5 Sensitivity Analysis, Calibration, and Validation
Sensitivity Analysis was performed on the model to identify which input parameters had
the most influence on the models output. A Latin Hypercube One-factor-At-a Time (LH-OAT)
method, embedded in SWAT, was executed to perform the sensitivity analysis (Van Griensven et
al., 2006). The analysis was completed for flow, sediment, nitrogen, and phosphorus, providing a
rank of parameters to be explored during the calibration process.
The calibration and validation was performed, using observed flow, sediment, and
nutrient data for current conditions. Calibration is the process of adjusting parameters to achieve
predictions or outputs that are within the common acceptable range when compared against
actual observed data, while reducing uncertainty. The validation is performed next, using the
determined parameters from calibration in order to measure model accuracy and credibility
(Moriasi et al., 2007). Parameters that were chosen to be involved in the calibration process
were identified during sensitivity analysis and also from knowledge of the study area. The model
was run from 2000 through 2005 with a two-year-warm-up (2000 - 2001). This period was
selected since it coincides with stream water quality observations. The calibration was performed
near the outlet (Figure 3) of the watershed with monthly observed data (24 monthly flow
samples) from 2002 through 2003. First, flow was calibrated, using observed data from USGS
site 04157000. Next, water quality outputs were calibrated individually, including sediments
(SED), total phosphorus (TP), nitrate (NO3), and total nitrogen (TN). Observed data for
sediment and nutrient calibration was obtained from EPA STORET (090177). The same sites
were used to gather data for validation period that covers years 2004 through 2005. The
65

calibration parameters used to calibrate the first SWAT model scenarios were then used for the
rest of the scenarios.
Both calibration and validation was evaluated using three statistical measures; Nash2

Sutcliffe efficiency (NSE), Percent bias (PBIAS), and coefficient of determination (R ), to
minimize the influence of biased single measures on performance. To determine if calibration
and validation was satisfactory or better, we used guidelines proposed by Moriasi et al. (2007).
The NSE statistic determines the goodness-of-fit of the model output by comparing
residual variance compared to the observed data variance (Moriasi et al., 2007). The NSE shows
how well observed data versus simulated data fits the 1:1 line. NSE can range from - ∞ to 1.0,
with 1.0 being a perfect fit and can be calculated based on equation 3, below:

[

∑
∑

(

)

(

)

]

(3)

where, is the observation for the constituent, is the simulated value for the constituent, is the
mean of observed data for the constituent, and n is the total number of observations.
PBIAS is a measure of the average tendency of the simulated data to be larger or smaller
than the observed data and has an optimum value of 0.0. Positive and negative values either
represent an underestimating or overestimating model, respectively (Gupta et al., 1999; Moriasi
et al., 2007). PBIAS can be computed through equation 4:

[

∑

(

)
∑

(

)

]

66

(4)

where, is the observation for the constituent, is the simulated value for the constituent, and n is
the total number of observations..
2

Another measure for goodness-of-fit is R , which determines collinearity between
simulated and measured data. The coefficient of determination can range from 0 to 1, with
2

higher values indicating a better fit. R is a commonly used statistic; however, drawbacks
include its sensitivity to extreme outliers and its absence in addressing whether the model is over
2

predicting or under predicting (Moriasi et al., 2007; Love and Nejadhashemi, 2011). R can be
calculated with equation 5:

(

∑
√∑

(

(

) (
)

√∑

)
(

)

(5)

)

where is the observation for the constituent, is the simulated value for the constituent, is the
mean of observed data for the constituent, is the mean of the simulated data for the constituent,
and n is the total number of observations.
1.10.6 Fish Data
Fish data (IBI and percent intolerant fish individuals) were used for 193 sites (Figure 3
and Table 2); including both wadeable and non-wadeable reaches, within the Saginaw River
watershed. The data were obtained from the Michigan DNR Fish Collection System and
Michigan River Inventory databases (Seelbach and Wiley, 1997). IBI was a main focus, because
of its representation of the overall biological health. The percent of intolerant individuals was of
interest because of its direct linkage to stressors within the reach, including water quality.
67

Wadeable sites were sampled along 80 to 960 m stretches, while sample lengths of non-wadeable
sites were 1610 m. Fish sampling for these sites was performed between 1982 and 2007 and was
collected through electrofishing methods including backpack, tow-barge, and boom units
depending on stream size. Sampling was performed by single-pass methods and all
measurements were taken in the field. Data were then summarized into indicator variables giving
insight into species thermal, feeding, habitat preferences; reproductive strategies; tolerances to
stressors; and taxonomic summaries. IBI scores were calculated depending on the temperature of
the reach and the size of the rivers. A modified procedure developed by the MDEQ (1997) was
used to determine IBI scores for wadeable warmwater streams. Wadable coldwater sites’ IBI
were calculated based on procedures described by Lyons et al. (1996). For transitional or cool
water sites, IBI scores were calculated based on both methods and the higher of the two was
obtained. Non-wadeable river IBI scores were calculated based on (Lyons et al., 2001).

68

Figure 3. Location of fish data and calibration site. For interpretation of the references to
color in this and all other figures, the reader is referred to the electronic version of this
thesis

69

Table 2. Summary of the fish and SWAT output data at the 193 sites
Variable
Description
Fish Measures
Fish Index of Biotic Integrity
IBI
% of intolerant fish individual
%INTOL
In-stream Variables
StreamGrad Stream segment gradient (m/m)
Average seasonal flow (cms)
SFLOW
% Average seasonal water yield
SPercBase
contributed by groundwater (%)
Average seasonal sediment
SSED
concentration (mg/L)
Average seasonal organic
SOrgN
nitrogen concentration (mg/L)
Average seasonal organic
SOrgP
phosphorus concentration (mg/L)
Average seasonal nitrate
SNo3
concentration (mg/L)
Average seasonal ammonium
SNH4
concentration (mg/L)
Average seasonal nitrite
SNo2
concentration (mg/L)
Average seasonal mineral
SMinP
phosphorus concentration (mg/L)
Average seasonal Algal biomass
SCHLA
(chl - a) concentration (mg/L)
Average seasonal carbonaceous
SCBOD
biochemical oxygen demand
concentration (mg/L)

Min

Max

Mean

10.0000
0.00000

54.00000
11.76471

53.33641
16.68774

100.0000
95.06173

22.38380 None
17.64305 BoxCox

0.00000
0.00165
0.00000

0.00130
0.32700
0.05183

0.00250
1.71600
0.13089

0.06120
24.30000
0.61609

0.00490 BoxCox
3.66366 BoxCox
0.15216 BoxCox

0.00581

30.70000

36.43494

667.00000

53.05459 BoxCox

0.00057

0.18224

0.59930

19.14941

1.59855 BoxCox

0.00049

0.04050

0.11592

4.42445

0.34060 BoxCox

0.00523

0.54439

0.84929

6.73634

1.00706 BoxCox

0.00081

0.07111

0.30408

16.59655

1.23761 BoxCox

0.00000

0.03725

0.13397

5.20531

0.41631 BoxCox

0.00007

0.06399

0.11104

2.14759

0.19370 BoxCox

0.00000

0.00028

0.00381

0.13178

0.01436 BoxCox

0.00000

0.01094

4.36289

340.83949

26.42415 BoxCox

70

Median

SD

Transform

Table 2 (cont’d)
SDO
STN
STP
AFLOW
APercBase
ASED
AOrgN
AOrgP
ANo3
ANH4
ANo2
AMinP
ACHLA
ACBOD

ADO

Average seasonal dissolved
oxygen concentration (mg/L)
Average seasonal total nitrogen
concentration (mg/L)
Average seasonal total
phosphorus concentration (mg/L)
Average annual flow (cms)
% Average annual water yield
contributed by groundwater (%)
Average annual sediment
concentration (mg/L)
Average annual organic nitrogen
concentration (mg/L)
Average annual organic
phosphorus concentration (mg/L)
Average annual nitrate
concentration (mg/L)
Average annual ammonium
concentration (mg/L)
Average annual nitrite
concentration (mg/L)
Average annual mineral
phosphorus concentration (mg/L)
Average annual Algal biomass
(chl - a) concentration (mg/L)
Average annual carbonaceous
biochemical oxygen demand
concentration (mg/L)
Average annual dissolved oxygen
concentration (mg/L)

16.4736

190.8016

192.8333

424.32580

0.05681

1.10778

1.88664

41.74690

3.38894 BoxCox

0.00141

0.11848

0.22696

6.57204

0.51208 BoxCox

0.00280
0.00000

0.55110
0.33706

2.34492
0.38045

28.67000
0.91104

4.44611 BoxCox
0.32031 BoxCox

0.04302

38.75000

43.78935

805.80000

63.73573 BoxCox

0.00618

0.21370

0.86339

11.42676

1.69843 BoxCox

0.00179

0.04138

0.14052

1.88790

0.26776 BoxCox

0.05866

0.66349

1.09702

4.89171

0.98626 BoxCox

0.00103

0.12063

0.40243

5.25043

0.71551 BoxCox

0.00000

0.04004

0.12701

1.48670

0.23073 BoxCox

0.00024

0.04952

0.10536

0.84138

0.13093 BoxCox

0.00000

0.00187

0.01258

0.61265

0.05069 BoxCox

0.00000

0.01513

5.94347

413.85163

34.10963 BoxCox

5.65440

188.4474

193.4237

432.68067

103.4019 None

71

93.41015 None

Table 2 (cont’d)
ATN
ATP

Average annual total nitrogen
concentration (mg/L)
Average annual total phosphorus
concentration (mg/L)

0.14780

1.44068

2.48985

17.03094

2.95091 BoxCox

0.00476

0.11615

0.24588

2.36214

0.36750 BoxCox

72

1.10.7 Data Analysis
With instream conditions beings the focus of this research, water quality and water
quantity outputs for each 13,831 individual reaches were generated from the model. These
outputs include flows, sediment loads, nutrient loads, and other water quality measurements
(Table 2). Flow data collected from SWAT comprised of flow in cubic meters per second (cms),
percent of flow contributed by groundwater (PercBase), along with the stream gradient
(StreamGrad) of each segment in meters per meter (m/m). Sediment and nutrient loads simulated
were converted to concentrations (mg/L) and included organic nitrogen (OrgN), nitrate (NO3),
nitrite (NO2), ammonium (NH4), organic phosphorus (OrgP), and mineral phosphorus (MinP).
Total nitrogen (TN) and total phosphorus (TP) was also calculated by summing all their
respective forms. Along with nutrients, other water quality outputs included dissolved oxygen
(DO), carbonaceous biochemical demand (CBOD), and algal biomass based on chlorophyll A
(CHLA). An annual average over the five years (2002-2006) was calculated for all variables.
Seasonal averages were also attained to represent conditions during the time of collection of fish
data and covered a three month time period of June through August. Once SWAT variables were
gathered, data from 193 reaches were extracted to link with fish community data at those specific
stretches where it was available.
Three methods were evaluated and compared in this study in order to identify the best
model that can be used to predict ecological condition of a stream (IBI and percent intolerant fish
individuals) using water quality and quantity measures. These methods include stepwise linear
regression, partial least squares regression (PLSR), and fuzzy logic. To address some of the
assumptions for two of the methods (stepwise and PLSR), normality was tested among all the
variables with the Shapiro-Wilk and Kolmogorov-Smirnov tests. In several cases, normality was
73

not achieved and transformations were made to best address this through the Box-Cox
transformation technique (Box and Cox, 1964). Box-Cox transformation can be used to find
potential nonlinear transformation based on a log-likelihood function (SAS/STAT, 2010).
Ten-fold cross validation was performed to validate the models, compare methods,
prevent over fitting, and select the best model (Mahmood and Khan, 2009). To do so, we divided
the 193 stream reaches with fish data into 10 approximately equal size mutually exclusive
subsets. This was performed through random sampling within R statistical software v. 2.12.1.
Once subsets were created, models can be trained based on 9 subsets and validated with the other
subset. This is performed 10 times, with each subset being used for validation once. Following
all analysis, the procedure for choosing the best method was conducted by comparing the mean
2
square error (MSE) and R based on validations. For each analysis method (Stepwise, PLSR,
2
Fuzzy), the MSE and R of the test data is averaged among the 10 models. These measurements

are compared among methods and the method with the lowest average MSE and highest average
2

R was identified as the best model. All models within each method will be tested against
validation datasets to identify the best model within each method using the criteria described
above.
1.10.7.1 Stepwise linear regression
Stepwise linear regression was first performed within SAS 9.2 to identify and model the
SWAT output variables that have significant influence on the two fish measures. Simple models
such as regression models have been considered reliable where there is a lack of knowledge or
limited data (Van Sickle et al., 2004). At the same the time, linear stepwise regression is an easy

74

approach. In this modified forward-selection process, variables are added one by one, depending
on the significance of the F statistic. After a variable has been added, the stepwise process goes
back and looks at the significances of the other variables already in the model and may remove
them (SAS/STAT, 2010). Once variables were selected, a test to detect collinearity among
variables was performed. We used a conservative variance inflation factor threshold of two
(VIF<2) (Maret et al., 2010) to identify whether or not to further investigate. This further
investigation was based on eigenvalues approaching zero (Morris et al., 1986) and condition
index values with high proportions of variation. If collinearity exists, the variable would be
removed and stepwise would be rerun. Furthermore, residuals were checked for normality to
satisfy the assumption of a normal distribution of the error.
1.10.7.2 Partial Least Squares Regression
Partial Least squares regression (PLSR) was also performed to identify significant
variables that have strong influence on fish measures using STATISTICA (STATSoft, 2011).
PLSR was performed because of advantages compared to general linear regression when dealing
with ecological processes, including its ability to deal with large array of variables and to handle
the complex interaction and redundancy among them (Carrascal et al., 2009). PLSR is a
multivariate method that linearly combines several predictor variables into latent factors that
maximize the explained variance in the response variable or variables (Carrascal et al., 2009).
Because it is an extension of multiple linear regression, similar assumptions with stepwise were
made and the transformed data were used in this analysis. Although, it is an option with PLSR to
look at how a set of predictor variables affect a set of response variables, within this study we
looked at each response variable independently. This was done so that comparison among the

75

methods could be made, due to the fact that the other methods could only handle one response
variable at a time.
1.10.7.3 Fuzzy Logic
Fuzzy logic was performed within MATLAB R2009A with Fuzzy Logic Toolbox and
was choose as an alternative method that could deal with complexity, non-linearity, and
uncertainty of the data that is commonly encountered in ecological studies (Chen and Mynett,
2003). When building the fuzzy logic models, the first step was to determine what variables to be
included in the model. This can often be a common problem in ecological modeling and can be
achieved through data-driven approaches or expert knowledge (Adriaenssens et al., 2004). In
order to better describe the complexity/non-linear relationships among the variables, Spearman’s
Rho correlations were performed within STATISTICA (STATSoft, 2011). Spearman's Rho rank
correlation is a nonparametric approach for correlating two sets of measurements by their rank
and is not limited by the assumption of a normal distribution, like when computing Pearson
product-moment correlation. Based on this results, all variables showing a significant correlation
(p<0.05) to IBI and percent intolerant individuals were considered to be incorporated in the
fuzzy logic analysis. In the next step, pairs of variables that exhibited a high correlation
coefficient (r >0.7) (Waite et al., 2010) were identified and the one with the weakest correlation
with the fish measure were removed (Wang et al., 2008). Following construction of the ranked
variables from the strongest correlated to fish measures to the weakest, a cutoff value r>0.18 was
assigned. This value allowed for inclusion of multiple variables, yet still kept the number of
variables to be included in the fuzzy logic model at a manageable size. This can be an issue in
fuzzy logic analysis, because rule sets can exponentially increase with the addition of input
variables (Chen and Mynett, 2003).
76

The creation and building of membership functions and inference rules can often be the
most difficult part involved in fuzzy modeling (Chen and Mynett, 2003) and can often be
subjective (Adriaenssens et al., 2004). This is especially true when working with fuzzy logic
models based on Mamdani-Assilian methods (Mamdani, 1977). When building a MamdaniAssilian type model, both the input and output variables are incorporated into membership
functions. The membership construction approach in this study was driven by scatter plots of
input and output variables, which had already been classified based on expert knowledge. For
example, IBI was broke into membership functions based on IBI score ratings (Karr et al., 1986;
Lyons, 1992). Natural breaks and clusters within the data lead to a starting point for the range
and shape of the membership functions. Next, adjustments were made based on trial-and-error
methods in order to optimize the model, which are often the only possible way (Adriaenssens et
al., 2004). In addition, fuzzy inference rules were also based on both the data and heuristic
knowledge. Once values were transformed (fuzzification) into linguistic terms, combinations
were identified along with their respective outputs. The most observed combinations for each
outcome helped in determining if-then rules.
1.10.8 Predictions
After selection of the best model for predictions of IBI and percent intolerant fish
individuals, the models were then used with the outputs from the different scenarios within the
SWAT model. Two pre-settlement scenarios were investigated; 1) with pre-settlement landuse
and 2) with pre-settlement landuse and pre-settlement climate. Predictions from all three
scenarios (current, pre-settlement landuse, pre-settlement landuse and climate) were then
compared through paired-t test and further investigation.

77

1.11 RESULTS
1.11.1 Calibration and Validation
The SWAT model was calibrated and validated for flow, sediments, total phosphorus,
nitrate, and total nitrogen. All components were successfully calibrated and validated based on
NSE of 0.50 or greater recommended by Moriasi et al. (2007). PBIAS also showed a satisfactory
or better performance with values within the range of ±25 for flow, ±55 for sediment, and ±70
2
for nutrients. In addition, R ≥ 0.5 has been used in other studies such as Nejadhashemi et al.

(2008) and Chinkuyu et al. (2004) as one of the criteria to evaluate a satisfactory model
performance. Model performances along with calibration and validation results are presented in
Table 3.
Table 3. Calibration and validation results
Parameter

Performance
measure

Flow

NSE
PBIAS(%)
2
R
NSE
PBIAS(%)
2
R
NSE
PBIAS(%)
2
R
NSE
PBIAS(%)
2
R
NSE
PBIAS(%)
2
R

Sediment

Total Phosphorus

Nitrate

Total Nitrogen

Calibration
Period

Combined

0.72
6.02
0.74

0.85
-1.75
0.85

0.81
1.48
0.82

0.71
-10.27
0.71

0.81
0.86
0.92

0.8
-3.02
0.88

0.64
18.08
0.68

0.67
-22.68
0.71

0.67
-6.87
0.7

0.76
22.7
0.86

0.82
12.87
0.83

0.8
17.01
0.83

0.5
47.34
0.87

78

Validation
Period

0.67
38.73
0.84

0.62
42.4
0.84

1.11.2 Model Results
Once the SWAT model was calibrated and outputs were generated, all three methods of
analyses, including stepwise linear regression, partial least squares regression, and fuzzy logic
were used to explore the relationships among variables and to find the best overall predictive
models for IBI and percent intolerant percent intolerant fish individuals.
1.11.2.1 Stepwise linear regression
When performing stepwise linear regression, AFLOW (average annual flow), ANO2
(average annual nitrite concentration), ANO3 (average annual nitrate concentration), ANH4
(average annual ammonium concentration), SFLOW (average seasonal flow for months of June
through August) were identified as significant predictors for all 10 models for IBI (Table 8).
Among these variables, flow was the most influential, by consistently provided the highest
2
partial R . Half of the models chose AFLOW, while the other half chose SFLOW as a predictor

of IBI. Since collecting flow data for long periods is costly, we decided to examine the
replacement of seasonal flow instead of annual flow. Therefore, as a test, annual flow was
manually removed from the dataset before performing the stepwise regression. When compared,
2
they performed very similarly based on MSE and R . All other variables that were identified by

stepwise were average annual concentrations of nitrogen, including NO3, NO2, and NH4.
Overall, through stepwise linear regression models, in-stream variables from SWAT explained
on average 32% of the variation in IBI.
Majority of models created to predict the percent intolerant individuals by stepwise
(Table 9) identified average annual total phosphorus concentration as the most influential
2

variable, with the largest partial R in 7 out of the 10 models. As for other variables identified,
79

this differed based on the 10-fold datasets and included; ANH4 (average annual ammonium
concentration), ASED (average annual sediment concentration), ANO3 (average annual nitrate
concentrations), SNO2 (average seasonal nitrite concentrations), SFLOW (average seasonal
flow). Four models identified seasonal flows, three identified average annual sediment
concentration, and three identified annual nitrate concentrations as significant parameters.
Overall, only 12% of variation in percent intolerant individuals was explained using stepwise
regression method.
1.11.2.2 Partial Least Squares Regression
By using stepwise linear regression, many variables were removed to reduce or eliminate
the collinearity and therefore, the interaction among certain variables may have not been
captured. To address this, PLSR was performed. In all models, no more than two significant
principal components were identified (Table 10). Within these components, seasonal and annual
flows were the most important variables recognized through PLSR. Behind flow, all other water
quality variables showed fairly equal importance, with none largely standing out. This shows the
high collinearity and redundancy among the variables. Although variables were similar in
importance, annual organic forms of nitrogen and phosphorus were found to be the next most
2

important variables. Overall, the average R for PLSR models built for IBI was nearly the same
as stepwise, at 0.32. Similarly, PLSR models created for percent intolerant fish individuals
(Table 11) identified important variables similar to those selected in stepwise procedure.
Although, all variables were comparable in importance, average annual total phosphorus and
average annual organic phosphorus concentrations were the most important. On average PLSR
model explained 14% of the proportion of variation in percent intolerant individuals when
validated, being slightly higher than that of stepwise.
80

1.11.2.3 Fuzzy Logic
Although, the above methods provided insightful results that were consistent with other
studies, it is recognized that the assumption of linear relationships may not be the most accurate.
To explore this, fuzzy logic models were built based on variables identified through spearman
rank correlations (Table 14, Table 12). The variables identified and used in fuzzy logic for
explaining IBI, from most correlated to least correlated, were AFLOW (average annual flow),
AOrgP (average annual organic phosphorus, SNO2 (average seasonal nitrite), SNO3 (average
seasonal nitrate), and StreamGrad (stream gradient). Models that were built, assuming non-linear
2

relationships, on average had an R of 0.44 when validated, which explained an additional 12%
of the variation in IBI than that of stepwise linear regression and PLSR methods. Fuzzy logic
also improved MSE values, which were 17% and 15% lower than stepwise and PLSR,
respectively. The same improvement of model performance was observed when building fuzzy
logic models for the percent intolerant fish (Table 13). The variables identified and used for the
models, from the highest correlated to the lowest, included, AOrgP (average annual organic
phosphorus), ANO2 (average annual nitrite), AFLOW (average annual flow), and ACHLA
2

(average annual chlorophyll-a). On average fuzzy logic models improved R to 0.25, which is
almost 78% improvement than other methods. In addition, MSE was also found to be lower in
fuzzy logic than the other models (Table 4).
1.11.3 Model Interpretations and Selection
Overall, all the methods described above consistently revealed significant relations
between average annual flow and both measures of stream health. Average annual flow was
positively correlated with IBI. This positive relation could result from the fact that, reaches with

81

smaller average annual flows are more sensitive to extreme weather conditions and the resulting
fluctuations in habitat have been shown to be negatively related to fish community metrics and
IBI (Poff and Zimmerman, 2010; Lammert and Allan, 1999). However, this positive relation
could also be explained by the significant correlation of average annual flow and stream size and
in turn how species richness has been shown to increase with stream size (Lyons, 1992). Given
that so many instream habitat variables change predictably with stream size (Vannote et al.,
1980) it is impossible to determine with our set of predictor variables, which interpretation is
correct. Adding a measure of stream size and instream physical habitat variables along with
other measures of watershed physiographic characteristics would likely provide the necessary
context to reveal the relative residual effects of flow and water quality on IBI. However, these
variables were not included in the scope of our project, but are being investigated in a current
study building upon our research.
IBI was negatively associated with nitrogen and phosphorus concentrations. This result
is consistent with several other studies including Wang et al. (2006) who showed total nitrogen
and total phosphorus explaining a high amount of variation in fish measures, including a
negative correlation with IBI. In addition, all three methods identified phosphorus concentrations
as being significant and of great influence on percent intolerant individuals. This negative
correlation among total phosphorus and intolerant fishes is consistent with the fact that a species
is classified as intolerant based on its sensitivity to environmental degradation and poor water
quality (Lyons, 1992). Once again, similar results were observed by Wang et al. (2006) between
total phosphorus and percent intolerant species.
Based on the results (Table 4), fuzzy logic models showed greater explanatory and
predictive performance, for both IBI and percent intolerant fishes. The final fuzzy logic model
82

chose for predicting IBI (from the 10 fuzzy logic model developed for the cross validation) had
2

an average R among all validation subsets of 0.48 and a MSE of 268. The final fuzzy logic
2

model performance for the prediction of percent intolerant had an average R of 0.21 and a MSE
of 275 among all validation subsets. The best models for all methods and the variables associated
with those models are presented in Table 5.
2
Table 4. Average MSE and R amongst models from 10-fold cross validation

Method

IBI

Average Test
MSE
Stepwise Linear Regression
363.9
PLS Regression
358.4
Fuzzy Logic
303.2

Average Test
2
R
0.32
0.32
0.44

83

% Intolerant
Average Test
MSE
334.2
331.7
261.3

Average Test
2
R
0.12
0.14
0.25

Table 5. Best performing models for all methods
Method

Response Variable

Predictor Variables

R

2

MSE

Stepwise

IBI

TSFlow
LANo2

0.36

326.8

PLSR

IBI

Component 1
Component 2

.34

335.0

Fuzzy Logic

IBI

AFlow
AOrgP
SNo2
SNo3
StreamGrad

.48

268

Stepwise

%INTOL

TATP
TASED
LSNo2

.20

299.8

PLSR

%INTOL

Component 1
Component 2

.16

311.7

Fuzzy Logic

%INTOL

AOrgP
ANo2
Aflow
ACHLA

.21

274.6

1.11.4 Evaluation of Landuse and Climate Changes Impacts on Stream Health
1.11.4.1 Impacts on IBI and percent intolerant individuals classes
The best fuzzy logic models choose in the previous section were used to predict the IBI
and percent intolerant fishes of each stream reach within the Saginaw River watershed under
different climate and landuse scenarios. As it was described earlier, the current condition
scenario evaluates basin-wide stream health under 2009 CDL landuse information. The second
scenario (Presettle1) consists of mid-1800 pre-settlement landuse data, while current climate
information was used. This scenario allows evaluation of landuse change effects to stream health
84

without considering climate variability. In addition, the last scenario (Presettle2) considered both
climate and landuse information from mid-1800. To further analyze and graphically present the
results, IBI scores were divided into five classes as very poor (0-19), poor (20-29), fair (30-49),
good (50-64), and excellent (65-100) (Lyons, 1992). Because the percent intolerant fish variable
does not have a rating system in connection to stream health, we used the classes adopted from
the fuzzy logic model (class I (x=0), class II (0<x≤10), class III (10<x≤20), class IV (20<x≤40),
and class V (40<x≤100). We evaluated the impacts of landuse and climate changes on fish
metrics for stream reaches with fish data(Figures 4-5) and those without fish data, basin-wide
(Figures 8-9).
Total reach lengths within different IBI score and rating classes is presented in Figure 4
for the 193 fish sampling locations. As shown in Figure 4 for the current IBI conditions, 40% of
the total stream length was classified as good, 28% as fair, 22% as excellent, 8% as poor, and 1%
as very poor. For the Presettle1, 44% of the total stream length was classified as good, 36% as
fair, 19% as excellent, 1% as poor, and 0% as very poor. For the Presettle2, 48% stream length
was classified as good, 33% as fair, 13% as excellent, 6% as poor, and 0% as very poor.
Although the distributions for each scenario are very similar, the pre-settlement scenarios
showed decreases in lengths of less poor, very poor, and excellent classes and increases in fair
and good classes. The result of pre-settlement conditions falling into poor and fair categories is
surprising in that we expect these sites to be in undisturbed and good conditions. This is a
reflection of the limitation of the model and how flows large influence within the model can lead
to possible misinterpretation. In addition, further examination of reaches that had unclear trend
between current and pre-settlement conditions showed that the land uses had been converted
from forestlands to wetlands during the two study periods. This likely contributed to the higher
85

IBI scores observed. Total reach lengths within different percent intolerant fish classes is
presented in Figure 5 for the 193 fish sampling locations. As shown in Figure 5, the model did
not predict any reach in either class I (x=0) or class V (40<x≤100). However, the percentages of
total stream length within class IV (20<x≤40) increased by 25% for Presettle1 and 15% for
Presettle2 from Current conditions. It is apparent that incorporating pre-settlement climate data
into the model mitigates the changes from Current scenario when compared to Presettle1.

86

180000

Total Length of Reaches (m)

160000
140000
120000

100000
Current
80000

Presettle1

60000

Presettle2

40000
20000
0
0-19; Very Poor

20-29; Poor
30-49; Fair
50-64; Good
IBI Scores and Ratings

65-100; Excellent

Figure 4. IBI scores and total reach lengths among 193 fish locations

87

Total Length of Reaches (m)

250000

200000

150000
Current

Presettle1

100000

Presettle2
50000

0
x=0

0>x≤10

10>x≤20
20>x≤40
% Intolerant Individuals

x>40

Figure 5. Percent intolerant individuals and total reach lengths among 193 fish locations

88

Total reach lengths within different IBI score and rating classes are presented in Figure 6
for the entire basin (watershed). Overall, at the basin scale, a similar trend was observed to the
193 fish sampling sites with the exception of the “good” class when incorporating historic
climate data (Presettle2). Results indicated a reduction of the total reach length classified as good
by 4% when compared to Presettle1. Meanwhile, total reach lengths within different percent
intolerant individuals classes is presented in Figure 7 for the entire watershed. Once again,
similar trends were observed to the 193 fish sampling sites with the exception of the class V
under Current scenario. As it was previously discussed, the majority of the areas classified as V
are currently covered by wetlands. The conversion from forested lands (pre-settlement) to
wetlands (current) could explain the models prediction of expansion to class V under the current
conditions. This provides insight into the potential benefits of wetlands and their ability to reduce
nutrients that provide better conditions to fish and overall stream health. The full impact of this
particular landuse change however cannot be fully interpreted from the models based on the
numerous other factors that would be altered as well that were not included within this study.
Overall, the results showed increases in IBI and percent intolerant individuals under presettlement conditions. This can be expected since pre-settlement scenarios represent less human
disturbance. Results also support the importance of including historic precipitation and
temperatures data in in evaluating land-use change impacts on aquatic communities.

89

Total Length of Reaches (m)

8000000
7000000
6000000
5000000
4000000

Current
Presettle1

3000000

Presettle2

2000000
1000000
0
0-19; Very Poor

20-29; Poor

30-49; Fair

50-64; Good

65-100;
Excellent

IBI Scores and Ratings

Figure 6. IBI scores and total reach lengths within Saginaw River watershed

90

14000000

Total Length of Reaches (m)

12000000
10000000
8000000
Current
6000000

Presettle1
Presettle2

4000000
2000000
0

x=0

0>x≤10

10>x≤20
20>x≤40
% Intolerant Individuals

x>40

Figure 7. Percent intolerant individuals and total reach lengths within Saginaw River watershed

91

1.11.4.2 Statistical significance analysis
Paired t-tests were performed to evaluate if and how landuse and climate factors affected
current predictor variables and ultimately IBI and percent intolerant at the basin-scale. Before
starting the analysis, it is important to provide some evidences about the fuzzy logic models
performance under Current condition. This was done by comparing our predicted fish measures
versus the actual observed values. The predicted IBI did not show a significant difference (pvalue = 0.034) from actual IBI observations at 1% level of significance. In addition, the percent
of intolerant individuals predicted from the fuzzy logic model did not a show a significant
difference (p-value = 0.69) from actual percentages observed values. This gives further insight
into the reliability of the fuzzy logic models’ predictions.
Next, the impacts of landuse and climate changes on 193 fish data sampling locations
were evaluated. Results showed no significant changes in index of biological integrity from
either of the pre-settlement conditions to current conditions. These insignificant changes could
be because all water quality variables within the model changed in a manner that predicts
decreases in IBI. Meanwhile, flow, the most influential variable within the IBI model, showed
significant increases with a corresponding predicted increase in IBI, which may counteract the
predicted decline from the water quality variables. However, a significant change (p <.01) was
observed in IBI when comparing the two pre-settlement scenarios, implying the effects of
climate change (Presettle1 to Presettle2). This resulted in a 4% decrease in the overall average
IBI with the addition of pre-settlement climate. The significant changes between these scenarios
resulted from the significant reduction in flow (due to precipitation reduction-Figure 2) while
mixed changes in water quality variables were observed. Once again, here we see a limitation of
the model where reduction in flow leads to a reduction in IBI. When the model was built, flows
92

inclusion was more than likely based on its reflection of stream size, with lower values
correlating with lower IBI scores. With flow reductions occurring between scenarios, IBI
predictions based on the models show reductions as well. Further inclusion of other variables,
such as stream size, can mitigate this limitation in the model.
Percent intolerant individuals on the other hand showed significant decreases from presettlement scenarios (Presettle1 and Presettle2) to the current scenario, with no significant
change between the two pre-settlement scenarios. As it was discussed earlier, AOrgP and ANO2
were found to be the most influential variables for predicting percent intolerant individuals.
Therefore, the significant reduction to percent intolerant individuals resulted from the SWAT
model predicted increases in AOrgP and ANO2 (between 65 to 73 and 74 to 82 percent in
average concentration, respectively). For the last scenario (Presettle1 and Presettle2), AOrgP and
ANO2 were significantly increased (average values were raised less than 50%). Meanwhile,
ACHLA (the fourth ranked variable for percent intolerant individuals) was significantly
increased while the average value was raised by 350%. The counteracting nature of these
variables may explain the insignificant change in percent intolerant individuals between the two
pre-settlement scenarios.
In order to predict the impacts of land-use and climate changes for stream reaches having
no fish observations, the above analysis were replicated for the entire basin (13,831 reaches).
Table 7 presents the results of paired t-tests and average changes between land-use and climate
scenarios. Overall, significant differences were observed in all predictor variables along with
IBI. In general, IBI significantly decreased from pre-settlement to current conditions. The basinwide average of IBI values decreased by 7% from Presettle1 to the Current scenario. This
predicted decrease in biotic integrity can be attributed to the significant increase in nutrients and
93

other water quality measures that are often negatively associated with agricultural and urban land
uses (Table 1). When climate factors were considered, the basin-wide average IBI decreased 2%
from Presettle1 to Current conditions as a result of AFLOW increasing. This result underscores
the importance of incorporating climate information when assessing temporal changes in stream
health conditions. This result is not consistent with conventional knowledge of likely changes in
biological integrity of streams since European settlement (Trautman, 1981; Harding et al., 1998),
which underscores the need for further investigation into mechanisms behind the relations within
our models and in particular the relations of predictors, like AFLOW, with other important
contextual variables like drainage area and stream size.
For the percent intolerant fishes, all predictor variables showed significant changes. The
only exception is that the average ANO2 concentration increased 4% from Presettle2 to Current
conditions. This is a result of climate change in combination of land-use changes. In addition,
significant increase (576%) in ANO2 was observed for Presettle1 to Presettle2 scenarios. For the
percent intolerant fish models, although ANO2 did not show significant changes from Presettle2
to Current conditions, the overall percent intolerant fishes significantly decreased from presettlement to current conditions. The basin wide average of percent intolerant fish values
decreased 35% from Presettle1 to Current conditions, which can be attributed to the significant
increases in nutrients and other water quality measures. When climate condition was
incorporated, the predicted basin-wide average percent intolerant fishes increased 32% between
Presettle2 and Current conditions, which can be attributed to less increases in AOrgP and the fact
that flow is not as influential to fish for stream reaches having on poor environmental conditions.
Such water quality change is combination of several factors, including basin-wide average
concentration changes in AOrgP (33%), ACHLA (221%), and ANO2 (576%).
94

Table 6. Paired t-test and changes among scenarios based on 193 fish sampling locations

Variables

Presettle1 to
Current

Presettle 2 to
Current

Presettle1 to Presettle2

% change
IBI
PCINTONB
AFLOW
AOrgP
ANO2
ACHLA
SNO3
SNO2

p-value

% change

p-value

% change

p-value

-2
-24
6
73
82
86
13
84

0.093
<.01
<.01
<.01
<.01
<.01
<.01
<.01

2
-23
23
65
74
36
14
92

0.111
<.01
<.01
<.01
<.01
0.169
<.01
<.01

-4
-1
-18
32
46
354
-4
-49

<.01
0.764
<.01
<.01
<.01
<.01
<.01
<.01

Table 7. Paired t-test and changes among scenarios considering all reaches within the
Saginaw River watershed
Variables

Presettle1 to
Current

Presettle2 to
Current

Presettle1 to Presettle2

% change
IBI
PCINTONB
AFLOW
AOrgP
ANO2
ACHLA
SNO3
SNO2

p-value

% change

p-value

% change

p-value

-7
-35
8
77
86
80
70
85

<.01
<.01
<.01
<.01
<.01
<.01
<.01
<.01

-2
-32
24
70
4
37
91
95

<.01
<.01
<.01
<.01
0.777
<.01
<.01
<.01

-5
-2
-17
33
576
221
-69
-67

<.01
<.01
<.01
<.01
<.01
<.01
<.01
<.01

1.11.4.3 Overall stream health changes among reaches
In this section, overall stream health for individual reaches at the basin level will be
evaluated. Figures 8-10 show whether any changes were predicted for fish measures within
individual reaches. In these figures, improvement was defined as any increase in the IBI scores

95

or percent intolerant individuals, while decline was defined as any decrease in the
aforementioned scores.
As land use changed from the Presettle1 to Current conditions (Figure 8a), the IBI scores
for the majority of stream reaches (59%) in the study region decreased, while 39% of reaches
increased, and 2% reaches had no change. When changes in IBI class is considered, 30% of the
reaches declined by one class or more, 17% improved in one or more classes, and 53% did not
change. This result highlights that although IBI scores in majority of reaches changed; many are
still classified as the same and may not experience substantial biological assemblage changes.
However, it is expected that under pre-settlement conditions, substantial biological changes
would occur compared to current conditions. This once again shows limitations of the models
and possible misinterpretations that may be made without considering other variables (stream
size, components of flow, physical habitat). The percentages of intolerant fishes decreased for
66%, increased for 31%, and unchanged for 3% stream reaches from pre-settlement to current
land use conditions (Figure 8b). Such changes are presumably attributed to the conversion of
forest to agricultural and urban land uses (Table 1), which could have led to increased nutrients
that had strong influence on percent intolerant fishes.
When pre-settlement climate condition was incorporated, the percentages of stream
reaches having negative (52%) and positive (47%) IBI scores changes were similar from
Presettle2 to Current conditions (Figure 9a). In contrast, the IBI classes were unchanged for the
majority of the stream reaches (50%), while IBI classes declined for 26% and increased for 24%
stream reaches. For the study region, both precipitation and anthropogenic land use increased
from pre-settlement to current conditions. The predicted changes in biological conditions are a
result of integrated influences of precipitation and land-use changes on the variables incorporated
96

in our models. These combined impacts of landuse and climate changes on percent intolerant
individuals were stronger than IBI with 70% stream reaches showing a decrease and 29%
showing an increase between presettle2 and Current conditions (Figure 9b).
The stream health changes when evaluated Presettle1 and Presettle2 models differed
substantially (Figures 10a and 10b), emphasizing the importance of incorporation of climate
information. Overall, the addition of historic climate data resulted in 63% reaches declining and
36% reaches increasing in IBI scores. However, when considering changes in IBI class, only
22% of the reaches declined while 10% showing improvement to a higher class. Unlike IBI, 59%
of reaches increased and 41% reaches decreased in percent intolerant fishes (Figure 10b), which
was largely due to the changes in the predictor variables (AOrgP, ANO2, AFLOW, and
ACHLA).

97

Figure 8. Improvements and declines in IBI (a) and percent intolerant individuals (b) from
Presettle1 to Current.

98

Figure 9. Improvements and declines in IBI (a) and percent intolerant individuals (b) from
Presettle2 to Current

99

Figure 10. Improvements and declines in IBI (a) and percent intolerant individuals (b)
from Presettle1 to Presettle2

100

1.12 CONCLUSION
Anthropogenic activities could have profound impacts on the health of aquatic ecosystems.
To rehabilitate or prevent these impacts, there is a need to fill critical gaps where data is not
available, to enhance our understanding of the complex relationships between biotic
communities and in-stream conditions, and to explore alternative methods of for predicting the
consequences.
This study employed a highly detailed and calibrated SWAT model for the Saginaw River
basin. SWAT successfully provided information to fill a knowledge gap in un-sampled locations
by supplying high-resolution in-stream predictions, such as flow, sediment, nutrient
concentrations, which may directly or indirectly determine the health of a stream. Overall, water
quality and quantity variables obtained from SWAT explained 48% of the variation in IBI and
21% of the variation in percent intolerant individuals. Since the goal of our study is to identify
the roles of in-stream water quality stressors in influencing fish indicators, the limited variations
in fish measures explained by the SWAT model outputs are expected. This is because many
other natural and human induced factors within the watershed and stream also influence the fish
indicators. Therefore, in order to have more comprehensive predictions and explanation of
ecological measures, additional variables such as physical, chemical, and natural factors should
be considered. However, this is not always possible and it is impractical to collect this data for
every reach within a watershed. Both linear (stepwise linear regression, partial least squares
regression) and non-linear (fuzzy logic) methods were employed to describe the relationship
between in-stream variables and fish measures (IBI and the percent of intolerant individuals).
The best model was selected through 10-fold cross validation analysis for the development of a
baseline evaluation of stream health using pre-settlement landuse and historical climate data.
101

In general, similar predictors of biological condition were identified and selected to develop
stepwise linear regression, partial least squares regression, and fuzzy logic models. Average
annual flow consistently revealed the strongest relation to IBI; with annual flows generally
positively related to this measure of stream health. Due to the significant positive correlation
between average annual flow and drainage area, the ecological significance of AFLOW to fish
communities in the Saginaw River watershed is difficult to decipher. Land uses that increase runoff, like agriculture and urbanization, can lead to increases in average annual flows yet also alter
other significant components of flow (e.g., reduced base flows and increased peak flows) which
have been consistently shown to reduce biological integrity (Poff and Zimmerman, 2010). More
detailed studies of these relations are currently underway to evaluate and better understand the
relative influence of watershed and instream variables on fish community metrics. On the other
hand, the percent of intolerant individuals was more influenced by nutrient concentrations and
average annual chlorophyll-a. This is the fact that those species are identified by their inability to
inhabit waters that experience poor water quality and environmental degradation.
Among the methods considered above, non-linear fuzzy logic method performed the best.
This reemphasizes the fact that non-linear models should be used when dealing with complex
and no-linear relations such as among water quality, water quantity, and stream health. The best
models were then employed to show the applicability of using SWAT as a means of collecting
essential data for predicting fish communities. Our results imply that human activities in our
study area have largely impacted overall stream health as measured by IBI and percent intolerant
fish individuals. Such impacts are largely resulted from increases in nutrient concentrations from
the conversion of forest to agricultural or urban lands (Table 1). The inclusion of climate data is
critical for this type of analysis because climate variables have direct influences on biological
102

communities through water quantity and indirect influences through its effects on nutrient,
sediment, and physical habitat. In contrast to the above mentioned improvements of presettlement landuse, some improvements were observed under Current conditions, which were
linked to the change from forests (pre-settlement) to wetlands (current). In addition, these
conflicting results may be attributed to the above mentioned limitations of the inputs to the
model (AFLOW) without considering highly correlated natural driving variables (stream size,
drainage area) that were not within scope of this project. Although the results of this study
provide insight into possible consequences of different landuse and climate scenarios, it is
important to acknowledge the uncertainty in the process of data collection, model input, model
structure, model parameter, and model output.
This research not only identified water quality variables and stream characteristics that can
be linked to stream health indicated by fish communities, but also demonstrated the possibilities
of working with a watershed model such as SWAT, as well as alternative, non-parametric
modeling methods. With such complex interactions and relationships involved in aquatic species
and their environments, there is a need to continue to explore alternative methods for capturing
and modeling these processes. Furthermore, understanding the relationship between fish
community metrics and variables obtained from watershed models, like SWAT, provides an
effective tool to more directly study how different landuse management and human disturbances
alter biological endpoints of aquatic ecosystems.
ACKNOWLEDGEMENTS
The authors would like to thank Ms. Minako Edgar at the Institute for Fisheries Research for
her assistance in preparing GIS layers, Dr. Sasha Kravchenko and Juan David Munoz at

103

Michigan States Department of Crop and Soil Science for their assistance with statistical
analysis, Sean Wozniki for the assistance throughout the project. Support was provided by The
Nature Conservancy’s Great Lakes Fund for Partnership in Conservation Science and
Economics. Funding for this project was provided by the United States Department of
Agriculture, Natural Resources Conservation Services under the Great Lakes Northern Forest
Cooperative Ecosystems Studies Unit (CESU) Umbrella Agreement Number 68-3A75-8-86.

104

STUDY AND MODEL THE EFFECTS OF CONSERVATION PRACTICES ON
STREAM HEALTH
Matthew D. Einheuser, A. Pouyan Nejadhashemi, Scott Sowa, Lizhu Wang, Yaseen Hamaamin

1.13 ABSTRACT
Anthropogenic activities such as agricultural practices can have large effects on the
ecological components and overall health of stream ecosystems. Therefore, having a better
understanding of those effects and relationships allows for better design of mitigating strategies.
Our study began with the development of a high-resolution watershed model for the Saginaw
River basin in Michigan for generating in-stream water quality and quantity data at stream
reaches with biological sampling data. These in-stream data was then used to explain
macroinvertebrate measures of stream health including family index of biological integrity
(Family IBI), Hilsenhoff biotic index (HBI), and the number of Ephemeroptera, Plecoptera , and
Trichoptera taxa (EPT taxa). Three methods (stepwise linear regression, kernel regression, and
adaptive neuro-fuzzy inference systems (ANFIS)) were evaluated for developing predictive
models for macroinvertebrate measures. The ANFIS method performed the best on average and
2

the final models displayed the most favorable R and mean squared error (MSE) for Family IBI
2

2

2

(R =0.50, MSE=29.80), HBI (R = 0.57, MSE=0.20), and EPT taxa (R =0.54, MSE=6.60).
Results suggest that nutrient concentrations have the strongest influence on all three
macroinvertebrate measures. Consistently, average annual organic nitrogen showed the most
significant association with EPT taxa and HBI. Meanwhile, the best model for Family IBI

105

included average annual ammonium and average seasonal organic phosphorus. The ANFIS
models were then used in conjunction with the Soil and Water Assessment Tool to forecast and
assess the potential effects of different best management practices (no-till, residual management,
and native grass) on stream health. Based on the model predictions, native grass resulted in the
largest improvement for all macroinvertebrate measures.
1.14 INTRODUCTION
A healthy stream can be described as one that is resilient, sustainable, and maintains its
ecological and societal values (Meyer, 1997). Often associated with the term stream health is the
term integrity, which refers to a quality or condition that is compared to an original, undisturbed,
condition (Karr, 1996). A common and accepted technique to quantify ecological health is the
use of biological indicators for measuring how communities react to disturbances. This can be
especially efficient when pollutants and physical habitat are otherwise impractical or hard to
quantify (Flinders et al., 2008; Barbour et al., 1999). Among biological indicators,
macroinvertebrate assemblages are one of the most widely used (Infante et al., 2008; Wang et al.,
2007; Flinders et al., 2008; Barbour et al., 1999). Due to their smaller migration habits and
sessile lifestyle, macroinvertebrates can accurately reflect localized site conditions (Barbour et
al., 1999; Flinders et al., 2008; USEPA, 2009). In addition, they often respond quickly to
stressors due to sensitive life stages, complex life cycles, and varying pollution intolerances.
Macroinvertebrates represent a range of trophic levels and being that they are an important food
source for fish (Barbour et al., 1999); they act as a link in the food web connecting multiple
organisms (USEPA, 2009). Over the years, numerous indices and metrics have been developed
and used for identifying and measuring stream conditions and integrity using macroinvertebrates
as indicators. There are two main informative types: multi-metric indexes, that measure specific
106

attributes of the assemblage; and pollution-tolerance indexes, which are based on taxon-specific
tolerance values (Fore et al., 1996). An example of a multi-metric index is the Family IBI which
is made up of metrics at a family taxonomic resolution. The Hilsenhoff biotic index (HBI;
Hilsenhoff, 1987) is an example of a pollution-tolerance index based on the tolerance values of
organic pollution for species and genera (Hilsenhoff, 1988; Barbour et al., 1999). Often, metrics
within indices include information on certain taxonomic orders of macroinvertebrates that are
highly intolerant to pollutants and can represent environmental conditions quite well. The three
common orders often looked at in rapid bioassessments are Ephemeroptera (mayflies),
Plecoptera (stoneflies), and Trichoptera (caddisflies), also known at EPT. These three orders of
macroinvertebrates are very sensitive to degradation and their occurrence in streams can be a
great indication of the water quality and overall health of the stream (Michigan DEQ, 1997).
Anthropogenic disturbances, such as agricultural production, can affect the ecological
processes and biota within lotic systems by altering flows, introducing point and non-point
source pollutants, modifying geomorphology, and influencing other components of aquatic
habitats (Wang et al., 2008; Dale and Polasky, 2007; Rowe et al., 2009). Such alteration can have
large impacts on macroinvertebrate communities and stream health in general (Bedoya et al.,
2011).
Flow is a primary characteristic of lotic systems and can be altered by different land uses.
In an indirect manner, flow can affect other characteristics including temperature, sediment,
nutrient and pesticide input, and dissolved oxygen levels. These conditions in turn affect aquatic
organisms. Flow can also affect the biota more directly, including habitat availability and
downstream drift (Bunn and Arthington, 2002; Borchardt, 1993). Abundance and diversity of

107

macroinvertebrates were found to be commonly reduced in studies with flow alterations (Poff
and Zimmerman, 2010).
With variable flows, higher peak rates, and higher velocities, often an increase in the
transport of sediments to a channel occurs. Bare soil and cultivated lands can contribute most of
this sediment to the channel (Zimmerman et al., 2003). These increases in sediments can impact
the ecology of stream in several direct and indirect ways. Sediments can affect
macroinvertebrates indirectly by affecting primary production (Wood and Armitage, 1997).
These influences on primary production can be linked to habitat structure alteration, food
availability, and dissolved oxygen levels (Griffith et al., 2009). Studies have also shown that
sediment impacts macroinvertebrate abundance, diversity, density, and community structure due
to covered and modified substrates, affecting their respiration, impairing feeding, and increasing
drift (Wood and Armitage, 1997).
Often bound to sediments, nutrients are also a significant factor originating from
agriculture activities within a watershed. Increases in nitrogen and phosphorus are well
documented in their association with agricultural practices (Bernot et al., 2006), especially the
use of fertilizers (Robertson and Vitousek, 2009; Gentry et al., 2007). These nutrients, although
essential to organisms, can exceed needed levels and have considerable negative impacts on the
biological communities and overall health of stream systems, commonly through eutrophication
(Carpenter et al., 1998). Macroinvertebrates have shown negative responses to nutrients as
demonstrated by Miltner and Rankin (1998), who observed a decrease in the Invertebrate
Community Index (ICI) score associated with NH3. It was also shown that EPT taxa decreased,
while some other taxa increased, with higher nutrient levels.

108

Several other factors can shape macroinvertebrate communities, including water quality
parameters such as dissolved oxygen (DO) and physical habitats such as substrate and cover
(Bedoya et al., 2011).
1.14.1 Approaches to Linking Environmental Variables and Biota
Efforts to link disturbances and stream variables to aquatic macroinvertebrates have been
made through multiple approaches. These range from simple linear models to complex soft
computing techniques such as fuzzy logic and artificial neural networks. Simple linear models,
such as stepwise regression are commonly used and are considered to be reliable when there is
lack of knowledge of the relationships or processes being examined (Van Sickle et al., 2004).
Waite and others (2010) used linear regression when looking at the relationship between
environmental variables and macroinvertebrate metrics as well as Maret et al. (2010) and Van
Sickle et al. (2004) who also employed standard regression models. Other authors such as
Cheimonopoulou et al. (2011) have used multidimensional approaches like Canonical
Correspondence Analysis (CCA) and Principle Component Analysis (PCA) to study the
relationships between macroinvertebrate communities and stressors. Meanwhile increasing
attention has been brought to addressing these relationships in a non-linear fashion. These
methods include piecewise linear regression (Maret et al., 2010), regression tree analysis (Wang
et al., 2007), and soft computing techniques such as artificial neural networks (Dedecker et al.,
2004).
Fuzzy logic is another soft computing method based on approximate reasoning that has
been increasingly used in ecological studies (Marchini, 2011). Fuzzy logic, introduced by Zadeh
(1965), is an effective approach for developing models used for prediction or decision support
(Adriaenssens et al., 2004). Fuzzy techniques can be interpreted easily because of its linguistic
109

nature and it efficiently deals with highly variable, complex, and uncertain data and processes
(Adriaenssens et al., 2004). Both data driven and expert knowledge techniques have been used
and several fuzzy methods exist. Fuzzy logic have been shown to be reliable for modeling
several aquatic biota related to stream health including algae (Chen and Mynett, 2003), fish
(Mouton et al., 2008; Jorde et al., 2000), and macroinvertebrates (Adriaenssens et al., 2006; Van
Broekhoven et al., 2006).
1.14.2 Challenges and Objectives
The first challenge when linking disturbances and in-stream water quality and quantity
conditions to ecological health is the lack of field data. There is not only a lack of high resolution
data, but it is often impractical and impossible to have complete and long term datasets for
streams networks, especially in a large basin as in this study. However, biophysical models can
play an important role in successfully filling this knowledge gap as well as forecasting
hypothetical conditions. Some studies have partially incorporated such models to help fill this
role including Wang et al. (2008) who used nitrogen and phosphorus estimates from a spatially
referenced regressions on watershed attributes (SPARROW) model to link with ecological
measures. It is very common for investigators to use the types and percentages of different land
uses in the watershed as predictor variables for biological measures of stream health. However,
using these variables in a region where watersheds are mainly dominated with one land-use type
(e.g. agriculture) can lead to poor predictions (Heitke et al., 2006; Stauffer et al., 2000; Johnson
and Host, 2010). In addition, relations base on the percentages of different land uses is
insensitive to climatological variables, which has an important influence on stream health. In
order to solve these problems, instead of using percentages of different land use or cover, the
geospatial information about land cover was incorporated in the Soil and Water Assessment Tool
110

(SWAT) model to determine local and regional stream conditions that are essential in
development of biological model. This also addressed the challenge of capturing the
confounding interactions among landscape factors that help in forming aquatic ecosystems
(Johnson and Host, 2010). Another challenge is finding a suitable modeling technique or method
that can capture the complex and non-linear relationships that exist between stream conditions
and ecological components (Wang et al., 2008; Sutela et al., 2010). Alternative soft computing
methods, such as fuzzy logic, may provide a possible answer for this challenge (Marchini, 2011).
Overall, the relationships between variables such as nutrients and macroinvertebrates need to be
further investigated and have not yet been well documented (Wang et al., 2007). Currently, there
are efforts through the Conservation Effects Assessment Project (CEAP) to quantify the
ecological benefits of best management practices (BMPs) and conservation practices through
modeling applications and tools (Maresch et al., 2008; Shields et al., 2006).
The objectives of this study were to: (1) bridge the gap between hydrologic models and
ecological conditions using the (SWAT) as a means of quantifying high resolution flow and
water quality elements; (2) identify the influential variables in explaining macroinvertebrate
indices and metrics, ultimately indicating overall stream health; (3) describe the relationships
between aquatic macroinvertebrates and local in-stream conditions (SWAT outputs) through
models built using statistical methods and adaptive neuro-fuzzy inference systems (ANFIS)
technique; and (4) show the applicability of SWAT coupled with the predictive
macroinvertebrate models to forecast and assess the effects of numerous agricultural BMPs on
stream health within the Saginaw River watershed.

111

1.15 METHODOLOGY
1.15.1 Study Area
The Saginaw River watershed (or basin) is Michigan’s largest 6-digit HUC (hydrologic
unit code 040802) basin with a size of 612,266 ha (Figure 11). Located in the central eastern
portion of the Lower Peninsula, this watershed consists of six 8-digit HUC watersheds, the
Tittabawassee (04080201), Pine (04080202), Shiawassee (04080203), Flint (04080204), Cass
(04080205), and Saginaw (04080206). The outlet of the Saginaw River basin discharges into
Lake Huron. Streams and rivers within the basin range in size and order; however, majority of
the basin is dominated by warm water streams. The basin lies within three Level-3 Ecoregions,
the Northern Lakes and Forests Ecoregion, the Southern Michigan/Northern Indiana Drift Plains
Ecoregion, and the Huron/Erie Lake Plains Ecoregion (USEPA, 2011b). The Saginaw River
watershed consists of approximately 43% of agricultural lands that are mainly comprised of corn,
soybean, and pasture fields. The remaining lands are 24% forest, 14% developed, 14% wetland,
4% rangeland, and 1% water. With a large percentage of its drainage area being agriculture and
developed, segments of river as well as the outlet of the basin have been impaired due to nutrient
loading, soil erosion, and contaminated sediments. Because of this, it has been identified as an
area of concern by the US EPA. These concerns are in regards to degraded fisheries, fish
consumption advisories, and loss of recreational values (USEPA, 2011a).

112

Figure 11. Saginaw River watershed (040802)
1.15.2 Soil and Water Assessment Tool
To obtain much needed water quality and quantity data at a large scale, an inexpensive
and informative approach is the use of models. The Soil and Water Assessment Tool (SWAT) is
one such model that is commonly used for watershed scale predictions that may be used by
decision makers and managers for dealing with water resource issues (Neitsch et al., 2005).
SWAT (ArcSWAT v.2.3.4) was a spatially explicit model developed by the USDA-ARS Temple
Texas. This physically based model was developed to simulate runoff, stream flow, soil erosion,

113

along with nutrient, pesticide, and sediment loadings and transportation based on input data
including landuse, weather, topography, soils, and scheduled management operations (Gassman
et al., 2007).
The SWAT model and watershed being simulated, at its basic unit, is made up of
hydrologic response units (HRUs) that represent areas with unique land use, soil, and landscape
slope (Neitsch et al., 2005). For each HRU, a hydrologic balance, sediment loads, and nutrient
losses are calculated and then accumulated within subbasins that have been delineated
throughout the watershed. At this time, routing of each variable occurs through the stream
network within the model.
1.15.2.1 Model Setup
In order to build an accurate and reliable model within SWAT, the use of accurate and
reliable input data is a must. As mentioned earlier, this input data includes, but is not limited to
landuse, topography, soils, and climate. For land-use data, the 2009 Cropland Data Layer (CDL)
map at 56-m resolution, acquired from the National Agriculture Statistics Service within the U.S.
Department of Agriculture (USDA) (Johnson and Mueller, 2010) was used. Classifications
within this layer are of a high resolution and are crop specific making it ideal for this dominated
agricultural watershed.
For topography data, the USGS National Elevation Dataset (NED) at 30-m resolution
was obtained through the Better Assessment Science Integrating point and Nonpoint Sources
(BASINS version 4.0) software (BASINS, 2007).
For soil data, the USDA State Soil Geographic dataset (STATSGO) at a scale of
1:250,000 was used. This data includes both physical and chemical properties for hydrologic
114

modeling, which is commonly used for management in large areas such as watersheds and basins
(USDA, 1995).
Precipitation and temperature data was obtained from the National Climatic Data Center
(NCDC). Within and around the basin, 21 stations and 16 stations provided precipitation and
temperature data, respectively. The acquired climate data spanned from 1990 to 2008. Additional
climatic data, including wind speed and relative humidity, was obtained by SWAT weather
generator (Neitsch et al., 2005).
Typically, watershed boundaries and stream networks are delineated and defined based
on the topography data. However, for this study, both the stream network and subbasins were
predefined layers inputted into the SWAT model. These layers were created using stream
network data from a 1:24k National Hydrography Dataset plus (NHDPlus), acquired by the
Michigan Institute for Fisheries Research. The 13,831 predefined subbasins and individual
stream reaches are presumably homogenous in physicochemical, geomorphological, and
biological characteristics.
Further setup of the model included defining HRUs based on dominant land use, soil, and
slope class (0 to 2%, 2 to 5%, 5 to 10%, >10%) within each subbasin. Additionally, management
operations, schedules, and crop rotations were modified from SWAT default values, as presented
by Love and Nejadhashemi (2011) for the study area.
1.15.2.2 Sensitivity Analysis, Calibration, and Validation
Following the model setup, sensitivity analysis and calibration was performed. Sensitivity
analysis was used to first identify input parameters within the SWAT model that had the most
influence on model outputs. This was done for model outputs including flow, sediments,
115

nitrogen, and phosphorus. SWAT employs a Latin Hypercube One-factor-At-a Time (LH-OAT)
method when performing sensitivity analysis (Van Griensven et al., 2006).
Calibration was then performed using parameters identified through sensitivity analysis
as well as parameters known to be sensitive based on knowledge of the study area. Calibration
was executed by adjusting those parameters in efforts to obtain outputs that meet an acceptable
accuracy when compared to actual data, while reducing the uncertainty. In addition, validation
was performed by using the adjusted parameters from calibration to determine model accuracy
and reliability (Moriasi et al., 2007). Both calibration and validation was performed at the basin
outlet (Figure 12) for flow, sediments, and nutrients. The model was run from 2000 through 2005
with the first two years being warm up years. This time frame was chosen according to
availability of measured streams and habitats data. In addition, the time frame includes both wet
and dry precipitation periods. Calibration and validations were performed on monthly basis.
Calibration period was 2002-2003 while the validation period was 2004-2005. Flow was
calibrated and validate against data from USGS (site # 04157000). In addition, the model was
calibrated and validated for water quality elements including total phosphorus, nitrate, and total
nitrogen based on observed data acquired from EPA STORET (site # 090177).
Evaluations of both calibration and validation were done using Nash-Sutcliffe efficiency
2
(NSE), percent bias (PBIAS), and coefficient of determination (R ). The evaluations and the

performance of the model were based on guidelines outlined in Moriasi et al. (2007). Multiple
measures were used to minimize biased results that might arise from using only one performance
measure. NSE is a commonly used performance measure in hydrologic models and reflects how
well simulated data versus observed data fit a 1:1 line. PBIAS on the other hand helps in

116

2

showing if the model is over-predicting and under-predicting. Meanwhile R is a commonly used
2

statistic that measures the goodness of fit between simulated and observed data. R however, can
be over sensitive to outliers and does not help in addressing over or under predicting of the
model (Moriasi et al., 2007; Love and Nejadhashemi, 2011). Further information and details on
the measures and evaluation guidelines used in this study can be found in Moriasi et al. (2007).
1.15.2.3 BMP Scenarios
Multiple best management practices (BMP) ,1) no tillage (NT),2) residue management
(RM 1000) (1000 kg/ha), and 3) native grass (NG) were simulated using SWAT model (applied
to all agricultural lands) to forecast the potential effects of these scenarios on the three
macroinvertebrate measures of stream health.
The no-tillage BMP refers to practice of keeping soils untilled and undisturbed until
agricultural planting. This not only reduces erosion of soils, but it also can increase soil
moisture.(USDA, 2010). To represent this in the SWAT model, custom operation schedules were
made to halt tillage operations.
Residue management refers to the practice of keeping a certain amount of crop residue on
the soil surface within an agricultural field. This can reduce both sheet and rill erosion, while
improving soil conditions (USDA, 2010). To model this within SWAT, no-till operations were
incorporated as well as modifying parameters including average moisture condition curve
number (CN2 (-2)), universal soil loss equation support practice factor (USLE_P (0.39)), and
Manning’s n value for overland flow (OV_N (0.2)) (Arabi et al. 2007). The modified values are
based on the amount of residue to be left, which was 1000 kg/ha for this study.

117

Native grass was the third BMP incorporated into the model and refers to the practice of
replacing agricultural row crops with native grasses (e.g. big bluestem) as done in programs such
as Conservation Reserve Program. This conservation practice can reduce sediment and nutrient
transport within watersheds. To represent this in the SWAT model, a custom operation was used
where row crops were converted to rangeland (Woznicki et al., 2011) and mowing was done
every 5 years according to the information obtained from the NRCS field office.
1.15.3 Macroinvertebrate Data
Macroinvertebrate data, including HBI, family IBI, and total EPT taxa, were obtained
from 262 sites (Figure 12 and Table 15) within the Saginaw River watershed. Both Family IBI
and HBI are indices of the overall stream health of the system while the total number of EPT
taxa reflects stream health based on those macroinvertebrates’ intolerance to degraded water. All
the macroinvertebrate data was obtained from the Michigan Department of Environmental
Quality (DEQ) and collected through standard DEQ procedures (Michigan DEQ, 1997).
Sampling was performed in all habitat types and in both high velocity and low velocity areas.
Methods of sampling included hand picking and dip nets, meanwhile samples were identified to
the family taxonomic level. Metrics were then calculated and index scores were obtained. Metric
measurements included taxa richness, numbers of taxa in the orders of Ephemeroptera,
Trichoptera, and Plecoptera, percent Ephemeroptera composition, percent Trichoptera
composition, percent contribution of the dominant taxon, percent isopods, snails, and leeches,
and percent surface dependent macroinvertebrates. Ephemeroptera taxa are pollution sensitive
and are often the first groups to disappear at impacted sites, Trichoptera taxa are often a
predominant component within the community and few species are tolerant to pollution, and
Plecoptera are sensitive to environmental quality (Michigan DEQ, 1997).
118

Figure 12. Location of macroinvertebrate data and calibration site
1.15.4 Data Analysis
Analysis of the data began with gathering and calculating in-stream concentrations and
conditions, both water quality and quantity variables, from the output of the watershed SWAT
model. The in-stream variables (flow, sediment loads, nutrient loads, etc.) generated from the
model was obtained for each of the 13, 831 individual reaches within the basin. The annual and
seasonal (June through August) averages were calculated over the five years when the model was
run (2002 to 2006). In addition, the bankfull cross sectional area of the reach as a trapezoidal
shape was attained based on bankfull depths and widths calculated through SWAT for each
stream segment. Flow related measurements included discharge (m3/second, flow), the

119

percentage of flow that is driven by groundwater (PercBase), and stream gradient (StreamGrad)
of the reach (m/m). Sediment and nutrient concentrations (mg/L) were then calculated based on
the loads and discharge within each reach. Nutrient forms and values that were obtained from
SWAT included organic nitrogen (OrgN), nitrate (NO3), nitrite (NO2), ammonium (NH4),
organic phosphorus (OrgP), mineral phosphorus (MinP), total nitrogen (TN), and total
phosphorus (TP). In addition, dissolved oxygen (DO), carbonaceous biochemical demand
(CBOD), and algal biomass based on chlorophyll A (CHLA) concentrations were acquired from
the model. These water quality and quantity variables along with the natural variable of stream
cross-sectional area were then specifically extracted for 262 reaches where the macroinvertebrate
data was available (Table 15). At these sites, the relationship between the above mentioned
variables and the macroinvertebrate measures were then investigated. Finally, three methods
were evaluated to explain these relationships.

120

Table 8. Summary of macroinvertebrate and SWAT output data at 262 sites
Variable
Description
Macroinvertebrate Measures
Macroinvertebrate Biotic Index
HBI
Family Index of Biotic integrity
FamilyIBI
Total number EPT taxa
EPTtaxa
In-stream Variables
Cross-sectional area of reach
CrossArea
(m2)
StreamGrad Stream segment gradient (m/m)
Average seasonal flow (cms)
SFLOW
% Average seasonal water yield
SPercBase
contributed by groundwater (%)
Average seasonal sediment
SSED
concentration (mg/L)
Average seasonal organic
SOrgN
nitrogen concentration (mg/L)
Average seasonal organic
SOrgP
phosphorus concentration
(mg/L)
Average seasonal nitrate
SNo3
concentration (mg/L)
Average seasonal ammonium
SNH4
concentration (mg/L)
Average seasonal nitrite
SNo2
concentration (mg/L)
Average seasonal mineral
SMinP
phosphorus concentration
(mg/L)
Average seasonal Algal biomass
SCHLA
(chl - a) concentration (mg/L)

Min

Max

Mean

3.60541
0.00000
0.00000

7.89024
36.00000
17.00000

5.10950
16.07634
6.13691

5.02465
15.0000
6.00000

0.68171 BoxCox
7.80202 None
3.82337 None

0.02100

573.0581

51.79563

14.0755

100.438 BoxCox

0.00000
0.00066
0.00000

0.03140
25.00000
0.58796

0.00179
1.87672
0.09320

0.00120
0.39700
0.04437

0.00267 BoxCox
4.04417 BoxCox
0.11958 BoxCox

0.08670

518.0000

41.41481

34.5000

44.5916 BoxCox

0.00294

15.42615

0.84168

0.29599

1.52020 BoxCox

0.00114

2.75150

0.16056

0.05695

0.28803 BoxCox

0.00079

10.76602

1.11424

0.72740

1.32166 BoxCox

0.00073

4.77614

0.35343

0.11764

0.61962 BoxCox

0.00000

2.21420

0.15899

0.05961

0.24940 BoxCox

0.00003

1.23628

0.13475

0.08921

0.15353 BoxCox

0.00000

1.04482

0.01192

0.00065

0.08284 BoxCox

121

Median

SD

Transform

Table 8 (cont’d)
SCBOD

SDO
STN
STP

AFLOW
APercBase
ASED
AOrgN
AOrgP

ANo3
ANH4
ANo2
AMinP

ACHLA

Average seasonal carbonaceous
biochemical oxygen demand
concentration (mg/L)
Average seasonal dissolved
oxygen concentration (mg/L)
Average seasonal total nitrogen
concentration (mg/L)
Average seasonal total
phosphorus concentration
(mg/L)
Average annual flow (cms)
% Average annual water yield
contributed by groundwater (%)
Average annual sediment
concentration (mg/L)
Average annual organic nitrogen
concentration (mg/L)
Average annual organic
phosphorus concentration
(mg/L)
Average annual nitrate
concentration (mg/L)
Average annual ammonium
concentration (mg/L)
Average annual nitrite
concentration (mg/L)
Average annual mineral
phosphorus concentration
(mg/L)
Average annual Algal biomass
(chl - a) concentration (mg/L)

0.00000

299.6894

5.45341

0.01764

26.8655 BoxCox

13.3485

424.3258

187.8763

179.877

84.4387 None

0.28433

23.05467

2.46834

1.67509

2.86633 BoxCox

0.00428

3.26787

0.29532

0.16339

0.38360 BoxCox

0.00133
0.00000

29.53000
0.91104

2.58448
0.31717

0.70005
0.33018

4.94849 BoxCox
0.28845 BoxCox

0.21060

660.7000

49.35930

42.5650

55.0759 BoxCox

0.00618

8.74909

1.17311

0.34997

1.77460 BoxCox

0.00179

1.27287

0.18353

0.06978

0.25342 BoxCox

0.04488

6.32230

1.37998

1.25396

1.02889 BoxCox

0.00122

3.13374

0.51953

0.23141

0.62542 BoxCox

0.00000

0.98350

0.15704

0.07660

0.19189 BoxCox

0.00017

0.52431

0.13105

0.09111

0.11607 BoxCox

0.00000

0.61265

0.01836

0.00331

0.05920 BoxCox

122

Table 8 (cont’d)
ACBOD

ADO
ATN
ATP

Average annual carbonaceous
biochemical oxygen demand
concentration (mg/L)
Average annual dissolved
oxygen concentration (mg/L)
Average annual total nitrogen
concentration (mg/L)
Average annual total phosphorus
concentration (mg/L)

0.00000

468.7425

8.71597

0.01659

43.9080 BoxCox

5.93453

432.6806

182.3555

175.107

97.8748 None

0.24720

12.05519

3.22965

2.31450

2.69881 BoxCox

0.01120

1.48351

0.31458

0.17047

0.32449 BoxCox

123

These methods included stepwise linear regression, kernel regression, and adaptive
neural-fuzzy inference system (ANFIS). Ten-fold cross validation was performed among all
methods. This allows the comparison among methods, facilitates in model validation, prevents
over-fitting, and helps with selection of the overall best predictive model for each of the response
variables (Mahmood and Khan, 2009). Random sampling was performed with R statistical
software v. 2.12.1 to divide the data for the 263 reaches into 10 approximately equally size
subsets. Models and relationships were built and trained based on 9 of the subsets and validated
with the remaining subset. This is done until each subset has been used for validation, resulting
in 10 models. Performance measures including mean square error (MSE) and coefficient of
2

determination (R ) were then calculated for the test or validation datasets and averaged across
the 10 models. This average performance based on validation is then compared among the three
2

methods (stepwise, kernel, and ANFIS) and the method with the lowest MSE and highest R was
recognized as the best method. Finally, models within the best method were tested against each
of the 10 folds or subsets and their average performance dictated the best model, which was later
used for predictions.
1.15.4.1 Stepwise Linear Regression
Stepwise linear regression models were developed using SAS 9.2 for each of the
macroinvertebrate response variables (HBI, family IBI, and EPT taxa) based on the significant
in-stream variables obtained from SWAT. Although linear models are a simple approach, they
are thought to be reliable and effective in cases where knowledge or data is limited (Van Sickle
et al., 2004). Variable normalties were evaluated using Shapiro-Wilk and Kolmogorov-Smirnov
tests. Box-Cox transformations (Box and Cox, 1964), which applies non-linear transformations

124

based on log-likelihood functions (SAS/STAT, 2010), were performed for those variables that
did not satisfy the normal assumption. In this stepwise procedure, a modified forward-selection
process was performed and variables were added based on significance of the F statistic.
Following the addition of a variable, the process goes back to check if other variables need to be
removed in the model based on their significances (SAS/STAT, 2010). Additionally, the
collinearity among predictive variables was addressed by looking at the variance inflation factor
(VIF), eigenvalues, and condition index. If a variable had a VIF of greater than two (Maret et al.,
2010) and either an eigenvalue approaching zero (Morris et al., 1986) or condition index value
with a high proportions of variation, then that variable was removed and stepwise would be
rerun.
1.15.4.2 Kernel Regression
Kernel regression was performed through R statistical software v. 2.12.1 in efforts to
explain and model macroinvertebrate measures with a non-linear and more flexible regression
function (Hastie et al., 2009). Kernel methods are memory-based and use a localized weighting
function. The specific analysis for this study uses a local linear regression technique due to
computational time and it’s un-biasness at boundaries of the domain, which is unlike some other
kernel methods. Three different kernels were considered (Tri-cube, Epanechnikov, and
Gaussian) to look at different local smoothing techniques. Within each kernel, leave-one out
cross validation was performed to define the best width of the local neighborhood (Hastie et al.,
2009). Environmental variables used within the kernel regression were selected with the same
methods as the ANFIS method based on Spearman’s Rho correlations. However, for kernel
method, there is no limit of the number of variables used in the models; therefore all variables
identified in the process were used.
125

1.15.4.3 Adaptive Neural-Fuzzy Inference Systems (ANFIS)
ANFIS is a fuzzy logic approach that uses adaptive learning techniques similar to that of
neural networks and was performed in MATLAB R2009A with Fuzzy Logic Toolbox
(MathWorks, 2010). This fusion technique is suitable for very suitable when data is already
available and when a predetermined model structure is unknown. Fuzzy logic is an alternative
soft computing method that can be used when dealing with non-linear and complex interactions
(Chen and Mynett, 2003). To determine the variables to be included in the ANFIS model,
Spearman’s Rho correlations were performed within STATISTICA (STATSoft, 2011) among all
variables. This non-parametric approach identified variables that were significantly correlated
(p<0.05) with either HBI, family IBI, or EPT taxa for further consideration. Meanwhile, where
pairs of variables had a high correlation coefficient (r >0.7) (Waite et al., 2010) amongst each
other, the one with the weakest correlation with the macroinvertebrate measure was removed
(Wang et al., 2008). The top three variables were then selected to be incorporated in the model
for each macroinvertebrate measure. In addition, stream size (cross-sectional area) was also
incorporated into each model to represent the natural longitudinal variation in stream
morphology, substrate, shading, and flow regime. Cross-sectional area was forced into the
models for the methods where it was possible (kernel and ANFIS). This was in attempt to model
the residual effects of flow, which is highly correlated with cross-sectional area, on
macroinvertebrate and ultimately stream health. This is another benefit to the fuzzy approach
compared to other methods in that variables can be included where otherwise they may be
rejected. The limit of four variables was used based on number of samples, which limits the
building of membership functions without exceeding the number of modified parameters. At this
point, ANFIS builds initial fuzzy inference system parameters based on grid partitioning and

126

continues to shape membership functions and rules with training data. Meanwhile, test data and
error of the model was tracked throughout the process to prevent over-fitting. Multiple
membership function shapes were tested as well as optimization options within the program
while building models. The output’s membership function could either be linear or constant. An
example of a linear membership function may be if the inputs are in specific membership
functions, then the output is a specific linear equation. The ending model was identified based on
minimizing error of both training and test datasets. Further information and a detailed
explanation of the ANFIS function can be found in the Fuzzy Logic Toolbox 2, User’s Guide
(MathWorks, 2010).
1.16 RESULTS
1.16.1 Calibration and Validation
Calibration and validation was performed within the SWAT model for water quantity and
quality components in the following order; flow, sediments, total phosphorus, nitrate, and total
nitrogen. Based on criteria explained in Moriasi et al. (2007), all variables were successfully
calibrated and validated based on NSE (>0.50) and PBIAS (within ±25 for flow, ±55 for
sediment, and ±70 for nutrients) (Table 16).

127

Table 9. SWAT model calibration and validation summary
Parameter

Performance
measure
NSE
Flow
PBIAS(%)
2
R
NSE
Sediment
PBIAS(%)
2
R
NSE
Total
Phosphorus PBIAS(%)
2
R
NSE
Nitrate
PBIAS(%)
2
R
NSE
Total
PBIAS(%)
Nitrogen
2
R

Calibration
Period
0.72
6.02
0.74

Validation
Period
0.85
-1.75
0.85

Combined

0.71
-10.27
0.71

0.81
0.86
0.92

0.8
-3.02
0.88

0.64
18.08
0.68

0.67
-22.68
0.71

0.67
-6.87
0.7

0.76
22.7
0.86

0.82
12.87
0.83

0.8
17.01
0.83

0.5
47.34
0.87

0.67
38.73
0.84

0.62
42.4
0.84

0.81
1.48
0.82

1.16.2 Model Results
Calibrated and validated outputs from the watershed model were obtained at sites where
macroinvertebrate data were available. The three methods of analyses (stepwise linear
regression, ANFIS, and kernel regression) were than performed to explore and identify in-stream
variables that can be used to explain and predict macroinvertebrate measures (HBI, FamilyIBI,
and EPTtaxa).
1.16.2.1 Stepwise Linear Regression
Average annual organic nitrogen concentration (AOrgN) was consistently identified as
the most influential variable for HBI in the Stepwise linear regression analysis. . Throughout the
2

ten-fold process, AOrgN had the highest partial R in explaining the variation of HBI for all
models. Throughout the stepwise technique and based on correlations, organic phosphorus also
128

exhibited high correlations with HBI, but it was removed because of its collinearity with AOrgN.
Additionally, all models identified average annual sediment as a significantly influential variable
for HBI. All other variables that were incorporated in the models were some forms of nitrogen
concentrations. However, one of the models built did identify dissolved oxygen as a significant
addition to the regression equation. Because HBI is based on macroinvertebrate pollution
tolerance scores, the selection of nutrient and sediment concentrations as predictors is consistent.
When applying the linear regression models to test datasets, on average in-stream water quality
variables (SWAT outputs) explained 25% of the variation in HBI (table 18).
Stepwise regression performed for FamilyIBI also recognized nitrogen concentrations as
a highly influential variable. Based on 9 out of the 10 models created during the 10-fold cross
validation, average annual total nitrogen concentration (ATN) was found to have the highest
2

partial R . Average annual flow was also identified consistently as a significant variable. Flow,
unlike ATN however, showed a positive correlation with FamilyIBI and explained an additional
four to eight percent of the variation. In majority of the models, no other variables significantly
2

increased the R or reduced MSE. Much like, HBI, the tolerances of macroinvertebrates to
nutrients is reflected in the FamilyIBI. However, this index also takes into consideration other
metrics not just based on tolerances, which may explain the selection of flow as an influential
variable, which may be a reflection of natural longitudinal variation within the watershed. Based
on stepwise linear regression techniques, on average, in-stream variables from SWAT explained
27% of the variation in FamilyIBI (Table 18).
For EPTtaxa, average annual nitrogen concentration (ATN) was also consistently
incorporated into the models. Similar to FamilyIBI average annual flow was also identified in
129

majorities of the models. The reason these two variables were selected in both FamilyIBI and
EPTtaxa is likely due to the fact that EPTtaxa is related to several of the metrics in FamilyIBI. In
addition to the above mentioned variables, average seasonal nitrite (SNO2) was also included in
majority of the models, explaining only an additional one to two percent of the variation in
EPTtaxa. Some models built based on certain folds of data, included other forms of nitrogen (e.g.
average annual nitrate and average seasonal ammonium) as well as the concentration of
chlorophyll a. The identification of nutrient values as being predictors of EPT taxa is consistent
based on the fact that EPT taxa are known to be sensitive to water quality. Overall, through the
stepwise process, water quantity and quality variables explained 34% of the variation in EPTtaxa
(table 18).
1.16.2.2 Kernel Regression
The first step of performing the kernel regression for HBI was identifying what
parameters should be used in the models. Unlike stepwise, this analysis does not have a set
selection process. The predictors identified by our spearman rank correlation analysis in the
order of importance include average annual organic nitrogen concentration (AOrgN), average
annual nitrite (ANO2), average seasonal total nitrogen (STN), average annual and seasonal
flows, average annual water yield contributed by groundwater (APercBase), average annual
carbonaceous biochemical oxygen demand concentration (ACBOD), average annual algal
biomass (chl - a) concentration (ACHLA), and average seasonal sediment concentrations
(SSED). Within the 10-fold process, no specific type of kernel for local smoothing consistently
outperformed the others and all three methods (Tri-cube, Epanechnikov, and Gaussian) were
used depending on the dataset. When applying these “black-box” models built through kernel
regression to test data, those predictors explained an average of 19% of the variation in HBI
130

(table 18). This low predictability was unexpected because it does not assume linear relationships
and performed worse than linear regression. This result could be contributed to how the variables
were selected and the fact that the kernel method used different variables than stepwise.
Kernel regression models for FamilyIBI used similar variables that were chosen for HBI.
Once again, concentrations of different nitrogen forms were included in every model such as
organic nitrogen, total nitrogen, nitrite, and nitrate. Also ACHLA, ACBOD, ASED, and
APerceBase were included in the kernel models. In addition to the common variables that have
been involved in majority analysis, average seasonal dissolved oxygen concentration (SDO) was
also an input variable in five of the ten models. One of the models based off of a specific dataset
within the 10-fold process used average seasonal ammonium (SNH4) and average seasonal
organic phosphorus (SOrgP) as inputs, replacing AOrgP and ANO2. The kernel regression
process for FamilyIBI consistently used a tri-cube kernel for local smoothing, which
outperformed the other kernels in every model. Kernel regression models explained on average
18% of the variation in FamilyIBI (table 18). Similar to HBI, once again kernel regression did
not show favorable results.
Similar variables were used for the kernel analysis of EPTtaxa. The highly correlated
variables were comparable to that of FamilyIBI. This likeness may be due to the fact that highly
related measures to EPTtaxa, are used to calculate FamilyIBI. However, some notable
differences were apparent in EPTtaxa model inputs and spearman rank correlations. Average
seasonal dissolved oxygen concentrations (SDO) were included in all models built based on the
10-fold datasets. Also, stream gradient was included in several of the models. Tri-cube kernels,
once again outperformed other local smoothing methods for EPTtaxa; however, Epanechnikov

131

kernel was also used for some of the models. On average, when applied to test data, the kernel
method explained 24% of the variation in EPTtaxa (table 18).
1.16.2.3 ANFIS
Average annual organic nitrogen concentration (AOrgN) was once again found to have a
strong association with HBI when variables were identified based on spearman rank and
modeled through the ANFIS approach. Meanwhile, concentrations of other forms of nitrogen
also were included in the models consistently, including average seasonal total nitrogen (100%
of the models) and average annual nitrite (50% of the models). Average annual flow was the
other variable that was incorporated into the ANFIS for some of the folds. Cross-sectional area
was selected for five of the models by spearman rank correlations, but was forced into all the
models. ANFIS model structures varied depending on the training dataset used throughout the
10-fold validation analysis. Variables were either broken down to two or three membership
functions and HBI’s membership function was also linear or constant, depending on the dataset.
Gaussian curve, triangular-shaped and generalized bell-shaped membership functions were the
common membership function structures identified within the HBI models. Based on test data
within 10-fold validation, ANFIS models on average explained 36% of the variation in HBI,
which is 11% more than linear regression and 17% more than kernel methods. MSEs were also
more favorable for ANFIS models, when compared to the other methods (table 18).
ANFIS models for FamilyIBI also included nitrogen concentrations in different forms as
highly influential inputs. In five out of the ten models built, average annual total nitrogen (ATN)
was included as a variable. Within the specific datasets used to build those models, ATN was
found to have the highest spearman correlation with FamilyIBI. Other models however, included
AOrgN (4 out of 10) and ANH4. Specifically, the model built that identified NH4, also included
132

average seasonal organic phosphorus (SOrgP) and was one of the better performing models.
Also, consistent among the data folds, average annual flow was selected and included into the
models. Cross-sectional area was forced into the models as a natural variable, although it was
correlated with flow in most of the models. The input variables were once again made up of two
or three membership functions and varied in shape. The common membership function shapes
that built the best models based on training datasets, included Gaussian (4 out of the 10 models),
triangular (2 out of the 10 models), and trapezoidal (2 out of the ten models). On average, ANFIS
models explained 28% of variation in FamilyIBI, similar to that of stepwise. MSEs based on test
2

data were also slightly more favorable for the ANFIS method (table 18). However, R values
obtained from training sets explain 12% more variation then stepwise methods (table 17).
For EPT taxa, ANFIS and spearman rank correlations selected AOrgN to be included in
all models throughout the 10-fold process. Average annual nitrite concentrations (ANo2) were
also selected for every model. Five out of the ten models included average annual flow, whereas
the other five included average seasonal flow. Cross-sectional area was forced into the model to
represent the natural variations, although it was selected for five of the models by spearman rank
correlation. The structure of the membership functions for all variables differed among models
depending on the training data. EPTtaxa membership function was found to be linear, instead of
constant, in nine of the ten models. Similar to the models built for HBI and FamilyIBI, input
variables for EPTtaxa were made up of two to three membership functions. The most common
shape used to build the membership functions within the EPTtaxa models was triangular. On
average, based on test data, the ANFIS models explained 39% of the variation in EPTtaxa (table
18).

133

1.16.3 Model Selection and Interpretation
Based on all the methods of analysis and model development, the effects of nutrients on
stream health were apparent. Nitrogen concentration, especially organic nitrogen, was
consistently identified as an influential variable for all of the macroinvertebrate measures. At the
same time, other variables such as nitrite, total nitrogen, and ammonium were also identified
having an effect on macroinvertebrate health. In most cases based on spearman rank correlations,
phosphorus measures were highly correlated with nitrogen concentrations and therefore removed
during the variable selection process. However, in some situations phosphorus, such as organic
phosphorus, showed strong correlations with macroinvertebrate measures. In fact, the best model
chosen for FamilyIBI included seasonal organic phosphorus concentration. Flow also was
identified as an influential factor for FamilyIBI and EPTtaxa. Although flow was used in the
models, a high correlation was revealed with cross-sectional area. Flow is likely to be selected
based on its correlation with this natural variation that captures several other factors. Flow,
however, was not found to be as influential in HBI, which is strictly based of pollutant tolerance
values. Both seasonal and annual outputs were highly correlated. In several situations, for the
same macroinvertebrate measure, both average seasonal and average annual nitrogen
concentrations and flows were selected to be in the model depending on the 10-fold dataset. On
average, EPTtaxa was explained the most by the in-stream variables for all three methods
(stepwise, kernel, and ANFIS). This could be a result where indices, like HBI and FamilyIBI
may be less sensitive because they are based off several metrics that may respond differently to
different stressors.
2

Overall, ANFIS displayed better performance (MSE and R ) for explaining and
predicting all three macroinvertebrate measures. ANFIS outperformed other methods for both
134

model training dataset (table 17) and model testing dataset (table 18). This highlights the
importance of recognizing the non-linear relationships that are involved in ecological processes.
At the time, it also displays the complexity and further variation that exists due to other variables
and interactions that are unknown or were not considered in the scope of this study. The
inclusion of these factors would lead to more reliable predictive models and may reduce
2

uncertainty and misinterpretation of results. Based on the average MSE and R from the test
dataset, the ANFIS method was chosen to further identify the best model for prediction. The best
2

model for each biological indicator was selected based on average MSE and R among all
validation subsets within the 10-fold validation process. The final predictive ANFIS model
2

identified for predicting HBI had a R of 0.57 and MSE of 0.20. The final model selected for
2

FamilyIBI had a R of 0.50 and MSE of 29.80. The final predictive model for EPTtaxa
2

performed with an average R of 0.54 and MSE of 6.60 (Table 19).

2
Table 10. Average R and MSE obtained from 10-fold cross validation based on training
data

Response
Variable

Stepwise Regression
Avg.
Training
MSE

HBI
FamilyIBI
EPTtaxa

0.357
43.836
9.006

Avg.
Training
2
R
0.27
0.28
0.35

Kernel Regression
Avg.
Training
MSE
0.364
47.570
10.113

135

Avg.
Training
2
R
0.21
0.22
0.31

ANFIS
Avg.
Training
MSE
0.236
36.270
7.519

Avg.
Training
2
R
0.49
0.40
0.48

Table 11. Average R2 and MSE obtained from 10-fold cross validation based on test data
Response
Variable

HBI
FamilyIBI
EPTtaxa

Stepwise Regression
Avg. Test
MSE
0.387
46.275
9.872

Avg. Test
2
R
0.25
0.27
0.34

Kernel Regression
Avg. Test
MSE
0.410
53.293
12.232

Avg. Test
2
R
0.19
0.18
0.24

ANFIS
Avg. Test
MSE
0.321
45.712
9.229

Avg. Test
2
R
0.36
0.28
0.39

Table 12. Predictor variables used for the best performing macroinvertebrate model
Method

Response Variable

Predictor Variables

Overall MSE

ANFIS

HBI

0.20

ANFIS

FamilyIBI

29.80

EPTtaxa

6.60

2

0.50

ANFIS

crossArea
AOrgN
STN
ANo2
crossArea
ANH4
AFLOW
SOrgP
crossArea
AOrgN
SFLOW
ANo2

Overall R
0.57

0.54

1.16.4 Evaluations of Best Management Practices’ Effectiveness on Stream Health
1.16.4.1 Significant Changes with Implementation
The best models selected in the above sections were used to assess the potential effects of
the implementation of three best management practices (BMPs) on stream health
(macroinvertebrate measures) within the Saginaw River watershed. This was not only to evaluate
the potential ecological impacts of conservation practices, but also to demonstrate the flexibility
and opportunities when using predictive models and connecting data with SWAT model outputs.
As mentioned earlier, the inclusion of other factors would provide more reliable predictions as
136

well as allow for more meaningful interpretations of the results obtained. A Wilcoxon rank
signed rank test, which is a non-parametric test used to compare populations based on the
medians, was performed for all variables to see if HBI, FamilyIBI, and EPTtaxa significantly
differed across the basin when applying the BMP scenarios. To begin, the current scenario
predictions, representing land-use conditions with no BMPs was compared with observed data
from the 262 sites within the basin. The predicted HBI scores across those sites showed no
significant difference from the observed population of scores (p-value = 0.7213). The FamilyIBI
scores from the predicted population also showed no significant difference from the scores
observed (p-value = 0.4093). In addition, the number of EPTtaxa did not significantly differ from
observed numbers across the sites (p-value = 0.2593). This provides further insight into the
reliability of the ANFIS model’s predictions.
For the 262 sites from which the predictive models were built and represented (Table
20),both HBI and EPTtaxa showed no significant change between no-tillage BMP and the
current conditions (p-values > 0.01). However, FamilyIBI did show a significant decrease in the
median value and increase in the mean with the addition of no-till operations. All water quality
variables showed significant small decreases with the implementation of no-till, except seasonal
organic phosphorus concentration (SOrgP), which increased. Increases in organic phosphorus
has been shown in other studies (Giri et al., 2011), and is likely due to an increase in organic
matter on fields from no-tillage. In addition, flow measures also showed small significant
increases between no till and the current condition. This could be resulted from reduced
infiltration and/or increased runoff, which can be a consequence of no-tillage operations (Jones
et al., 1994). The significant change in FamilyIBI, as opposed to EPTtaxa and HBI, could be due
to the inclusion and interaction of seasonal organic phosphorus concentrations with other
137

variables in the predictive model. When observing the effects of residue management at the 262
sites, significant differences from current conditions were displayed. HBI values significantly
decreased, while EPTtaxa and FamilyIBI showed a significant increase, all indicating that stream
health improved. Both annual and seasonal flows along with nutrient concentrations showed
significant decreases with the implementation of residue management. However, SOrgP showed
a significant increase. As assumed, all significant changes were amplified when compared to the
effects of no-tillage. This was expected because residue management includes no-tillage
operations plus a decrease in runoff. This trend continues when exploring the effects of
implementing conservation practices like the Conservation Reserve Program and converting
lands to native grass. Significant increases were predicted for both FamilyIBI and EPTtaxa. At
the same time, HBI scores showed a significant improvement. Significant decreases in nutrient
concentrations (greater than 25% change in mean and median values) were observed and likely
the biggest contribution to significant changes in macroinvertebrate measures. Both annual and
seasonal flow measures showed a significant reduction when comparing the native grass scenario
to current conditions.
The same analysis was performed to evaluate the BMP effects throughout the basin
(13,831 reaches), including un-sampled sites where no observed data are available (Table 21).
Similar results are observed when looking at the difference among the whole basin compared to
the differences at the 262 sites. No-Till practices did not seem to make a significant difference
for HBI and EPTtaxa throughout the basin (p-values > 0.01). However, FamilyIBI did show a
significant increase. Although small, significant changes were observed among all water quality
and quantity variables. When focusing on the effects of residue management, all three
macroinvertebrate measures significantly improved. HBI showed a significant decrease and both
138

EPT taxa and Family IBI increased. All water quality and quantity variables across the basin
responded in a similar fashion to the 262 sampled sites. When comparing native grass to current
conditions, significant improvement were detected for all three macroinvertebrate measures. It is
important to note that the results from the Wilcoxon rank test are primarily used to evaluate the
statistical significant effects of BMPs’ implementation on the three ecological health measures
on the study area as a whole. The percent changes, however, can be misleading because they are
based on the mean and medians, which in some cases contradict each other (e.g. No-Till effects
on HBI and FamilyIBI).

Table 13. Wilcoxon rank sum test and percent changes variables before and after BMP
implementation scenarios for 262 sampling locations
Variables

Current to No Till

% change
Mean median pvalue
HBI
0.46% -0.14% 0.3961
FamilyIBI 0.90% -1.93% <0.01
EPTtaxa
-3.0% -1.18% 0.041
AFLOW
0.34% 0.66% <0.01
SFLOW
0.71% 0.22% <0.01
AOrgN
-6.9% -5.89% <0.01
ANo2
-7.6% -4.53% <0.01
ANH4
-8.1% -7.15% <0.01
SOrgP
1.48% 5.50% <0.01
STN
-7.4% -8.11% <0.01

Current to Residue
Mgt.
% change
Mean Median pvalue
-2.5% -0.79% <0.01
1.86% -0.32% <0.01
4.90% 0.28% <0.01
-2.4% -1.46% <0.01
-3.9% -2.07% <0.01
-9.6% -14.8% <0.01
-13% -16.2% <0.01
-15% -15.3% <0.01
7.51% 2.44% <0.01
-1.8% -12.3% <0.01

139

Current to Native Grass
% change
Mean Median pvalue
-4.59% -1.48% <0.01
4.94% 3.63% <0.01
17.24% 10.49% <0.01
-5.03% -5.95% <0.01
-12.3% -7.73% <0.01
-41.5% -35.6% <0.01
-48.1% -45.7% <0.01
-46.6% -42.9% <0.01
-33.9% -25.3% <0.01
-42.8% -39.8% <0.01

Table 14. Wilcoxon rank sum test and percent changes variables before and after BMP
implementation scenarios for the entire basin
Variables

Current to No Till

Current to Residue Mgt.

Current to Native
Grass
% change
% change
% change
Mean median pvalue Mean Median pvalue Mean Median pvalue
HBI
0.62% 0.12% 0.1169 -1.86% -0.33% <0.01 -1.1% -0.22% <0.01
FamilyIBI 1.47% 0.33% <0.01 1.46% 1.07% <0.01 7.11% 7.99% <0.01
EPTtaxa
-0.5% 0.07% 0.0344 4.31% 4.52% <0.01 1.67% 2.54% <0.01
AFLOW
0.32% 0.31% <0.01 -2.49% -3.32% <0.01 -7.0% -7.83% <0.01
SFLOW
0.76% 0.95% <0.01 -4.11% -5.47% <0.01 -9.5% -14.6% <0.01
AOrgN
-4.5% -0.33% <0.01 -4.29% -3.71% <0.01 -44% -24.3% <0.01
ANo2
-7.9% 0.01% <0.01 -16.9% -1.58% <0.01 -51% -39.6% <0.01
ANH4
-8.2% -1.34% <0.01 -19.0% -9.15% <0.01 -56% -66.4% <0.01
SOrgP
0.01% 4.72% <0.01 12.01% -0.64% <0.01 -31% -28.3% <0.01
STN
-6.7% -1.87% <0.01 -0.10% -7.25% <0.01 -39% -39.4% <0.01

1.16.4.2 Stream Health for Individual Reaches
In this section, the predicted effects of BMP implementation for individual reaches
throughout the basin are evaluated. Overall, the practice of no-tillage increased FamilyIBI in
27% of the reaches within the basin. Meanwhile 18% showed a decline in FamilyIBI and the
remaining 55% showed no change at all (Figure 13, Table 22). Of the reaches that showed no
impact, only 5% of them were within a subbasin that had a BMP. Within the reaches that showed
increases in Family IBI, roughly 50% were located within subbasins that implemented no-tillage.
However, of the reaches that showed decreases, roughly 50% also were in subbasins that
implemented BMPs. The water quality and quantity variables that were included in predicting
FamilyIBI also changed similar among improved and declined reaches. This could be a result of
the non-linearity responses that are captured in the ANFIS models and indicates the importance
of interactions among variables. It also may be a limitation of the models that were built off of
data that was affected by a large amount of other factors that are not captured in the scope of this
study. The number of EPTtaxa was increased by no-till operations in 20% of the subbasins,
140

while 62% showed no change, and the remaining 18% showed a decline (Figure 14, table 22). Of
the reaches that showed no change, most were in subbasins that were classified as pasture, forest,
and wetlands. Only 13% of them were in subbasins where no-tillage was implemented. Of the
reaches that showed a decline in EPTtaxa, 45% of them were found in subbasins where no-tillage
was practiced. Meanwhile, 52% of the reaches that showed increases were found in subbasins
where no-tillage was present. A distinction that was made between the reaches that increased and
decreased in EPTtaxa was that degraded reaches on average showed an increase in ANo2
concentration, whereas improved reaches showed a decrease in ANO2. In addition, higher
percentage (40%) of the reaches with declined EPTtaxa showed an increase in AOrgN
concentrations) compared to reaches with increased EPTtaxa (15% of the reaches increased in
AOrgN). When focusing on no-tillage’s effects to HBI, 21% of the reaches in the basin showed
an improvement, 62% showed no change, and 17% showed a negative change (Figure 15, Table
22). The majority (86%) of the reaches that did not respond in HBI were those with no-BMP. Of
those reaches with improved HBI scores, 66% reaches showed a decrease in all three nitrogen
concentration variables (STN, AOrgN, and ANo2). For reaches with declined HBI scores, 93%
showed an increase in at least one nitrogen measure.
The residue management BMP implementation in the subbasin also resulted mixed
results. About 31% of the reaches showed an increase in FamilyIBI, 54% showed no change, and
15% showed a decrease (Figure 16, Table 22). For reaches that showed no change, 97% had no
row-crop lands applicable for residue management. However, the majority (60%) of subbasins
that had residue management implementation had increased FamilyIBI value. Reaches that had
decreased FamilyIBI values showed an average increase in SOrgP. The EPTtaxa also showed
more of an increase with residue management. About 26% of the reaches showed an increase in
141

EPTtaxa, 59% showed no change, and 15% showed decreases (Figure 17, Table 22). Of the
reaches that showed no change, majority (57%) were in subbasins dominated by forest or
forested wetland, and only 10% were in subbasins with residue management. A major difference
between reaches that improved and declined was reflected on how AOrgN responded. For
reaches where EPTtaxa declined, on average the AOrgN showed an increase, while for reaches
that EPTtaxa improved, AOrgN showed a decrease on average. In addition, a larger percentage
of reaches decreased in AOrgN for the improved reaches when compared to the degraded ones.
For HBI, about 25% of the reaches improved, 58% did not respond, and 17% responded
negatively to the Residue Management (Figure 18, Table 22). Of the reaches that did not
respond, 93% of them were located in subbasins that were not disturbed through residue
management operations. For reaches with degraded HBI condition, more than half of the reaches
showed an increase in at least one nitrogen measure (STN, AOrgN, or ANo2). In addition, the
proportion of reaches that had increased STN, AOrgN, and ANO2 for the degraded sites was
higher than the proportion in the improved sites. This could potentially explain why some sites
actually degraded from residue management.
In the case where native grass was applied to the agricultural lands, 50% of reaches
increased in FamilyIBI score, 22% had no change, and 28% decreased (Figure 19, Table 22). Of
the reaches that showed no change, only 1% was within subbasins that received the native grass
conversion and the subbasins of those reaches were largely forests, wetlands, and pastures that
were not altered between two scenarios. Reaches with increased FamilyIBI had a smaller
proportion (16%) of reaches with increased ANH4. For EPTtaxa, about 46% of the reaches
increased, 24% had no change, and 30% declined by native grass BMP (Figure 20, Table 22). Of
the reaches that showed no change in EPTtaxa, 92% were in subbasins that had no BMP
142

implementation and very little differences in water quality and quantity variables between
scenarios. The reaches improved by native grass BMP had larger average decrease in AOrgN (2.57 mg/L) than reaches degraded by native grass had (-0.85 mg/L). Native grass’s effect on HBI
benefitted for 44% of the reaches within the basin, had no effect for 25% of the reaches, and had
negative for 31% of the reaches (Figure 21, Table 22). Among the reaches that had declined
condition, 60% showed an increase in at least one of the three nutrient measures.

143

Table 15. The impact of BMP implementation scenarios at watershed scale based on macroinvertebrate measures
BMP

No-Till
Residue Mgt.
Native Grass

Family IBI
% total reaches
Improve Decline
No
Change
27%
31%
50%

18%
15%
28%

55%
54%
22%

EPT taxa
% total reaches
Improve Decline
No
Change
20%
26%
46%

144

18%
15%
30%

62%
59%
24%

HBI
% total reaches
Improve Decline
No
Change
21%
25%
44%

17%
17%
31%

62%
58%
25%

Figure 13. Improvements and declines in Family IBI from current scenario after no-till
implementation

145

Figure 14. Improvements and declines in EPT taxa from current scenario after no-till
implementation

146

Figure 15. Improvements and declines in HBI from current scenario after no-till
implementation

147

Figure 16. Improvements and declines in Family IBI from current scenario after residue
management implementation

148

Figure 17. Improvements and declines in EPT taxa from current scenario after residue
management implementation

149

Figure 18. Improvements and declines in HBI from current scenario after residue
management implementation

150

Figure 19. Improvements and declines in Family IBI from current scenario after native
grass implementation

151

Figure 20. Improvements and declines in EPT taxa from current scenario after native grass
implementation

152

Figure 21. Improvements and declines in HBI from current scenario after native grass
implementation

1.17 CONCLUSION
As demands for agricultural production and landscape conversion into urban environment
continue, there is a need to assess the effects of these activities for developing possible
mitigating strategies to protect the health of water resources. To effectively do so, there is a need
to continue to explore alternative methods in predicting and understanding the complex
relationships involved. The objectives of this study included linking hydrologic models and
ecological conditions using the Soil and Water Assessment Tool (SWAT as a means of building
153

and collecting high resolution flow and water quality elements. The study then identifies the
influential variables for explaining macroinvertebrate indices. The relationships between aquatic
macroinvertebrates and local in-stream conditions (SWAT outputs) are explored through models
built using statistical and alternative methods. Those models were then used to forecast and
assess the effects of numerous agricultural BMPs on stream health.
Within this study, a high resolution and detailed SWAT model was built and calibrated for
the Saginaw River watershed to provide data and predictions of in-stream conditions, where they
otherwise would not be available. Outputs obtained from the SWAT model included in-stream
variables such as sediment and nutrient loads that often originate from anthropogenic activities.
In addition, stream size was included in efforts to capture the natural longitudinal variation from
headwaters to downstream rivers. The variables were then linked with three macroinvertebrate
measures (HBI, FamilyIBI, and EPTtaxa), which were used as indicators of overall stream
health.
Overall, in-stream variables obtained through the SWAT model explained 54% of the
variation in EPTtaxa, 57% of the variation in HBI, and 50% of the variation in FamilyIBI. The
limited variance explained by the in-stream water quality and quantity variables within this study
exhibits the influence of other factors, which are not available for every reach or within the scope
of this study. Linear (stepwise linear regression) and non-linear (ANFIS and kernel regression)
methods were explored to describe the linkage between in-stream variables and
macroinvertebrate measures. The best model was identified through 10-fold cross validation
analysis for the prediction of stream health under current conditions and under scenarios
involving the implementation of agricultural conservation practices. Among the methods, the
non-linear ANFIS method performed the best for all three stream health measures. This
154

highlights the non-linear and complex relationships between environmental and ecological
components.
Throughout all three methods, similar in-stream variables were consistently identified as
having a large association with the macroinvertebrate measures. Average annual organic nitrogen
2

had the highest partial R and highest spearman correlation with HBI; with lower concentrations
generally associated with better stream health. This relationship was expected due to the fact that
HBI is calculated based on pollution tolerance values. The best model for predicting HBI also
included average seasonal total nitrogen and annual nitrite. EPTtaxa was also consistently
explained through nitrogen concentrations, with AOrgN being included in the best model along
with nitrite, average seasonal flow, and cross-sectional area. FamilyIBI was also commonly
explained through nutrient concentrations more than any other variables. The best model for
Family IBI included average annual ammonium and average seasonal organic phosphorus in
addition to average annual flow and cross-sectional area.
The predicative models were then used in connection with outputs from SWAT models for
forecasting the effects of BMP implementation within the Saginaw River watershed. The three
best management practices looked at represented three different efficiencies when it comes to
reducing the effects to our water quality and quantity, with no-tillage being minimal, residue
management being medium, and native grass being the maximum. This was partially mirrored in
the predicted effects on stream health (HBI, FamilyIBI, and EPTtaxa). Significant changes were
observed with all three practices from current conditions with the exception of HBI and EPTtaxa
under no-tillage scenario. The improvements observed from the implementation of the BMPs
were highly caused by their reduction of nutrient concentrations, especially within subbasins

155

where they were implemented. Meanwhile, there were a smaller percentage of reaches that
showed declines in stream health with the implementation of BMPs. This was often connected
with increases in one or more nutrient concentrations. In addition, these results could be
misinterpretations because of limitations in the models and variables included. While the results
help enhance the understanding of the potential benefits from large scale BMP implementation,
uncertainty in the data collection and model components still exists, which should be explored in
future studies. This study employed a SWAT model that was calibrated at a single location, and
future studies using a more spatially extensive calibration may reduce this uncertainty. In
addition, the use of more biological data, both spatially and temporally, may also improve
understanding.
This study identified in-stream variables and conditions that have significant relations with
macroinvertebrate communities and ultimately stream health. At the same time, it made these
connections through a hydrologic model and the use of non-linear alternative methods. The
benefit of linking watershed models, such as SWAT, to stream health is that it allows us to
forecast the effects of different management practices and anthropogenic activities.
Acknowledgements
We thank Ms. Minako Edgar at the Institute for Fisheries Research for her assistance throughout
the project, Dr. Sasha Kravchenko, Juan David Munoz, and Wenning Feng for their assistance
with statistical analysis, and Sean Wozniki for his assistance throughout the project. Funding for
this project was provided by the United States Department of Agriculture, Natural Resources
Conservation Services under the Great Lakes Northern Forest Cooperative Ecosystems Studies
Unit (CESU) Umbrella Agreement Number 68-3A75-8-86.

156

CONCLUSIONS
The research presented here evaluated the potential for a Soil and Water Assessment Tool
(SWAT) model to provide influential in-stream variables to explain fish and macroinvertebrate
measures beyond the biological sampling points within the Saginaw River watershed, located in
Michigan. The high resolution SWAT model was built that includes detailed information at
spatial units having homogenous physicochemical, geomorphological, and biological features for
all stream segments within the study area. The in-stream variables provided by the model
included flow, sediment, nutrient concentrations, which may directly or indirectly influence the
health of a stream. Relationships between in-stream variables (SWAT outputs) and fish (Index of
Biological Integrity (IBI) and Percent Intolerant Individuals) and macroinvertebrate (Family
index of Biological Integrity (Family IBI), Hilsenhoff Biotic Index (HBI), and Number of EPT
taxa) measures were explored and predictive models were built using statistical methods and
alternative soft computing techniques. Scenarios representing landuse conversion, climate
change, and management practices were then presented within the SWAT model and their
potential effects on stream health were estimated. The first section of this research focused on the
connection between in-stream conditions and fish measures. In addition, historical reference
conditions for streams in the Saginaw River watershed were predicted under pre-settlement landuse and climate data. The second study explored the relationships between in-stream conditions
and macroinvertebrate measures and assessed different agricultural best management practices’
(BMPs) performances on improving ecological health measures. The following can be concluded
from the results of both studies:


The Soil and Water Assessment Tool can be an effective tool to provide influential instream data where it otherwise would not exist

157



In-stream variables obtained from SWAT explained 48% of the variation in IBI, 21% of
the variation in the percent of intolerant individuals, 50% of the variation in Family IBI,
57% of the variation in HBI, and 54% of the variation in the number of EPT taxa.



Fuzzy logic and adaptive neural-fuzzy inference system techniques outperformed other
statistical methods (stepwise linear regression, partial least squares regression, and kernel
regression) and provide a valuable and practical approach to connecting environmental
and water quality conditions to ecological health measures.



Average annual flow rate had the strongest correlation with IBI, whereas nutrient
concentrations (e.g. organic phosphorus, organic nitrogen) showed the largest influence
on the percentage of intolerant individuals and all three macroinvertebrate measures.
However, adding a measure of stream size and in-stream physical habitat variables along
with other measures of watershed physiographic characteristics would likely improve the
accuracy of model predictions.



Results showed overall significantly higher IBI and percent intolerant individuals under
the pre-settlement landuse scenario. This implies that landuse change from pre-settlement
to current has profound negative impacts on stream health.



Pre-settlement climate factors had a strong influence on stream flow and water quality
measures that interactively affect stream health as indicated by fish measures. These
results suggest that efforts to model historic baseline habitat conditions and to provide
context for stream health assessments should include both pre-settlement land use and
climate conditions.

158



Among the studied best management practices, native grass resulted in the most
improvement in stream health as indicated by macroinvertebrate measures, followed by
residue management and no-tillage operations.

159

RECOMMENDATIONS FOR FUTURE RESEARCH
This research provides results and valuable insight into the efforts of linking watershed
models and in-stream variables with ecological health. In addition the information looks at
alternative methods in capturing these relationships and building predictive models that can act
as effective tools for decision makers in quantifying the impacts of landuse and climate change,
along with management strategies. However, continuing research must be done that builds upon
these studies and addresses issues outside the scope of this project. The following are
suggestions for future research:


There is a great deal of uncertainty in the data collection and model components that
exists. This uncertainty should be explored and quantified to better aid managers and
decision makers.



In order to have more comprehensive predictions and explanation of ecological measures,
additional variables such as physical, chemical, and natural factors should be considered.
Adding measures, such as stream size and instream physical habitat variables would
likely provide more robust predictions and the necessary context to reveal the relative
residual effects of flow and water quality on aquatic communities.



With such complex interactions and relationships involved in aquatic species and their
environments, there is a need to continue to explore alternative methods for capturing and
modeling these processes. This may include continuing research employing different
fuzzy logic techniques or fusion-methods.



Ideally, both fish and macroinvertebrates should be measured together because they can
react differently, vary in sensitivity to different stressors, and mirror conditions at
different scales. Assessing multiple indices of stream health and doing so through
160

methods such as analytic hierarchy processes may provide more comprehensive insight
into the impacts of landuse and climate change.


There are a number of potential best management practices that may be implemented
within agricultural regions and at different scales, which go beyond the scope of this
project. Additional studies addressing these scenarios will provide a more comprehensive
assessment of BMP implementation within the Saginaw River watershed.



Potential future changes in climate and landuse (e.g. biofuel crop expansion) and their
impacts on ecological health need to be addressed. Future research, applying similar
strategies as this study, can be performed to assess the potential associated risk involved.



Relationships and measures studied in this project were within a highly agricultural and
warm-water dominated watershed. Additional studies in different climatological and
physiographical regions will likely provide diverse results and are needed to better
understand the interactions involved and implications of landuse and climate change
across a larger scale.

161

APPENDIX

162

APPENDIX
Table 16. Stepwise linear regression models for IBI
10
Fold
Set

Response
Variable

Predictor
Variables

1

IBI

2

IBI

3

IBI

4

IBI

5

IBI

6

IBI

7

IBI

8

IBI

9

IBI

10

IBI

*1,*2
*1,*3,*4
*1,*2
*1,*2,*3
*1,*2,*3
*1,*2
*5,*2
*5,*4
*5,*2
*5,*2

Training Training
2
MSE
R

Test
MSE

Test
2
R

Overall Overall
2
MSE
R

303.07

0.39

662.05

0.02

333.69

0.33

336.94

0.35

301.67

0.19

326.48

0.34

323.89

0.35

399.22

0.29

326.27

0.34

335.53

0.33

332.12

0.34

328.24

0.34

321.30

0.35

452.56

0.31

327.57

0.34

345.94

0.35

251.83

0.32

331.30

0.33

333.05

0.32

314.53

0.50

326.05

0.34

326.71

0.37

399.29

0.08

329.15

0.34

338.90

0.31

270.62

0.63

326.55

0.34

340.19

0.32

255.86

0.49

326.16

0.34

* All Variables after transformation
*1 AFLOW; *2 ANO2; *3 ANO3; *4 ANH4; *5 SFLOW

163

Table 17. Stepwise linear regression models for percent intolerant individuals
10
Fold
Set

Response
Variable

Predictor
Variables

1

%INTOL

2

%INTOL

3

%INTOL

4

%INTOL

5

%INTOL

6

%INTOL

7

%INTOL

8

%INTOL

9

%INTOL

10

%INTOL

*1,*2
*3,*4
*1,*4
*3,*2,*5
*3,*4
*3,*6
*3,*6
*3,*6
*3,*6
*1,*2

Training Training
2
MSE
R

Test
MSE

Test
2
R

Overall Overall
2
MSE
R

280.37

0.14

580.81

0.03

309.95

0.12

320.11

0.11

272.30

0.02

315.40

0.10

328.93

0.10

236.20

0.01

319.80

0.09

304.54

0.18

270.68

0.00

301.21

0.13

332.19

0.10

105.14

0.19

309.84

0.10

297.22

0.10

528.69

0.18

320.01

0.10

336.18

0.09

107.66

0.33

313.68

0.10

326.03

0.11

227.89

0.05

315.86

0.10

309.13

0.07

445.93

0.39

323.31

0.10

284.93

0.15

566.83

0.01

314.14

0.12

Test
MSE

Test
2
R

Overall Overall
2
MSE
R

606.05
320.04
374.17
290.32
472.06
246.22
294.15
403.52
308.11
269.80

0.04
0.16
0.31
0.42
0.29
0.27
0.56
0.07
0.56
0.46

* All Variables after transformation
*1 ANH4; *2 ASED; *3 ATP; *4 ANO3; *5 SNO2; *6 SFLOW
Table 18. PLSR models for IBI
10
Fold
Set
1
2
3
4
5
6
7
8
9
10

Response
Variable

Predictor
Variables

IBI
IBI
IBI
IBI
IBI
IBI
IBI
IBI
IBI
IBI

PC1, PC2
PC1, PC2
PC1, PC2
PC1, PC2
PC1, PC2
PC1, PC2
PC1, PC2
PC1, PC2
PC1, PC2
PC1, PC2

Training Training
2
MSE
R
313.39
343.04
331.88
345.17
325.90
347.33
342.41
335.04
342.54
342.12

0.35
0.33
0.32
0.30
0.32
0.32
0.29
0.34
0.29
0.31

164

342.20
340.77
336.04
339.77
340.29
337.37
337.66
342.14
338.97
334.63

0.31
0.31
0.32
0.31
0.31
0.32
0.32
0.31
0.31
0.32

Table 19. PLSR models for percent intolerant individuals
10
Fold
Set
1
2
3
4
5
6
7
8
9
10

Response
Variable

Predictor
Variables

%INTOL
%INTOL
%INTOL
%INTOL
%INTOL
%INTOL
%INTOL
%INTOL
%INTOL
%INTOL

PC1, PC2
PC1, PC2
PC1, PC2
PC1, PC2
PC1, PC2
PC1, PC2
PC1, PC2
PC1, PC2
PC1, PC2
PC1, PC2

Training Training
2
MSE
R
288.58
326.99
328.14
328.53
336.69
298.75
341.22
331.41
312.05
291.16

0.11
0.09
0.10
0.10
0.09
0.09
0.08
0.09
0.07
0.13

165

Test
MSE

Test
2
R

669.48
238.87
221.78
193.01
95.61
536.97
100.49
230.69
449.42
581.00

0.00
0.12
0.02
0.11
0.28
0.10
0.38
0.06
0.37
0.00

Overall Overall
2
MSE
R
326.08
318.31
317.67
315.19
312.95
322.20
317.52
320.97
326.28
321.19

0.08
0.09
0.09
0.10
0.09
0.09
0.09
0.09
0.09
0.09

Table 20. Fuzzy logic models for IBI
10
Fold
Set
1

IBI

2

IBI

3

IBI

4

IBI

5

IBI

6

IBI

7

IBI

8

IBI

9

IBI

10

IBI

Response
Variable

Predictor
Variables

Training
MSE

Training
2
R

Test
MSE

Test
2
R

Overall
MSE

Overall
2
R

*1,*2,*3,
*4,*5
*1,*2,*3,
*4,*5
*1,*2,*3,
*4,*5
*1,*2,*3,
*4,*5
*1,*2,*3,
*4,*5
*1,*2,*3,
*4,*5
*1,*2,*3,
*4,*5
*1,*2,*3,
*4,*5
*1,*2,*3,
*4,*5
*1,*2,*3,
*4,*5

295.31

0.41

494.5

0.20

314.92

0.37

319.65

0.40

240.6

0.47

274.90

0.40

288.05

0.43

350.0

0.39

271.59

0.43

341.08

0.37

237.4

0.56

330.88

0.39

295.88

0.43

438.5

0.42

276.86

0.42

283.89

0.47

219.8

0.31

250.64

0.46

329.69

0.37

262.0

0.56

323.03

0.39

280.83

0.49

387.7

0.25

263.11

0.46

274.37

0.45

205.1

0.68

243.17

0.48

298.17

0.41

196.2

0.61

287.62

0.43

*1 AFLOW; *2 AOrgP; *3 SNO2; *4 SNO3; *5 StreamGrad

166

Table 21. Fuzzy logic models for percent intolerant individuals
10
Fold
Set
1

Response
Variable

Predictor
Variables

Training
MSE

Training
2
R

Test
MSE

Test
2
R

Overall Overall
2
MSE
R

%INTOL

250.40

0.16

537.6

0.03

278.68

0.14

2

%INTOL

286.30

0.13

172.1

0.38

275.07

0.14

3

%INTOL

300.58

0.11

148.2

0.34

285.59

0.12

4

%INTOL

292.57

0.12

178.2

0.12

281.32

0.12

5

%INTOL

317.79

0.11

92.63

0.30

295.62

0.11

6

%INTOL

266.98

0.12

380.9

0.29

278.20

0.13

7

%INTOL

298.36

0.12

92.17

0.41

278.06

0.14

8

%INTOL

304.37

0.12

196.7

0.12

293.22

0.12

9

%INTOL

273.57

0.10

320.8

0.53

278.47

0.13

10

%INTOL

*1,*2,*3,
*4
*1,*2,*3,
*4
*1,*2,*3,
*4
*1,*2,*3,
*4
*1,*2,*3,
*4
*1,*2,*3,
*4
*1,*2,*3,
*4
*1,*2,*3,
*4
*1,*2,*3,
*4
*1,*2,*3,
*4

250.01

0.17

493.8

0.01

275.27

0.14

*1 AOrgP; *2 ANO2; *3 AFLOW; *4 ACHLA

167

Table 22. Spearman rank correlations with red indicating significance
IBI %INTOL StreamGrad SFLOW
0.67
1.00
0.01
0.23
%INTOL
0.01
1.00
-0.52
StreamGrad -0.18
0.43
0.23
-0.52
1.00
SFLOW
0.02
-0.08
-0.34
0.42
SSED
-0.38
-0.35
0.12
-0.43
SOrgN
-0.40
-0.36
0.12
-0.41
SOrgP
0.20
0.05
-0.48
0.66
SNo3
-0.35
-0.32
0.01
-0.18
SNH4
-0.29
-0.30
-0.13
0.09
SNo2
0.04
-0.12
-0.41
0.51
SMinP
-0.02
-0.10
-0.63
0.49
SCHLA
-0.15
-0.14
0.07
-0.27
SCBOD
0.13
0.08
-0.18
0.35
SDO
-0.04
-0.17
-0.31
0.26
STN
-0.13
-0.25
-0.22
0.13
STP
0.03
0.10
-0.07
-0.08
SPercBase
0.47
0.27
-0.49
0.98
AFLOW
0.03
-0.05
-0.29
0.33
ASED
-0.41
-0.38
0.03
-0.44
AOrgN
-0.43
-0.40
0.06
-0.45
AOrgP
0.08
-0.04
-0.40
0.34
ANo3
-0.30
-0.34
-0.13
-0.05
ANH4
-0.26
-0.32
-0.20
0.13
ANo2
-0.09
-0.21
-0.34
0.35
AMinP
-0.15
-0.25
-0.61
0.33
ACHLA
-0.17
-0.15
0.16
-0.37
ACBOD
0.09
0.12
-0.14
0.21
ADO
-0.17
-0.26
-0.27
0.05
ATN
-0.24
-0.34
-0.21
-0.01
ATP
0.05
0.15
-0.06
-0.05
APercBase

SSED
-0.08
-0.34
0.42
1.00
0.30
0.30
0.74
0.51
0.59
0.84
0.65
0.26
0.21
0.78
0.74
-0.39
0.35
0.95
0.41
0.39
0.74
0.71
0.73
0.88
0.59
0.16
0.13
0.74
0.71
-0.36
168

SOrgN
-0.35
0.12
-0.43
0.30
1.00
0.99
-0.06
0.83
0.60
0.23
0.23
0.63
-0.24
0.50
0.65
-0.50
-0.52
0.34
0.88
0.89
0.12
0.67
0.52
0.32
0.34
0.56
-0.19
0.50
0.62
-0.48

SOrgP
-0.36
0.12
-0.41
0.30
0.99
1.00
-0.03
0.85
0.63
0.26
0.24
0.58
-0.23
0.52
0.67
-0.49
-0.49
0.34
0.87
0.88
0.14
0.68
0.53
0.33
0.34
0.50
-0.18
0.50
0.62
-0.45

SNo3
0.05
-0.48
0.66
0.74
-0.06
-0.03
1.00
0.25
0.44
0.88
0.63
-0.05
0.26
0.77
0.59
-0.16
0.61
0.69
0.08
0.07
0.83
0.46
0.55
0.77
0.53
-0.12
0.14
0.65
0.51
-0.13

SNH4
-0.32
0.01
-0.18
0.51
0.83
0.85
0.25
1.00
0.84
0.53
0.35
0.39
-0.08
0.66
0.73
-0.35
-0.26
0.54
0.81
0.81
0.35
0.86
0.75
0.57
0.44
0.35
-0.11
0.63
0.70
-0.28

SNo2
-0.30
-0.13
0.09
0.59
0.60
0.63
0.44
0.84
1.00
0.63
0.43
0.22
0.02
0.65
0.67
-0.31
0.02
0.59
0.61
0.62
0.45
0.81
0.90
0.65
0.53
0.16
-0.03
0.60
0.63
-0.23

SMinP
-0.12
-0.41
0.51
0.84
0.23
0.26
0.88
0.53
0.63
1.00
0.67
0.12
0.21
0.89
0.81
-0.28
0.44
0.78
0.32
0.32
0.77
0.67
0.72
0.90
0.61
0.03
0.10
0.75
0.71
-0.24

Table 22. (cont’d)
SCHLA
-0.10
%INTOL
-0.63
StreamGrad
0.49
SFLOW
0.65
SSED
0.23
SOrgN
0.24
SOrgP
0.63
SNo3
0.35
SNH4
0.43
SNo2
0.67
SMinP
1.00
SCHLA
0.24
SCBOD
0.17
SDO
0.62
STN
0.57
STP
-0.32
SPercBase
0.43
AFLOW
0.63
ASED
0.28
AOrgN
0.28
AOrgP
0.55
ANo3
0.45
ANH4
0.48
ANo2
0.65
AMinP
0.86
ACHLA
0.14
ACBOD
0.11
ADO
0.55
ATN
0.54
ATP
-0.35
APercBase

SCBOD
-0.14
0.07
-0.27
0.26
0.63
0.58
-0.05
0.39
0.22
0.12
0.24
1.00
-0.23
0.32
0.41
-0.58
-0.32
0.30
0.60
0.58
0.10
0.38
0.25
0.21
0.32
0.95
-0.16
0.35
0.43
-0.69

SDO
0.08
-0.18
0.35
0.21
-0.24
-0.23
0.26
-0.08
0.02
0.21
0.17
-0.23
1.00
0.06
0.00
0.04
0.36
0.15
-0.25
-0.25
0.11
-0.02
0.06
0.16
0.14
-0.26
0.85
-0.02
-0.05
0.04

STN
-0.17
-0.31
0.26
0.78
0.50
0.52
0.77
0.66
0.65
0.89
0.62
0.32
0.06
1.00
0.94
-0.41
0.17
0.74
0.56
0.55
0.76
0.75
0.71
0.83
0.57
0.23
-0.02
0.87
0.82
-0.39

STP
-0.25
-0.22
0.13
0.74
0.65
0.67
0.59
0.73
0.67
0.81
0.57
0.41
0.00
0.94
1.00
-0.51
0.05
0.70
0.65
0.65
0.62
0.77
0.71
0.78
0.54
0.30
-0.06
0.80
0.84
-0.48
169

SPercBase AFLOW
0.10
0.27
-0.07
-0.49
-0.08
0.98
-0.39
0.35
-0.50
-0.52
-0.49
-0.49
-0.16
0.61
-0.35
-0.26
-0.31
0.02
-0.28
0.44
-0.32
0.43
-0.58
-0.32
0.04
0.36
-0.41
0.17
-0.51
0.05
1.00
-0.01
-0.01
1.00
-0.40
0.27
-0.43
-0.53
-0.44
-0.54
-0.16
0.30
-0.37
-0.14
-0.32
0.06
-0.32
0.28
-0.31
0.29
-0.44
-0.40
0.05
0.22
-0.30
-0.03
-0.42
-0.10
0.87
0.00

ASED
-0.05
-0.29
0.33
0.95
0.34
0.34
0.69
0.54
0.59
0.78
0.63
0.30
0.15
0.74
0.70
-0.40
0.27
1.00
0.48
0.45
0.74
0.75
0.74
0.86
0.59
0.22
0.08
0.76
0.72
-0.37

AOrgN AOrgP
-0.38
-0.40
0.03
0.06
-0.44
-0.45
0.41
0.39
0.88
0.89
0.87
0.88
0.08
0.07
0.81
0.81
0.61
0.62
0.32
0.32
0.28
0.28
0.60
0.58
-0.25
-0.25
0.56
0.55
0.65
0.65
-0.43
-0.44
-0.53
-0.54
0.48
0.45
1.00
0.99
0.99
1.00
0.34
0.31
0.83
0.83
0.66
0.66
0.49
0.49
0.42
0.41
0.56
0.54
-0.20
-0.19
0.71
0.69
0.81
0.81
-0.41
-0.41

Table 22. (cont’d)
%INTOL
StreamGrad
SFLOW
SSED
SOrgN
SOrgP
SNo3
SNH4
SNo2
SMinP
SCHLA
SCBOD
SDO
STN
STP
SPercBase
AFLOW
ASED
AOrgN
AOrgP
ANo3
ANH4
ANo2
AMinP
ACHLA
ACBOD
ADO
ATN
ATP
APercBase

ANo3
-0.04
-0.40
0.34
0.74
0.12
0.14
0.83
0.35
0.45
0.77
0.55
0.10
0.11
0.76
0.62
-0.16
0.30
0.74
0.34
0.31
1.00
0.62
0.61
0.82
0.53
0.03
0.07
0.85
0.68
-0.13

ANH4
-0.34
-0.13
-0.05
0.71
0.67
0.68
0.46
0.86
0.81
0.67
0.45
0.38
-0.02
0.75
0.77
-0.37
-0.14
0.75
0.83
0.83
0.62
1.00
0.92
0.83
0.56
0.33
-0.05
0.84
0.89
-0.32

ANo2
-0.32
-0.20
0.13
0.73
0.52
0.53
0.55
0.75
0.90
0.72
0.48
0.25
0.06
0.71
0.71
-0.32
0.06
0.74
0.66
0.66
0.61
0.92
1.00
0.84
0.59
0.19
-0.01
0.76
0.79
-0.26

AMinP ACHLA ACBOD
-0.21
-0.25
-0.15
-0.34
-0.61
0.16
0.35
0.33
-0.37
0.88
0.59
0.16
0.32
0.34
0.56
0.33
0.34
0.50
0.77
0.53
-0.12
0.57
0.44
0.35
0.65
0.53
0.16
0.90
0.61
0.03
0.65
0.86
0.14
0.21
0.32
0.95
0.16
0.14
-0.26
0.83
0.57
0.23
0.78
0.54
0.30
-0.32
-0.31
-0.44
0.28
0.29
-0.40
0.86
0.59
0.22
0.49
0.42
0.56
0.49
0.41
0.54
0.82
0.53
0.03
0.83
0.56
0.33
0.84
0.59
0.19
1.00
0.67
0.13
0.67
1.00
0.23
0.13
0.23
1.00
0.07
0.08
-0.19
0.86
0.58
0.29
0.85
0.60
0.36
-0.29
-0.35
-0.59
170

ADO
0.12
-0.14
0.21
0.13
-0.19
-0.18
0.14
-0.11
-0.03
0.10
0.11
-0.16
0.85
-0.02
-0.06
0.05
0.22
0.08
-0.20
-0.19
0.07
-0.05
-0.01
0.07
0.08
-0.19
1.00
-0.04
-0.07
0.06

ATN
-0.26
-0.27
0.05
0.74
0.50
0.50
0.65
0.63
0.60
0.75
0.55
0.35
-0.02
0.87
0.80
-0.30
-0.03
0.76
0.71
0.69
0.85
0.84
0.76
0.86
0.58
0.29
-0.04
1.00
0.93
-0.29

ATP
-0.34
-0.21
-0.01
0.71
0.62
0.62
0.51
0.70
0.63
0.71
0.54
0.43
-0.05
0.82
0.84
-0.42
-0.10
0.72
0.81
0.81
0.68
0.89
0.79
0.85
0.60
0.36
-0.07
0.93
1.00
-0.40

APercBase
0.15
-0.06
-0.05
-0.36
-0.48
-0.45
-0.13
-0.28
-0.23
-0.24
-0.35
-0.69
0.04
-0.39
-0.48
0.87
0.00
-0.37
-0.41
-0.41
-0.13
-0.32
-0.26
-0.29
-0.35
-0.59
0.06
-0.29
-0.40
1.00

REFERENCES

171

REFERENCES
Adriaenssens, V., De Baet,s B., Goethals, P., De Pauw, N. (2004) Fuzzy rule-based models for
decision-support in ecosystem management. Science of the Total Environment; 319: 112.
Adriaenssens, V., De Baets, B., Goethals, P., De Pauw, N. (2006) Fuzzy knowledge-based
models for prediction of asellus and gammarus in watercourses in Flanders (Belgium).
Ecological Modelling; 195: 3–10.
Arabi, M., Frankenberger, J.R., Engel, B.A., Arnold, J.G. (2008) Representation of agricultural
conservation practices with SWAT. Hydrological Processes; 22 (16): 3042–3055.
Arnold, J.G., Srinivasan, R., Muttiah, R.S., Williams, J.R. (1998) Large area hydrologic
modeling and assessment part I: model development. Journal of American Water
Resources Association; 34 (1): 73–89.
Barbour, M.T., Gerritsen, J., Snyder, B.D., Stribling, J.B. (1999). Rapid bioassessment protocols
for use in streams and wadeable rivers: periphyton, benthic macroinvertebrates and fish,
second edition. EPA 841-B-99-002.U.S. Environmental Protection Agency; Office of
Water; Washington, D.C.
BASINS. (2007) Better Assessment Science Integrating Point and NonpointSources.
Washington, D.C.: U.S. Environmental Protection Agency; Available at:
www.epa.gov/waterscience/basins.
Bedoya, D., Manolakos, E., Novotny, V. (2011) Prediction of biological integrity based on
environmental similarity – Revealing the scale-dependant link between study area and
top environmental predictors. Water Research; 45: 2359-2374.
Bernot, M.J., Tank, J.L., Royer, T.V., David, M.B. (2006) Nutrient uptake in streams draining
agricultural catchments of the midwestern united states. Freshwater Biology; 51: 499 –
509.
Bingner, R.L., Locke, M.A. (2009). AnnAGNPS - A United States Department of Agriculture
Watershed Conservation Management Planning Tool for Non-Point Source Pollution
Control. Water Quality Research to Evaluate the Effects of Agricultural Conservation
Practices Utilized in the United States and India Workshop Proceedings. Allahabad
Agricultural Institute-Deemed University, Allahabad, India. p. 8-9.
Borchardt, D. (1993) Effects of flow and refugia on drift loss of benthic macroinvertebrates:
implications for habitat restoration in lowland streams. Freshwater Biology; 29: 221-227.
Booker, D.J., Dunbar, M.J. (2004) Application of physical habitat simulation (PHABSIM)
modeling to modified urban river channels. River Res Applic; 20: 167-183.

172

Box, G.E.P., Cox, D.R. (1964) An analysis of transformations, Journal of the Royal Statistical
Society, Series B; 26:211-252.
Boyer, D.G. (2008) Fecal coliform dispersal by rain splash on slopes. Agricultural and Forest
Meteorology; 148 :1395 – 1400.
Brenden, T.O., Wang, L., Seelbach, P.W. (2008). A river valley segment classification of
Michigan streams based on fish and physical attributes. Trans. Am. Fish. Soc; 137(6):
1621-1636.
Bunn, S.E., Arthington, A.H. (2002) Basic principles and ecological consequences of altered
flow regimes for aquatic biodiversity. Environmental Management; 30(4): 492-507.
Caissie, D. (2006) The thermal regime of rivers: a review. Freshwater Biology; 51: 1389-1406.
Caraco, N.F., Cole, J.J. (2002) Contrasting impacts of a native and alien macrophyte on
dissolved oxygen in a large river. Ecological Applications; 12(5): 1496-1509.
Carleton, J.N., Park, R.A., Clough, J.S. (2009) Ecosystem modeling applied to nutrient criteria
development in rivers. Environmental Management; 44 :485-492.
Carpenter, S.R., Caraco, N.F., Correll, D.L., Howarth, R.W., Sharpley, A.N., Smith, V.H.
(1998) Non point pollution of surface waters with phosphorus and nitrogen. Ecological
Application; 8(3): 559-568.
Carrascal, L.M., Galván, I., Gordo, O. (2009) Partial least squares regression as an alternative to
current regression methods used in ecology. Oikos; 118: 681–690.
Casper, A.F., Dixon, B., Earls, J., Gore, J.A. (2011) Linking a spatially explicit watershed model
(SWAT) with an in-stream fish habitat model (PHABSIM): a case study of setting
minimum flows and levels in a low gradient, sub-tropical river. River Res. Applic.; 27(3):
269-282.
Cheimonopoulou, M.T., Bobori, D., Theocharopoulos, I., Lazaridou, M. (2011) Assessing
ecological water quality with macroinvertebrates and fish: a case study from a small
mediterranean river. Environmental Management; 47 (2): 279 -290.
Chinkuyu, A. J., Meixner, T., Gish, T.J., Daughtry, C.S. (2004) The importance of seepage zones
in predicting soil moisture content and surface runoff using GLEAMS and RZWQM.”
Transactions of ASABE; 47(2): 427-438.
Chu, C., Jones, N.E. Mandrak, N.E., Piggott, A.R., Minns, C.K. (2008) The influence of air
temperature, groundwater discharge, and climate change on the thermal diversity of
stream fishes in southern Ontario watersheds. Can. J. Fish.Aquatic Science; 65: 297–308.

173

Chen, Q.W., Mynett, A.E. (2003) Integration of data mining techniques and heuristic knowledge
in fuzzy logic modelling of eutrophication in Taihu lake. Ecological Modelling; 162 (12), 55–67.
Cooper, M., Rediske, R., Uzarski, D.G., Burton, T.M. (2009).Sediment contamination and faunal
communities in two subwatersheds of mona lake, michigan. J. Environmental Quality;
38: 1255-1265.
Covert, A. (1997) Comparison of U.S. Geological Survey and Ohio Environmental Protection
Agency Fish-Collection Methods Using the Index of Biotic Integrity and Modified Index
of Well-Being, 1996–97. USGS-Water Resources Investigations Repot 00-4255.
Dale, V.H., Polasky, S. (2007) Measures of the effects of agricultural practices on ecosystem
services. Ecological Economics; 64: 286-296
Del Rio-Rodriguez, R.E., Inlis, V., Millar, S.D. (1997) Survival of escherichia coli in the
intestine of fish. Aquaculture Research; 28: 257-264.
Davis, W.S. (1995) Biological assessment and criteria: building on the past. Biological
Assessment and Criteria. Edited by Wayne S. Davis and Thomas Simon; Lewis
Publishers.
Dedecker, A.P., Goethals, P.L.M., Gabriels, W., De Pauw, N. (2004) Optimization of artificial
neural network (ANN) model design for prediction of macroinvertebrates in the Zwalm
river basin (Flanders, Belgium), Ecol. Model.; 174: 161–173.
Degerman, E., Sers, B., Törnblom, J. and Angelstam, P. (2004). Large woody debris and brown
trout in small forest streams – towards targets for assessment and management of riparian
landscapes. Ecol. Bull.; 51: 233–239.
DeShon, J.E. (1995) Development and application of the invertebrate community index (ICI).
Pages 217–243 in W. S. Davis and T. P. Simon, editors. Biological assessment and
criteria. Lewis, Boca Raton, Florida, USA.
Diana, M., Allan, J.D., Infante, D. (2006). The influence of physical habitat and land cover on
stream fish assemblages in Southeastern Michigan. American Fisheries Society
Symposium; 48: 359-374.
Diebel, M.W., Maxted, J.T., Jensen, O.P., Zanden, M.J.V. (2010) A spatial autocorrelative
model for targeting stream restoration to benefit sensitive nongame fishes. Can. J. Fish.
Aquatic Sci.; 67: 165-176.
Domagalski, J., Ator, S., Coupe, R., McCarthy, K., Lampe, D., Sandstrom, M., Baker, N. (2008)
Comparative study of transport processes of nitrogen, phosphorus, and herbicides to
streams in five agricultural basins, USA. J. Environ. Qual.; 37: 1158-1169.

174

Duriancik, L.F., Bucks, D., Dobrowolski, J.P., Drewes, T., Sckles, S.D., Jolley, L., Kellogg,
R.L., Lund, D. Makuch, J.R., O’Neill, M.P., Rewa, C.A., Walbridge, M.R., Parry, R.,
Weltz, M.A. (2008) The first five years of the conservation effects assessment project.
Journal of Soil and Water Conservation; 63(6): 185-188.
Fattal, B., Dotan, A., Tchorsh, Y. (1992) Rates of experimental microbiological contamination of
fish exposed to polluted water. Water Research; 26(12): 1621-1627.
Fiener, P., Auerswald, K. (2002) Effectiveness of grassed waterways in reducing runoff and
sediment delivery from agricultural watersheds. Journal of Environmental Quality;
32:927-936.
Flinders, C., Horwitz, R., Belton, T. (2008) Relationship of fish and macroinvertebrate
communities in the mid-atlantic uplands: implications for integrated assessments.
Ecological Indicators; 8: 588-598.
Fore, L.S., Karr, J.R., Wisseman, R.W. (1996) Assessing invertebrate responses to human
activities: evaluating alternative approaches. J.N. Am. Benthol. Soc.; 15(2):212-231.
Fowler, H.J., Blenkinsop, S., Tebaldi, C. (2007) Linking climate change modelling to impacts
studies: recent advances in downscaling techniques for hydrological modelling.
International Journal of Climatology; 27(12): 1547-1578.
Francis-Floyd, R. (1992) Dissolved oxygen for fish production. University of Florida
Extension.Institute of Food and Agricultural Sciences. Fact Sheet FA-2.
Gagliardi, J.V., Karns, J.S. (2000) Leaching of escherichia coli O157:H7 in diverse soils under
various agricultural management practices. Applied and Environmental Microbiology;
66(3): 877–883.
Garvey, J.E., Whiles, M.R., Streicher, D. (2007) A hierarchical model for oxygen dynamics in
streams. Can. J. Fish Aquat. Sci.; 64: 1816-1827.
Gassman, P.W., Reyes, M.R., Green, C.H., Arnold, J.G. (2007) The soil and water assessment
tool: historical development, applications, and future research directions. Transactions of
ASABE; 50: 1211–1250.
Geldreich, E.E., Clarke, N.A. (1966) Bacterial pollution indicators in the intestinal tract of
freshwater fish. Applied Microbiology; 14 (3): 429-437.
Gentry, L.E., David, M.B., Royer, T.V., Mitchell, C.A., Starks, K.M. (2007) Phosphorus
transport pathways to streams in tile-drained agricultural watersheds. J. Environ. Quality;
36: 408-415.
Gillium, J.W., Parsons, J.E., Mikkelsen, R.L. (1997) Nitrogen dynamics and buffer zones. In
buffer zones: their processes and potential in water protection. Edited by Nick Haycock,
Tim Burt, Keith Goulding, and Gilles Pinay.
175

Gillium, R.J. (2007) Pesticides in U.S. streams and groundwater. Environmental Science and
Technology; 3409-3414.
Giri, S., Nejadhashemi, A.P., Woznicki, S. (2011) Evaluation of targeting methods for
implementation of best management practices in the saginaw river watershed. Journal of
Environmental Management; submitted (2011).
Griffith, Michael., Daniel, B., Morrison, M.A., Troyer, M.E., Lazorchak, J.M., SchubauerBerigan, J.P. (2009). Linking excess nutrients, light and fine bedded sediments to impacts
on faunal assemblages in headwater agriculture streams. JAWRA; 45:6.
Gupta, H.V., Sorooshian, S., Yapo, P.O. (1999) Status of automatic calibration for hydrologic
models: Comparison with multilevel expert calibration. Journal of Hydrologic
Engineering; 4 (2): 135-143.
Harding, J.S., Benfield, E.F., Bolstad, P.V., Helfman, G.S., Jones, E.B.D. (1998) Stream
biodiversity: The ghost of land use past. Proceedings of National Academy of Science
USA; 95: 14843–14847.
Harman, W.L., Wang, E., Williams, J.R. (2004) Reducing atrazine losses: water quality
implications of alternative runoff control practices. J. Environmental Quality; 33:7-12.
Hastie, T., Tibshirani, R., Friedman, J. (2009) Kernel smoothing methods. The elements of
statistical learning. Springer Series in Statistics; 1-28.
Heathman, G.C., Flanagan, D.C., Larose, M., Zuercher, B.W. (2008) Applications of the soil and
water assessment tool and annualized agricultural non-point source models in the st.
joseph river watershed. Journal of Soil and Water Conservation; Nov/Dec: 63, 6.
Heathwaite, A.L., Johnes, P.J. (1996) Contribution of nitrogen species and phosphorus fractions
to stream water quality in agricultural catchments. Hydrological Processes; 10: 971 –
983.
Heimlich, R. (2003) USDA agricultural resources and environmental indicators. Agricultural
Handbook No. AH722. 2.3, 32. Found at
http://www.ers.usda.gov/publications/arei/ah722/.
Heitke, J.D., Pierce, C.L., Gelwicks, G.T., Simmons, G.A., Siegwarth, G.L. (2006) Habitat, land
use, and fish assemblage relationships in Iowa streams: preliminary assessment in an
agricultural landscape; 287–304 in R. M. Hughes, L. Wang, and P. W. Seelbach
(editors).Influences of landscapes on stream habitats and biological
assemblages.Symposium 48.American Fisheries Society, Bethesda, Maryland.
Helfrich, L.A., Weigmann, D.L., Hipkins, P., Stinson, E.R. (2009) Pesticides and aquatic
animals: a guide to reducing impacts on aquatic systems. Virgina Tech University.

176

Hill, W.R., Fanta, S.E., Roberts, B.J. (2009) Quantifying phosphorus and light effects in stream
algae. Limnol.Oceanogr; 54(1), 368–380
Hill, W.R., Ryon, M.G., Smith, J.G., Adams, S.M., Boston, H.L., Stewart, A.J. (2010) The role
of periphyton in mediating the effects of pollution in a stream ecosystem. Environmental
Management; 45:563-576.
Hilsenhoff, W.L. (1987) An improved biotic index of organic stream pollution. Great Lakes
Entomologist; 20:31-39.
Hilsenhoff, W.L. (1988) Rapid field assessment of organic pollution with a family-level biotic
Index. J.N. Am. Benthol. Soc.; 7(1):65-68.
Hinz, L.C., Wiley, M. J. (1998) Growth and production of juvenile trout in michigan streams:
influence of potential ration and temperature. MDNR Fisheries Report. 2042.
Hoar, T., Nychka, D. (2008) Statistical downscaling of the community climate system model
(CCSM) monthly temperature and precipitation projections.White paper.Boulder, Colo.:
Institute for Mathematics Applied to Geosciences, NationalCenter for Atmospheric
Research (IMAGe/NCAR); Available at: www.gisclimatechange.org/Downscaling.pdf.
Hoffmann, C.C., Kjaergaard, C., Uusi-Kamppa, J., Hansen, H., Kronvang, B. (2009) Phosphorus
retention in riparian buffers: review of their efficiency. J. Environ. Qual.; 38: 1942-1955.
Huang, Y.B., Lan, Y.B., Thomson, S.J., Fangc, A., Hoffmannb, W.C., Laceyd, R.E. (2010)
Development of soft computing and applications in agricultural and biological
engineering.Computers and Electronics in Agriculture; 71: 107–127.
Hutchens, J.J., Schuldt, J.A., Richards, C., Johnson, L.B., Host, G.E., Breneman, D.H. (2009)
Multi-scale mechanistic indicators of midwestern USA stream macroinvertebrates.
Ecological Indicators; 9: 1138-1150.
Icaga Y. (2007) Fuzzy evaluation of water quality classification. Ecological Indicators; 7 (3):
710-718.
Infante, D., Allan, D., Linke, S., Norris, R.H. (2008) Relationship of fish and macroinvertebrate
assemblages to environmental factors: implications for community concordance.
Hydrobiologia; 623: 87-103.
Izagirre, O., Serr, A., Guasch, H., Elosegi, A. (2009) Effects of sediment deposition on
periphytic biomass, photosynthetic activity and algal community structure. Science of the
Total Environment; (2009) 5694–5700.
Jamieson, R., Gordon, R., Sharples, K., Stratton, G. and Madani, A. (2002). Movement and
persistence of fecal bacteria in agricultural soils and subsurface drainage water. Can.
Biosystems Eng.; 44: 6.1–6.10.
177

Jensen, M., Reynolds, K., Andreasen, J., Goodman, I. (2000) A knowledge-based approach to
the assessment of watershed condition. Environmental Monitoring and Assessment; 64
(1): 271-283.
Johnson, L.B., Host, G.E. (2010) Recent developments in landscape approaches for the study of
aquatic ecosystems. Journal of the North American Benthological Society; 29:41–66.
Johnson, L.B., Richards, C., Host, G., Arthur, J. (1997) Landscape influences on water chemistry
in Midwestern stream ecosystems. Freshwater Biology; 37: 193–208.
Johnson, D.M., Mueller, R. (2010) The 2009 Cropland Data Layer; Available on the Internet.
http://www.nass.usda.gov/research/Cropland/docs/JohnsonPE&RS_Nov2010.pdf
Jones, O.R., Hauser, V.L., Popham, T.W. (1994) No-tillage effects on infiltration, runoff, and
water conservation on dryland. Transactions of the ASAE; 37(2): 473-479.
Jorde, K., Schneider, M., Zoellner, F. (2000) Analysis of in-stream habitat quality—preference
functions and fuzzy models. In: Wang and Hu, Editors, Stochastic Hydraulics, Balkema,
Rotterdam; 671–680.
Justus, B.G., Petersen, J.C., Femmer, S.R., Davis, J.V., Wallace, J.E. (2010) A comparison of
algal, macroinvertebrate, and fish assemblage indices for assessing low-level nutrient
enrichment in wadeable Ozark streams. Ecological Indicators; 10: 627–638.
Karr, J. (1981). Assessment of biological integrity using fish communities.
Fisheries.<http://www.epa.gov/bioiweb1/pdf/AssessmentofBioticIntegrityUsingFishCom
munities.pdf
Karr, J.R., Dudley, D.R. (1981). Ecological perspective on water quality goals. Environmental
Management ; 5: 55- 68.
Karr, J.R., Fausch, K.D., Angermeier, P.L., Yant, P.R. (1986) Assessing biological integrity in
running waters: a method and its rationale. Spec. Pub 5. Champaign, IL: Illinois Natural
History Survey; 28.
Karr, J. (1991) Biological integrity: A long-neglected aspect of water resources Management.
Ecological Applications; 1 (1): 66 – 84.
Karr, J. (1996) Ecological integrity and ecological health are not the same. from Engineering
with Ecological Constraints. National Acadamy Press.
Karr, J.R. (1999) Defining and measuring river health. Freshwater Biology; 41: 221-234.
Kerans, B.L., Karr, J. R. (1994) A benthic index of biotic integrity (B-IBI) for rivers of the
tennessee valley. Ecological Applications; 4(4): 768-785.

178

Kuhnle, R.A., Bingner, R.L., Langendoen, E.J., Simon, A., Wilson, C.G., Alonso, C.V., Shields,
F.D. (2005) Goodwin creek experimental watershed-assessment of conservation and
environmental effects. 2005 ASAE Annual International Meeting.Paper # 052130.
Lammert, M., Allan, D. (1999) Assessing biotic integrity of streams: effects of scale in
measuring the influence of land use/cover and habitat structure on fish and
macroinvertibrates. Environmental Management; 23:2.
Lenat, D.R. (1993) A biotic index for the southeastern united states: derivation and list of
tolerance values, with criteria for assigning water-quality ratings. J.N. Am. Benthol. Soc.;
12(3): 279-290.
Langendoen, E.J. (2000) CONCEPTS-conservation channel evolution and pollutant transport
system. USDA-ARS Research Report No. 16.
Langendoen, E.J., Lowrance, R.R., Simon, A. (2009) Assessing the impact of riparian processes
on streambank stability. Ecohydrology; 2: 360-369.
Liu, Y., Yang, W., Wang, X. (2007) GIS-based integration of SWAT and REMM for estimating
water quality benefits of riparian buffers in agricultural watersheds. Transactions of the
ASABE; 50(5): 1549-1563.
Liu, X., Zhang, X., Zhang, M. (2008) Major factors influencing the efficacy of vegetated buffers
on sediment trapping: a review and analysis. J. Environ. Qual.; 37: 1667-1674.
Loperfido, J.V., Just, C.L., Schnoor, J.L. (2009) High-frequency diel dissolved oxygen stream
data modeled for variable temperature and scale. Journal of Environmental Engineering;
135:12.
Love, B.J., Nejadhashemi, A.P. (2011) Environmental impact analysis of biofuel crops
expansion in the saginaw river watershed. Journal of Biobased Materials and Bioenergy;
5: 1-25.
Lowrance, R., Altier, L.S., Williams, R.G., Inamdar, S.P., Inamdar, S.P., Sheridan, J.M., Bosch,
D.D., Hubbard, R.K., Thomas, D.L. (2000) REMM: the riparian ecosystem management
model. Journal of Soil and Water Conservation; 55(1): 27-34.
Lyautey, E., Lu, Z., Lapen, D.R., Wilkes, G., Scott, A., Berkers, T., Edge, T.A., Topp, E. (2010)
Distribution and diversity of escherichia coli populations in the south nation river
drainage basin, eastern Ontario, Canada. Applied and Environmental Microbiology;
76(5): 1486–1496.
Lyons, J. (1992).Using the index of biotic integrity (IBI) to measure environmental quality in
warmwater streams of Wisconsin. U.S. Department of Agriculture, Forest Service, North
Central Experiment Station, St. Paul, Minnesota, General Technical Report NC-149.

179

Lyons, J., Wang, L., Simonson, T.D. (1996) Development and validation of an index of biotic
integrity for cold-water streams in Wisconsin. North Am J Fisheries Manage.; 16:241–
256
Lyons, J., Weigel, B.M., Paine, L.K., Undersander, D.J. (2000) Influence of intensive rotational
grazing on bank erosion, fish habitat quality, and fish communities in southwestern
Wisconsin trout streams. Journal of Soil and Water Conservation; 55 (3): 271.
Lyons, J., Piette, R.R., Niermeyer, K.W. (2001) Development, validation, and application of a
fish-based index of biotic integrity for Wisconsin’s large warmwater rivers. Transactions
of the American Fisheries Society; 130: 1077–1094.
Lyons, J., Zorn, T. Stewart, J., Seelbach, P., Wehrly, K. Wang, L. (2009) Defining and
characterizing coolwater streams and their fish assemblages in Michigan and Wisconsin,
USA. North American Journal of Fisheries Management; 29: 1130-1151.
Mahmood, Z., Khan, S. (2009) On the use of k-fold cross-validation to choose cutoff values and
assess the performance of predictive models in stepwise regression. Biostatistics; 5 (1):
25.
Mamdani, E.H. (1977) Application of fuzzy logic to approximate reasoning using linguistic
systems.IEEE Transactions on Computers; 26: 1182–1191.
Marchini, A. (2011) Modelling Ecological processes with fuzzy logic approaches. Modelling
Complex Ecological Dynamics; Part 2, 133-145, DOI: 10.1007/978-3-642-05029-9_10.
Maresch, W., Walbridge, M.R., Kugler, D. (2008) Enhancing conservation on agricultural
landscapes: a new direction for the conservation effects assessment project. Journal of
Soil and Water Conservation; 63: 6.
Maret, T.R., Robinson, C.T., Minshall, G.W. (1997) Fish assemblages and environmental
correlates in the least-disturbed streams of the upper Snake River basin.Transactions of
the American Fisheries Society; 126: 200–216.
Maret, T., Konrad, C., Tranmer, A. (2010) Infleunce of environmental factors on biotic
responses to nutrient enrichment in agricultural streams. Journal of the American Water
Resources Association; 46 (3): 498-513.
MathWorks. (2010) MATLAB Fuzzy Logic Toolbox 2 User’s Guide 2010.
Mausbach, M.J., Dedrick, A.R. (2004) The length we go: measuring environmental benefits of
conservation practices. Journal of Soil and Water Conservation; 59: 5.
Meyer, J.L. (1997) Stream health: incorporating the human dimension to advance stream
ecology. Journal or North American Benthol. Soc.; 16(2): 439-447.

180

MDEQ. (1997). GLEAS Procedure #51 Survey Protocols for Wadable Rivers. Chapter 25A in
Schneider, James C. (ed.) 2000. Manual of fisheries survey methods II: with periodic
updates. Michigan Department of Natural Resources, Fisheries Special Report 25, Ann
Arbor.
Miltner, R.J. (2010) A method and rational for deriving nutrient criteria for small rivers and
streams in Ohio. Environmental Management; 45: 842-855.
Miltner, R.J., Rankin, E.T. (1998) Primary nutrients and the biotic integrity of rivers and
streams.Freshwater Biology; 40, 145–158.
Minnesota Pollution Control Agency. (2003) Establishing relationships among in stream nutrient
concentrations, phytoplankton abundance and composition, fish IBI and biochemical
oxygen demand in Minnesota USA rivers final report to USEPA region V.
Minnesota Pollution Control Agency. (2008) Bacteria: sources, types, impact on water quality - a
general overview. Water Quality/Impaired Waters #3.20.
Mishra, A., Benham, B., Mostaghimi, S. (2008) Bacterial transport from agricultural lands
fertilized with animal manure. Water Air Soil Pollut.; 189: 127-134.
Moriasi, D.N., Arnold, J.G., Van Liew, M.W., Binger, R.L., Harmel, R.D., Veith ,V. (2007)
Model evaluation guidelines for systematic quantification of accuracy in watershed
simulations. Transactions of the ASABE; 50(3): 885-900.
Morris, J., Mansfield, E., Sherman, J. (1986) Failures to detect moderate effects with ordinary
least squares-moderated multiple regression: some reasons and a remedy. Psychological
Bulletin 1986; 99 (2): 282-288.
Mouton, A., Schneider, M., Peter, A., Holzer, G., Muller, R., Goethals, P., De Pauw, N. (2008)
Optimisation of a fuzzy physical habitat model for spawning European grayling
(Thymallusthymallus L.) in the Aareriver (Thun, Switzerland). Ecological Modelling;
215: 122132.
MSU Weed Science. (2010) 2010 Weed Control Guide for Crops. Introduction.
http://msuweeds.com/publications/2010_weed_guide/
Natural Resources Conservation Services (NRCS). (2001) Buffer strips: common sense
conservation. http://www.nrcs.usda.gov/feature/buffers/
Natural Resources Conservation Service (NRCS). (2010). Conservation effects assessment
project (CEAP). http://www.nrcs.usda.gov/technical/NRI/ceap/
Natural Resources Conservation Service (NRCS). (2006) National conservation practice
standards – NHCP. http://www.nrcs.usda.gov.

181

Nayak, P.C., Sudheer, K.P., Ramasastri, K.S. (2005) Fuzzy computing based rainfall-runoff
model for real time flood forecasting. Hydrological Processes; 19 (4): 955– 968.
Neitsch, S.L., Arnold, J. G., Kiniry, J.R., Williams, J.R. (2009) Overview of soil and water
assessment tool (SWAT) model. in soil and water assessment tool: Global Applications.
Edited by Jeff Arnold and others.WASWC Special Publication #4.
Neitsch, S.L., Arnold, J.G., Kiniry, J.R., Williams, J.R. (2005) Soil and water assessment tool
theoretical documentation, version 2005. Temple, Tex.: USDA-ARS Grassland, Soil and
Water Research Laboratory; Available on the Internet.
www.brc.tamus.edu/swat/doc.html.
Nejadhashemi, A.P., Shirmohammadi, A., Montas, H., Sheridan, J., Bosch, D. (2008) Watershed
physical and hydrological effects on baseflow separation. Journal of Hydrologic
Engineering; 13(10): 971-980.
Newcombe, C.P., Jensen, J.O. (1996) Channel suspended sediment and fisheries: a synthesis for
quantitative assessment of risk and impact. North American Journal of Fisheries
Management; 16: 693-727.
Nuhfer, A.J., Baker, E.A. (2004) A long-term field test of habitat change predicted by
PHABSIM in relation to brook trout population dynamics during controlled flow
reduction experiments. Michigan DNR, Fisheries Research Report 2068, 2004.
Ohio EPA. (1989) Biological criteria for the protection of aquatic life: volume III: standardized
biological field sampling and laboratory methods for assessing fish and
macroinvertebrate communities. Ecological Assessment Section Division of Water
Quality Planning and Assessment.
Parajuli, P.B., Mankin, K.R., Barnes, P.L. (2008) Applicability of targeting vegetative filter
strips to abate fecal bacteria and sediment yield using SWAT. Agricultural Water
Management; 95(10): 1189-1200.
Parjuli, P.B., Nelson, N.O., Frees, L.D., Mankin, K.R. (2009) Comparison of AnnAGNPS and
SWAT model and simulation results in USDA-CEAP agricultural watersheds in southcentral Kansas. Hydrological Processes; 23: 748-763.
Park, R.A., Clough, J.S., Wellman, M.C. (2008) AQUATOX: modeling environmental fate and
ecological effects in aquatic ecosystems. Ecological Modeling; 213: 1-15.
Pimentel, D. (2005). Environmental and economic costs of the application of pesticides primarily
in the United States, Environment, Development and Sustainability; 7: 229–252.
Plafkin, J.L., Barbour, M.T., Porter, K.D., Gross, S.K., Hughes, R.M. (1989) Rapid
bioassessment protocols for use in stream and rivers: benthic macroinvertebrates and fish.
EPA/444/4-89-011.US Environmental Protection Agency, Washington, DC, 150 pp.
182

Poff, N.L., Allan, J.D., Bain, M.B., Karr, J.R., Prestegaard, K.L., Richter, B.D., Sparks, E.,
Stromberg, J.C. (1997) The Natural flow regime: a paradigm for river conservation and
restoration. BioScience; 47:11, 769-784.
Poff, L.N., Zimmerman, J.K. (2010) Ecological responses to altered flow regimes: a literature
review to inform the science and management of environmental flows. Freshwater
Biology; 55: 194-205.
Poole, G., Berman, C. (2001) An ecological perspective on in-stream temperature: natural heat
dynamics and mechanisms of human-caused thermal degradation. Environmental
Management; 27(6): 787-802.
Poole, G. (2010) Stream geomorphology as a physical science basis for advances in stream
ecology. Journal of the North American Benthological Society; 29: 12–25.
Power, G., Brown, R.S., Imhof, J.G. (1999) Groundwater and fish – insights from northern North
America.Hydrological Processes; 13:401-422.
Rankin, E.T. (1989) The qualitative habitat evaluation index [QHEI]: rationale, methods, and
application. Ohio EPA.Ecological Assessment Section.
Rashleigh, B., Barber, M.C., Walters, D.M. (2009) Foodweb modeling for polychlorinated
biphenyls (PCBs) in the twelvemile creek arm of lake hartwell, South Carolina, USA.
Ecological Modeling; 220: 254-264.
Robertson, G.P., Groffman, P.M. (2007) Nitrogen transformation. Pages 341-364 in E.A. Paul,
ed. Soil Microbiology, Biochemistry, and Ecology.Springer, New York, New York,
USA.
Robertson, P., Vitousek, P. (2009) Nitrogen in agriculture: balancing the cost of an essential
resource. Annu. Rev. Environ. Resour.; 34: 97–125.
Roodsari, R.M., Shelton, D.R., Shirohammadi, A., Pachepsky, Y.A., Sadeghi, A.M., Starr, J.L.
(2005) Fecal coliform transport as affected by surface condition.Transactions of ASAE;
48(3): 1055−1061.
Rossi, C.G., Dybala, T.J., Moriasi, D.N., Arnold, J.G., Amonette, C. Marek, T. (2008) Hydologic
calibration and validation of soil and water assessment tool for the leon river watershed.
Journal of Soil and Water Conservation; Nov/Dec; 63: 6.
Roth, N., Allan, D., Erickson, D.L. (1996). Landscape influences on stream biotic integrity
assessed at multiple spatial scales. Landscape Ecology; 11(3): 141-156.
Rowe, D., Pierce, C., Wilton, T.F. (2009) Physical habitat and fish assemblage relationships with
landscape variables at multiple spatial scales in wadeable Iowa streams. N. American
Journal of Fisheries Management; 29: 1333-1351.
183

Royer, T.V., David, M.B., Gentry, L.E. (2006) Timing of riverine export of nitrate and
phosphorus from agricultural watersheds in Illinois: implications for reducing nutrient
loading to the Mississippi river. Environ. Sci. Technol.; 40: 4126-4131.
Santhi, C., Srinivasan, R., Arnold, J.G., Williams, J.R. (2006) A modeling approach to evaluate
the impacts of water quality management plans implemented in a watershed in Texas.
Environmental Modelling and Software; 21: 1141-1157.
SAS/STAT. (2010) SAS/STAT 9.2 User’s Guide, Second Edition; Avaliable on the Internet.
http://support.sas.com/documentation/cdl/en/statug/63033/HTML/default/viewer.htm#sta
tug_transreg_sect015.htm.
Schafer, R.B., Caquet, T., Siimes, K., Mueller, R., Lagadic, L., Liess, M. (2007) Effects of
pesticides on community structure and ecosystem functions in agricultural streams of
three biogeographical regions in Europe. Science of the Total Environment; 382: 272 –
285.
Schneider, K.N., Winemiller, K.O. (2008) Structural complexity of woody debris patches
influences fish and macroinvertebrate species richness in a temperate floodplain- river
system. Hydrobiology; 610 (1).
Schulz, R., Liess, M. (1999) A field study of the effects of agriculturally derived insecticide
input on stream macroinvertebrate dynamics. Aquatic Toxicology; 46: 155-176.
Seager, J., Milne, I., Mallett, M., Sims, I. (2000) Effects of short-term oxygen depletion on fish.
Environmental Toxicology and Chemistry; 19(12): 2937-2942.
Seelbach, P.W., Wiley, M.J. (1997) Overview of the Michigan rivers inventory project.Michigan
Department of Natural Resources, Fish Division Technical Report 97-3; Ann Arbor,
Michigan.
Sharpley, A.N., Gburek, W.J., Folmar, G., Pionke, H.B. (1999) Sources of phosphorus exported
from an agricultural watershed in Pennsylvania. Agricultural Water Management; 41:7789.
Shields, F.D., Langendoen, E., Doyle, M.W. (2006) Adapting existing models to examine effects
of agricultural conservation programs on stream habitat quality. Journal of the American
Water Resources Association; 42(1): 25-33.
Sims, J.T., Simard, R.R., Joern, B.C. (1998) Phosphorus loss in agricultural drainage: historical
perspective and current research. J. Environmental Quality ; 27: 277-293.
Solomon, K.R., Carr, J.A., Du Preez, L.H., Giesy. J.P., Kendall, R.J., Smith, E.E., Van Der
Kraak, G.J. (2008) Effects of atrazine on fish , amphibians, and aquatic reptiles: a critical
review. Critical Reviews in Toxicology; 38: 721-772.
184

Sood, A., Ritter, W.F. (2010) Evaluation of best management practices in millsboro pond
watershed using soil and water assessment tool (SWAT) model. J. Water Resource and
Protection; 2: 403-412.
Sparks, R.E. (1995) Need for ecosystem management of large rivers and their floodplains.
BioScience; 45(3): 168-182.
StatSoft. (2011) STATISTICA 2011.
Stauffer, J.C., Goldstein, R.M., Newman, R.M. (2000) Relationship of wooded riparian zones
and runoff potential to fish community composition in agricultural streams. Canadian
Journal of Fisheries and Aquatic Sciences; 57: 307–316.
Steen, P., Wiley, M., Schaeffer, J. (2010) Predicting future changes in muskegon river watershed
game fish distributions underfuture land cover alteration and climate change scenarios.
Transactions of the American Fisheries Society; 139: 396–412.
Stewart, J.S., Wang, L., Lyons, J. Horwatich, J.A., Bannerman, R. (2001) Influences of
watershed, riparian-corridor, and reach-scale characteristics on aquatic biota in
agricultural watersheds. Journal of American Water Resources Association; 37(6): 1475
– 1487.
Sullivan, B.E., Rigsby, L.S., Berndt, A., Jones-Wuellner, M., Simon, T.P., Lauer, T., Pyron, M.
(2004) Habitat influence on fish community assemblage in an agricultural landscape in
four east central Indiana streams. J. of Freshwater Ecology; 19(1): 141-148.
Sullivan, T.J., Moore, J.A., Thomas, D.R., Mallery, E., Snyder, K.U., Wustenberg, M., Mackey,
S.D., Moore, D.L. (2007) Efficacy of vegetated buffers in preventing transport of fecal
coliform bacteria from pasturelands. Environmental Management; 40:958–965.
Sutela, T., Vehanen, T., Jounela, P. (2010) Response of fish assemblages to water quality in
boreal rivers. Hydrobiologia; 641: 1-10.
Takagi, T., Sugeno, M., (1985) Fuzzy identification of systems and its application to modeling
and control. IEEE Trans. Syst. Man Cybernet; 15 (1), 116-132.
Trautman, M.B. (1981) The fishes of Ohio. Ohio State University Press, Columbus.
USDA.State Soil Geographic (STATSGO) Data Base. (1995) Data use information;
http://www.nrcs.usda.gov/technical/techtools/statsgo_db.pdf
USDA.National conservation practice standards.Conservation Practices. (2010);
http://www.nrcs.usda.gov/wps/portal/nrcs/detailfull/?ss=16&navtype=SUBNAVIGATIO
N&cid=nrcs143_026849&navid=120160340000000&pnavid=120160000000000&positi
on=Not%20Yet%20Determined.Html&ttype=detailfull&pname=Conservation%20Practi
ces%20|%20NRCS

185

USEPA, Nature Conservancy (1994). Conservation of biological diversity in the great lakes
basin ecosystem: issues and opportunities.
http://www.epa.gov/glnpo/ecopage/glbd/issues/
USEPA (2000) Guiding principles for constructed treatment wetlands.
http://www.epa.gov/owow/wetlands/pdf/constructed.pdf
USEPA (2006a).Fecal Bacteria. Monitoring and assessing water quality.
http://www.epa.gov/volunteer/stream/vms511.html.
USEPA (2006b) Dissolved oxygen and biochemical oxygen demand. Monitoring and assessing
water quality. http://www.epa.gov/volunteer/stream/vms52.html
USEPA (2009a) Agriculture landuse overview.
http://www.epa.gov/oecaagct/ag101/landuse.html
USEPA (2009b) Pesticide types. About pesticides.
http://www.epa.gov/pesticides/about/types.htm
USEPA (2009c) Invertebrates as indicators. Biological indicators of watershed health.
http://www.epa.gov/bioindicators/html/invertebrate.html
USEPA (2009d) Water quality criteria for nitrogen and phosphorus pollution.
http://www.epa.gov/waterscience/criteria/nutrient/basic.htm.
USEPA (2009e) Soil preparation. Ag 101. http://www.epa.gov/oecaagct/ag101/cropsoil.html.
USEPA (2011a) Rivers and streams. http://water.epa.gov/type/rsl/index.cfm
USEPA (2011b) Water quality criteria for nitrogen and phosphorus. Water: nutrients;
http://water.epa.gov/scitech/swguidance/standards/criteria/nutrients/
USEPA (2011c) Great lakes areas of concerns. Saginaw river and bay area of concern;
http://www.epa.gov/grtlakes/aoc/sagrivr.html.
USEPA (2011d) Ecoregions of Michigan. Western ecology division;
http://www.epa.gov/wed/pages/ecoregions/mi_eco.htm
USGS (2000) Pesticides in stream sediment and aquatic biota. Fact Sheet 09200.http://water.usgs.gov/nawqa/pnsp/pubs/fs09200/fs09200.pdfBased on the book by
L.H. Nowell, P.D. Capel, and P.D. Dileanis, 1999, Pesticides in Stream Sediment and
Aquatic Biota—Distribution, Trends, and Governing Factors: Boca Raton, Fla., CRC
Press, Pesticides in the Hydrologic System series, v. 4, 1040.
USGS (2010) Physical habitat simulation (PHABSIM) software. USGS Fort Collins Science
Center; http://www.fort.usgs.gov/products/software/phabsim/

186

USGS (2011) The effects of urbanization on water quality;
http://ga.water.usgs.gov/edu/urbanquality.html
Van Broekhoven, E., Adriaenssens, V., De Baets, B., Verdonschot, P.F. (2006) Fuzzy rule-based
macroinvertebrate habitat suitability models for running waters, Ecol. Model.; 198: 7184.
Van Griensven, A., Meixner, T., Grunwald, S., Bishop, T., Diluzio, M., Srinivasan, R. (2006) A
global sensitivity analysis tool for the parameters of multi-variable catchment models.
Journal of Hydrology; 324 (1-4): 10-23.
Van Sickle, J., Baker, J., Herlihy, A., Bayley, P., Gregory, S., Haggerty, P., Ashkenas, L., Li, J.
(2004) Projecting the biological condition of streams under alternative scenarios of
human land use. Ecological Applications; 14: 368–380.
Vannote, R.L., Minshall, G.W., Cummins, K.W., Sedell, J.R., Cushing, C.E. (1980) The river
continuum concept.Canadian Journal of Fisheries and Aquatic Sciences; 37: 130–137.
Vidon, P., Campbell, M., Gray, M. (2007) Unrestricted cattle access to streams and water quality
in till landscape of the Midwest. Agricultural Water Management; 95: 322 – 330
Waite, I., Brown, L., Kennen, J., May, J., Cuffney, T., Orlando, J., Jones, K. (2010) Comparison
of watershed disturbance predictive models for stream benthic macroinvertebrates for
three distinct ecoregions in western US. Ecological Indicators; 10 (6): 1125-1136.
Wang, H., Hondzo, M., Poole, C.X., Spacie, A. (2002) Dissolved oxygen dynamics of streams
draining an urbanized and an agricultural catchment. Ecological Modeling; 160: 145 –
161.
Wang, L., Seelbach, P.W., Lyons, J. (2006) Effects of levels of human disturbance on the
influence of catchment, riparian, and reach-scale factors on fish assemblages. American
Fisheries Society Symposium; 48: 199-219.
Wang, L., Robertson, D., Garrison, P. (2007) Linkages between nutrients and assemblages of
macroinvertibrates and fish in wadeable streams: implication to nutrient criteria
development. Environ. Manage; 39: 194-212.
Wang, L., Brenden, T., Seelbach, P.W., Cooper, A., Allan, D., Clark, R.J., Wiley, M. (2008)
Landscape based identification of human disturbance gradients and references for streams
in Michigan. Environmental Monitoring and Assessment; 141: 1–17.
Wang, X. Gassman, P.W., Williams, J.R., Kemanian, A.R. (2008) Modeling the impacts of soil
management practices on runoff, sediment yield, maize productivity, and soil organic
carbon using APEX. Soil and Tillage Research; 101(1-2): 78-88.
Webb, B.W., Hannah, D.M., Moore, R.D., Brown, L.E., Nobilis, F. (2008) Recent advantages in
stream and river temperature research. Hyrdological Processes; 22: 902-918.
187

Wehrly, K., Wiley, M., Seelbach, P.W. (2003) Classifying regional variation in thermal regime
based on stream fish community patterns. Transactions of the American Fisheries
Society; 132: 18-38.
Weigel, B.M., Robertson, D.M. (2007) Identifying biotic integrity and water chemistry relations
in nonwadeable rivers of Wisconsin: Toward the development of nutrient criteria.
Environmental Management; 40: 691–708.
Wilkerson, E., Hagan, J. M., Siegel, D., Whitman, A.A. (2006) The effectiveness of different
buffer widths for protecting headwater stream temperature in Maine. Forest Science;
52(3).
Williams, J.G. (2010) Lost in space, the sequel: spatial sampling issues with I-D PHABSIM.
River Research and Applications; 26: 341-352.
Williams, J.R., Jones, C.A., Dyke, P.A. (1984) A modeling approach to determine the
relationship between erosion and soil productivity.Transactions of the American Society
of Agricultural Engineering; 27: 129 – 144.
Williams, J.W., Izaurralde, R.C., Steglic, E.M. (2008) Agricultural policy/environmental
extender model: Theoretical Documentation. BREC Report3 2008-17.
Wohl, N., Carline, R. F. (1996) Relations among riparian grazing, sediment loads,
macroinvertebrates, and fishes in three central Pennsylvania streams. Can. J. Fish.Aquat.
Sci.; 53: 260-266.
Wood, P.J., Armitage, P. (1997) Biological effects of fine sediment in the lotic environment.
Environ. Manage; 21(2): 203 – 217.
Woznicki, S., Nejadhashemi, A.P., Smith, C.M. (2011) Assessing best management practice
implementation strategies under climate change scenarios. Transactions of the ASABE;
54(1): 171-190.
Yamada, T., Logsdon, S.D., Tomer, M.D., Burkart, M.R. (2007) Groundwater nitrate following
installation of a vegetated riparian buffer. Science of the Total Environment; 385: 297309.
Yankey, R., Potter, S., Maret, T., McLaughlin, J., Carter, D., Brockway, C. (1991) Rock creek
rural clean water program: ten year report. U.S. Department of Agriculture and Twin
Falls and Snake River Soil Conservation Districts, Twin Falls, Idaho.
Yoder, C. (1987) A modification of the index of well being for evaluating fish communities.
Appendix C-1.Ohio EPA Division of Water Quality Monitoring and Assessment.

188

Yoder, C.O., Smith, M.A. (1999) Using fish assemblages in a state biological assessment and
criteria program: essential concepts and considerations. In: Assessing the Sustainability
and Biological Integrity of Water Resources Using Fish Communities.Edited by Thomas
Simon.
Yuan, Y., Bingner, R., Williams, R., Lowrance, R., Bosch, D., Sheridan, J. (2007) Integration of
the models of AnnAGNPS and REMM to assess riparian buffer system for sediment
reduction. International Journal of Sediment Research; 22(1) 60-69.
Yuan, Y., Locke, M.A., Bingner, R.L. (2008) Annualized agricultural non-point source model
application for mississippi delta Beasley Lake watershed conservation practices
assessment. Journal of Soil and Water Conservation; 63(3): 542.
Zhang, L., Zhao, W., He, Z., Liu, H. (2009) Application of the takagi-sugeno fuzzy system for
combination forecasting of river flow in semiarid mountain regions. Hydrological
Processes; 23: 1430-1436.
Zadeh, L.A. (1965) Fuzzy sets. Information and Control; 8(3): 338–353.
Zimmerman, J., Vondracek, B., Westra, J. (2003) Agricultural land use effects on sediment
loading and fish assemblages in two Minnesota (USA) watersheds. Environmental
Management; 32(1): 93-105.

189