1.. ‘va {F.Hlsh .u“
a V .afiisrfl. gang... .
. .3
. J, .
Laiefinnrfi.

sl- ' .

. I a .sn:;:
.0:

.Dé .7 T .

7-15.41... . .

3......
Eu: 2
1.1.2.2.
.i
W 1.7%.”.
n at! :I.
p 5.1
n

m...“

.
. 4 i
.51.. . . ‘ 4h! Wrm
6.3.3“; A :1. .fl
1. , 1! :NH

It."

 

xtetl‘
i 4 In... [1.2.12.1

 

4:133 tn 9...!
0‘ r!)

 

“#99:; ..... ~
63.1.2.2. . :61» a}.
Qanurliii
I'.<’ 0.43- If:
.u ‘fAhnfluh... .1.
q ‘ . .
10:4! .5... 4

.WF

 

 

 

‘l Isl“. l‘q-trf '4

ill. . .

| 165.. hiya-.7!

 

‘. 1 p‘liryir
441.P:F...O. . . . ‘ ‘ I. .r.. ’ ‘ .

[m

2030

LIBRARY
Michigan State
University

This is to certify that the
thesis entitled

LAND USE / LAND COVER AND WATER QUALITY IN
THE MUSKEGON RIVER WATERSHED, MICHIGAN: A

M.A.

 

CASE STUDY

presented by

Ranjeet John

has been accepted towards fulfillment
of the requirements for the

degree in

6&8th

/ Major rofessor’s Signature

 

09/06/05

Date

MSU is an Affirmative Action/Equal Opportunity Institution

 

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

 

DATE DUE

DATE DUE

DATE DUE

 

WM 31 ion

 

 

 

 

 

 

 

 

 

 

 

 

 

 

2/05 czfimmjlfld-pJS

LAND USE / LAND COVER AND WATER QUALITY IN THE
MUSKEGON RIVER WATERSHED, MICHIGAN: A CASE STUDY

BY

Ranjeet John

A THESIS

Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of

MASTER OF ARTS

Department of Geography

2005

ABSTRACT

LAND USE / LAND COVER AND WATER QUALITY IN THE MUSKEGON
RIVER WATERSHED, MICHIGAN: A CASE STUDY

By

Ranjeet John

Water quality in the Muskegon River Watershed is a function of land use such as
agriculture, residential development, industry, and transportation. Ongoing residential
development around the lakes as well as nutrient-rich runoff from urban and
agricultural landscapes affect water quality. Landsat 7 (ETM+) imagery was used to
obtain an LULC map in the MRW through an unsupervised classification. Surface
hydrological modeling was performed on a SRTM DEM to delineate first order sub-
watersheds, which are more susceptible to non-point pollution as they have no
upstream contributing flow. This study compares LULC (2001-2002) in the MRW with
water quality indices such as total nitrogen (TN), total phosphorus (T P), specific
conductivity, sensitive insect species (EPT taxa) and total invertebrate taxa. Results
indicate that there is significant correlation between an increase in proportions of
agricultural / urban land use within the watershed and water quality indices such as
total phosphorus concentration. In addition, there was a negative correlation between
the percentage of urban land use within the sub-watershed and sensitive insect taxa
as well as invertebrate populations. TN and TP concentrations were also influenced

by distance to urban and agricultural areas respectively.

ACKNOWLEDGEMENTS

I would like to express my appreciation to members of my research committee for
their patience and helpfulness. Special thanks to my graduate advisor, Dr. Jiaguo
Qi, who helped me improve my professional abilities by giving me the opportunity
to work on different projects including the MRW project. I would like to thank Dr.
Joseph Messina and Dr. Ashton Shortridge for reading and editing previous
drafts of this thesis. This work would not have been possible without the timely

comments and suggestions made by the committee.

I would like to thank my colleagues at the Center for Global Change and Earth
Observations (CGCEO) and the Department of Geography, Michigan State
University for their support. Special thanks to Eraldo Matricardi and Dr. Cuizhen
Wang for their patience in answering my questions and for being role models. I
would also like to thank Dr. David Lusch for his helpfulness over the past few
years. This work was supported by the Muskegon River Watershed Ecological
Assessment Project (MRWEAP) and the NASA LULCC grant (NNGOSGD496) at

Michigan State University.

iii

TABLE OF CONTENTS

 

 

List of Tables ................................................................................................... vii
List of Figures ................................................................................................... ix
List of Abbreviations ........................................................................................ xii
CHAPTER 1 INTRODUCTION - - l3
1 .1 Background ................................ ' ........................................................... 1 3
1.2 Problem statement and objectives ........................................................ 14
1.3 Study area ............................................................................................ 16

1 .4 Hypothesis ............................................................................................ 18

1 .5 Sub-hypothesis ..................................................................................... 18
1.6 Benefits of this research ....................................................................... 18
CHAPTER 2 LITERATURE REVIEW - 20
2.1 Land use and water quality ................................................................... 20
2.1.1 First order watersheds ............................................................................... 25
2.1.2 Land use in the Muskegon River Watershed ......................................... 27

2.2 EPA monitoring programs in the US ..................................................... 31

2.2.1 The Environmental Monitoring and Assessment Program (EMAP)... 31

2.2.2 EPT taxa ...................................................................................................... 32
2.3 Surface hydrological modeling .............................................................. 33
2.3.1 Advantages of SRTM DEMs ..................................................................... 33
2.3.2 Digital terrain modeling and topographic attributes .............................. 34

iv

CHAPTER 3 METHODOLOGY 36

 

3.1 Surface hydrological modeling .............................................................. 36
3.1.1 Acquisition and pre-processing of SRTM derived DEMs ..................... 36
3.1.2 Delineation of Strahler stream order using D-8 flowline derived stream

network ......................................................................................................... 37
3.1.3 Delineation of first order watersheds, percent slope and wetness
index ............................................................................................................. 40

3.2 Classification of ETM+ imagery using unsupervised method ................ 43
3.2.1 Imagery inventory ....................................................................................... 43
3.2.2 Image pre-processing and unsupervised classification ....................... 46
3.2.3 LULC within MRW ...................................................................................... 56
3.2.4 Mapping of surface water quality indicators ........................................... 57
3.2.5 Proximity to LULC types ............................................................................ 58
3.2.6 Scale issues in relating LULC to water quality ...................................... 58

CHAPTER 4 RESULTS AND DISCUSSION- _- 60

4.1 LULC and water quality in the MRW ..................................................... 60
4.1.1 Specific conductivity and percentage of LULC ...................................... 60
4.1.2 TN concentration and percentage of LULC ........................................... 63
4.1.3 TP concentration and percentage of LULC ............................................ 79
4.1.4 Total invertebrate taxa and percentage of urban land 'use .................. 94
4.1.5 Total EPT taxa and percentage of urban land use ............................... 95
4.1.6 Proximity to urban areas ........................................................................... 97
4.1.7 Proximity to agricultural areas .................................................................. 99

CHAPTER 5 CONCLUSIONS 101

 

 

APPENDICES
APPENDIX-A:
APPENDIX B:
BIBLIOGRAPHY

FIGURES USED IN THE STUDY

 

DATA USED IN STATISTICAL ANALYSIS

 

 

vi

106

- 107

116

119

Table 3-1.
Table 3-2.
Table 3-3.
Table 3-4.
Table 3-5.
Table 3-6.
Table 3-7.

Table 4-1.

Table 4-2.

Table 4-3.

Table 4-4.

Table 4-5.

Table 4-6.

Table 4-7.

Table 4-8.

Table 4-9.

List of Tables

Satellite data ...................................................................................... 44
ETM+ spectral radiance range in w/ (m2.sr.um) ................................. 48
ETM+ solar spectral irradiances ......................................................... 49
ETM+ scenes and their geo-referencing parameters ......................... 50
Accuracy assessment for MRW LULC mosaic ................................... 54
Confusion matrix for MRW LULC mosaic ........................................... 54
LULC within MRW .............................................................................. 56
Correlation between specific conductivity (u siemens) and LULC typtgs)
Urban land use in sub-watersheds greater than 1% .......................... 62

Correlation between specific conductivity (p siemens) and urban land
use within sub-watershed ................................................................ 62

Correlation between TN concentrations and LULC within sub-
watershed ........................................................................................ 63

Correlation between TN concentrations and LULC within first order
watershed ........................................................................................ 64

Percentage of LULC (within sub-watersheds) vs. TN concentrations 66

Percentage of LULC (within first order watersheds) vs. TN
concentrations ................................................................................. 70

Correlation between TP concentrations and LULC within sub-
watersheds ...................................................................................... 79

Correlation between TP concentrations and LULC within first order
watersheds ...................................................................................... 80

Table 4-10. Percentage of LULC (within sub-watersheds) vs. TP concentrations

......................................................................................................... 81

Table 4-11. Percentage of LULC (within first order watersheds) vs. TP

concentrations ................................................................................. 85

vii

Table 4-12. Percentage of urban land use vs. total invertebrate population within
sub-watersheds ............................................................................... 94

Table 4-13. Percentage of urban land use correlated with sensitive insect

population within sub-watersheds .................................................... 96
Table 4-14. Proximity to urban areas & TN concentrations within MRW ............. 97
Table 4-15. Proximity to urban areas & TP concentrations within MRW ............. 98
Table 4-16. Proximity to urban areas vs. TN concentrations within MRW ........... 99

Table 4-17. Proximity to agricultural areas & TN concentrations within MRW ....99

Table 4-18. Proximity to agricultural areas & TP concentrations within MRW...1OO

Table 5-1. Water quality indicators within MRW sub-watersheds ..................... 116
Table 5-2. Water quality indicators within MRW sub-watersheds ..................... 117
Table 5-3. Percentages of LULC types within sub-watersheds ......................... 118

viii

Figure 1-1.
Figure 2-1.
Figure 3-1.

Figure 3-2.

Figure 3-3.

Figure 3-4.

Figure 3-5.

Figure 3-6.

Figure 3-7.

Figure 4-1.

Figure 4-2.

Figure 4-3.

Figure 4-4.

Figure 4-5.

Figure 4-6.

Figure 4-7.

List of Figures

Muskegon River Watershed .............................................................. 17
MRW: ecological classification .......................................................... 28
Flow chart to extract watersheds from DEM ..................................... 39
DEM flow direction artifacts created due to the nature of the D8 flow
line algorithm ................................................................................... 41
Muskegon first order watershed coverage obtained using binary mask
of Strahler stream network ............................................................... 41
Spatial extent of MRW overlaid with WRS-2 (Landsat) footprint ....... 43
Spatial extent of MRW overlaid with county boundaries and Digital
Ortho Quarter Quads (DOQQ’s) ...................................................... 45
Spatial extent of MRW overlaid with county boundaries and pan
sharpened lKONOS imagery ........................................................... 46
Flow chart for image processing of DN to surface reflectance .......... 47

Percentage of agricultural land use (within sub-watersheds) vs. TN
concentrations ................................................................................. 66

Percentage of agricultural & urban land use (within sub-watersheds)
vs. TN concentrations ...................................................................... 68

Percentage of human use index (urban + agriculture + bare soil within
sub-watersheds) vs. TN concentrations ........................................... 69

Percentage of agricultural land use (within first order-watersheds) vs.
TN concentrations ............................................................................ 71

Percentage of agricultural & urban land use (within first order
watersheds) vs. TN concentrations .................................................. 72

Percentage of human use index (urban + agriculture + bare within
first order watersheds) vs. TN concentrations .................................. 73

Percentage of forests (within sub-watersheds) vs. TN concentrations
......................................................................................................... 74

ix

Figure 4-8. Percentage of forests (within first order watersheds) vs. TN
concentrations ................................................................................. 75

Figure 49 TN and TP concentration distribution overlaid on % agricultural land
use ................................................................................................... 78

Figure 4-10. Percentage of agricultural land (within sub-watersheds) vs. TP
concentrations ................................................................................. 82

Figure 4-11. Percentage of urban & agricultural land (within sub-watersheds) vs.
TP concentrations ............................................................................ 83

Figure 4-12. Percentage of human use index within sub-watersheds vs. TP
concentrations ................................................................................. 84

Figure 4-13. Percentage of agriculture (within first order -watersheds) vs. TP
concentrations ................................................................................. 86

Figure 4-14. Percentage of urban & agriculture (within first order-watersheds) vs.
TP concentrations ............................................................................ 87

Figure 4-15. Percentage of human use index (urban + agriculture + bare within
first order watersheds) vs. TP concentrations .................................. 88

Figure 4—16. Percentage of forests vs. TP concentrations within sub-watersheds

......................................................................................................... 89
Figure 4-17. Percentage of forests vs. TP concentrations within sub-watersheds

......................................................................................................... 90
Figure 4-18. TN and TP concentrations overlaid on % human use index ........... 92
Figure 5-1. Imagery from WRS Path 21 Row 30 Landsat 7 (ETM+) ................. 107
Figure 5-2. Imagery from WRS Path 22 Row 30 Landsat 7 (ETM+) ................. 108

Figure 5-3. MRW-LULC map derived through unsupervised classification ...... 109

Figure 5-4. MRW—LULC map derived through unsupervised classification and

smoothed with 3x3 majority filter .................................................... 110
Figure 5—5. SRTM derived Digital Elevation Model (DEM) ................................ 111
Figure 5-6. SRTM derived DEM overlaid with Strahler stream order ................ 112

Figure 5-7. Wetness index derived from SRTM DEM overlaid with MRW wetlands

coverage ........................................................................................ 1 13
Figure 5-8. Urban development around Lake Houghton (2001) ........................ 114
Figure 5-9. Urban development around Lake Higgins (2001) ........................... 115

Images in this thesis are presented in color.

xi

List of Abbreviations

AOI
CAFO
CGCEO
DEM

DN
DOQQ
DTED
EPA
EPT
ETM+
H.U.lndex
ISA
ISODATA
IWI
LULC
MDNR
MIRIS
MMU
MRW

' MRWEAP
NAWQA
NRCS
NPS
SRTM
TOA

TN

TP
TRFIC
UTM
WGS
WI

Area Of Interest
Confined Animal Feeding Operation
Center for Global Change and Earth Observation
Digital Elevation model
Digital Number
Digital Ortho Quarter Quads
Digital Elevation Terrain Data
Environmental Protection Agency
Ephemeroptera, Plecoptera & Trichoptera,
Enhanced Thematic Mapper
Human Use Index
lmpervious Surface Area
Iterative Self-Organizing Data Analysis Technique
Index of Watershed Indicators
Land Use / Land Cover
Michigan Department of Natural Resources
Michigan Resource lnforrnation System
Minimum Mapping Unit
Muskegon River Watershed
Muskegon River Watershed Ecological Assessment Program
National Water Quality Assessment project
National Resource Conservation Service
Non-Point Source Pollution
Shuttle Radar Topography Mission
Top Of the Atmosphere
Total Nitrogen
Total Phosphorus
Tropical Rainforest Information Center
Universal Transverse Mercator
World Geodetic System
Wetness Index

xii

CHAPTER 1 INTRODUCTION

1.1 Background

Human induced changes in the Muskegon River Watershed (MRW) in
north-central Michigan have had an adverse effect on the quality of water in its
lakes and streams. Water quality in the MRW is a function of the different land
uses such as agriculture, residential development, industry, and transportation.
Residential development around the lakes as well as the construction of roads
and parking lots affect water quality. Water quality also be affected because of
algal blooms and high sediment loads carried by runoff from urban and
agricultural landscapes. The lakes and wetlands of the MRW have important
recreational and economic uses, but some are in danger of hyper-eutrophication,
a phenomenon caused by excessive nutrients in the runoff from urban and
agricultural landscapes. Eutrophication is characterized by water-bodies being
dominated by the same set of nuisance species that can tolerate the eutrophic
conditions and leads to the subsequent reduction in the diversity of species
(Carpenter et al., 1995). The increase in stream runoff is due to the cumulative
impact of urban growth along the eastern shores of Lake Michigan (Pijanowski et
al., 2002). This urban growth encompasses urban, sub-urban and near shore

residential development, an d has increased in 2002.

13

1.2 Problem statement and objectives

There is an increase in urban development within the watershed, Which
affects water quality due to increase in surface runoff. The increase in impervious
surface area due to urban development leads to a subsequent increase in water
temperature in streams fed by surface runoff, which modify the stream hydrology
through incision. Stream incision is defined as “the rate of incision in detachment-
limited systems and by definition, determined by the stream's ability to erode the

bed, usually by a combination of abrasion and plucking” (Whipple et al., 2000).

Water quality is degraded in certain sub-watersheds by inputs of nutrients
such as nitrogen and phosphorus from towns and agricultural areas in the form of
non-point source pollution (NPS). Confined animal feeding operations (CAFO’s)
act as a point source for phosphorus input. Excessive sedimentation, another
reason for deterioration of water quality, is attributed to stream bank erosion,

roads, and farm practices like drainage.

To obtain a firm understanding of the spatial and temporal nature of water
quality, an ecological assessment of the watershed using remote sensing and
GIS is necessary, as it is cost effective. The ecological assessment is carried out
at the sub-watershed level rather than a basin wide study. This allows a cross
comparison of the sub-watersheds in terms of their vulnerability to deterioration

in water quality.

14

Ecological assessments are necessary in the protection, maintenance and
restoration of ecological systems (US EPA, 1996). Land use indicators such as
percentage of urban I built up and agriculture combined with topographic
attributes such as elevation and slope help identify the anthropogenic activities

that generate stress in the ecosystem under study.

The objectives of the research are:
1) The mapping of urban built up areas and agricultural land use at the sub-
watershed level as well as areas susceptible to non-point source pollution.
2) Statistical analysis to correlate percentage of land use with water quality

indicators

This research seeks to correlate land use / land cover (LULC) in the sub-
watersheds with water quality indicators. In addition, an inventory of the LULC types
was carried out in areas that were susceptible to non-point sources of pollution such
as first order watersheds. The percentage of urban and agricultural land use within
the sub-watersheds as well as the first order watersheds were regressed against
various water quality indices such as total nitrogen, total phosphorus, conductivity as
well as biological indicators such as sensitive insect populations (EPT taxa) and total

invertebrate taxa to build predictive models.

15

1.3 Study area

The Muskegon River is 219 miles long from its start at Houghton and
Higgins lakes down to its mouth at Muskegon Lake, and eventually, Lake
Michigan. The watershed is located between latitude 43° to 44°30’N and

longitude 84°30’ to 86°W (Figure 1-1).

The MRW covers an area of almost 7000 square km and includes 94
tributaries, 183 stream segments and 95 dams (O'Neal, 1997). The watershed is
within the counties of Wexford, Missaukee, Roscommon, Kalkaska, Crawford,
Lake, Osceola, Clare, Newaygo, Mecosta, Montcalm and Muskegon. The primary
tributaries of the Muskegon river include the West Branch of the Muskegon River, the
Clam River, the Middle Branch River, the Hershey River, the Little Muskegon River,
Bigelow Creek, Brooks Creek, and Cedar Creek (O'Neal, 1997). Some of the

important cities in the watershed are Big Rapids, Newaygo and Muskegon.

l6

    

Muskegon River Watershed

 

Figure 1-1. Muskegon River Watershed.

17

1.4 Hypothesis

The purpose of this research is to test whether there is a correlation
between water quality and percentage of LULC within the MRW. The specific

hypothesis is that LULC affects water quality within the MRW.

1.5 Sub-hypothesis

1) LULC within the MRW can be accurately mapped through an un-
supervised classification of Landsat-7 (ETM+) imagery and aggregated to the sub-

watershed level.

2. Water quality within the MRW, measured through a set of water quality
indicators and aggregated to the sub-watershed level, can be correlated with the

percentage of LULC.

1.6 Benefits of this research

This study, based on correlations between specific LULC types and water
quality would not only assess the ecological integrity but also identify the sub-
catchments within the MRW with greater proportions of urban and agricultural land
use and thus help in better management of watershed. Once identified, the
relationships between the various land uses and water quality indicators can be
applied to other watersheds in the State of Michigan as well. This study of land—

water interactions within the MRW is especially important at a time when urban

18

sprawl threatens the ecological integrity of Michigan’s water and natural

resources.

19

CHAPTER 2 LITERATURE REVIEW

2.1 Land use and water quality

Anthropogenic activities on land can have a detrimental effect on riparian
bodies such as rivers and streams. The National Water Quality Assessment Program
(NAWQA) was initiated by the USGS in 1991 to understand how human activities
and natural processes affect water quality in this nation (USGS, 1999). This study
came about partly due to the growing public concerns about the quality of the
nation’s water resources. The passing of the Clean Water Act in 1977, whose
purpose was “to restore and maintain the chemical, physical, and biological integrity
of the waters of the United States” led to a strong campaign by the public and private
sectors to limit contaminants from point sources from entering streams (USGS,
1999). NAWQA findings indicate that streams in watersheds and basins with
significant agricultural and urban development have higher levels of nutrients and
pesticides, which contribute to higher growth of algal growth. The increase in
impervious surfaces like paved lots and urban pavements increased surface runoff
(USGS, 1999). Also, the NAWQA studies showed that streams in basins with steep
slope and clayey soils were vulnerable to contamination due to stream runoff. In the
continental United States, urban streams had the highest concentrations of
pesticides such as chlordane and dieldrin (USGS, 1999). It was also found that
concentrations of phosphorus were higher in urban areas than in rural areas and this
in part due to the effluent from wastewater plants (USGS, 1999). The runoff from
urban areas have elevated levels of phosphorus and nitrogen and caused the

eutrophication in lakes, streams and reservoirs. Cities are important contributors to

20

non-point pollution and homeowners with lawns apply just as much fertilizer and
pesticides per unit area as farmers would on their farms. Studies conducted in the
upper Midwest suggest that lawn care, through the application of nutrients rich in
nitrogen and phosphonrs contributes to nutrient rich mnoff and that nutrient
concentrations are more than those originating from impervious surface areas like
roofs and paved surfaces like streets and driveways (Bannennan et al., 1993;

Waschbusch et al., 2000; Steuer et al., 1997).

The sources of water pollution typically fall into two categories, 1) point source
pollution and 2) non-point source pollution. Point source pollution originates from
discrete locations that are spatially explicit and some examples are sewage
treatment plants, industrial effluents and land disposal sites. Non-point sources of
pollution originate from various diffuse sources that occur over a larger and broader
geographical area. Some examples of non-point source of pollution include
agricultural runoff, storm water, urban runoff and atmospheric deposition. Because of
its diffuse nature, non-point source pollution cannot be isolated in a spatially explicit
manner (USGS, 1999).

Agriculture is the most important source of non-point source pollution
according to the US EPA (2000). Agricultural practices like the spraying of pesticides
and herbicides, irrigation, planting and harvesting, and confined animal feedlots all
contribute to NPS pollution. Another significant form of agricultural NPS pollution

seems to be siltation (Rabeni & Smale, 1995). The nutrients most often considered in

21

land-water interaction studies are nitrogen and phosphoms (Turner et al., 2001).
Nitrogen concentrations in rivers are sensitive to land use patterns, the riparian zone
structure and river flow (Cirrno and McDonnell, 1997). Accumulation of excess
phosphorus in rivers and streams has been recognized as the cause for
eutrophication (Carpenter et al., 1998). Similarly, concerns about nitrogen inputs into
aquatic ecosystems have been raised (Mueller and Helsel, 1996; Vitousek and
Howarth, 1991). Farmers apply nutrients such as nitrogen and phosphorus to their
farm plots, but not all of it is absorbed by the plants. The nutrients in the soil are
leached through runoff and find their way into lakes and streams where they cause
eutrophiwtion. Recent advances in technologies such as GIS and remote sensing
have seen various studies conducted to assess agricultural NPS pollution at different
geographical scales that include catchment, watershed, basin and landscape level
assessments (Richards et al., 1993; Allan et al., 1997; Johnson & Gage, 1997;

Harding et al., 1999; Lammert and Allan, 1999).

Regression analysis has been used to determine relationships between
land use patterns and nitrogen l phosphorus concentrations (Osborne and Wiley,
1988). The land use patterns in the Salt River Basin, Illinois, were mapped from
aerial photos and results indicated that urban land use and its distance to the
stream was the most important variable in predicting nutrient concentrations in
stream water (Osborne and Wiley, 1988). A study in the Minneapolis—St. Paul
Metropolitan region demonstrated that lakes within watersheds dominated by

forests and intact wetlands tended to be less eutrophic and have lower levels of

22

chlorides and lead (Detenbeck et al., 1993). On the other hand, lakes within
agricultural watersheds were more likely to be eutrophic. There was also a
positive correlation between the percentage of urban land use and phosphorus in
the Minneapolis area (Detenbeck et al., 1993). In another study, LULC within 62
catchments in the Saginaw River, Michigan was related to stream water
chemistry (Johnson et al., 1997). The results of the study demonstrated that the
land use / land cover had a strong influence on water quality but the predictive
power of specific water quality indicators varied by season (Johnson et al., 1997).
The spatial distribution and proportion of different land use I land cover types has
been found to directly affect water quality (Hunsaker & Levine, 1995). Forested
riparian zones had better water quality than deforested riparian zones with similar
agricultural land use (Hunsaker & Levine, 1995). Soranno (1996) found
significant relationships between land use and concentrations of phosphorus in

the Lake Mendota watershed, Wisconsin. A GIS based predictive model was built
where the phosphorus export coefficient varied by land use type. The study also
measured the contribution of phosphorus to a lake as a function of distance. In
addition, the results emphasized the importance of riparian vegetation in

reducing forest runoff (Soranno et al., 1996).

Horton (1933) suggested that rainfall within a watershed either infiltrates
the soil or is transported overland to streams as storm flow. Overland flow or
sheet flow occurs when the precipitation intensity exceeds the soil infiltration

capacity. However, hortonian overland flow, which is widely accepted as the case

23

for degraded watersheds that are predominately agricultural, does not seem to
hold for forested watersheds where abundant canopy cover prevents erosion and

facilitates infiltration.

Urbanization is a significant land use today as it is associated with
population, the economy and the conversion of other LULC types. The
proliferation of impervious surface areas (ISA), in areas that are heavily
vegetated reduces carbon sequestration (Milesi et al., 2003). In addition, the
increase in impervious surface areas results in the alteration of sensible and
latent heat fluxes leading to the formation of urban heat islands (Changnon,
1992). Urban development plays an important role in influencing the rate of runoff
and erosion (Goudie, 1990). Urban growth undergoes several stages (Kibler,
1982). In the early stage of urban development, the logging of trees and
vegetation may result in the decrease of evapotranspiration, interception and
increase siltation as well as total suspended solids. The latter stages of growth
include an increase in the construction of houses, streets and storm drains which
in turn leads to a decrease in infiltration, increase in storm flows and a lower
ground water table. As the number of residential and commercial buildings
increase, there is a subsequent increase in paved surfaces liked roads and
parking lots which in turn decrease the time of concentration which is the amount
of time required for water to move from the most distant part of the watershed or
catchment to the outlet and is a function of the percentage of impervious area

and slope. The increase in impervious surfaces will result in higher peak

24

discharges after rainfall events because of a decrease in infiltration and result in
an increase in surface and storm water runoff (Booth, 1991). The effects of an
increase in ISA are found when 10% of the watershed is covered with impervious
surfaces, leading to the alteration of stream channels; rising water temperatures;
a reduction in the diversity of aquatic insects and fish; and the degradation of
wetlands and riparian zones (Beach, 2002). The response of rivers in terms of
discharge due to increase in sediment loading due to changes in land use

patterns such as deforestation has been documented (Ligon et al., 1995).

2.1.1 First order watersheds

First order streams are the uppermost, stream channels that do not have
any upstream reaches and have perennial or intermittent flow (Gomi et al., 2002).
First order watersheds make up a large portion (60-70%) of the catchment area
(Siddle et al., 2000; Meyer and Wallace, 2001). The first order watersheds are
important sources of nutrients organic matter and sediments for the higher order
streams and their catchments (Gomi et al., 2002). The movement of organic
matter and invertebrate species from the first order to higher order watersheds
supports the fish population downstream (Wipfli and Gregorvich, 2002). In
addition, leaf litter and large woody debris alter and control the stream
morphology and provide habitats like riffles and pools for invertebrates and fish
fry (Zimmerman and Church, 2001). Invertebrate species found in the first order
streams serve as food for aquatic biota (erfli, 1997). The large woody debris

also dams the stream channels and alters the channel reaches such as

25

cascades and step pools (Halwas and Church, 2002). The riparian forest canopy
in the first order streams attenuates incoming solar radiation and so controls the

water temperature as well as the amount of light (Gomi et al., 2002).

Headwater systems that include hill slopes and first order watersheds
control stream flow generation (Tsukamoto et al., 1982) and water chemistry in
the stream (Likens et al., 1977). These headwater systems contain four
topographic units (Hack and Goodlett, 1960) which are 1) hill slopes, 2) zero-
order basins, 3) ephemeral channels emerging from the zero order basins called
transitional channels, and 4) first order streams. The hillslopes do not have
channelized flow. The zero order basins could be defined as an unchannelized
hollow with converging contour lines (T sukamoto et al., 1982). Temporary
channels or transitionary channels may connect the zero order basin and first
order streams (T sukamoto et al., 1982). These ephemeral channels do not
support the complete life cycles of macro invertebrate biota. Storm flow
generation within the river basin Is more rapid in first order watersheds owing to
the small storage capacity. There is also a greater variation in peak flow
discharge as compared to higher order watersheds (Gomi et al., 2002). The
increase in impervious surface area due to urban development in the first order
watersheds could increase the peak flow discharge and so are especially
vulnerable to NPS pollution from urban landscapes (Gomi et al., 2002). In spite of

the significant role of first order watersheds within the larger basin or catchment,

26

their processes have been extensively modified by land use (Meyer and Wallace,

2001)

2.1.2 Land use in the Muskegon River Watershed

The Muskegon River Watershed (MRW) can be broadly classified into three
major ecological zones: 1) the outwash bowl, 2) the morainal valleys, and 3) the

freshwater estuary (Figure 2-1).

The headwaters of the Muskegon river originate in an outwash plain
formed by deposits of stratified debris from glacial meltwater streams. The
morainal valley which constitutes the mid-river section was created by deposits
of unsorted debris (gravel, sand and boulders) left behind by retreating glaciers.
The freshwater estuary or the mouth of the Muskegon river is a drowned river
valley created due to coastal subsidence or through inundation by glacial melt

water (Christopherson, 1995).

27

 

Figure 2-1. MRW: ecological classification

The history of land use within the MRW is relevant to the study as it is
important to consider the past consequences of significant alterations to the
watershed system. Human settlements in the MRW expanded as population
increased drastically between 1810 to 1840 (O'Neal, 1997). This was largely
influenced by intensive logging operations for copper and white pine. The logging
caused extensive damage to the existing aquatic habitat. The water quality in the
MRW was also affected by anthropogenic activities that increased in the early
1900's and reached a peak in the 1950’s and 1960’s. Nutrient and sediment
pollution was common as well as extensive wetland reclamation in the vicinity of

Muskegon lake (O'Neal, 1997).

28

Agricultural and urban land uses cause the greatest impact on water quality
(Mueller and Helsel, 1996). Agricultural land use predominated in the MRW in 1997,
and accounted for 33.4% of the total area with urban areas taking up 0.6% of the
watershed (O'Neal, 1997). Agricultural lands in Michigan also contribute to poor
water quality through erosion of sediments into streams (O'Neal, 1997). It is
estimated that the soil erosion from crop and pasture lands might be 14 to 21 times
higher than erosion rates on forest land. Roads also contribute to erosion of
sediments through an increase in runoff over paved surfaces (Alexander et al.,

1995)

The Muskegon river basin lies within two major land resource area
classifications of the National Resource Conservation Service (NRCS). These
are the northern lower Michigan sandy drift and the southern lower Michigan drift
plain (O'Neal, 1997). Soil erosion in the form of annual sheet and rill erosion is
0.84 tons/acre for crops-pasture land and 0.04 tons/acre for forests in the
northern lower Michigan sandy drift and 2.09 tons/acre for crops-pasture land
and 0.15 tons/acre for forests (O'Neal, 1997). To help prevent the soil erosion,
the NRCS has considered drainage, and forage improvements. The draining of
land by deepening of existing streams destroys and eliminates many aquatic
habitats. The removal of trees and other riparian vegetation leads to a
degradation of canopy cover over the streams and leads to increased water

temperatures (O'Neal, 1997).

29

Dams were constructed on Muskegon river at Big Rapids during 1866 and
Newaygo in 1900 and dismantled in 1966 and 1969, respectively. Four major dams
still remain on the Muskegon river and they include Reedsburg dam, Rogers dam,
Hardy dam and Croton dam. Of these, Reedsburg dam is a wildlife flooding and the
rest are hydroelectric dams (O’Neal, 1997). The dams pose serious environmental
problems to the river as the natural flow regime of the river is altered and also change
their physical, chemical and biological characteristics (Poff et al., 1997; Poff and Hart,
2002). The dams cause fish mortalities as they get caught in the hydroelectric
turbines. Of the fish mortalities, as much as 70% are game fish (O’Neal, 1997). The
dams also prevent the movement of aquatic insect larva, which serve as food for the
various fish species. Changes in water quality and rising temperatures have been
attributed to the major hydroelectric dams, namely, Croton, Hardy and Rogers’s

dams.

The current pattern of land use in Michigan indicates an increase in
urban/built up area and a subsequent decrease in agricultural and pastureland. In
1952, the land use under agriculture in Michigan was approximately 72, 843 km2
(Veatch, 1953). The area under agriculture decreased to 50,000km2 and 49, 446 km2
in 1987 (Natural Resources Inventory, 1987). The farmland loss can be attributed to

urban sprawl or as in some cases in the MRW, re-growth of vegetation.

Though the population growth is less than the national average (6.9% as

compared to 13.1%), there has been significant population shifts within Michigan’s

30

borders. The general trend is that people seem to move from urban areas to the
suburbs and rural areas (Machemer et al., 1999). The increase in urban development
seems to outstrip the population growth and is characteristic of urban sprawl. The
increase could be attributed to the low cost of land in niral areas as compared to the
cities. The increasing urbanization is alarming as Michigan’s natural resources
account for a large percentage (29%) of Michigan’s economy (Machemer et al.,
1999). The primary effect of urban development in the suburbs is fragmentation of

forests from large contiguous tracts to small patches.

2.2 EPA monitoring programs in the US
2.2.1 The Environmental Monitoring and Assessment Program (EMAP)

The Environmental Monitoring and Assessment Program (EMAP) is a
research program undertaken by the Environmental Protection Agency (EPA) to
monitor and assess the status and trends of the ecological resources of the United
States (Paulsen et al, 1991; US EPA, 1997). The EPA published a report titled
“Ecological assessment of the mid-atlantic region, a landscape atlas”. The atlas
described ecological conditions across the mid-Atlantic region of the United States
that included the states of Delaware, the District of Columbia, Maryland,
Pennsylvania, Virginia, and West Virginia. The report was based on information
derived from satellite imagery as well as other sources of geo-spatial information. In
October 1997, the EPA released another report, which was its first index of
watershed indicators (US EPA, 1997) that shared many similarities with the EMAP

mid-Atlantic landscape atlas. The primary difference between the two reports is that

31

the index of watershed indicators primarily deals with water quality issues but the
mid-Atlantic landscape atlas documents the impact of land use / land cover on water
quality. This study uses some of the methods mentioned in both reports to address

the possible effects of land use / land cover on water quality.

2.2.2 EPT taxa

Environmental monitoring groups across the United States have adopted EPT
(Ephemeroptera, Plecoptera, Trichoptera) taxa richness as a useful measure of
stream water quality. The EPT taxa that include mayflies, caddis flies and stone flies
evolved in streams with high levels of oxygen and in fast flowing waters. Any
reduction in flow, depleted oxygen supply or increase in temperature results in a
decrease in population. The widespread use of EPT taxa as a stressor indicator
might be owing to its ease of use and effective tracking of water quality as well its
habitat specific impact (Wallace et al., 1996). The use of EPT taxa is a follow—on to
the historic use of benthic macro-invertebrates to evaluate water quality that dates
back to early studies in the Illinois River (Richardson, 1928). Macro-invertebrate taxa
were increasingly used to monitor water quality from the 1950’s onwards and studies
from the mid-century period began to cite EPT taxa as intolerant (Gaufin and

Tarzwell, 1952).

32

2.3 Surface hydrological modeling

Topography is an important factor in determining the stream flow in
forested uplands (Wolock and Price, 1994). It also defines the movement of
water within a catchment area due to gravity, the spatial distribution of soil
moisture (Burt and Butcher, 1985) and the water chemistry of the stream flow
(Wolock et al., 1990). Surface hydrology is defined as “the spatial and temporal
storage and redistribution of rainfall as it falls on or enters into the soil“ (Engman,
1997). The prediction of spatial patterns and the rate of surface runoff require a
hydrologic model and a categorization of the land surface (Zhang and
Montgomery, 1994). Digital elevation model (DEM) data are arrays of regularty
spaced elevation values referenced horizontally either to a Universal Transverse
Mercator (UTM) projection or to a geographic coordinate system (USGS, 2000).
Digital elevation models are being used in hydrological modeling studies
(Bruneau et al., 1995) and for a variety of engineering as well as planning

applications (Zhang and Montgomery, 1994).

2.3.1 Advantages of SRTM DEMs

The Shuttle Radar Topography Mission (SRTM) derived DEM, currently
the highest resolution global DEM, was sampled at 1 are second or 30m (Rabus
et al., 2002). Until now, high resolution DEMs were obtained mostly from optical
stereo data acquired from aerial photographs or satellite imagery. The DEMs

obtained from these were not homogeneous due to the quality of image contrast.

33

Also, the presence of clouds or the lack of sunlight in stereo-pairs resulted in
image artifacts (Rabus et al., 2002). The SRTM data set, on the other hand was
obtained by a single technique, i.e., interferometic Synthetic Aperture Radar
(lnSAR) sampling in 11 days by the Space Shuttle Endeavour (Rabus et al.,
2002). The absolute vertical accuracy is +/- 16 m and the absolute horizontal
(90% circular error) accuracy is 20m (Sun et al., 2003). The SRTM global data
set was obtained from the post processing of the C band radar interferometry
which introduces an error in the vertical accuracy as the C band radar return
effectively samples the height at the top of the canopy and unlike radar bands
with longer wavelengths (P and L bands), does not reach the ground beneath. A
validation of SRTM DEM’s vertical accuracy using the Shuttle Laser Altimeter
(SLA-02) showed that areas with sparse vegetative cover had absolute vertical
accuracy that exceeded the mission specifications of 16m. Surface slope
comparisons between the SRTM and Digital Elevation Terrain Data (DTED) DEM
showed that slope derived from the SRTM DEM was superior to the slope

derived from the DTED Level 1 (3 arc sec) DEM (Sun et al., 2003).

2.3.2 Digital terrain modeling and topographic attributes

Topographic attributes include primary attributes like elevation and slope as
well as secondary attributes that are derived from combinations of primary data. The
primary attributes like slope, catchment area, and specific catchment area are
significant because they influence overland flow velocity and runoff rates, mnoff

volume, and steady state runoff rate respectively (Moore et al., 1991). The catchment

34

area is a measure of the surface runoff of the landscape and it combines the effect of
upslope contributing area as well as the local convergent and divergent flows within

(Moore et al., 1991).

Secondary or compound data describe the spatial distribution and
variability of specific processes that occur within a landscape such as wetness
index (soil water content) or stream power (Moore et al., 1991). Wetness index is
a second order derivative of slope and relates to the spatial distribution of soil
saturation zones (Moore et al., 1991). Stream power is the erosive power of
stream flow (Moore et al., 1991). The stream power index can be used to identify
places where soil conservation practices that reduce the erosive power of the

stream can be implemented (Moore et al., 1991).

Depending on the scale of regional planning, the fundamental units for
water resource conservation are the basin, watershed or sub-watershed (Moore
et al., 1991). The traditional mapping and delineation of watershed and sub
watershed boundaries was centered on the stream network and therefore
considered a conservative representation (Mark, 1983). The past decade has
seen the rapid development in the field of hydrological modeling through the use
of DEMs. The ARC GRID module uses an algorithm (O’Callaghan and Mark,
1984; Jenson and Domingue, 1988) that determines the flow direction of each
element in 3x3 matrix to one of its eight neighbors in the direction of steepest

descent (Moore et al., 1991).

35

CHAPTER 3 METHODOLOGY
3.1 Surface hydrological modeling
3.1.1 Acquisition and pro-processing of SRTM derived DEMs

SRTM derived 30m DEMs were used to model surface flow in the MRW.
These DEMs were downloaded from the USGS seamless web server
(http://seamless.usgsgovl) and merged to obtain coverage for the entire watershed.
The grids were downloaded as floating point and in order to conserve disk space
were converted to 16 bit unsigned integer by the use of the GRID command, INT

(ESRI, 1994).

Errors in DEMs some times manifest in the form of “sinks” or “pits”. These
production artifacts are regions that have lesser elevation than the area around it.
These sinks were identified with the SINK command in ARC GRID. The output
grid was found to contain nodata cells caused during acquisition / production of
the SRTM data. Cells that do not have a valid value assigned are termed as
nodata cells (ESRI, 1994). The nodata cells were subsequently filled with a
FOCAL MEAN command. After, the nodata cells were removed the individual
grids (six in number) were projected UTM, Zone 16, datum & spheroid WGS 84.
The SINK command (ESRI, 1994) was run again on the merged grid, as there
might still be legitimate sinks of natural origin rather than artifacts due to the
nature of radar backscatter. After the sinks had been identified, the FILL
command, based on the algorithm developed by Jenson and Domingue (1988)

was used to fill the artificial production artifacts.

36

3.1.2 Delineation of Strahler stream order using D-8 flowline derived
stream network

In DEM analysis for hydrologic application, it is often assumed that all flow
from a cell is directed towards one and only oneiof its neighbors and this
assumption is referred to as the “deterministic eight-neighbors” or D-8 model
(Fairfield and Leymarie, 1991). The D-8 flow-line algorithm is a standard
algorithm that is used by the Environmental Systems Research Institute (ESRI)
software to derive flow direction from a surface. The algorithm uses an input
DEM to depict the direction of flow from each cell within a 3x3 window. There are
eight valid directions or eight adjacent cells into which the flow can travel (ESRI,
1994). The flow direction determines the direction of flow from every cell or
element in the grid and is determined by finding the direction of steepest flow
(ESRI, 1994). The flow accumulation function computes accumulated flow as the
weight of all the cells flowing into each down-slope cell in the output grid (ESRI,
1994). Cells with high flow accumulation have concentrated flow and are used to
determine the stream network. A threshold value can be applied to the flow
accumulation product to obtain the stream network within a watershed (Jenson

and Domingue, 1988).

The surface hydrology commands such as flow direction & flow accumulation
were an on the MRW grid (Fig 3-1). To obtain the stream network, a threshold value
was derived by adding the mean of the flow accumulation grid to the standard

deviation and re-classed such that, all values greater than the mean + 1 standard

37

deviation would be 1 othenrvise 0. This process was repeated iteratively until the
raster stream network derived from the SRTM DEM matched vector stream
networks in two different datasets namely Michigan Resource lnfonnation System
(MIRIS) as well as Digital Line Graphs (planimetric information obtained from
various USGS maps). The threshold number, 3326, was found to be near the
mean of the flow accumulation distribution. The threshold number was then
multiplied by 0.003x 0.003 to obtain the area required to generate a first order

stream (2.99 sq km) within the Muskegon river watershed.

The STREAMORDER command (Strahler) was run on the output grid. In the
Strahler stream order, order increases when streams of the same order intersect
(Strahler, 1957). Therefore, the intersections of two first order streams will give
rise to a second order stream and intersection of two second order streams
would give rise to a third order stream. However, the intersections of two different
orders will not result in increase in order. This method is the most commonly

used model for stream networks (ESRI, 1994).

The stream order grids were then converted to are coverages using the
STREAMLINE command. The WATERSHED command, which uses the streamline
arc coverages as well as the flow direction grid as input was used to generate the
sub-watersheds. These watersheds differ slightly from the MIRIS watershed as the
stream order was derived from various sources (digitizing of existing maps or from

aerial photos).

38

 

 

 

 

Merge DEM

 

 

 

 

Flow direction

 

 

FILL sinks

 

 

 

 

Flow
accumulation

 

 

 

 

Thresholded
flow

 

 

 

 

Stream order

 

 

 

 

Streamline
coverage

 

 

I

 

 

Watersheds

 

 

Figure 3-1. Flow chart to extract watersheds from DEM

39

 

3.1.3 Delineation of first order watersheds, percent slope and wetness
index

In order to delineate the first order watersheds, a binary mask was applied on
the Strahler stream order grid. The higher order streams were masked out leaving
only the first order streams. The STREAMLINE command in ARC GRID was used on
the first order stream grid along with the flow direction force grid to obtain an Arc-
coverage of the first order streams. The WATERSHED command in ARC GRID was
then run using the first order streams coverage and flow direction force grid. The first
order watershed grid thus obtained was converted to a vector layer. The D-8 flow line
algorithm models flow direction at 45° angles and so the conversion of the raster
watershed layer to a vector layer results in the formation of a number of spindle
shaped polygon artifacts at the pour points (Figure 3-2). To correct for this anomaly,
the first order watershed arc coverage was edited using the are tools module in the
ARC INFO software. The post processing of the first order watershed was time
consuming, but absolutely necessary to delineate the first order watersheds within

the sub—watersheds (Figure 3-3).

40

 

 

Figure 3-2. DEM flow direction artifacts created due to the nature of the D8 flow
line algorithm

 

 

 

 

 

 

l
0 20 40 80 Kilometers

Figure 3-3. Muskegon first order watershed coverage obtained using binary mask
of Strahler stream network

41

In addition to first order watersheds, percent slope and wetness index (W.l)
were also obtained from the SRTM derived DEM. Slope identifies the maximum rate
of change between each cell and its neighbors (ESRI, 1994). Percent slope was
obtained using the SLOPE command using the percent rise option instead of degree
slope and Is useful in identifying areas of steep slope, which might be vulnerable to

erosion.

The wetness index or topographic index is the log of the ratio of the catchment
area and the percent slope (Wolock and Price, 1994). The wetness index helps in
identification and delineation of wetlands (Moore et al., 1991). The equation for

wetness index is

WI = In (a/tanb)
or Wl = In (a / (rise / run) + 0.0001) (3—1)

where a = flow accumulation
b = slope and tan b = slope percentage x 0.01

Implementing the wetness index for raster processing requires some
adjustment. Wolock and Price (1994) suggest that flow accumulation or upslope
contributing cells might be scaled by contour length, and in this case the cell side
(Wolock and Price, 1994). The cell side was the resolution of an ETM+ pixel, i.e.

30m. In GRID, the wetness index was obtained using the following command

Wl = LN ((FA + 1) * cell side) / (slope (percent rise) * 0.01 + 0.0001)

42

The wetness index was used to aid in the image classification of the ETM+ imagery

by overlaying the wetness index grid over the classified image.

3.2 Classification of ETM+ imagery using unsupervised method

3.2.1 Imagery inventory

Four ETM+ (Landsat-7) images in the World Reference System (WRS-2)
Path 21, Rows 29 & 30 and Path 22, Rows 29 & 30 were obtained from 2001 and
2002 (Table 3-1). The imagery was obtained from the TRFIC archive at the
Center for Global Change and Earth Observations (CGCEO) at Michigan State

University.

 

    
         

Pith 021 Row 029

Path on Rowlnli
Path 021 row03ll

 

 

 

Figure 3-4. Spatial extent of MRW overlaid with WRS-2 (Landsat) footprint

43

 

Landsat 7 Date

 

ETM+ Path/row 22/29 9/7/2002
ETM+ Path/row 22/30 7/2/2001

ETM+ Path/row 21/29 6/25/2001
ETM+ Path/row 21/30 7/14/2002

 

 

 

 

Table 3-1. Satellite data

The spatial extent of the MRW is such that it covers portions of four Landsat
scenes (Figure 3-4). The portions of the ET M+ imagery mentioned above were then
clipped using the subset option in ERDAS IMAGINE. The spatial database thus
obtained includes subsets from 2001/2002 (ETM+) which would facilitate an
inventory of the land use / land cover within the forty sub watersheds through the

process of image classification.

In addition, two lKONOS-scenes from 2002 (September & October) were
obtained over the MRW at Houghton Lake and over the Big Rapids-Haymarsh
area to provide ground truth for accuracy assessment of the image classification
process (Figure 3-6). The 4m multi-spectral data was pan-sharpened using the
1m panchromatic band. Digital Ortho-Quarter Quads (DOQQ’s) were
downloaded from the Michigan Department of Natural Resource’s (MDNR),
Spatial Data Library. The DOQQ’s were then mosaicked and reprojected to UTM

Zone 16 to be used as ground truth reference (Figure 3-5).

44

 

 

K‘UUGHA WORD

 

 

 

m... I

 

 

 

-=-_:——_—— IGIorneters
0 10 20 40 60 80

Figure 3-5. Spatial extent of MRW overlaid with county boundaries and Digital
Ortho Quarter Quads (DOQQ’s)

45

 

 

 

 

 

 

 

 

_ _ lfilometers
0 1O 20 40 60 80

Figure 3-6. Spatial extent of MRW overlaid with county boundaries and pan
sharpened lKONOS imagery

3.2.2 Image pre-processing and unsupervised classification

The Landsat data record is extremely important to earth sciences as it
marks over three decades of earth observation. Its advantages include a high
spatial resolution, medium temporal resolution, and an extensive swath (Teillet et
al., 2001). However, in order to benefit from this impressive dataset, it is
absolutely essential to process the digital numbers collected by the sensor by
radiometric calibration to an absolute scale, in physical units (Teillet et al., 2001).
Image pre—processing and radiometric correction was carried using ERDAS

IMAGINE version 8.6 (Fig 3-7).

46

 

Raw DN

I

Radiance

 

 

 

 

 

 

 

 

 

l Radiometric
Correction

 

 

Reflectance
(TOA)

MODTRAN
4 Reflectance

(Surface)

I

Rectified Geometric
Image Correction

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Figure 3-7. Flow chart for image processing of DN to surface reflectance
The raw digital numbers are first converted into physical units of radiance (L1,) using
wlibration coefficients (Table 3-2) in a given spectral band
L), = [(LMAX - LMIN)]/255] x DN + LMIN (3-2)
Where LMAX = spectral radiance scaled to DNmax in WI (m2.sr.pm)

LMIN = spectral radiance scaled to DNmin in w/ (m2.sr.pm)
DN = Digital Numbers

47

 

 

 

 

 

 

BAND ETM+

in w/(m2.er.rm)

LMAX LMIN
1 191.6 -6.2
2 196.5 -6.4
3 152.9 -5.0
4 241.1 -5.1
5 31.06 -1.0
7 10.80 -0.35

 

 

Table 3-2. ETM+ spectral radiance range in w/ (m2.sr.pm)

When there is a need to compare imagery from different sensors, it is

advantageous to use reflectance values instead of using radiance. The cosine

effect of different sun angles can be removed and the differences due to

variability in exo-atmospheric radiances between spectral bands can also be

accounted for. Unitless planetary reflectance was obtained by the following

equafion

__ n-L, OdZ

Esun,1 0 cos 9,

 

where pp = Unitless planetary reflectance

L). = Spectral radiance at the sensor’s aperture
93= Solar zenith

d = Earth-sun distance in astronomical units

Esun 1 = Mean solar exo-atmospheric irradiances from table below

48

(3-3)

 

Band ETM+ Solar Spectral
Irradiances
Units: in w/(m2.sr.pm)

1970.00
1843.00
1555.00
1047.00
227.00
80.53

 

 

 

 

\ICfl-thA

 

Table 3-3. ETM+ solar spectral irradiances

The ETM+ radiance values were radiometrically corrected to at-sensor
reflectance using the calibration parameters in the metadata and then to surface
reflectance after correcting for distorting atmospheric effects. Haze in the upper
atmosphere often causes satellite imagery to look faded due to Rayleigh
scattering and ozone absorption. The top of the atmosphere (T OA) reflectance
was then converted into surface reflectance by using the coefficients of slope and

intercept, which in turn were obtained by running the MODTRAN 4 program.

Geodetic accuracy or the geographic navigation accuracy for a particular
pixel is absolutely essential especially in the case of change detection studies.
The ETM+ level 1G imagery products are systematically corrected, i.e.,
registration is performed without ground control (Goward et al., 2001). After
systematic correction, ETM+ imagery has a geodetic error of less than 250m with
1 standard deviation (Masek et al., 2001). The ETM+ images were geometrically
corrected to ensure proper alignment and scale and were rectified to the
Universal Transverse Mercator (UTM) projection (Zone 16) and WGS 84 datum

using a first order polynomial and nearest neighbor resampling. To correct for

49

horizontal displacement and inaccuracy, GPS co-ordinates recorded at the road
intersections within the watershed were used as ground control Points (GCPs) to
serve as a reference in the process of geometric correction to geo-reference the

imagery. The accuracy was within one pixel (Table 3-4).

 

 

 

 

 

L-7 Path IROW X Residual Y Residual RMS Error
22130 (ETM+) 0.0678 0.0510 0.0849
21/30 (ETM+) 0.0676 0.2258 0.2357
21129 (ETM+) 0.1882 0.3466 03944
22129 (ETM+) 0.2024 0.2024 0.2862

 

 

 

 

 

 

Table 3-4. ETM+ scenes and their gee-referencing parameters

An unsupervised classification was carried out on the ETM+ scenes through
the use of the Iterative Self-Organizing Data Analysis Technique (ISODATA)
algorithm. The ISODATA clustering method uses minimum spectral distance to
iteratively classify pixels until the spectral distance patterns emerge in the form of
clusters (T ou and Gonzalez, 1974). To perform ISODATA classification, the user
needs to specify the maximum number of clusters thought necessary , the number of
iterations to be run and the convergence threshold, which defines the maximum

percentage of pixels that are allowed to remain unchanged between iterations.

The ISODATA algorithm was run on the ETM+ scenes after specifying 100
clusters, 80 iterations, and a convergence threshold of 0.98. The National Land
Cover Database (NLCD 92) thematic map was created by an unsupervised
classification using 100 clusters and so the same number of clusters was used in

the generation of a classified product for this research (Vogelmann et al., 1998).

50

The classification was performed after geo-referencing so that the resultant

thematic map would be more accurate (ERDAS, 2004).

Each image reached the convergence threshold before the maximum
level of iterations. Visual interpretation helped in identification of eight distinct
LULC classes from the Michigan Land Cover/ Use Classification System (2000),
namely 1) deciduous, 2) coniferous, 3) grass, 4) water, 5) wetland, 6) bare soil,
7) urban, and 8) agriculture. These classes were then recoded to an Anderson’s
level I classification (Anderson, 1976) with seven classes, 1) urban, 2)

agriculture, 3) grass, 4) forest, 5) water, 6) wetland and 7) bare soil.

The seven classes were defined as follows:

1) Urban: Developed areas that were characterized by 30 % or more area
under constructed material (Asphalt & concrete).

2) Agriculture: Areas characterized by croplands, which include row crops
(corn & soybean), small grain (wheat & barley) as well as pasture/hay.

3) Grass and shrublands: These include areas that have 25% or more area
covered by grasses and shrubs

4) Forest: Areas that had 20% or more covered by broadleaved forest as well
as coniferous forests.

5) Water: Areas under open water

51

6) Wetlands: Areas where the soil or substrate is periodically saturated with or
covered with water (Cowardin et al., 1979). These areas are characterized
by both forested and non-forested wetlands.

7) Bare: Area covered by bare rock, silt, sand or clay with little or no

vegetation.

The accuracy assessment of the ETM+ derived LULC maps through
image classification required the comparison of the classified imagery with
reference or ground truth data (Congalton, 1988). The high- resolution (1m) pan-
sharpened lKONOS images and DOQQ’s were used as ground truth images and
as a photo-interpretation key in the accuracy assessment process. The accuracy
assessment utility in ERDAS IMAGINE 8.6 was used to generate 300 random
points with 40 or more points per class using stratified random sampling.
Congalton (1991) suggests that a minimum of 40-50 samples be collected for
each land cover in the error matrix to adequately sample the dataset to be
evaluated. Random sampling was used in an earlier assessment but was
replaced by a stratified random sampling scheme as the former tends to under
sample small but important classes such as urban areas in watersheds. It is for
this reason that stratified random sampling was chosen where a minimum
number of sample points (40-50) were collected for each class type (Congalton,

1 988).

52

The randomly sampled co-ordinates were imported into Arcview 3.2 as a
tab delimited file and saved as a point shapefile. This shapefile was overlaid over
the high-resolution reference images, namely the 2002 lKONOS images and the
Michigan DNR DOQQ’s (1998). The reference targets were identified using
visual interpretation and entered in the reference column of the accuracy

assessment utility.

The relationship between the classified map product and the reference
images are summarized in an error matrix (Table 3-4). The overall classification
accuracy or the percentage of correctly classified pixels of the LULC product was
83.33% and was obtained by dividing the sum of the (correctly classified) pixels
in the major diagonal by the total number of samples (Table 3-5). The producers
accuracy (which is a measure of the reliability of the map) was derived by the
total number of correct pixels for a land cover type divided by the total number of
pixels of that land cover type derived from reference data, in other words, the
column total. The users accuracy (a measure of the adequacy for each category)
was derived by dividing the total number of correct pixels for a land cover type
divided by the total number of pixels that were accurately classified in that land
cover class. The Kappa (Km) statistic, a measure of percent accuracy within an
overall measurement of classifier accuracy (Congalton, 1991), was 0.8054. The
Kappa statistic is a better measure of accuracy than an overall assessment as it

considers inter class agreement (Fitzgerald and Lees, 1994).

53

where r is the number of rows in the matrix, N is the total number of

observations, xii is the number of observations in row i and column i, x». and x+,-

K

hat —

r r
NZ xii - 20514 x x+i
_ i=l i=1

 

r
2
N _ Z<xi+ X x+i
i=1

are the marginal totals for row i and column irespectively.

(3-4)

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Class Reference Classified Number Producers Users Kappa
Name Totals Totals Correct Accuracy Accuracy Statistic
Urban 31 41 26 83.87% 63.41% 0.592
Agriculture 46 43 32 69.57% 74.42% 0.6979
Herbaceous 46 44 41 89.13% 93.18% 0.9195
Forest 51 48 46 90.20% 95.83% 0.9498
Water 40 41 40 100.00% 97.56% 0.9719
Wetland 39 43 33 84.62% 76.74% 0.7327
Bare soil 47 40 32 68.09% 80.00% 0.7628
Table 3-5. Accuracy assessment for MRW LULC mosaic
Reference
Class Urb. AL Herb. For. Wat. Wet. Ba Row UA
U. 26 7 0 0 0 0 8 41 63.41
Ag 1 32 1 0 0 2 7 43 74.42
Herb. 0 2 41 0 0 1 0 44 93.18
For. 0 0 0 46 0 2 0 48 95.83
Wat. 0 0 0 0 40 1 0 41 97.56
Wetl. 0 1 4 5 0 33 0 43 76.74
Bare. 4 4 0 0 0 0 32 40 80
col tot 31 46 46 51 40 39 47 250 Sum
Prod 83.87 69.56 89.1 90.19 100 84.61 68 08 300 Total

 

 

Table 3-6. Confusion matrix for MRW LULC mosaic

54

 

 

The reference totals column (Table 3-6) refers to the number of pixels in
each land cover identified within the high-resolution reference imagery. The
classified totals are the number of pixels in each land cover classification mosaic
from 2001. The “number correct” is the number of pixels in the classification that
match the high-resolution reference imagery. Overall classification accuracy

produced from this assessment was 83.3% with an overall kappa statistic of 0.80.

The errors of omission and commission were most apparent within the
urban, agriculture, grassland and wetland land cover types. Recently harvested
agricultural fields exhibit high reflectance, are similar to highly reflective surfaces
within an urban area. The construction materials in urban structures like cement
and concrete have similar spectral properties to the bare soil in an agricultural
area that has been recently harvested. Therefore, an agricultural plot can often
be mistaken for a urban area owing to the spectral properties Similarly there
were instances of omission and commission between fallow fields that exhibited
re-growth and land with 50% or less herbaceous cover. Another land cover type
that had errors of omission and commission was the wetland cover type. Lowland
forests that are seasonally flooded can be labeled as forest or wetland depending
the date of acquisition. Leafon imagery resulted in a spectral signature that was
representative of forests whereas a leafoff imagery especially in late fall resulted
in a classified product that depicted a forested wetland. The labeling of the
unsupervised clusters to wetlands was verified by overlaying the image with a

vector layer of wetlands obtained from the MNDR Spatial Library.

55

3.2.3 LULC within MRW

The classified imagery from the ETM+ scenes was converted into Arc
Grids using the Import/Export module in ERDAS IMAGINE. The ARC GRID files
were overlaid with an AOI (Area of Interest) of the sub-watershed and the
catchment area under each LULC type was obtained and tabulated. The raw
count was converted into an area measurement by multiplying the raw count with
0.09 (the area of the pixel in hectares) for conversion into hectares. The area in
hectares was then converted into percentage values to obtain the proportion of
land use/land cover within the sub-watershed. The minimum mapping unit (MMU)
was defined by size of the 3x3 majority filter (nine 30m pixels) used to remove
the salt and pepper effect caused due to noise in the imagery. The proportion of
land use and land cover in the Muskegon River Basin is 3.677% urban, 18.11 %
agricultural, 40.5% forested, 15.89 % under herbaceous, 0.26 % bare, 4.06 %

water and 17.48% consists of wetlands.

 

 

 

2001 2001
Area (in ha) Area (in %)
Urban 26010.18 3.677342

 

Agriculture 128145.24 18.11729
Herbaceous 1 12410.45 15.89269

 

 

 

 

 

 

 

 

 

 

Forest 286471.62 40.50161
Water 28754.46 4.065331
Wetland 123643.35 17.48081
Bare soil 1873.89 0.264932
Total 707309.19 100

 

Table 3-7. LULC within MRW

56

3.2.4 Mapping of surface water quality indicators

Surface water quality indicators were collected by teams of student and
research assistants lead by Dr. Jan Stevenson, Department of Zoology, Michigan
State University. Between 2001 and 2003, 256 sites were sampled in spring and
summer seasons. These sites include 107 streams/rivers, 85 wetlands and 64
lakes. The water quality measurements include, total nitrogen (TN), total
phosphorus (T P), trophic state index based on phosphorus (T SIP), trophic state
Index based on cholorophyll (TSlchl), trophic state Index based on secchi depth

(T Slsd) and specific conductivity.

Biological indicators such as total insect population, sensitive insect
population (EPT taxa), species richness of Algae, zooplankton, invertebrates,
fish, mussels and plants were also collected at various sampling locations that
were re-visited every year (2001-2003). Each water quality measurement had
geographic co-ordinates (WGS 84 datum) recorded by a GPS at the time of
collection. The water quality measurements were entered in MS Excel
worksheets and saved either as .txt files or .dbf files. The water quality
measurements were then imported into ESRI Arc Map using the add XY
interface. The water quality measurements were mapped in geographic co-
ordinates and then projected into UTM zone 16 co-ordinates (datum WGS 84) so

that they could be overlaid on the land use / land cover layer.

57

3.2.5 Proximity to LULC types

The proximity to urban and agricultural land use was calculated using the
Euclidean distance function in ARCGRID (ESRI, 1994). Binary raster layers that
represented urban areas and agricultural land use were used as source Grids
The Euclidean distance surfaces this obtained were overlaid with total nitrogen /
total phosphorus sampling points to obtain the distance to urban areas and

agricultural land use for each sampling point.

3.2.6 Scale issues in relating LULC to water quality

In this study, two spatial scales were considered. One was the sub-
watershed level study where the LULC in hectares was aggregated to the sub-
catchment level and the other being at the level of the first order watersheds.
This methodology was based on the premise that hydrological units, namely the
sub-watersheds and the first order watersheds would delimit sampling locations

and define the areas that influence the measurements taken at those locations.

In addition, the immediate vicinity around the sampling points was
correlated with water quality indices through the creation of variable buffers of
arbitrary width. This approach did not yield significant results when the width of
the buffers was small. Increasing the buffer size beyond 500m-1 km meant that
areas upstream as well downstream would be assumed to influence the nutrient

concentration at the sampling point. This methodology was discontinued since

58

areas that were 500m—1 km downstream of the sampling location would not have

a significant impact on the nutrient concentration.

There have been many studies with opposing views on the scale at which
land-use and water interactions should be studied. Some suggest that TP
concentrations in stream water could be explained by LULC patterns within 100m
of a stream while other studies (Omemik et al., 1981) demonstrated that the
more distant LULC types at the catchment extent had an impact on water quality
along with riparian land use in the immediate vicinity. Comparisons between
different case studies are complicated owing to differences between nutrient
concentrations, the size of the watershed, and the temporal nature of the

measurements (Turner et al., 2001).

59

CHAPTER 4 RESULTS AND DISCUSSION
4.1 LULC and water quality in the MRW
4.1.1 Specific conductivity and percentage of LULC

Specific conductivity is the measure of the ability of any solution to carry
an electric current and increases with increasing mineral content (Prevost et al.,
2000). Spearrnan’s rank correlation was used to test for correlations between the
percentages of land use and specific conductivity. Specific conductivity within the
40 sub-watersheds was assumed to be positively correlated with the percentages
of urban land use, agricultural land, combined percentages of urban and
agricultural land as well as the percentage of human use index (combined
percentages of urban, agricultural and bare land). It was also assumed that forest

cover and specific conductivity measurements were negatively correlated.

 

 

 

 

 

 

Conductivity vs. LULC Correlation tvalue tcritical, c =
Conductivity vs. urban 0.606 4.5069 2:695
Conductivity vs. agriculture 0.533 3.7267 1.695
Conductivity vs. urban + agriculture 0.725 6.2274 1.695
Conductivity vs. Human Use Index 0.720 6.1379 1.695
Conductivity vs. forest -0.358 -2.2683 -1.695

 

 

 

 

 

 

Table 4-1. Correlation between specific conductivity (u siemens) and LULC types

60

As hypothesis, it was assumed that specific conductivity would increase
as the percentage of anthropogenically modified LULC within the sub-watershed
and first-watersheds increases. The null hypothesis was that there is no
correlation between the two variables (t st-critical at a = 0.05). As alternative
hypothesis, it was assumed that there is a positive correlation between specific
conductivity and the percentage of land use (t > t-critical at a = 0.05). The t

critical value = 1.695 at a = 0.05 (95% probability) and df = 34.

Ho: Null hypothesis: there is no correlation between specific conductivity
and the percentage of land use
H1: Alternative hypothesis: there is a correlation between specific

conductivity and the percentage of land use

The null hypothesis was rejected as the tvalues for the different land use
/ land cover types (table 4-1) were greater than t critical value = 1.695. The
correlation between urban, agricultural LULC types as well as the percentage of
combined land use and specific conductivity, which ranged between 0.60 and
0.72 was statistically significant at the 95 % confidence level. This could be
attributed to the percentage of urban land use greater than 1% in some of the

watersheds (Table 4-2).

61

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Sub-watershed City/T own
1 Higgins lake/l -127
3 Houghton/l -127
12 Cadillac
14 Lake city
23 Reed city/Evart
24 Big rapids/US 31
25 Big rapids/US 31
31 Howard city
35 Newaygo (Brooks & Hess lakes)
36 Freemont
38 Muskegon wastewater ponds
39 Muskegm city
40 Muskegon ciy

 

Table 4-2. Urban land use in sub-watersheds greater than 1%

 

 

 

 

L Model summary Parameter estimates I
Equation I R Square F df1 df2 Sig. Constant b1
|Linear | .205 8.789 1 34 .006 329.099 11.040|

 

 

 

 

 

 

Table 4-3. Correlation between specific conductivity (u siemens) and urban land
use within sub-watershed

A simple linear regression characterized a positive linear association between the
percentage of urban land use (independent variable) and specific conductivity, the
dependent variable (Table 4-3). The model explained 20 % of the dependent variable

(specific conductivity) at a = 0.05.

62

4.1.2 TN concentration and percentage of LULC

Total nitrogen concentrations (measured in parts per billion) within the
sub-watershed were assumed to be positively correlated (using Spearrnan’s rank
correlation) with the percentages of urban land use, agricultural land, combined
percentage of urban and agricultural land as well as percentage of human use
index (combined percentages of urban, agricultural and bare land). It was also
assumed that the percentage of forest cover and TN concentrations were

negatively correlated.

 

 

 

 

 

 

TN vs. LULC Correlation tvalue tcritical, a = 0.05
TN vs. urban 0.189 1.1386 1.695
TN vs. agriculture 0.4 2.5819 1.695
TN vs. urban + agriculture 0.454 3.0144 1.695
TN vs. human use index 0.464 3.0988 1.695
TN vs. forest -0.47 -3.1501 -1.695

 

 

 

 

 

 

Table 4-4. Correlation between TN concentrations and LULC within sub-
watershed

63

 

 

 

 

 

 

TN vs. LULC Correlation tvalue tcritical, a = 0.05
TN vs. urban 0.169 1.014 1.695

TN vs. agriculture 0.364 2.2833 1.695

TN vs. urban + agriculture 0.351 3.0228 1.695

TN vs. human use index 0.360 2.9811 1.695

TN vs. forest -0.437 -3.4339 -1.695

 

 

 

 

 

 

Table 4-5. Correlation between TN concentrations and LULC within first order
watershed

In addition, TN concentrations within the first order watersheds were
assumed to be positively correlated with the percentages of urban and
agricultural LULC types as well as their combined percentages. It was also
assumed that the TN concentrations were negatively correlated with the

percentage of forests within the first order watersheds.

As hypothesis, it was assumed that TN concentrations increase as the
percentage of anthropogenically modified land within the sub watershed and first-
watersheds increases. The null hypothesis was that there is no correlation
between the two variables (t st—critical at a = 0.05). As alternative hypothesis, it
was assumed that there is a positive correlation between TN concentrations and
the percentage of land use (t > t-critical at a = 0.05). The t critical value = 1.695

at a = 0.05 (95% probability) and df = 34.

64

Ho null hypothesis: there is no correlation between TN concentrations and
the percentage of land use
H1 alternative hypothesis: there is a correlation between between TN

concentrations and the percentage of land use

The null hypothesis was rejected as the tvalues for the different land use /
land cover types (table 4-4) were greater than t critical value = 1.695. However,
the correlation between the percentage of urban land (within the sub-watershed
as well the first order watershed) and increasing TN concentrations were not

statistically significant at the 95 % confidence level.

Logarithmic regression was undertaken to explain the relationship and
variance in a regression model between the percentage of LULC (dependent
variable) and TN concentrations (dependent variable). A non-linear regression
was considered, as a linear least square regression did not explain any variance.
A logarithmic model was fitted to TN concentrations and the percentage of urban
use within the sub-watersheds which yielded an R2 = 0.003 (Table 4-6). The
model could only explain 0.03% of the dependent variable (TN concentrations)

and was also not statistically significant at the 95 % confidence level.

F critical value = 4.13 at a = 0.003 (95 % probability) and df = 35

F statistic = 0.091, therefore 0.091 < 4.13 or F < F critical

65

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Independent Logarithmic Model for total nitrogen Parameter estimates
(dependent)
°/o LU R
Square F df1 df2 Sig. Constant b1

urban .003 .091 1 34 .764 25.006 6.740
_agriculture .024 0.847 1 34 .364 1351.463 253.085

urb. +39. .017 0.584 1 34 0.584 916.198 358.346

H.U. Index .016 0.538 1 34 .468 944.906 345.990

forest .024 0.851 1 34 .363 6527.997 1264.905

 

 

Table 46 Percentage of LULC (within sub-watersheds) vs. TN concentrations

Percentage of ag vs. TN (sub—watershed)

 

14000.0 "I

12000.0 .4

10000.0 -

8000.0 -

6000.0 -

4000.0 -

2000.0 —

0.0-

 

O

 

 

I
0.00

I
10.00

I
20.00

I
30.00

I
40.00

Percentage agriculture

50.00

60.00

Figure 4-1. Percentage of agricultural land use (within sub-watersheds) vs. TN

66

concentrations

 

Then, a logarithmic model fitted to TN concentrations and the percentage
of agricultural land use within the sub-watersheds, yielded an R2 = 0.024 (Figure
4.1). The model explained only 2.4 % of the dependent variable (total nitrogen)

and was not statistically significant at a = 0.05 (Table 4-6).

F critical value = 4.13 at a = 0.05 (95 % probability) and df = 35

F statistic = 0.847, therefore 0.847 < 4.13 or F statistic < F critical

When the percentages of urban and agricultural land use (Figure 4-2)
were summed and regressed with the TN concentrations in the sub watersheds,
the model explained 1.7 % of the dependent variable (total nitrogen) and was not

statistically significant (Table 46).

F critical value = 4.13 at a = 0.05 (95 % probability) and df = 35

F statistic = 0.584, therefore 0.584 < 4.13 or F statistic < F critical

67

Percentage of urban 8: ag vs. TN

 

 

 

 

 

14000.0 4 o
12000.0 .—
10000.0 -I
8000.0 -' 0
6000.0 '-
0
4000.0 -
o
0
2000.0 - 8 o O
63> o
o $9 0°
0.0-I
I I I I I T I
0.00 10.00 20.00 30.00 40.00 50.00 60.00

Percentage urban + agriculture

Figure 4-2. Percentage of agricultural & urban land use (within sub-watersheds)
vs. TN concentrations

68

Percent of human use index vs. TN

 

14000.0 4 O

12000.0 _

10000.0 -

8000.0 - 0

6000.0 -

4000.0 -
O

 

O

8
£06) 90 O 00

2000.0 — O

O
0.0- (9

I I I I I I l
0.00 10.00 20.00 30.00 40.00 50.00 60.00

 

 

 

Percentage of human use index

Figure 4-3. Percentage of human use index (urban + agriculture + bare soil within
sub-watersheds) vs. TN concentrations

The percentage of human use index, i.e., the combined percentages of

urban, agricultural and bare-soil, was regressed with TN concentration (Figure 4-

3). However, the model did not significantly explain 1.6 % of the dependent

variable, total nitrogen (Table 4-6).

F critical value = 4.13 at c = 0.05 (95 % probability) and df = 35

F statistic = 0.538, therefore 0.538 < 4.13 or F statistic < F critical

69

Logarithmic regression was also undertaken to explain the relationship
and variance in a regression model between the percentages of land use
(independent variable) and TN concentrations (dependent variable) aggregated
to the first order watershed level. The percentages of land use types include the
percentages of urban land use, agricultural land, the combined percentage of
urban and agricultural land as well as the percentage of human use index

(combined percentages of urban, agricultural and bare land).

 

 

 

 

 

 

 

Independent Logarithmic model for total nitrogen Parameter estimates
(degendent)
% LU-first R
square F df1 df2 Sig. Constant b1

urban .000 .000 1 35 .990 1895.525 9.586
igriculture .175 7.223 1 34 .011 11.760 8.444

urban + ag. .071 2.675 1 35 .111 -380.810 790.953

h.u. index .071 2.666 1 35 .111 431.927 803.948

Forest .049 1.791 1 35 .189 8007.71 -1645.04

 

 

 

 

 

 

 

 

 

Table 4-7. Percentage of LULC (within first order watersheds) vs. TN
concentrations

A logarithmic model (Table 4-7) fitted to TN concentrations and the
percentage of urban land use within the first order-watersheds yielded an R2 =
0.000. The model could not explain any variance in the dependent variable (TN

concentration) and was not statistically significant at the 95 % confidence level.

F critical value = 4.13 at a = 0.003 (95 % probability) and df = 35

F statistic = 0.000, therefore 0.000 < 4.13 or F < F critical

70

 

Percentage agriculture within first order watershed vs.
TP

 

140.0—

120.0—

100.0—

80.0#

60.0—1

40.0—

20.0—

 

 

 

 

l l I l
0.00 10.00 20.00 30.00 40.00 50.00 60.00
Percentage agriculture

Figure 4-4. Percentage of agricultural land use (within first order-watersheds) vs.
TN concentrations
F critical value = 4.13 at a = 0.05 (95 % probability) and df = 35

F statistic = 7.223, therefore 7.223 > 4.13 or F statistic > F critical

Another logarithmic model fitted to TN concentrations and the percentage
of agricultural land use within the first order-watersheds yielded an R2 of 0.175
(Figure 4-4). The model explained 17% of the dependent variable, TN
concentration, but was not statistically significant at the 95 % confidence level

(Table 4-7).

71

The combined percentages of urban and agricultural land use (Figure 4-5)
in the first order watersheds were regressed against the TN concentrations and
explained 7.1 % of the dependent variable, TN concentration, but was not

statistically significant at the 95 % confidence level (Table 4-7).

F critical value = 4.12 at a = 0.05 (95 % probability) and df = 34

F statistic = 2.675, therefore 2.675 > 4.12 or F statistic > F critical

Percentage of urban + ag (first order watersheds)
vs. TN

 

140000.. 0
12000.0—
10000.0—
8000.0—i 0
6000.0-

4000.0—

     

20000..

O
CI) 0
0C5D 0
.0

 

 

 

l l l l T 1 n
0.00 10.00 20.00 30.00 40.00 50.00 60.00
Percentage of urban & agriculture

Figure 4-5. Percentage of agricultural & urban land use (within first order
watersheds) vs. TN concentrations

72

Percentage of human use index
( first order watershed) vs. TN

 

14000.0— 0
12000.0—
10000.04
8000.0— 0
6000.0—

4000.0—

    

2000.0-

   

O
O 06300
co 00

0.0— O

 

 

 

l l l I I T r
0.00 10.00 20.00 30.00 40.00 50.00 60.00
Percentage human use index

Figure 4-6. Percentage of human use index (urban + agriculture + bare within
first order watersheds) vs. TN concentrations

The combined percentages of urban and agricultural land uses as well as
bare soil were regressed with TN concentrations within the first order watersheds
(Figure 4-6). The logarithmic model explained 7.1 % of the dependent variable,
total nitrogen, but was not statistically significant at the 95 % confidence level

(Table 4-7).

F critical value = 4.13 at a = 0.05 (95 % probability) and df = 35

F statistic = 2.666, therefore 2.666 < 4.13 or F statistic < F critical

73

The TN concentrations could be influenced by land use types such as
urban, agricultural and bare soil. However, this is to be expected as the three
LU LC types are all created and modified by humans. Therefore, we would expect
to find an inverse or negative correlation between natural cover such as forests
and nitrogen concentrations to prove that nutrient loading in the MRW is a
function of land use. A logarithmic model (Figure 4-7) in which the percentage of
forest cover was regressed with TN concentrations within the sub-watersheds
resulted in R2 = 0.024, explaining 2.4 % of the dependent variable (TN

concentration), but was not statistically significant at a = 0.05 (Table 4-6).

Percentage of forests vs. TN (sub-watershed)

 

14000.0 - o
12000.0 -
10000.0 -

8000.0 -‘ O

 

6000.0 ‘

4000.0 q
0

2000.0 - 0 C6

0 C9 00 8000) O@ O
0.0~

I T I I
0.00 20.00 40.00 60.00 80.00

 

 

 

Percentage of forest

Figure 4-7. Percentage of forests (within sub-watersheds) vs. TN concentrations

74

F critical value = 4.13 at a = 0.05 (95 % probability) and df = 35

F statistic = 0.851, therefore, 0.851< 4.13 or F statistic < F critical

Percentage of forest cover vs.TN

 

14000.0. 0

12000.0-—

10000.0—

8000.0-9

6000.0—

40000.4

2000.0—

 

 

 

 

l l T l l
0.00 20.00 40.00 60.00 80.00 100.00
Percentage of forest

Figure 4-8. Percentage of forests (within first order watersheds) vs. TN
concentrations

Another logarithmic model (Figure 4-8) in which the percentage of forest
cover was regressed with TN concentrations within the first order watersheds
resulted in R2 = 0.049 explaining 4.9 % of the dependent variable, TN

concentration, but was not statistically significant at c = 0.05 (Table 4-7).

75

critical value = 4.13 at c = 0.05 (95 % probability) and df = 35

F statistic = 0.851, therefore, 0.851< 4.13 or F statistic < F critical

The proportion of urban land use within a watershed gives us an indication
of urban development. However, a more accurate picture of the effects of
anthropogenic modification can be realized through the use of a human use
index. The human use index combines the proportion of land use types such as
urban, agriculture and bare soil. The index allows us to identify areas that have
been converted from their natural state (O’Neil et al., 1988) instead of
considering individual LULC categories. The logarithmic regressions between
total nitrogen distributions and percentages of LULC indicate an initial increase in
TN concentrations with an increase in the percentages of agricultural and urban
land use and then remain unchanged regardless of any increase in LULC

percentages.

Denitrification, sedimentation and uptake by aquatic biotic communities
would account for some of the variability in the regression model (Hill, 1981;
Burns 1998). In addition to the above, nutrient transport within a watershed is a
function of stream discharge and travel time. Travel time is a measure of stream
length divided by stream velocity. Stream velocity itself is directly proportional to
stream discharge. As stream discharge drops, velocity also drops resulting to an
increase in travel times from the source to the stream system. As a result, a

lesser volume of water entering the watershed system would mean a greater loss

76

of nutrients such as nitrogen and phosphorus. Total nitrogen distribution,
depicted by graded symbols across the watershed was overlaid on a GIS layer
depicting the percentage of agricultural land using ESRI Arc Map. The resultant
map shows a strong spatial correlation between increasing total nitrogen
concentrations and the percentage of agricultural land use (Figure 4-9). It is
interesting to note that the upper Midwest has the highest application rates of
fertilizer (greater than 7 tons per square mile) when compared to the rest of the

Nation (USGS circular, 1999).

77

Total Nitrogen
19.9 - 734.0
734.1 - 1221.3
1221.4 - 2080.0
2080.1 - 4850.0

4850.1 - 13902.2

Percent Agriculture
" ‘ 0.10-0.88
0.89-2.41
" 2.42.455
:' ‘,' 4.86-8.05
[:3 8.081397
in 13.98- 18.54
a 18.55- 22.52
- 22.53 -27.52
- 27.53 -35.90
- 35.91 -51.46

I

   

Total phosphorus
- 0.00- 7.10

e 7.11 - 20.70
0 20.71 - 39.60

. 39.61 - 74.10

0 74.11 - 129.60

Figure 4-9. TN and TP concentration distribution overlaid on % agricultural land
use

78

4.1.3 TP concentration and percentage of LULC

Total phosphorus concentrations (measured in parts per billion) within the
sub-watershed were assumed to be positively correlated (using Spearman’s rank
correlation) with the percentages of urban land use, agricultural land, the
combined percentage of urban and agricultural land, as well as the percentage of
human use index (combined percentages of urban, agricultural and bare land). It

was also assumed that TP concentrations were negatively correlated with forest

 

 

 

 

 

 

cover.
TP vs. LULC Correlation tvalue tcritical, a = 0.05
TP vs. urban 0.384 2.425 1.695
TP vs. agriculture 0.564 3.982 1.695
TP vs. urban + agriculture 0.651 5.000 1.695
TP vs. urban + ag + bare 0.658 5.095 1.695
TP vs. forest -0.496 -3.330 -1.695

 

 

 

 

 

 

Table 4-8. Correlation between TP concentrations and LULC within sub-
watersheds

In addition, TP concentrations within the first order watersheds were
assumed to be positively correlated with percentages of urban and agricultural
LULC types as well as their combined percentages. It was also assumed that the
TP concentrations were negatively correlated with the percentage of forests in

the first order watersheds.

79

 

 

 

 

 

 

TP vs. LULC Correlation tvalue tcritical, d = 0.05
TP vs. urban 0.295 1.8002 1.695

TP vs. ag 0.568 4.0241 1.695

TP vs. urban + agriculture 0.640 4.8567 1.695

TP vs. urban + ag + bare 0.629 4.7178 1.695

TP vs. forest -0.499 -3.3575 -1.695

 

 

 

 

 

 

Table 4-9. Correlation between TP concentrations and LULC within first order
watersheds

As hypothesis, it was assumed that TP concentrations increase when the
percentage of anthropogenically modified land within the sub-watersheds and the

first order watersheds increases. The null hypothesis was that there is no

correlation between the two variables (t s t-critical a - 0.05). As alternative
hypothesis, it was assumed that there is a positive correlation between total
invertebrate taxa and percentage of urban land use (t > t-critical at a = 0.05). The

t critical value was 1.695 at a = 0.05 (95% probability) and df = 34.

Ho Null hypothesis: there is no correlation between TP concentrations and
percentage of land use
H1 Alternative hypothesis: there is a correlation between TP

concentrations and percentage of land use

The null hypothesis was rejected as the tvalues for the different cases (Table 4-

8) were greater than the t critical value = 1.695.

80

Logarithmic regression analysis was also undertaken to explain the variance
in a regression model between percentage of LULC (dependent variable) and TP
concentration in the sub-watershed as well as within the first order watersheds. A
non-linear regression was considered, as a linear least square regression did not
explain any variance in the independent variable. A logarithmic model (Table 4—10)
fitted to TP concentrations and the percentage of urban land use within the sub-
watersheds which yielded an R2 = 0.020. The model explained 2.0% of the
dependent variable, TP concentrations and was not statistically significant at a =

0.05.

F critical value = 4.12 at a = 0.05 (95 % probability) and df = 34

F statistic = 0.701 therefore 0.701 < 4.12 or F statistic < F critical

 

 

 

 

 

 

 

 

 

 

Independent Logarithmic model for total phosphorus Parameter estimates
(dependent)

% LU-sub R Square F df1 df2 Sig. Constant b1
urban .020 0.701 1 34 .408 25.006 6.740
agriculture .187 7.836 1 34 .000 10.884 9.1 63
urb.+ag. .307 15.053 1 34 .000 -20.436 ’ 18.713
h.u. index .309 15.206 1 34 .000 -21.401 18.948
forest .404 23.012 1 34 .000 238.851 -56.44

 

 

 

 

 

 

Table 4-10. Percentage of LULC (within sub-watersheds) vs. TP concentrations

A logarithmic model (Figure 4-10) fitted to TP concentrations and the
percentage of agricultural land use within the sub-watersheds yielded an R2 =
0.187 and significantly explained 18.7 % of the dependent variable, TP

concentrations (Table 4-10).

81

 

F critical value = 4.12 at a = 0.05 (95 % probability) and df = 34
F statistic = 7.836

Therefore 7.836 4.12 or F statistic > F critical

Percentage agriculture within sub-watershed

 

140.00—

120.00....

100.00-

80.00 -

60.00 -l

 

40.00 -

20.00 -1

000-4

 

 

 

 

l l l l l l
0.00 10.00 20.00 30.00 40.00 50.00 60.00
Percentage agriculture

Figure 4-10. Percentage of agricultural land (within sub-watersheds) vs. TP
concentrations

82

Percent of urban + agriculture vs. TP (sub-watershed)

 

140.00 —
120.00 -
100.00 0
80.00 A
60.00 A
40.00 J
20.00 a

0.00—

 

 

 

 

l 1 l 1 i l 1
0.00 10.00 20.00 30.00 40.00 50.00 60.00
percent urban + agriculture

Figure 4-11. Percentage of urban & agricultural land (within sub-watersheds) vs.
TP concentrations

A logarithmic model (Figure 4-11) fitted to TP concentrations and the combined

percentages of urban and agricultural land use raised the R2 to 0.307,

significantly explaining 30.7 % of the dependent variable, TP concentrations

(Table 4-10).

F critical value = 4.12 at a = 0.05 (95 % probability) and df = 34

F statistic = 15.053 therefore 15.053 > 4.12 or F statistic > F cn'tical

83

Percentage of human use index vs. TP

 

140.00—

120.00..

100.00—

80.00 -

60.00 —

40.00 —

20.00 —

0.00 —

 

 

 

 

l l l l l l l
0.00 10.00 20.00 30.00 40.00 50.00 60.00

Percentage of human use index
Figure 4-12. Percentage of human use index within sub-watersheds vs. TP
concentrations
A logarithmic model (Figure 4-12) in which the combined percentages
of urban land use, agricultural land use, and bare soil were regressed with TP
concentrations within the sub-watersheds resulted in R2 = 0.309, significantly

explaining 30.9 % of the dependent variable, TP concentrations (T able 4-10).

F critical value = 4.12 at a = 0.05 (95 % probability) and df = 34

F statistic = 15.206. therefore 15.206 > 4.12 or F statistic > F critical

84

Logarithmic regression was also undertaken to explain the relationship
and variance in a regression model between the percentages of LULC
(dependent variable) and TP concentrations (dependent variable) aggregated to
the first order watershed level. The percentages of land use types include the
percentages of urban land use, agricultural land, the combined percentages of
urban and agricultural land as well as the percentage of human use index

(combined percentages of urban, agricultural and bare land).

 

 

 

 

 

 

 

 

 

 

Independent Logarithmic model for total phosphorus Parameter
(dependent) estimates
% LU-first R Square F df1 df2 Sig. Constant b1
urban .018 .639 1 34 .430 25.593 6.612
agriculture .175 7.223 1 34 .011 1 1.760 8.444
urban+ag. .282 13.32 1 34 .001 -18.886 17.636
human index .284 13.50 1 34 .001 -20.333 18.031
forest .367 19.69 1 34 .000 218.885 -50.37

 

 

 

 

 

 

Table 4-11. Percentage of LULC (within first order watersheds) vs. TP
concentrations

A logarithmic model (Table 4-11) fitted to TP concentrations and the
percentage of urban land use within the first order -watersheds yielded an R2 =
1.8. The model explained 1.8% of the dependent variable (T P concentration) and

was not statistically significant at a = 0.05.

F critical value = 4.12 at a = 0.05 (95 % probability) and df = 34

F statistic = 0.639 therefore 0.639 < 4.12 or F statistic < F critical

85

 

Percent agriculture within first order watershed vs. TP

 

    

 
 

 

 

 

140.0—
0
120.0—
0
100.0—
80.0—
0 o
60.0—
0 O
O O
40.0—
o O O
ochbO O
0.0—
l l l I l l l
0.00 10.00 20.00 30.00 40.00 50.00 60.00

Percent agriculture

Figure 4-13. Percentage of agriculture (within first order -watersheds) vs. TP
concentrations

A logarithmic model (Figure 4-13) fitted to TP concentrations and the
percentage of agricultural land use within the first order-watersheds yielded an
R2 = 0.175 and significantly explained 17.5 % of the dependent variable, TP

concentrations (Table 4-11).

F critical value = 4.12 at a = 0.05 (95 °/o probability) and df = 34
F statistic = 7.223

Therefore 7.223 4.12 or F statistic > F critical

86

Percentage of urban + ag within first order watershed vs.

 

 

 

 

 

TP
140.04
0
120.04
0
100.0—
80.0—
0 o
60.0—
40.0—
20.0“
0.0—
l l l l l 1 r
0.000 10.000 20.000 30.000 40.000 50.000 60.000

Percent of urban + ag LULC

Figure 4-14. Percentage of urban & agriculture (within first order-watersheds) vs.
TP concentrations

A logarithmic model (Figure 4-14) fitted to TP concentrations and the

combined percentages of urban and agricultural land use resulted in an R2 of

0.282, significantly explaining 28.2% of the dependent variable, TP

concentrations (Table 4-11).

F critical value = 4.12 at a = 0.05 (95 % probability) and df= 34

F statistic = 13.328 therefore 13.328 > 4.12 or F statistic > F critical

87

Percentage of human use index within first order watershed vs.

 

 

 

 

 

TP
140.0—
0
1200—
0
100.0—
80.0—
0 o
60.0—
40.0—
20.04
0.0—
I I I I T I T
0.000 10.000 20.000 30.000 40.000 50.000 60.000

Percentage human use Index

Figure 4-15. Percentage of human use index (urban + agriculture + bare within
first order watersheds) vs. TP concentrations

Another logarithmic model (Figure 4-15) in which the combined

percentages of urban, agricultural and bare soil were regressed with TP

concentrations within the first order watersheds resulted in R2 = 0.284,

significantly explaining 28.4 % of the dependent variable, TP concentrations

(Table 4-11).

F critical value = 4.12 at a = 0.05 (95 % probability) and df = 34

F statistic = 13.500, therefore 13.500> 4.12 or F statistic > F critical

88

The amount of TP concentrations increased with the percentages of urban &
agricultural land use and then remained the same irrespective of any increase when
bare soil cover was added. This might be accounted for by mitigation of TP
concentrations due to sedimentation, and stream flow. The TP concentrations were
positively correlated with the land use types such as urban, agricultural and bare
soil. However, this is to be expected as the three LULC types are all created and
modified by humans. Therefore, we would expect to find an inverse or negative
correlation between natural cover such as forests and phosphorus
concentrations to prove that nutrient loading in a watershed is a function of land
use.

Percentage of Forests vs.TP (sub-watersheds)

 

140.00—

120.00..

100.00-l

80.00 -1

60.00 —

40.00 —

20.00 —

 

 

0.00—

 

 

F T l l
0.00 20.00 40.00 60.00 80.00

Percentage of forest

Figure 4-16. Percentage of forests vs. TP concentrations within sub-watersheds

89

A logarithmic model (Figure 4-16) in which the percentage of forest cover
was regressed with TP concentrations within the sub-watersheds resulted in R2 =
0.404, significantly explaining 40.4% of the dependent variable, TP

concentrations (Table 4-11).

F critical value = 4.13 at a = 0.05 (95 % probability) and df = 34

F statistic = 23.012, therefore 23.012 > 4.13 or F statistic > F critical

Percentage of forest cover within first order
watersheds vs. TP

 

 

 

 

140.0—
0
120.0—
0
100.04
80.0—
0 O
60.0_
0 O
0 O
40.0— o
o
8 o
_ o
. 20.0 o o 0 9
o QO
0
00—-
l I I I I I
0.000 20.000 40.000 60.000 80.000 100.000

Percentage of forest

Figure 4-17. Percentage of forests vs. TP concentrations within sub-watersheds

Another logarithmic model (Figure 4-17) in which the percentage of forest
cover was regressed with TP concentrations within the first order-watersheds
resulted in R2 = 0.367, significantly explaining 36.7% of the dependent variable,

TP concentrations (Table 4-11).

90

The regression between total phosphorus and forest cover within the sub-
watershed as well as the first order watersheds confirmed the negative
correlation between the variables. This negative relationship confirms the

premise that land use patterns influence TP concentrations within the MRW.

The nutrient export within the watershed as downstream of the
headwaters is a function of various physical and anthropogenic factors. Factors
such as inter-annual changes in precipitation, cropping practices, fertilizer
applications relative to rainfall events, local geology, and the density and spatial
distribution of impervious surfaces play an important role in influencing total
nitrogen and total phosphorus transport (Wickham et al., 2003). Concentrations
of nutrients in watersheds with a greater proportion of agricultural areas are
higher in spring and summer months owing to higher flow conditions as well as

fertilizer application in the peak—growing season.

91

Legend

TN

0 19.9-7340
o 7341-12213
. 12214-20800
. 2080.1 - 4850.0

. 4850.1 - 13902.2

Percent Human Use Index

T—_ 1.40-3.77
lf 3.78 - 5.39
[1 _ 5.40- 9.62
I: ‘3 963-1505
[1‘5 15.06-18.25
18.26- 22.78
.; 22.79-26.88
- 28.89- 30.73
- 30.74- 42.17
- 42.18-54.88

 

Total phosphorts
- 0.00- 7.10
e 7.11 - 20.70
0 20.71 - 39.60
. 39.61 - 74.10

. 74.11 - 129.60

Figure 4-18. TN and TP concentrations overlaid on % human use index

 

The differences between sub-watershed nutrient concentrations could be
explained by different cropping practices such as conservation tillage versus non-
conservation tillage (Wickham et al., 2003). A sub-watershed where conservation
tillage was practiced would have greater phosphorus concentrations than
compared to a forested watershed. On the other hand, a sub-watershed where
non-conservation tillage was practiced would have greater phosphorus

concentrations than a sub-watershed which practices conservation tillage.

As compared to total nitrogen, a smaller proportion of phosphorus is lost
to nutrient decay. Total phosphorus originates from livestock manure and
agricultural fertilizer and wastewater treatment plants. Even though less
phosphorus is leached from agricultural field than nitrogen, there is a higher
probability that it will reach concentrations (greater than 0.1 mg/liter) enough to
cause eutrophication in streams and lakes through aquatic plant growth (USGS

circular, 1999).

Total phosphorus distribution depicted by graded symbols across the
watershed in ESRI ArcMAP, was overlaid on the percentage of agricultural land
use (Figure 4-18). The total phosphorus distribution was also mapped on the
total proportion of the landscape modified by human activities, ie, the human use
index (urban + agriculture + bare). Again, a strong spatial correlation was seen

between the two variables. A visual assessment indicates that total phosphorus

93

concentrations that are downstream of sub-watersheds with greater percentage

of agricultural land use have higher values (Figure 4-18).

4.1.4 Total invertebrate taxa and percentage of urban land use

Total invertebrate taxa within the sub-watershed were negatively
correlated with the percentage of urban land use (Spearman’s rank correlation: -
0.416, a = 0.05). The null hypothesis was that there is no correlation between the
two variables (t st-critical at a = 0.05). As alternative hypothesis, it was assumed
that there is a negative correlation between total invertebrate taxa and the
percentage of urban land use (t > t-critical at a = 0.05). The t critical value = -

1.695 at a = 0.05 (95% probability) and (if = 34.

Ho Null hypothesis: there is no correlation between total invertebrate taxa
and the percentage of urban land use
H1 Alternative hypothesis: there is a correlation between total invertebrate

taxa and the percentage of urban land use

The null hypothesis was rejected as the t = -2.675 and was greater than —1.695.

 

Total invertebrate taxa & LULC Correlation tvalue tcritical, a = 0.05

 

Total invertebrate taxa & % urban -0.416 -2.675 -1.695

 

 

 

 

 

 

Table 4-12. Percentage of urban land use vs. total invertebrate population within
sub-watersheds

94

Linear regression of the total invertebrate taxa and the percentage of urban land
use failed to explain the variance in the data and so a log-transforrn was used on

the percentage of urban land use to linearize the model.

. 0.5 1 (4-1)
' =1 N* +— 1— +—
y1 og[ y N( y) 2]

where y = % urban land use

The log-transformed model (Bailey and Gatrell, 1995) significantly described 22%
of the variance (p < 0.003) between total invertebrate taxa and the percentage of
urban land use. The linear regression performed after the log transform showed a
negative trend between the diversity of invertebrate taxa and the percentage of urban

land use (constant = 5.5661 and slope = -0.022).

4.1.5 Total EPT taxa and percentage of urban land use

Total EPT taxa within the sub-watershed were negatively correlated with
the percentage of urban land use (Spearman’s rank correlation: -0.381, a = 0.05).
The null hypothesis was that there is no correlation between the two variables (t s t-
critical, 0 = 0.05). As alternative hypothesis, it was assumed that there is a negative
correlation between total EPT and the percentage of urban land use (t > t-cn'tical at a

= 0.05). The t critical value = -1.695 at a = 0.05 (95% probability) and df = 34.

95

Ho Null hypothesis: there is no correlation between total EPT taxa and
the percentage of urban land use
H1 Alternative hypothesis: there is a correlation between total EPT taxa

and the percentage of urban land use

The null hypothesis was rejected as the t = —2.393 and was greater than t-critical.

 

Sensitive insect taxa & LULC Correlation tvalue tcritical, a = 0.05

 

EPT taxa & % urban -0.381 -2.393 -1 .695

 

 

 

 

 

 

Table 4-13. Percentage of urban land use correlated with sensitive insect
population within sub-watersheds

EPT taxa within the sub-watershed were regressed with the percentage of
urban use. As in the case of total invertebrate taxa, a linear regression failed to
explain the variance in the data and so the percentage of LULC was log-
transformed in an attempt to linearize the model. A linear regression of the log-
transformed urban land use yielded an R2 = 0.15 (p < 0.011) and shows that
there is a slight decrease in EPT taxa with increase in percentage of urban land

use (constant = 5.264 and slope = -0.036).

This negative correlation can be explained by the fact that the mayflies,
caddisflies and stoneflies that make up the EPT taxa evolved in oxygen rich, cool

waters. Depleted oxygen levels as well as increasing water temperatures due to

96

increased runoff and short travel times over increasing impervious surfaces might
explain the reason in EPT population decline as the percentage of urban land

use increases.

4.1.6 Proximity to urban areas

A correlation test between total nitrogen concentrations and distance to
urban areas found the two variables to be negative correlated (Table 4-14). Total

nitrogen concentrations dropped with increasing distance from urban areas.

 

TN concentration & distance to LULC Correlation, a = 0.05

 

TN concentration & Urban -0.137

 

 

 

 

Table 4-14. Proximity to urban areas & TN concentrations within MRW

The null hypothesis was that there is no correlation between the two
variables (t st-critical at a = 0.05). As alternative hypothesis, it was assumed that
there is a negative correlation between proximity to urban areas and TN
concentrations within the watershed (t > t-critical at a = 0.05). The t critical value

= -1.645 at a = 0.05 (95% probability) and df = 217.

97

 

Ho Null hypothesis: there is no correlation between proximity to urban
areas and TN concentrations
H1 Alternative hypothesis: there is a correlation between proximity to

urban areas and TN concentrations

The null hypothesis was rejected as the t = -2.0375 and was greater than
t-critical. A correlation test was also carried out between total phosphorus
concentrations and distance to urban areas found the two variables to be
positively correlated but with a very low correlation co-efficient, which was not

statistically significant (at a = 0.05).

 

TP concentration & distance to LULC Correlation, a = 0.05

 

TP concentration & urban 0.057

 

 

 

 

Table 4-15. Proximity to urban areas & TP concentrations within MRW

The null hypothesis was that there is no correlation between the two
variables (t st-critical at a = 0.05). As alternative hypothesis, it was assumed that
there is a negative correlation between proximity to urban areas and total
phosphorus concentrations within the watershed (t > t-critical at a = 0.05). The t

critical value = 1.645 at a = 0.05 (95% probability) and df = 127.

Ho Null hypothesis: there is no correlation between proximity to urban

areas and TP concentrations

98

H1 Alternative hypothesis: there is a correlation between proximity to urban

areas and TP concentrations

The alternative hypothesis was rejected as the t = 0.6434 and was less than t-

critical. Regression analysis characterized a positive linear association between the

dependent variable (total nitrogen concentration) and the independent variable,

distance to urban areas resulted in a negative relationship (Table 4—16). However, it

was not statistically signifimnt at a = 0.05.

 

 

 

 

I Model summary Parameter estimates I
Equation I RSquare F df1 df2 Sig. Constant b1 I
I Linear I .008 1.735 1 217 .189 1043.300 -.397 I

 

 

 

 

 

 

Table 4-16. Proximity to urban areas vs. TN concentrations within MRW

4.1.7 Proximity to agricultural areas

A correlation test carried out between total nitrogen concentrations and

distance to agricultural areas found the two variables to be negatively correlated,

but with a very low correlation co-efficient, which was not statistically significant

at a = 0.05 (Table 4-17).

 

TN concentration & distance to LULC

Correlation

 

 

TN concentration & agriculture

 

-0.0124

 

 

Table 4-17. Proximity to agricultural areas & TN concentrations within MRW

99

A correlation test between TP concentrations and distance to agricultural
areas found the two variables to be negative correlated (Table 4-18). Total
phosphorus concentrations dropped with increasing distance from agricultural

areas.

 

TP concentration & distance to LULC Correlation, a = 0.05

 

TP concentration & agriculture -0.324

 

 

 

 

Table 4-18. Proximity to agricultural areas & TP concentrations within MRW

The null hypothesis was that there is no correlation between the two
variables (t st-critical at a = 0.05). As alternative hypothesis, it was assumed that
there is a negative correlation between proximity to agricultural areas and TP
concentrations within the watershed (t > t-critical at a = 0.05). The t critical value

= 1.645 at a = 0.05 (95% probability) and df = 127.
Ho Null hypothesis: there is no correlation between proximity to agricultural
areas and TP concentrations
H1 Alternative hypothesis: there is a correlation between proximity to

agricultural areas and TP concentrations

The null hypothesis was rejected as t = -3. 8594 and was greater than t—critical.

100

CHAPTER 5 CONCLUSIONS

The objective of this research was to obtain an accurate inventory of LULC
within the 40 sub—watersheds of the Muskegon River Watershed and correlate the
LULC types with water quality indices such as total nitrogen, total phosphorus,
specific conductivity, populations of sensitive insect species (EPT taxa) and total
invertebrate taxa. The specific hypothesis was that LULC affects water quality within
the MRW.

The results demonstrate that there is a positive correlation between types of
land use (urban, agriculture and bare soil) and water quality indicators. This positive
correlation was determined by the spatial scale of study. The two spatial scales
considered were the sub-watershed level study and the first order watershed
level. In addition, negative correlations were found between forest cover and
water quality indicators at the sub-watershed level. This confirms the fact that it is
land use within the watershed that affects the water quality and that forest cover

within catchments is beneficial.

The sub-hypothesis conclusion is that the LULC within the MRW was
accurately mapped through an unsupervised classification of ETM+ imagery
(Landsat-7) and with an overall classification accuracy of 83%. In addition, water
quality indicators measured at 256 sampling locations were aggregated to the sub-

watershed level and correlated with LULC types within the 40 sub-watersheds.

101

Statistical relationships were established between water quality measures and
percentage of land use I land cover within the watershed. Among the water quality
indicators, total phosphorus and total nitrogen distribution as well as biological
indiwtors such as EPT taxa and total invertebrate taxa, were influenced by the

percentages of agricultural and urban land uses within the sub-watershed.

Significant positive correlations were found between total nitrogen
concentrations and percentage of urban and agricultural land uses (0.18 and 0.4
respectively). The correlations increased slightly when the combined
percentages of urban, agriculture and bare soil were considered. Positive
correlations were also found between total phosphorus concentrations and
percentage of urban and agricultural land uses (0.38 and 0.65 respectively).
These correlations are important as the phosphorus concentrations cause
eutrophication in lakes and steams within the watershed. The sources of the
nutrients might be from fertilizer application on farms but might also be from
confined animal feeding operations (CAFO’s) or wastewater treatment plants. An
inverse or negative correlation was found between natural cover such as forests
and nitrogen/phosphorus concentrations (0.40 and 0.49 respectively) to prove

that nutrient loading in a watershed is a function of land use.

The variance in total phosphorus concentrations could be explained more by

the percentage of land use within the sub—watersheds as compared to the variance in

total nitrogen concentrations. Though the statistical relationships between the

102

 

dependent variables (water quality indices) and the independent variables (LULC
types) were significant at the 95% confidence level, the predictive power of the

logarithmic models was very low with R2 ranging from 0.018 to 0.40.

The low predictive power of the logarithmic regression models for nitrogen
concentrations can be attributed to various physical and anthropogenic factors. For
instance, the variability in total nitrogen concentrations could be attributed to
denitrifioation, sedimentation and uptake by aquatic biotic communities. The transport
of nutrients in a watershed is also influenced by stream velocity and travel time. The
variability of the nutrient concentrations within the sub-watersheds could also be
governed by factors such as changes in precipitation, cropping practices, fertilizer
applications relative to rainfall events, local geology, the density and the spatial
distribution of impervious surfaces. Total nitrogen concentrations were influenced
more by distance to urban areas as opposed to total phosphorus concentrations,

which were influenwd by distance to agricultural areas.

Specific conductivity was positively correlated with individual land use types
such as urban and agricultural land use or the combined percentages of human
modified LULC types with correlations ranging from 0.60 and 0.72. This could be
attributed to a greater proportion of urban land use in some of the sub-watersheds.
Also, the percentage of forest cover was negatively correlated with specific
conductivity. This negative correlation makes a strong case for the influence of urban

and agricultural land use on water quality. In addition, sensitive insect populations

103

 

(EPT taxa) as well as the total invertebrate taxa were negatively correlated with
urban land use (0.38 and 0.41 respectively). These correlations were significant at

the 95 % confidence level.

My research studies the land-water interactions at two spatial scales, i.e., the
sub-watershed level and the first order watershed level. It is based in part on
Omemik's study (1981) that demonstrated that land use in the distant uplands had as
much effect on water quality as land use in the vicinity. in addition, my research
considers the proportions of land use within hydrologically distinct units such as sub-
watersheds as opposed to the immediate vicinity, which is the case in most
ecological studies. These studies suggest that water quality is influenced by the land

use within a buffer zone that is approximately 30—100m from the water’s edge.

The beautiful natural surrounding and easy access to recreation have resulted
in an increase in urban development around the lakes in Michigan. In recent times,
land value has gravitated towards residential development rather than agricultural
land use. Also, it is important to note that the upper Midwest has the highest
application rates of fertilizers on agricultural land when compared to the rest of the
nation. Agricultural land use still plays a major role in affecting water quality. This
research is important as it documents the correlation between urban and agricultural
land use and water quality of Michigan’s second largest watershed. The statistical
analysis provides proof that sub-watersheds with a greater proportion of natural

cover such as forest cover are beneficial and negatively correlated to nutrient

104

concentrations as compared to sub-catchments with a greater proportion of urban /
agricultural land use. Local decision makers could be empowered to act in their own
interests and conserve the natural surrounding which in turn would ensure qood
water quality. Sub-watersheds that are predominantly urban or agricultural in nature
could mitigate their impact on the water quality through the maintenance of riparian
buffer strips. In addition to acting as filters and reducing nutrient concentration within
lakes and streams, riparian buffers prevent stream bank erosion. Ecologists who
believe that landscape patterns influence water quality often prescibe riparian buffers
as a conservation measure. Future research that concerns land-water interactions in
the MRW might consider correlating landscape metrics (e.g., contagion, dominance)

that describe the spatial configuration of the landscape with water quality indicators.

105

1'

APPENDICES

106

APPENDIX-A: Figures used in the study

 

Figure 5-1. imagery from WRS Path 21 Row 30 Landsat 7 (ETM+)

107

 

Figure 5-2. Imagery from WRS Path 22 Row 30 Landsat 7 (ETM+)

108

 

 

Figure 5-3. MRW—LULC map derived through unsupervised classification

109

   
     

Projection: UTM
Zone 16
Datum WGS 84

051)

 

Figure 5-4. MRW—LULC map derived through unsupervised classification and
smoothed with 3x3 majority filter

110

   
 

Projection: UTM
Zone 16
Datum WGS 84

 

Elevation (in meters)
. High : 528

.. _ Low : 177

Figure 5-5. SRTM derived Digital Elevation Model (DEM)

111

 

Figure 5-6. SRTM derived DEM overlaid with Strahler stream order

112

 

 

Figure 5-7. Wetness index derived from SRTM DEM overlaid with MRW wetlands
coverage

113

 

Figure 5-8. Urban development around Lake Houghton (2001)

114

“.4 I. .. 1

.. 42A dtp‘ut

NA 9.. a.

.. IX. .r
n.
g-
I

s.
N

. I

-x ,‘ao- ‘

e v.
N. ._ t.
a . . a _
.I._ . s. 4.
s .7 A M _
.r..l...» . . . .r‘ 3 I;
. 4 s... 1.. a... .
A a. v0 ’ I .
a .. .7. J
o .. EN... .2‘
.. ..\. .u r L
... .\w as 5.011.
. .. m 1%
.pl. :6? .. .
. 2 .1
4 .6 . n.
.. 10.3.).
. a..- .. Car. .4
._ a _.
.... .. 1 ._
.. a .. w
. 7. It}-.. t: .
, .5 , 6..

 

. ~ . _
7)., .

  

{romaifsdfi Left. . .ih.l.r...}..A&1...vi.TlJili If

A 2% fluthurqwrivfl L.£.hr..u.«wnfli.t Hi.” / .
I . .. .1;1..fi..3n:.u....a .
/. . \ . . .1-.. .
.. /(\ v
r. ~ m
_ ..
l .. .2

 

2

 

 

0051

Figure 5-9. Urban development around Lake Higgins (2001)

115

APPENDIX B: Data used in statistical analysis

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Sub-watersheds TN TP

1 270.80 no data
2 1204.20 7.10
3 1460.00 16.17
4 1549.70 12.40
5 1320.00 27.83
6 1006.20 19.80
7 no data no data
8 1 721 .40 48.70
9 1 1 18.78 26.38
10 695.92 1 1 .50
1 1 637.04 1 1 .91
12 1447.30 41 .10
13 1478.90 31.50
14 1200.00 22.30
15 no data no data
16 2060.10 129.60
17 1046.20 1 1 .70
18 848.60 14.00
19 1 193.00 67.20
20 1503.10 27.10
21 890.70 56.20
22 906.60 16.91
23 1325.40 31 .47
24 937.90 32.90
25 13902.20 30.13
26 586.40 13.30
27 2829.40 22.23
28 7811.00 18.00
29 3595.40 57.00
30 4850.00 109.80
31 2483.40 21 .30
32 no data no data
33 641 .30 28.30
34 809.30 15.00
35 1808.70 65.70
36 593.00 14.20
37 1084.40 17.60
38 1070.00 40.95
39 555.60 1 1.60
40 2080.00 49.30

 

 

Table 5-1. Water quality indicators within MRW sub-watersheds

TN- total nitrogen, TP- total phosphorus (measured in parts per billion)

116

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Sub-watersheds Conductivity EPT taxa Tot invert. taxa
1 no data no data no data
2 236.0 16.0 43.0
3 268.5 10.0 34.0
4 224.0 9.0 42.0
5 294.0 23.0 66.0
6 268.8 12.0 36.0
7 no data no data no data
8 292.5 4.0 22.0
9 221.0 11.0 42.0
10 181.0 20.0 50.0
11 324.0 6.0 26.0
12 no data no data no data
13 338.8 13.0 43.0
14 355.9 15.0 39.0
15 no data no data no data
16 398.1 15.0 41.0
17 349.3 17.0 47.0
18 290.8 25.0 58.0
19 359.3 11.0 41.0

20 344.4 15.0 49.0
21 371.6 14.0 42.0
22 521.0 22.0 48.0
23 414.3 13.0 32.0
24 415.7 11.0 37.0
25 455.0 8.0 24.0
26 355.4 17.0 38.0
27 330.0 7.0 33.0
28 519.3 16.0 41.0
29 472.2 14.0 43.0
30 514.0 1.0 9.0
31 525.3 5.0 30.0
32 496.5 6.0 23.0
33 325.0 2.0 26.0
34 287.8 18.0 42.0
35 521 .3 27.0 57.0
36 496.5 6.0 23.0
37 273.5 9.0 34.0
38 444.9 12.0 49.0
39 286.5 6.0 28.0
40 631.0 2.0 16.0

 

 

 

 

Table 5-2. Water quality indicators within MRW sub-watersheds
Conductivity- specific conductivity, EPT taxa- ephemeroptera, plecoptera &

trichoptera, tot invert.taxa - total invertebrate taxa

117

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

sub- urban (log
watersheds urban agriculture urban+ agric. bare ag.+urban+bare transformed)
1 4.33% 0.11% 4.44% 0.12% 4.56% 0.808
2 2.62% 0.88% 3.50% 0.28% 3.78% 0.443
3 3.60% 1.18% 4.78% 0.19% 4.97% 0.667
4 1.64% 2.42% 4.06% 0.05% 4.11% 0.156
5 1.49% 8.05% 9.54% 0.09% 9.63% 0.101
6 1.87% 13.98% 15.85% 0.31 % 16.15% 0.231
7 0.80% 0.60% 1 .40% 0.00% 1 .40% —0.183
8 2.12% 24.32% 26.44% 0.45% 26.89% 0.306
9 2.30% 1.04% 3.34% 0.11% 3.45% 0.359
10 2.34% 1.28% 3.62% 0.08% 3.71% 0.370
11 1.71% 3.56% 5.27% 0.10% 5.37% 0.180
12 6.56% 12.55% 19.11% 1.05% 20.16% 1.142
13 2.65% 21 99% 24.64% 0.29% 24.93% 0.452
14 3.67% 15.75% 19.43% 0.23% 19.66% 0.683
15 3.35% 42.68% 46.03% 0.45% 46.47% 0.616
16 2.95% 46.24% 49.18% 0.53% 49.71% 0.525
17 1.85% 16.13% 17.97% 0.29% 18.26% 0.223
18 1.00% 2.00% 3.00% 0.10% 3.10% -0.091
19 2.69% 26.93% 29.62% 0.23% 29.84% 0.461
20 2.40% 1 1.95% 14.35% 0.06% 14.41% 0.387
21 2.86% 34.61% 37.46% 0.13% 37.59% 0.503
22 2.56% 13.32% 15.88% 0.13% 16.01% 0.429
23 3.83% 21.00% 24.83% 0.15% 24.99% 0.715
24 3.45% 35.90% 39.35% 0.15% 39.49% 0.637
25 4.07% 18.55% 22.62% 0.16% 22.78% 0.761
26 2.07% 11.43% 13.50% 0.03% 13.52% 0.291
27 2.50% 22.53% 25.02% 0.27% 25.30% 0.412
28 2.84% 32.07% 34.91% 0.14% 35.05% 0.500
29 3.15% 27.53% 30.68% 0.05% 30.73% 0.572
30 3.25% 51.47% 54.72% 0.17% 54.89% 0.594
31 6.50% 35.37% 41.87% 0.31% 42.17% 1.135
32 5.22% 0.17% 5.39% 0.00% 5.39% 0.955
33 2.57% 4.85% 7.42% 0.01% 7.44% 0.432
34 1.89% 12.18% 14.07% 0.07% 14.13% 0.236
35 3.98% 25.82% 29.80% 0.13% 29.94% 0.744
36 1 .00% 2.00% 3.00% 0.43% 3.43% -0.091
37 3.10% 14.90% 18.00% 0.20% 18.21% 0.561
38 3.98% 20.62% 24.60% 0.25% 24.84% 0.743
39 13.75% 1.02% 14.77% 0.28% 15.06% 1.793
40 26.74% 0.69% 27.42% 2.25% 29.67% 2.416

 

 

 

 

 

 

118

Table 5-3. Percentages of LULC types within sub-watersheds

 

Bibliography

Allan, J. D., Erickson, D. L., and Fay, J. (1997). The influence of catchment land
use on stream integrity across multiple scales. Freshwater Biol. 37:149—
161 .

Alexander, 6. R., Fenske, J. L., and Smith, D. W. (1995). A fisheries
management guide to stream protection and restoration. Michigan
Department of Natural Resources, Fisheries Division. Special Report
Number 15, Ann Arbor, Michigan.

Anderson, J. R., Hardy, E. E., Roach, J. T., and Witmer, W. E. (1976). A land use
and land-cover classification system for use with remote sensor data.
USGS Professional Paper 964. Reston, VA: US. Geological Survey.

Bailey, TC, and Gatrell, AC (1995) Interactive Spatial Data Analysis. Longman.

Bannerman, R., Owens, D., Dodds, R., and Homewer, N. (1993). Sources of
Pollutants in Wisconsin Storrnwater. Water Science and Technology.
28(3-5): 241 -259.

Beach, D. (2002). Coastal Sprawl: The effects of urban design on aquatic
ecosystems of the United States, Pew Oceans Commission, Arlington, Va.

Booth, D. B. (1991). Urbanization and the natural drainage system-impacts,
solutions and prognoses. The Northwest Environmental Journal 7293-1 18.

Bruneau, P., Gascuel-oduux, 0., Robin, P., Merot, P., Beven, K. (1995).
Sensitivity to Space and Time resolution of a hydrological model using
Digital Elevation Data. Hydrological Processes, 9, 69-81.

Burns, D. A., (1998). Retention of NO-3 in an upland stream environment: a
mass balance approach. Biogeochemistry 40: 73-96.

Burt, T. P., and Butcher, D. P. (1985). Topographic controls of soil moisture
distributions, J of Soil sci, 36, 469-486.

Carpenter, S. R., Caraco, N. F., Corell, D. L., Howarth, R. W., Sharpley, A. N.,
Smith, V. H. (1998). Nonpoint pollution of suface waters with phosphorus
and nitrogen. Ecological Applications 82559-568.

Carpenter, 8., Frost, T., Persson, L., Power, M., and Soto, D. (1995). Freshwater
ecosystems: linkages of complexity and processes. In H.A. mooney, ed.
Functional Roles of Bio-Diversity: A Global Perspective, Chapter 12. John
Wiley & Sons, New York, New York, USA.

119

Changnon, S. A. (1992). Inadvertent weather modification in urban areas:
Lessons for global climate change. Bull.Am.MeteoroI.Soc., 73, 619-627.

Christopherson, R. W. (1995). Elemental Geosystems: A Foundation in Physical
Geography, Prentice Hall .

Cirmo, C. P., and Mc Doneli, J. J. (1997). Linking the hydrologic and
biogeochemical controls of nitrogen transport in near-stream zones of
temperate forested catchments. A review. Joumal of Hydrology 199:88-
1 20.

Congalton, R. G. (1988). Using spatial autocorrelation analysis to explore the
errors in maps generated from remotely sensed data. Photogrammetn'c
Engineering and Remote Sensing, 54, 587.

Congalton, R. G. (1991). A review of assessing the accuracy of classifications of
remotely sensed data. Remote Sensing of the Environment, Vol. 37, pp.
35-46.

Cowardin, L. M., Carter, V., Golet, F. C., LaRoe, E. T. (1979). Classification of
Wetlands and Deepwater Habitat of the United States, Fish and Wildlife
Service, US. Department of the Interior, Washington, DC.

Detenbeck, N., Johnston, C. A., and Niemi, G. (1993). Wetland effects on lake
water quality in the Minneapolis/St Paul metropolitan area. Landscape
Ecology 8:39-61 .

Engman, E. T. (1997). Soil moisture, the hydrologic interface between surface
and ground waters. Remote Sensing and Geographic lnforrnation
Systems for Design and Operation of Water Resources Systems
(Proceedings of Rabat Symposium S3), 242, 129— 140.

ERDAS Field Guide (2004), Leica Geosystems GIS & Mapping.
ESRI (1994), Cell Based Modeling with GRID.

Fairfield, J., and Leymarie, P. (1991). Drainage networks from grid digital
elevation models. Water Resources, 27(5), 109— 717.

Fitzgerald, R. W., and Lees, B. G. (1994). Assessing the classification accuracy of
multisource remote sensing data. Remote Sensing of the Environment, Vol.
47. pp. 362-368.

Gaufin, A. R., and Tarzwell, C. M. (1952). Aquatic invertebrates as indicators of
stream pollution. Public Health Reports 67: 57-64.

120

Gomi, T., Sidle, R, C., Richardson, J. S. (2002). Understanding Processes and
downstream Linkages of Headwater Systems. Bioscience, 52, (10), 905-
916.

Goudie, A. (1990). The human impact on the natural environment, 3rd Ed. The
MIT Press, Cambridge, Massachusetts.

Goward, S. N., Masek, J. G., Williams, D. L., Irons, J. R., and Thompson, R. J.
(2001). The Landsat 7 mission: Terrestrial research and applications for
the 21 st century. Remote Sensing of Environment, 78, 3-12.

Halwas, K. L, Church, M. (2002). Channel units in small, high gradient streams
on Vancouver Island, British Columbia. Geomorphology 43: 243-256.

Harding, J. S., Young, R. G., Hayes, J. W., Shearer, K. A., and Stark, J. D.
(1999). Changes in agricultural intensity and river health along a river
continuum. Freshwater Biol. 42:345—357.

Horton, R. E. (1933), The role of infiltration in the hydrologic cycle. Trans. Amer.
Geophys. Union 14:446-460.

Hill, A. R. (1981). Nitrate-nitrogen flux and utilization in a stream ecosystem
during low summer flows. Canadian Geographer 35: 225—239.

Hunsaker, C. T., and Levine, D. A. (1995). Hierarchical approaches to the study
of water quality in rivers. Bioscience, 45(3), 183—192.

Jenson, S., K., and Domingue, J., O. (1988). Extracting topographic structure
from digital elevation data for geographic information system analysis.
Photogrammetric Engineering and Remote Sensing, 54, 1593-1600.

Johnson, L. B., and Gage, S. H. (1997). Landscape approaches to the analysis
of aquatic ecosystems. Freshwater Biology 37:1 13—132.

Johnson, L. B., and Richards, 0., Host, 6., and Arthur, J. W. (1997). Landscape
influences on water chemistry in Midwestern streams. Freshwater Biology
37: 193-208.

Kibler, D. F. (ed.), (1982), Urban storrnwater hydrology. American Geophysical
Union, Washington, DC.

Lammert, M., and Allan, J. D. (1999). Assessing biotic integrity of streams:

Effects of scale in measuring the influence of land use/cover and habitat
structure on fish and macroinvertebrates. Environ.Manage. 23:257-270.

121

Ligon, F. K., Dietrich, W. E., & Trush, W. J. (1995). Downstream ecological
effects of dams: A geomorphic perspective. Bioscience, 45, 183-192.

Likens, G. E, Bonnann, F. H, Pierce, R. S, Eaton, J.S, Johnson, N. M (1977).
Biogeochemistry of a Forested Ecosystem. New York: Springer-Verlag.

Machemer, P.L., Kaplowitz, M, D., Edens, T, C. (1999). Managing Growth and
Addressing Urban Sprawl, An Overview. 1999, Research Report,
Michigan Agricultural Experiment Station, Michigan State University.

Mark, D. M. (1983). Automated detection of drainage networks from digital terrain
models, Proc.Auto-Carto 6, Automated Cartography: lntemational
Perspectives on Achievements and Challenges. Ottawa, Ontario, 288-298.

Masek, J. G., Honzak, H., Goward, S. N., Liu, P., and Pak, E. (2001). Landsat-7
ETM+ as an observatory for land cover: Initial radiometric and geometric
comparisons with Landsat-5 Thematic Mapper. Remote Sensing of
Environment, 78, 1 18-130.

Meyer, J. L., Wallace, J. B. (2001). Lost linkages and Iotic ecology:
Rediscovering small streams. Pages 295—317 in Press MC, Huntly NJ,
Levin S, eds. Ecology: Achievement and Challenge. Oxford (United
Kingdom): Blackwell Scientific.

Milesi, C., Elveridge, C. D., Nemani, R. R, and Running, S. W. (2003). Assessing
the impact of urban land development on net primary productivity in the
South eastern United States. Remote Sensing of the Environment, 86,
401 -41 0.

Michigan Land Cover IUse Classification System (2000).

Moore, i. D., Grayson, R. B., Ladson, A. R. (1991). Digital terrain modeling: A
review of Hydrological, Geomorphologicai, and Biological Applications.
Hydrological Processes, Vol.5, 3-30.

Mueller, D. K and Helsel, D. R. (1996). Nutrients in the Nation’s waters- too much
of a good thing? US. Geological Survey Circular 1136.

Natural Resources Inventory. (1987). Michigan Data. USDA Soil Conservation
Service, East Lansing, Michigan.

Neckel, H and D. Labs. (1984). The solar radiation between 33000 and 125000.
Solar Physics, 90:205-258.

O’Callaghan, J., F ., and Mark, D. M. (1984). The extraction of drainage networks

from digital elevation data, Computer Vision, Graphics and Image
Processing, 28, 323-344.

122

O'Neai, R. P. (1997). Muskegon River Watershed Assessment. Michigan
Department of Natural Resources, Fisheries Division. Fisheries Special
Report 19. 187 pp.

O’Neill, R. R., Christensen, S. W., Dale, V. H., De Angelis, D. L., Gardner, R. H.,
Graham, R. L, Jackson, 8., Krummel, J. R., Milne, B. T., Sugihara, G.,
Turner, M. G., Zygmyunt, B. (1988). lndices of Landscape pattern,
Landscape Ecology, 1(3): 153-162.

Omemick, J. M., Abernathy, A., R., and Male, L. M. (1981 ). Stream nutrient
levels and proximity of agricultural and forest land to streams: some
relationships. Joumal of Soil and Water Conservation 36:227-231.

Osborn, L. L., and Wiley, M. J. (1988). Empirical relationships between land
use/land cover and stream quality in an agricultural watershed. Joumal of
Environmental Management 26: 9-27.

Paulsen, S. G., Larsen, D. P., Kaufmann, P. R., Whitier, T. R., Baker, J. R., Peck,
D. V., McGue, J., Hughes, R. M., McMullen, D., Stevens, D., Stoddard, J.
L., Lazorchak, J., Kinney, W., Selle, A. R., and Hjort, R. (1991).
Environmental Monitoring and Assessment Program: EMAP surface
waters monitoring and research strategy- fiscal year 1991. EPN600/3-
91/022.United States Environmental Protection Agency. Washington, D.
C.

Pijanowski, B. C., Shellito, B., Pithadia, S., Alexandridis, K. (2002). Forecasting
and assessing the impact of urban sprawl in coastal watersheds along
eastern Lake Michigan, Lakes & Reservoirs: Research and Management
7: 000—000 (in press).

Poff, N. L., Allan, J. D., Bain, M. B., Karr, J. R., Prestegaard K. L., Richter, B. D.,
Sparks, R.E., Stromberg, J.C. (1997). The natural flow regime: A
paradigm for conservation and restoration of river ecosytems. BioScience
47: 769—784.

Poff, N. L., Hart, D. D. (2002). How dams vary and why it matters for the
emerging science of dam removal. BioScience 52: 659—668.

Prevost, M., Plamondon, A. P., Belleau, P. (1999). Effects of drainage of a
forested peatland on water quality and quantity. Journal of Hydrology/214:
130-143.

Rabeni, C. F., and Smale, M. A. (1995). Effects of siltation on stream fishes and

the potential mitigating role of the buffering riparian zone. Hydrobiologia
303:211—219.

123

 

Rabus, B., Eineder, M., Roth, A., Bamier, R. (2003). The shuttle radar
topography mission—a new class of digital elevation models acquired by
spacebome radar. ISPRS Journal of Photogrammetry & Remote Sensing,
57, 241— 262.

Richards, 0., Host, G. E., and Arthur, J. W. (1993). Identification of predominant
environmental factors structuring stream macroinvertebrate communities
within a large agricultural catchment. Freshwater Biol. 29:285—294.

Richardson, R. E. (1928). The bottom fauna of the Middle Illinois River, 1913-
1925. Illinois Natural History Survey Bulletin 17: 387-475.

Sidle, R. C, Tsuboyama, Y., Noguchi, S., Hosoda, I., Fujieda, M., Shimizu, T.
(2000). Streamflow generation in steep headwaters: A linked hydro-
geomorphic paradigm. Hydrological Processes 14: 369—385.

Soranno, P. A., Hubler, S. L., Carpenter, S. R., and Lathrop, R. C. (1996).
Phosphorus loads to surface waters: a simple model to account for spatial
patterns of land use. Ecological Applications 6: 865-878.

Steuer, J., Selbig, W., Homewer, N., and Prey, J. (1997). Sources of
Contamination in an Urban Basin in Marquette, Michigan and an Analysis
of Concentrations, Loads, and Data Quality. US. Geological Survey,
Water-Resources Investigations Report 97-4242.

Strahler, A. N. (1957). Quantitative Analysis of Watershed Analysis of Watershed
geomorphology. Transactions of the American Geophysical Union.

Sun, G., Ranson, K. J, Kharuk, V. l., Kovacs, K. (2003). Validation of surface
height from shuttle radar topography mission using shuffle laser altimeter,
Remote Sensing of Environment, 88, 401-411.

Teillet, P. M., Barker, J. L., Markham, B. L., Irish, R. R., Fedosejevs, G. and
Storey, J. C. (2001). Radiometric cross-calibration of the Landsat-7 ETM+
and Landsat-5 TM sensors based on tandem data sets. Remote Sensing
of Environment, 78, 39-54.

Tou, J. T., and Gonzalez, R. C. (1974). Pattern Recognition Principles. Reading,
Massachusetts: Addison-Wesley Publishing Company.

Turner, M. G., Gardner, R. H., O'Neill, R. V. (2001). Landscape Ecology In
Theory and Practice: Pattern and Process. New York: Springer Verlag.

Tsukamoto, Y., Ohta, T., Noguchi H. (1982). Hydrological and geomorphological

study of debris slides on forested hillslope in Japan, ed. Recent
Developments in the Explanation and Prediction of Erosion and Sediment

124

Yield. lntemational Association of Hydrological Sciences publication no.
137, 89-98, Walling D E.

U. S. Environmental Protection Agency. (1996). Strategic Plan for the Office of
Research and Development. EPA/600/R-96/059. Washington, D. C.

U. S. Environmental Protection Agency. (1997). Environmental Monitoring and
Assessment Program (EMAP) (http://www.epa.gov/emaplhtml/about.html).

U. S. Environmental Protection Agency. (1997). Index of Watershed indicators: A
Overview (http://www.epa.gov/iwi/iwi-overview.pdf).

U.S.Geological Survey. (1999). The Quality of Our Nation’s Waters- Nutrients
and Pesticides: U.S.Geological Survey Circular 1225, 82p.

U.S.Geological Survey. (2000). US GeoData Digital Elevation Models, 040-00
(April 2000) http://erg.usgs.gov/isblpubs/factsheets/st4000.html.

Veatch, J. O. (1953). Soil and land of Michigan. Memoir (Michigan State College.
Agicultural Experiment Station), no.7.

Vitousek, P. M. and Howarth, R. W. (1991). Nitrogen limitation on land and in the
sea: how can it occur? Biogeochemistry 13:87-115.

Vogelmann, J. E, Sohl, T., Howard, S. M. (1998). Regional Characterization of
Land Cover Using Multiple Sources of Data. Photogrammetric Engineering
& Remote Sensing, Vol.64, 1, 45-57.

Wallace, J. B., Grubaugh, J. W., and Whiles, M. R. (1996). Biotic indices and
stream ecosystem processes: results from an experimental study.
Ecological Applications 6: 140-151.

Waschbusch, R., Selbig, W., and Bannerman, R. (2000). Sources of phosphorus
in storrnwater and street dirt from two urban residential basins in Madison,
Wisconsin, 1994-1995. in: National Conference on Tools for Urban Water
Resource Management and Protection. US EPA February 2000: pp. 15-
55.

Whipple, K. X., Hancock, G. S., and Anderson, R. S. (2000). River incision into
bedrock: Mechanics and relative efficacy of plucking, abrasion, and
cavitation. Geol. Soc. Am. Bull, 112, 490-503.

Wickham, J. D., Wade, T. G., Riitters, K. H., O'Neill, R. V., Smith, J. H., Smith, E.

R., Jones, K. B., and Neale, A. C. (2003). Upstream-to-downstream
changes in nutrient export risk, Landscape Ecology. 18: 195-208.

125

 

Wipfli, M. S. (1997). Terrestrial invertebrates as salmonid prey and nitrogen
source in streams: Contrasting old-grth and young-grth riparian
forests in southeastern Alaska, USA. Canadian Journal of Fisheries and
Aquatic Sciences 54: 1259—1269.

Wipfli, M. S. and Gregovich, D., P. (2002), Invertebrates and detritus export from
fishless headwater streams in southeastern Alaska: Implications for
downstream salmonid production. Freshwater Biology 47: 957—970.

Wolock, D., M., Homberger, G. M., and Musgrove, T., J. (1990). Topographic
effects on flow path and surface chemistry of the Llyn Brianne catchments
in Wales. J.Hydrol., 115, 243-259.

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

Zimmerman, A. and Church, M. (2001). Channel morphology, gradient profiles
and bed stresses during flood in a step-me channel. Geomorphology, 40:
311-327.

Zhang, W. and Montgomery, D. R. (1994). Digital elevation model grid size,
landscape representation and hydrologic simulations. Water Resources
Research, Vol 30, 4, 1019-1028.

126

 

IIIIIIIIIIII