LIBRARY
Michigan State
University

 

 

 

PIACE 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 Du; L DATE DUE DATE DUE

 

JUL ”$089“ if

 

 

 

 

 

 

 

 

 

 

 

 

 

 

6/01 cJCIRC/DanDmpGS-sz

WATER CLARITY/ TROPHIC CONDITION MONITORING

USING SATELLITE REMOTE SENSING DATA

By

Narumon Wiangwang

A THESIS
Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of
MASTER OF SCIENCE
Department of Geography

2002

ABSTRACT

WATER CLARITY/ TROPHIC CONDITION MONITORING USING
SATELLITE REMOTE SENSING DATA

By

Narumon Wiangwang

Human activities impact the quality of water in lakes and bays in many ways.
Eutrophication, one of the most pervasive problems that had been of concerned in both
Michigan and Thailand, primarily caused by excessive nutrient export in runoff from
urban and agricultural landscapes. Traditional water quality monitoring techniques are
costly and time-consuming, thereby often resulting in only a sporadic sampling over time
and space. Satellite sensors, such as Landsat TM and ETM+, has the potential to enhance
the environmental resource management as it was found to well correlated with Secchi
Disc Transparency (SDT). The purpose of this study is to test the accuracy and the
applicability of the satellite-based water clarity/ trophic condition monitoring in Michigan
lakes and the Bight of Bangkok, Thailand.

The results of this study showed that the accuracy of using satellite data for water
clarity/ trophic status measurement was quite high (1’2 values 0.57 — 0.70; p < 0.001) in
Michigan and even higher (r2 > 0.83; p < 0.001) in Bangkok Bight, Thailand. These
accuracies were achieved by converting digital numbers to radiance before correlating to
water clarity. Therefore, this method can be used with both Landsat-5 TM and Landsat-7
ETM+ allowing a finer temporal scale monitoring. An advantage of this study is to

provide a potentially water monitoring on a regional scale within a single day at low cost.

ACKNOWLEDGEMENTS

First, I would like to thank my thesis committee for their help and guidance. I
deeply thank Dr. David Skole, my advisor, for his expertise, his valuable time and
advices. I am also grateful for his support in all the data used in this study. I extend
special appreciations to my committee members, Dr. Jiaguo Qi and Dr. Joseph Messina,
for reading previous drafts of this thesis and providing valuable comments. Without their
academic expertise, understanding and encouragement this work would not have been
possible.

Special thanks to my fellow graduate students and friends, Steve Cameron, Eraldo
Matricardi, Yushuang Zhou, Cuizhen Wang, Stacy Nelson, and Andrea Silva, for always
providing thoughtful comments, assistance, and encouragement when I needed it the
most. I am grateful to CGCEO staffs, Deana Haner for helping with the schedules, Jay
Samek and Oscar Castaneda for their finest comments and technical data supports.

I would also like to thank for several grants of funding that support this research.
I thank the government of Thailand and the department of fisheries for giving me the
great opportunity to be here, to contribute this research, and to meet lots of grateful
people. I thank the Center for Global Change and Earth Observation (CGCEO) for
providing me the best working environment and for covering all fieldwork expenses in
Thailand; Graduate Office Fellowship MSU Dept. of Geography for the funds for the
Bight of Bangkok imagery; the NASA-funded Upper Midwest Regional Earth Science
Applications Center (RESAC) project and USGS/DEQ granted through CGCEO for

Michigan imagery and in-situ data.

iii

I would also like to extend my special appreciations to my previous undergraduate
advisor, Dr. Charlie Navanugruha, and my fellow, Siam Lawavirojwong, for their
constant support and inspiration. Their encouragement keeps me work until I achieve my
goaL

Last, but not least, I deeply thank my husband Perry for his love and support
during months of fieldwork in Thailand and weeks of the thesis writing. His patience and
support are part of this thesis. I would like to express the deepest gratitude for my
parents and my brother for their constant love and care. I thank them for always

believing in me and being supportive in everything in my life.

iv

TABLE OF CONTENTS

List of Figures .............................................................................. ix

List of Tables ............................................................................... xiii
CHAPTER I: INTRODUCTION ...................................................... 1
1.1 Background ......................................................................... l
1.2 Comparison of ground-based and satellite monitoring approaches ......... 7
1.3 Study area ........................................................................... 9
1.4 Hypothesis .......................................................................... 9

1.5 Sub-hypothesis ..................................................................... 10

1.6 Benefits of the research ........................................................... 10

References ........................................................................... 10

CHAPTER 1]: LITERATURE REVIEW ............................................. 12

2.1 Landuse .............................................................................. 12
2.1.1 Nature of land use in Michigan ......................................... 13

2.1.1.1 History ofland use in Michigan ............................ 14

2.1.1.2 Soils and land ................................................. 14

2.1.1.3 Importance of Michigan agriculture ........................ 15

2.1.1.4 Changes in land use .......................................... 15

2.1.1.5 Impacts from land use change .............................. 16

2.1.2 Nature of land use in Thailand .......................................... 17

2.2 Trophic State Index (TSI) ........................................................ 21

2.2.1 Advantage of Carlson’s TSI ............................................. 24

2.2.2 Potential errors of TSI ................................................... 25

2.3 Field observation data ............................................................ 26

2.3.1 Factors affecting lake color ..............................................
2.3.2 Factors affecting ocean color ...........................................
2.3.3 Monitoring programs in Michigan .....................................

2.3.3.1 The Lake Water Quality Assessment (LWQA)

Monitoring Program ..........................................

2.3.3.2 The Cooperative Lakes Monitoring Program (CLMP)...

2.4 Remote sensing technologies for water quality monitoring ..................
2.4.1 Landsat sensor cross-calibration (Landsat-5 TM and Landsat-7
ETM+) .....................................................................
2.4.2 Use of satellite data in water monitoring applications ..............
2.5 Radiometric and atmospheric effects ..........................................
2.5.1 Radiometric normalization ..............................................
References ...........................................................................
CHAPTER III: METHODOLOGY ...................................................
3.1 Ground observed water clarity data .............................................
3.1.1 Field observations in Michigan .........................................
3.1.2 Field observations in Thailand ..........................................
3.2 Satellite data ........................................................................
3.2.1 Image pre-processing and radiometric correction ....................
3.2.2 Water-only image .........................................................
3.2.3 Signature acquisition .....................................................
3.2.4 Multiple regression analysis .............................................
3.2.5 Lake water clarity/ trophic condition map ............................
References ..........................................................................

vi

27

27

28

28

32

33

34

36

37

39

40

45

45

45

47

49

50

53

55

57

61

63

CHAPTER IV: MICHIGAN LAKES WATER MONITORING .................. 65

4.1 Introduction ........................................................................ 65
4.2 Results .............................................................................. 66
4.2.1 Comparisons between DN and radiance images ..................... 66

4.2.2 Comparisons between Landsat-5 TM and Landsat-7 ETM+ ....... 76

4.2.3 Analysis of ground observation data ................................... 86
4.2.3.1 Geographic distribution of the datasets ..................... 86

4.2.3.2 Comparisons of Secchi disc transparency distribution

and temporal distribution of the datasets ................... 88

4.2.4 Comparison of ground observation data between MDEQ dataset

and CLMP dataset ........................................................ 93

4.3 Conclusion .......................................................................... 96
4.3.1 Main hypothesis conclusion ............................................. 96

4.3.2 Sub hypothesis conclusions ............................................. 97

4.3.3 Factors affecting predictability of the models ........................ 99
References .......................................................................... 100

CHAPTER V: THE BIGHT OF BANGKOK WATER MONITORING 103

5.1 Introduction ........................................................................ 103
5.2 Results .............................................................................. 104
5.2.1 Comparisons between DN and radiance images. . . . . 104
5.2.2 Comparisons between Landsat-5 TM and Landsat-7 ETM+ ....... 113

5.2.3 Seasonal changes of Secchi disc transparency, Chlorophyll a,

and suspended sediment ................................................. 118

5.2.4 Correlation between Secchi disc transparency, Chlorophyll a,

and suspended sediment ................................................. 124

vii

5.2.5 Changes in water clarity/ trOphic status ............................... 125

5.2.6 Integrated TSI value ...................................................... 127
5.3 Conclusion .......................................................................... 128
5.3.1 Main hypothesis conclusion ............................................. 128
5.3.2 Sub hypothesis conclusions ............................................. 129
5.3.3 General conclusions ...................................................... 1 30
References .......................................................................... 1 32

CHAPTER VI: RESEARCH CHALLENGES AND FUTURE NEEDS. . . . . 133
APPENDICES ........................................................................... 136

Appendix A - Detailed information of 123 ground observation data from

CLMP and MDEQ ................................................................ 137
Appendix B — Pictures of ETM+ and TM images used in the study ....... 140
Appendix C - Data used in the regression model .............................. 141

viii

Figure 2.1

Figure 2.2

Figure 3.1

Figure 3.2
Figure 3.3
Figure 3.4
Figure 3.5
Figure 3.6
Figure 3.7
Figure 3.8

Figure 3.9

Figure 3.10
Figure 3.11

Figure 3.12

Figure 4.1

Figure 4.2

Figure 4.3

LIST OF FIGURES

F ive-year rotational cycles of field observations under LWQA

monitoring program (Bednarz et al. 2001) ............................
Lakes over 101,171 m2 (25 acres) in size (Bednarz et al. 2001). ..

Inland lake field sample locations in Landsat path 2] row 29, 30,
and 31 ......................................................................

The Bight of Bangkok, Thailand .......................................
Flow diagram of conversion process from DN to radiance. . . . .
Flow diagram of water-only image .....................................
Example of water-only image ..........................................
Example of unsupervised classification of the water-only image.
Example of A01 selection ...............................................
Example statistical output ...............................................

Flow diagram of conversion from DN/radiance image to

regression model image .......................... ' ........................
Flow diagram of conversion fiom regression model image to TSI
Example of pixel-level water clarity/ trophic condition map .......
Example of lake-level water clarity/ trophic condition map ........

Statistical analysis of 85 lakes using data from Landsat-7 ETM+

digital number image .....................................................

Statistical analysis of 85 lakes using data from Landsat-7 ETM+

radiance image ............................................................

Statistical analysis of 33 lakes using data from Landsat-5 TM

digital number image row 29 and 30 ...................................

ix

30

31

46

48

52

53

54

55

57

59

60

60

62

62

69

70

72

Figure 4.4

Figure 4.5

Figure 4.6

Figure 4.7

Figure 4.8

Figure 4.9

Figure 4.10

Figure 4.11

Figure 4.12

Figure 4.13

Figure 4.14

Figure 4.15

Statistical analysis of 33 lakes using data from Landsat-5 TM

radiance image row 29 and 30 ..........................................

Statistical analysis of 12 lakes using data from Landsat-5 TM

row 31 digital number image ...........................................

Statistical analysis of 12 lakes using data from Landsat-5 TM

row 31 radiance image ...................................................

Statistical analysis of 43 sub-sampled lakes using data from

Landsat-7 ETM+ radiance image ......................................

Statistical analysis of 43 sub-sampled lakes using data from

Landsat-5 TM radiance image ..........................................

Statistical analysis for 32 lakes using data from Landsat-7 ETM+

radiance image row 29 and 30 ..........................................

Statistical analysis for 32 lakes using data from Landsat-5 TM

radiance image row 29 and 30 ..........................................

Statistical analysis for 11 lakes using data from Landsat-7 ETM+

radiance image row 31 ...................................................

Statistical analysis for 11 lakes using data from Landsat-5 TM

radiance image row 31 ...................................................

Geographic location of 43 lakes sampled in both sensors and 41
lakes sampled only in Landsat-7 full dataset 84 lakes ...............

Distribution of Michigan lake samples by date of acquisition and
Secchi disc transparency distribution of 84 lakes used in ETM+

model .......................................................................

Distribution of Michigan lake samples by date of acquisition and
Secchi disc transparency distribution of 45 lakes used in TM

model .......................................................................

73

74

75

78

79

81

82

83

84

87

89

90

Figure 4.16

Figure 4.17

Figure 4.18

Figure 4.19

Figure 5.1

Figure 5.2

Figure 5.3

Figure 5.4

Figure 5.5

Figure 5.6

Figure 5.7
Figure 5.8
Figure 5.9

Figure 5.10

Distribution of Michigan lake samples by date of acquisition and
Secchi disc transparency distribution of sub-sampled 43 lakes

used in ETM+ model .....................................................

Distribution of Michigan lake samples by date of acquisition and
Secchi disc transparency distribution of sub-sampled 43 lakes
used in TM model ........................................................

Statistical analysis of 15 MDEQ lakes from ETM+ radiance

image .......................................................................

Statistical analysis of 70 CLMP lakes from ETM+ radiance

image .......................................................................

Statistical analysis of 30 samples using data from Landsat-7
ETM+ 09/02/01 digital number image .................................

Statistical analysis of 30 samples using data from Landsat-7
ETM+ 09/02/01 radiance image ........................................

Statistical analysis of 13 samples using data from Landsat-5 TM
07/24/01 digital number image .........................................

Statistical analysis of 13 samples using data from Landsat-5 TM
07/24/01 radiance image ................................................

Statistical analysis of 30 samples using data from Landsat-5 TM
08/25/01 digital number image .........................................

Statistical analysis of 30 samples using data from Landsat-5 TM
08/25/01 radiance image ................................................

TSI map from Landsat-5 TM 07/24/01 ................................
TSI map from Landsat-5 TM 08/25/01 ................................
TSI map from Landsat-7 ETM+ 09/02/01 ............................

TSI map from Landsat-7 ETM+ 01/08/02 ............................

xi

91

92

94

95

105

106

108

109

110

111

114

115

116

117

Figure 5.11
Figure 5.12
Figure 5.13
Figure 5.14
Figure 5.15
Figure 5.16
Figure 5.17

FigureB -1

Example of land use near the Bight of Bangkok ..................... l 18

Ground sample sites ...................................................... 119
Seasonal changes of Secchi disc transparency in the study area... 121
Seasonal changes of Chlorophyll a in the study area ................ 122
Seasonal changes of suspended sediments in the study area ....... 123
TSI change maps ......................................................... 126
Integrated TSI trend ...................................................... 128
Pictures of ETM+ and TM images used in the study ................ 140

Images in this thesis are presented in color.

xii

Table 2.1

Table 2.2

Table 2.3

Table 2.4

Table 3.1

Table 3.2

Table 3.3

Table 4.1

Table 4.2

Table 4.3

Table 5.1

Table 5.2

Table 5.3

Table A - 1

Table C - 1

Table C - 2

LIST OF TABLES

Summary of the limits to define trophic states of waters in

traditional system (Zilioli and Brivio, 1997) ............................

Completed trophic state index and associated parameters (Carlson,

1977) ..........................................................................

Empirically derived slope and intercept values enabling

radiometric conversion of Landsat-7 ETM+ DN values to Landsat-
5 TM DN values (Vogelmann et al. 2001) ..............................
Calibration coefficients (gain and offset) of Landsat-5 TM data. .

Information for 123 ground observation data from CLMP and
MDEQ data sources ........................................................

Information of satellite data used in the study ..........................
Calibration coefficients of Landsat-5 TM and Landsat-7 ETM+. . ..
Preliminary efficacy test between digital number and radiance. .
Summary of the multiple regression results for Michigan lakes. .
Summary of the multiple regression results from sub-sample lakes.

Summary of the multiple regression results for the Bight of
Bangkok .....................................................................

Correlations among Secchi disc transparency, Chlorophyll a, and

suspended sediment .........................................................
Integrated TSI value ........................................................

Detail information of 123 ground observation data from CLMP
and MDEQ data sources ...................................................

Data of 84 lakes used in Landsat-7 ETM+ digital number model...

Data of 84 lakes used in Landsat-7 ETM+ radiance model ...........

xiii

21

23

36

39

47

49

52

67

76

85

113

124

127

137

141

143

Table C - 3

Table C - 4

Table C - 5

Table C - 6

Table C - 7

Table C - 8

Table C - 9

Table C-lO

Table C-ll

Table C-12

Data of 33 lakes used in Landsat-5 TM row 29 and 30 digital

number model ...............................................................

Data of 33 lakes used in Landsat-5 TM row 29 and 30 radiance

model .........................................................................

Data of 12 lakes used in Landsat-5 TM row 31 digital number

model .........................................................................
Data of 12 lakes used in Landsat-5 TM row 31 radiance model. .

Data of 43 lakes sampled in both sensors from ETM+ radiance

model .........................................................................
Data of 43 lakes sampled in both sensors from TM radiance model

Data of 32 lakes sampled in both sensors from ETM+ row 29 and

30 radiance model ..........................................................

Data of 32 lakes sampled in both sensors from TM row 29 and 30

radiance model ..............................................................

Data of 11 lakes sampled in both sensors from ETM+ row 31

radiance model ..............................................................

Data of 11 lakes sampled in both sensors from TM row 31

radiance model ..............................................................

xiv

145

146

147

147

148

149

150

151

152

152

CHAPTER I

INTRODUCTION

1.1 Background

Human activities, including disturbances in landscapes, impact the quality
of water in lakes and bays in many ways. Eutrophication is one of the most
pervasive problems affecting lakes; the primary cause is excessive nutrient export
in runoff from urban and agricultural landscapes (Mueller and Helsel, 1996).

Lakes in the Upper Midwest region of the US. typically eutrophy due to
runoff from farms, urban areas, industry, and construction sites (Weier, 2002).
Storm water runoff often carries chemicals and sediments from these locations
into nearby lakes. Sediments can increase turbidity, while nutrients in the
sediments, particularly phosphorous, can cause undesired algal growth (Weier,
2002)

The highest phosphorus concentrations in surface waters are usually found
close to agricultural and urban areas (Mueller and Helsel, 1996). These high
concentrations increase concerns about decreased oxygen in the water, toxicity to

fish, and accelerated eutrophication (Mueller and Helsel, 1996).

Nutrients that cause eutrophication

Nutrients are chemical elements that are essential to plant and animal
nutrition (Mueller and Helsel, 1996). Important nutrients that are often found to
be the limiting factors to aquatic species are nitrogen (N) and phosphorus (P).
However, these nutrients are considered to be contaminants in water if
concentrations are high. N and P can have several forms, both soluble and non-
soluble in water. Chemical and biological processes can change the forms of N
and P, and can transfer them to or from water, soil, biological organisms, and the
atmosphere (Mueller and Helsel, 1996). Nutrient concentrations in water for
nitrogen and phosphorus are generally reported in milligrams per liter (mg/ L).

Phosphate, the compound of phosphorus with oxygen and hydrogen, is the
only major form of dissolved phosphorus in natural water. Phosphate, compared
to nitrate, is not very mobile in ground water. It is only moderately soluble and
tends to remain attached to soil particles. However, considerable amounts of
“particulate” phosphate can be transported to streams, lakes, and oceans through
erosion (Mueller and Helsel, 1996).

Eutrophication is a natural process that results from accumulation of
excessive nutrients in lakes or other water bodies. Natural eutrophication is the
process by which lakes gradually age and become more productive over thousands
of years (Oglesby, 1977). Human activity, however, has greatly accelerated this
process in thousands of lakes across the globe. Phosphate and Nitrate are

frequently found to be the limiting factors of algae growth. In freshwater,

phosphorus is often the nutrient responsible for accelerated eutrophication. On the
other hand, eutrophication in estuaries and coastal marine water are more
commonly related to high nitrate concentrations (Mueller and Helsel, 1996).

Algae feeding on these nutrients grow into unsightly scums on the water
surface, decreasing recreational value and clogging water-intake pipes. Moreover,
dead algae can result in a variety of water quality problems, including noxious
tastes and odors, clogged pipelines, restricted recreation, and oxygen deficiency
that occasionally results in fishkills (Mueller and Helsel, 1996).

Humans add excessive amounts of plant nutrients (primarily phosphorus,
nitrogen, and carbon) to streams and lakes in various ways. Runoff from
agricultural fields, field lots, urban lawns, and golf courses is one source of these
nutrients. Untreated, or partially-treated, domestic sewage is another major
source. Sewage is an important source of phosphorus, particularly through
detergents containing large amounts of phosphates. Phosphates can be powerful
stimulants to algal grth when they are flushed into lakes. In 1992, the EPA
reported that accelerated eutrophication was one of the leading problems facing
the Nation’s lakes and reservoirs (Mueller and Helsel, 1996).

The three variables most commonly used to indicate trophic state are total
phosphorus (TP), chlorophyll a (Chl-a), and Secchi disc transparency (SDT)
(Kloiber, 2002). Lake management agencies and organizations frequently use the
measurements of these variables, together with water quality indices, i.e. Carlson’s

trophic state indices (Kloiber et a1. 2002). Among the three variables, SDT has

been widely incorporated into volunteer monitoring programs and is used
frequently to identify trends in lake trophic conditions (Heiskary et a1. 1994).

In the Midwest, poorly drained soils limit water movement to the lower soil
layers and artificial drainage provides a quick path for nutrient-rich runoff to reach

streams and lakes (Mueller and Helsel, 1996).

Environmental importance of water quality monitoring

Lakes are valuable environmental, economic and recreational resources in
this region. With a unique geographical location, the Upper Great Lakes region—
Michigan, Minnesota, and Wisconsin house more than 30,000 inland lakes. Lakes
are considered a high value natural resource in many aspects, i.e. recreation,
wildlife habitats, ecological studies, tourism dependent economy, and thus they
are important to monitor and maintain quality.

Algal blooms, caused by nutrient contamination of water, have been of
national concerns for several decades (Carlson, 1977; Oglesby, 1977; Mueller and
Helsel, 1996). In affected lakes, substances in the water, algae, bacteria, and
chemicals, can affect the health of people and animals who fish and swim in the
lake. Moreover, Upper Great Lakes communities that rely on tourist dollars could
be dramatically affected by algae-filled lakes that turn away summer vacationers
(Weier, 2002).

Routine monitoring of lakes and streams is essential in providing

information on expected ranges in water quality, temporal trends, regional

differences, and relationships between the landscape and water quality. This
information can be used in water quality criteria development and decision
planning for many land and state agencies (Kloiber et a1. 2000). Unfortunately,
monitoring water quality for the 30,000 plus lakes in the Upper Great Lakes region
has never been possible due to the deficiency of budgets and staffs.

The three water quality parameters most commonly used to indicate trophic
state are total phosphorus (TP), chlorophyll a (Chl-a), and Secchi disc
transparency (SDT). These parameters, along with various lake productivity
indices such as Carlson’s Trophic State Indices (TSI), are widely used by
limnologists, and lake management agencies (Kloiber et a1. 2002). Among the
three variables, SDT has been most widely used in volunteer monitoring programs
and is used frequently to identify trends in lake conditions due to its simple and
inexpensive measurements (Heiskary et al. 1994).

Although Secchi depths provide an excellent monitoring system, taking
such measurements is labor intensive. While Secchi disk measurements work well
for small area samples or small numbers of lakes, they are not practical for
monitoring the over 11,000 lakes in Michigan or large areas such as the Bight of
Bangkok.

Thailand, with roughly of 514,000 km2 in area has a population of 67
million people. Forty seven percent of the population depend upon agricultural
activities. Annual algal blooms in the Gulf of Thailand appear due to nutrient

runoff from agricultural and adjacent aquacultural areas.

Algal blooms and water quality are of great concern among people and
monitoring agencies. However, comprehensive monitoring programs require long
term funding commitments. For Thailand, a country with long coastline and many
rivers, an effective water quality monitoring program is absolutely crucial. Yet,
conventional survey techniques in the open sea around Thailand have always been
difficult due to cost and labor shortages.

Globally, the importance of water quality monitoring is well known, but the
associated costs of sampling and analysis have been prohibitive. Scientists have
recognized the potential of satellite technologies to provide a cost effective means
of monitoring lakes, coastal waters and the adjacent lands over the long term.

Satellite data can provide synoptic views of physical features (i.e. currents,
sea ice), water quality parameters (i.e. phytoplankton, turbidity), and watershed
characteristics (i.e. land classification, mangrove detection) (Johnson, 2001).
Estimation of water quality of over thousands of lakes in Michigan or the entire
Eight of Bangkok can be achieved for a single date by using satellite imagery.
The approach requires ground observation data of only a small percentage of
representative water bodies from the region (Kloiber et al. 2002) or a small
number of sea-truth data samples in the Bight of Bangkok.

Regression models developed for city-wide water clarity/trophic state
measures using a single image in previous studies have resulted in high r2 values
(0.80 — 0.90, p < 0.05) (Lathrop and Lillesand, 1986; Lavery et al. 1993; Giardino

et al. 2001). These strong relationships suggest a potentially useful SDT

measurement using satellite remote sensing techniques in a larger number of lakes
across a regional scale such as the state of Michigan. However, multiple image
based studies of lakes have not yet been performed.

Many studies in the past used Landsat-5 Thematic Mapper (TM) to study
the relationship between satellite signatures and ground observed SDT, however,
few researchers have used Landsat-7 Enhanced Thematic Mapper Plus (ETM+).
Landsat—7 ETM+ is the latest in a series of Landsat satellites that have provided a
continuous set of calibrated Earth Science data to national and international users
since 1972. The system performance and data quality are outstanding, particularly
with respect to radiometry, image geometry and geographic registration. Further,
the Landsat-7 satellite is planned for a lifespan through the year 2005 whereas
Landsat-5 TM is being phased out. Lake monitoring methodologies using

Landsat-7 need to be explored

1.2 Comparison of ground-based and satellite monitoring approaches
Ideally, a water monitoring program would collect continuous data for
every water body to provide complete information across spatial and temporal
scales. Such programs are not only time-consuming but also cost-prohibitive.
Thus, current programs often fail to represent the variability of the distributions of
parameters in the water (Kloiber et al. 2000). On the other hand, remote sensing

technologies can provide a synoptic view and simultaneously assess feasible

alternatives for water quality monitoring, mapping and management (Khorram,
1991)

Ground-based monitoring collects samples from specific locations on a
water body. Researchers must rely on interpolations between sample points or
assume that they represent the entire water body (Khorram, 1991). This
assumption may be valid for small lakes or tiny bays but is doubtful for large lakes
and bays with complex morphometry (Kloiber et al. 2000). Satellite-based
monitoring can provide a broader perspective because it has the capability to
provide water quality information with shore-to-shore coverage, leading to a
broader perspective (Kloiber et a1. 2000).

Monitoring programs that cover wider geographic areas (i.e. state or
national-scale programs) are often spread across several agencies. This leads to
inconsistencies in data collection and quality due to changes in staff and budget. It
is difficult to maintain uniform sampling and analytical methods, especially over
long time periods (Kloiber et al. 2000). Collection of remote sensing data is
potentially more consistent than the ground based data collection.

There are many potential uses for satellite data, including assessments of
regional lake status and trends (geographic and temporal), and evaluation of
relationships between landscape alteration and water quality.

Although satellite imagery has several advantages as a tool for assessing
water quality, it has not been widely adopted by aquatic scientists. A possible

reason for the limited use of this tool is the lack of familiarity with the technology

among these scientists. Fortunately, present day desktop computers and several
sophisticated software packages (i.e. ERDAS Imagine and ENVI) are

comparatively inexpensive and easy to use.

1.3 Study area

The study area is located in two sites: inland lakes in Michigan and the
Bight of Bangkok, Thailand. I will test the accuracy and the applicability of the
satellite-based approach being developed for a city-wide inland lake monitoring to
the statewide level monitoring in Michigan. Then apply the same methodology to
the Bight of Bangkok in order to determine the applicability of the method in a

tropical marine environment.

1.4 Hypothesis

The purpose of this research is to test the accuracy and the applicability of
the satellite-based water clarity/ trophic condition monitoring method as an
alternative for water quality monitoring.

The specific hypothesis is water clarity/ trophic condition in both Michigan
inland lakes and the Bight of Bangkok can be assessed using satellite remote

sensing in a common methodology. ‘/

1.5 Sub-hypothesis

1. Using standard radiance values from Landsat imagery has a potential to
produce more accurate results than using DN values.

2. The method can be used with both Landsat-5 TM and Landsat-7 ETM+
images.

3. The method developed in Upper Great lakes region can be used in Bight

of Bangkok without significant difference in accuracy.

1.6 Benefits of the research

The major advantage of the method is to provide a cost effective way to
obtain trophic condition of water at a statewide scale, or whole Bight monitoring,
that would otherwise be expensive using conventional sampling and measurement

methods.

References

Carlson, R. E. (1977). A trophic state index for lakes. Limnology and
oceanography, 22 (2), 361 — 368.

Giardino, C., Pepe, M., Brivio, P. A., Ghezzi, P., and Zilioli, E. (2001). Detecting
chlorophyll, secchi disk depth and surface temperature in a sub-alpine lake
using Landsat imagery. The Science of the Total Environment, 268, 19 — 29.

Heiskary, S., Lindbloom, J ., and Wilson, C. B. (1994). Detecting water quality
trends with citizen volunteer data. Journal of Lake and Reservoir

Management, 9(1), 4-9.

10

Johnson, D. “Coastal Remote Sensing Science”. CSCOR/ Coastal Ocean
Program. Dec 2001. 4 May 2002,
<http://www.cop.noagov/FacLSheets/CRS.htm>

Khorram, S. and Cheshire H. M. (1991). Water quality mapping of Augusta Bay,
Italy from Landsat-TM data. International Journal of Remote Sensing, 12(4),
803-808.

Kloiber, S. M., Brezonik, P. L., Olmanson, L. G., and Bauer, M. E. (2002). A
procedure for regional lake water clarity assessment using Landsat
mutispectral data. Remote Sensing of Environment, 82(1), 38-47.

Kloiber, S.M., Anderle, T.H., Brezonik, P.L., Olmanson, L., Bauer, ME, and
Brown, DA. (2000). Trophic state assessment of lakes in the Twin Cities
(Minnesota, USA) region by satellite imagery. Archive Hydrobiologie Special
Issues Advances in Limnology, 55,137-151.

Lathrop, R. G. and Lillesand, T. M. (1986). Utility of Thematic Mapper data to
assess water quality. Photogrammetric Engineering and Remote Sensing, 52,
671-680.

Lavery, P., Pattiaratchi, C., Wyllie, A., and Hick, P. (1993). Water quality
monitoring in estuarine waters using the Landsat Thematic Mapper. Remote
Sensing of Environment, 46, 268-280.

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.

Oglesby, R. T. (1977). Relationships of fish yield to lake phytoplankton standing
crop, production, and morphedaphic factors. Journal of Fisheries Research
Board of Canada, 34, 2271-2279.

Weier, J. “Testing the Waters”. Earth Observatory, NASA. Mar 2002. 13 May
2002,
<http://earthobservatory.nasagov/Study/WaterQuality/water_qualitv2 .html>

11

CHAPTER II

LITERATURE REVIEW

In order to provide sufficient background knowledge on water monitoring
and satellite technology, I reviewed literature from six, topics: landuse, nutrients
and eutrophication, trophic state index, field observation data, satellite data, and

radiometric and atmospheric effects.

2.1 Landuse

Activities on the land can have a considerable effect on water bodies and
streams. These activities are usually grouped into categories of “landuse,” i.e.
forest land, agricultural land, urban land, and wetland, etc. Landuse classification
is used to help select sample sites that reflect the variety of Land uses found within
the watershed and link human activity to changes in water quality (Mueller and
Helsel, 1996).

Agricultural land and urban land are the landuse categories that cause the
greatest impacts on water quality (Mueller and Helsel, 1996). Runoff from
intensively fertilized farmlands and urban sewage discharges have a significant
potential to accelerate eutrophication of receiving water bodies.

Recent advances in hydrologic modeling and GIS technologies provide the

tools to identify where nutrient problems are likely to occur, using soil drainage

12

and geological information (Mueller and Helsel, 1996). By definition, soil
drainage is the ability of soil to transmit water (Mueller and Helsel, 1996).
Movement of water from the land surface to streams and bodies of water is
affected, in part, by this factor. In tile-drained areas, nutrient concentrations are
usually higher in streams than in ground water. In contrast, nutrients in naturally
well-drained areas are often higher in ground water than in streams. In nature,
nutrient concentration in streams is generally the main cause of eutrophication in
receiving water bodies. Geology is identified by the type of geologic formations
through which ground water passes and can affect how easily water and nutrients

move downward through (Mueller and Helsel, 1996).

2.1.1 Nature of land use in Michigan

Michigan’s unique geographical location has provided its citizens with rich
freshwater resources including over 11,000 inland lakes. In addition to ecological
value, lakes also provide tremendous aesthetic and recreational value for people in
Michigan.

Located in the Midwestern Corn Belt, Michigan has significant agricultural
land. Most farmers apply nutrients to maximize their crop yields, but excessive
nutrients carried by water runoff become pollutants downstream. The tile-drained
soil of Michigan enhances the severity of the nutrient pollution problem by
blocking nutrients from moving vertically through the ground. As a result,

considerable amounts of fertilizers accumulate into nearby lakes.

13

2.1.1.1 History of land use in Michigan

During the past 100 or 125 years the natural landscape of Michigan has
been altered by human actions (Veatch, 1953). It is true that humans cannot
change the major elements of the environment, i.e. the climate, the land forms, the
composition of the Glacial drift covering, or the bedrock, yet they can make
alterations in the cover of vegetation, the fauna, the soil, and the waters. The
effects of man’s activities after 100 years are not geographically uniform
throughout the state, partly because of the differences in regional climate,
topography, soil and other resources. Those variable impacts result in part from
the spatial distribution of the population in Michigan. Historically, approximately
90 percent of the people in the state lived in the southern half of the region with
only 5 percent in the entire Upper Peninsula (Veatch, 1953).

Out of the approximately 149,734 km2 of the land area of Michigan, 72,843
km2 (48.6 percent) were classified by the US. Census (1953) as farmland and as
much as 8,094 km2 appropriated for cities, industrial sites, highways, and rural

homes (Veatch, 1953).

2.1.1.2 Soils and land

Roughly 47,348 km2 or 31.6 percent of Michigan is comprised of poorly
drained soil (Veatch, 1941). The terms “clay soil,” or “clay land,” are lands that
include some actual clay in the surface texture, but the terms more often refer to

land underlain by clay at a depth of a few inches to approximately one foot. This

14

broad group of soil constitutes the greater part of the highly productive and
durable agricultural land in the State (Veatch, 1953). In addition, an estimated 70
— 75 percent of the original wet or shrub land underlain by clay had been cleared

and drained for some sort of agricultural use by 1941 (Veatch, 1941).

2.1.1.3 Importance of Michigan agriculture

Agriculture, Michigan’s second largest industry, contributed more than 37
billion dollars annually to the state economy in 1994 (Michigan Farmland and
Agriculture Development Task Force, 1994). Michigan has many agricultural
advantages, i.e. an abundance of inland fresh water, fertile soils and a mild climate
moderated by the Great Lakes. Because of its unique micro-climates, the state of
Michigan was ranked second in the nation with agricultural diversity (Michigan
Farmland and Agriculture Development Task Force, 1994). The State has
produced more than 100 different food and fiber products and has been the leading
producer of tart cherries, blueberries, flowers and edible beans such as navy, fruit
and black turtle beans. This amazing diversity has given the State agricultural
economy great stability (Michigan Farmland and Agriculture Development Task

Force, 1994).

2.1.1.4 Changes in land use
Current trends of land use in Michigan indicate decreases in agricultural

land and pasture and an increase in urban build-up. Veatch (1953) reported the

15

area of farm land as approximately 72,843 km2 in 1953. The area had decreased to
50,001 km2 in 1982 and 49,446 km2 in 1987 (National Resources Inventory,1987).
In 1992, there were approximately 40,873 km2 of farmland in Michigan with
32,780 km2 of tillable cropland. The fastest decline in farmland occurred from
1954 to 1974, with approximately of 25,900 km2 of farmland converted to other
uses (Michigan Farmland and Agriculture Development Task Force, 1994). This
decrease in agricultural acreage occurred in concert with increases in urban and
built-up land (National Resources Inventory, 1987).

Agriculture was directly impacted by recent trends in land use patterns.
Michigan has not experienced significant increases in population during the last
15-20 years, although a dramatic shift in the location of residential development
had occurred (Michigan Farmland and Agriculture Development Task Force,
1994). As a result, the amount of land used for residential housing has continued
to increase at a rapid rate, placing additional pressures on agricultural land. When
suburbs expand, they often invade lands previously planed in agricultural corps.
Yesterday’s comfields have changed into today’s shopping centers (Michigan

Farmland and Agriculture Development Task Force, 1994).
2.1.1.5 Impacts from land use change

As the population density increases in agricultural areas, the impact upon

agricultural operations can increase exponentially. The impact of increasing

16

residential development is felt not only in the loss of farmland, but also impacts
existing farm operations.

Increases in crop production can be obtained by intensification of
cultivation and improvement of farm management on existing acreage, without the
addition of new land. Where the natural fertility is not favorable, increased use of

commercial fertilizers may help to boost production (Veatch, 1953).

2.1.2 Nature of landuse in Thailand

As the human population of Thailand grows and expands, more territory is
developed and modified for human use. Coastal regions are often the prime
targets for food production, industrial and residential development.

Bangkok city, the capitol of Thailand, has grown steadily over the past 20
years. The 1980 population in Bangkok was approximately 4.5 million people and
is currently estimated to be 7 million people (National Statistical Office Thailand,
2002). With population growth, the non-point nutrient pollution sources from
agricultural units dominated most parts of the country and untreated residential
wastewater has increased tremendously. All these components are released into
the major rivers; the Chao Praya, Bang Pakong and Tha Chin rivers, and

eventually flow into the Gulf of Thailand.

17

The Gulf of Thailand

The Gulf of Thailand is a semi-enclosed sea, as defined by the United
Nation’s 1982 Law of the Sea (International Cooperative Study of the Gulf of
Thailand, 2001). The Gulf is located in Southeast Asia, immediately to the west
of the South China Sea. It is bordered by several nations, Thailand, Cambodia,
Vietnam, and Malaysia, which have each historically profited from the Gulfs
wealth of living and mineral resources (International Cooperative Study of the
Gulf of Thailand, 2001).

The Chao Phraya, Bang Pakong, and Tha Chin rivers enter the gulf near
its head. There are also several rivers and streams that flow along the coast.
Through these sources, substantial amounts of organic and inorganic pollutants
enter the gulf.

Land-based pollution near the gulf of Thailand is derived mainly from
agricultural and industrial areas, but also includes residential units (International
Cooperative Study of the Gulf of Thailand, 2001). Pollution can be discharged
directly into the sea, or flow into the coastal water through rivers. There are two
major sources of pollutants, point sources and non-point sources.

Point sources of pollution can be traced to a specific location, i.e. industrial
and sewage treatment plants. Industrial development along the Gulf of Thailand
has increased over the past ten years. Industrial estates found adjacent to the
coastal areas of the Bight of Bangkok (upper part of the gulf of Thailand) include

Map Ta Phut, Laem Chabang, and Bang Poo. Map Ta Phut houses petrochemical

18

plants and a fertilizer complex. Agro-based industries are found in Laem Chabang
and Bang Poo. Industrial wastes are discharged from these plants along the river
and are often deposited in the Gulf. Point sources account for only a fraction of
the land-based sources of pollution affecting coastal and marine environments of
the Gulf of Thailand (International Cooperative Study of the Gulf of Thailand,
2001)

Non-point pollution sources are the sources that are more difficult to
identify geographically, as the sources can be located far from coastal waters, i.e.
agricultural and urban lands upstream. Thailand, with 67 percent of its population
working in agriculture (National Statistical Office Thailand, 2002), is one of the
world’s major food producers. Most of the agricultural and aquaculture activities
are located where rivers or streams can be accessed. Heavy use of farmland
fertilizers and chemicals usually results in water pollution. Chao Phraya, Bang
Pakong and Tha Chin rivers pass through vast agriculture lands, and carry the
majority of pollution from non-point sources into the Gulf of Thailand.

Since the world’s conventional fish catch from oceans and lakes continues
to decline due to overfishing and environmental damage, aquaculture has a
significant potential augment native sources in meeting human needs (World
Resource Institute, 2002). Aquatic animals grown in aqua-farms are used
basically for commercial purposes. Serious concern over the environmental
impacts of aqua-farming remains, especially in intensive shrimp production and

other high-value species (World Resource Institute, 2002). Aquaculture uses

19

substantial amounts of food and chemical nutrients to enhance phytoplankton
growth, which is then fed upon by farm animals. Discharges from the farms are
heavily contaminated by algae, chemical agents, uneaten fish feed, and excretory
wastes, and when flushed into coastal or river waters can lead to oxygen depletion
and harmful algal blooms. In Thailand alone, shrimp ponds discharge
approximately 1.3 billion cubic meters of effluent into coastal waters each year
(World Resource Institute, 2002).

Aqua-farms in Thailand are bounded in earth ponds, blocked coastal
waters, or cages in open ocean water. Cage and net-pen aquaculture systems have
the most potential for causing environmental degradation (International
Cooperative Study of the Gulf of Thailand, 2001). Land based farms are usually
located along coastlines or in estuaries, especially close to mangroves because of
the naturally high levels of nutrients necessary for aquaculture and the diluted salt
water favorable for young aquatic species (International Cooperative Study of the
Gulf of Thailand, 2001).

Nutrient-pollution from land based aqua farms enters the Gulf directly, via
runoff from storms that overflow farm ponds, or by normal daily or weekly
discharges from the adjacent farms (International Cooperative Study of the Gulf of
Thailand, 2001). Phytoplankton blooms stimulated by these high nutrient level

discharges occasionally cause fish stress or fish kills.

20

2.2 Trophic State Index (TSI)
The concept of trophic state is widely used for integrative descriptions of
lake water quality (Zilioli and Brivio, 1997). A summary of the traditional

classification systems used to denote status of water, with respect to the main

physical and bio-chemical indicators, is shown in Table 2.1.

Table 2.1 Summary of the limits to denote trophic status of waters in traditional

systems (Zilioli and Brivio, 1997).

 

 

 

 

Secchi disk Chlorophyll Phosphorus
Trophic status (m) (mg/m3) (mg/m3)
Mean Maximum Mean Maximum Mean
Ultra-oligotrophic 2 12 2 6 s 1 S 2.5 s 4
OligotrOPhic 2 6 2 3 s 2.5 s 8 s 10
Mesotrophic 6 — 3 3 — 1.5 2.5 — 8 8 — 25 10 — 35
Eutrophic 3 — 1.5 1.5 — 0.7 8 — 25 25 — 75 35 - 100
HYPeT'eWOPhiC s 1.5 s 0.7 2 25 2 75 2 100

 

 

 

 

 

 

 

This traditional trophic state description has been use for several decades,
worldwide. The index contains 5 categories, which is generally not sufficient for
describing trophic status in detail. Carlson’s Trophic State Index (TSI) was
developed in 1977 to facilitate communication with the public about the status of

lakes (Carlson, 1977). A lake may be considered oligotrophic by one criterion and

21

eutrophic by another; this problem is sometimes rectified by classifying lakes that
show characteristics of both oligotrophy and eutrophy as mesotrophic (Carlson,
1977)

Several indices, single or multiple parameter, were proposed to describe the
trophic condition of lakes. A multiparameter index is limited by the fact that a
multitude must be measured. Single criterion indices, on the other hand, have
been the alternative to the trophic monitoring concept because it has the potential
to be both explicit and sensitive to change (Carlson, 1977).

The actual algal biomass values are not appropriate to use in the index
because the term “biomass” is not well defined and is usually estimated by one or
more parameters such as dry or wet weight, cell volume, particulate carbon,
chlorophyll, or SDT (Carlson, 1977). SDT provides values that are easily
transformed into a convenient scale, and is one of the simplest and most frequently
used measures in aquatic studies. Another benefit is that its values are easy to
understand.

Carlson (1977) used the doubling of algal biomass as the boundary between
different trophic states, meaning that a new trophic state will be identified each
time the concentration of algal biomass doubles from some base value. Biomass
doubling, i.e. SDT values of 8 m, 4 m, 2 m, 1 m, etc., would be represented by
transforming Secchi disc values to the logarithm to the base 2 (Carlson, 1977)
(Table 2.2).

TSI = 10 (6 —1og2 SDT) (2 - 1)

22

Table 2.2 Completed trophic state index and associated parameters (Carlson,

 

 

1977)
TSI Secchi disk Surface Phosphorus Surface Chlorophyll
(m) (mg/m3) (mg/m3)

0 64 0.75 0.04
10 32 1.5 0.12
20 16 3 0.34
30 8 6 0.94
40 4 12 2.6
50 2 24 6.4
60 l 48 20
70 0.5 96 56
80 0.25 192 154
90 0.12 384 427
100 0.062 768 1,183

 

 

 

 

 

 

The values range from 0 to 100 rather than 0 to 10. Generally, the
theoretical limit is infinite, but the practical limit is 100 or 110 (transparency

values of 6.4 and 3.2 cm) (Carlson, 1977).

In addition to the TSI computed from SDT, the index can be computed

from Chlorophyll or total phosphorus. The computational forms of the equations

are

23

 

TSI (SDT) = 10(6 — 9%11) (2 — 2)
TSI (Chl) = 10(6 — 2'04 " (’Ig'”(a”)) (2 — 3)
Inf—8—
: _ _7_—E _
TSI (TP) 10(6 In 2 ) (2 4)

Since SDT measurement is extremely simple and usually provides a TSI
value similar to that obtained from chlorophyll, it is the most efficient and widely

used in limnological observations (Carlson, 1977).

2.2.1 Advantage of Carlson ’s T SI

Carlson’s TSI has several advantages over previous attempts at trophic
classification, these include:

1. The significant increase in the number of classes allows sensitivity
analysis for describing trophic changes. The numerical scale, as opposed to a
nominal scale, allows a large number of continuous lake classes rather than three
to five distinct ones. Also, the logical basis, doubling in algal biomass, used for
major trophic classes makes the trophic state classification more acceptable
theoretically.

2. .It provides the opportunity to compare trophic conditions between lakes
where different parameters were collected. Although the index is constructed

from a single parameter, the mathematical correlation between parameters allows

24

a selection of the best one for a given situation. Calculation of the index for
more than one parameter provides relationships between parameters.

3. Trohic state classification using SDT values alone can be collected at
little expense by nonscientists in public-participation programs.

Carlson (1977) noted in his study that the concept of a trophic state index is
not the same as a water quality index because the term “quality” implies a
subjective judgment. Eutrophic is often related with poor water quality, and this is
a major point of confusion with the existing terminology. Individual attitudes and
the intended use of water determine whether water quality is excellent or poor.
The definition of trophic state and its index should remain neutral to such
subjective judgments. The TSI can be a useful tool for aquatic management and a

scientific tool for standard productivity investigations (Carlson, 1977).

2.2.2 Potential errors of T S]

Although TSI has been used as an effective tool to describe trophic state,
there are some other important factors aside from algal turbidity. These non-
trophic state factors, such as humic color and soil-derived suspended sediment,
may affect SDT and give erroneous values (Carlson, 1977). Lorenzen (1980)
illustrated that using data for chlorophyll a and SDT from different lakes can
result in very misleading interpretations due to the non-chlorophyll light
absorption materials in water bodies. For that reason, Lorenzen (1980) suggested

that the value of non-chlorophyll light absorption should be obtained for any

25

particular lake under study before predicting the chlorophyll concentrations via

transparency (Lorenzen, 1980)

2.3 Field observation data

Monitoring programs worldwide measure water clarity using the Secchi
disc method (Bukata et al. 1988; Wallin and Hakanson, 1992; Lee et al. 1995).
Although SDT provides an excellent classification system for limnologists,
acquiring the measurements for necessary regional scale monitoring across a large
numbers of lakes is labor intensive. For this reason, the method is not practical for
monitoring the approximately 30,000 lakes in the Upper Mid-West region. This
problem was somewhat relieved by the citizen-volunteer lake monitoring
programs (CLMP). The quality of such data has been reported by several
researchers to have no statistical difference in averages of SDT compared to
professional measurements, and to be comparable to data collected by professional
monitors (Heiskary et al. 1994; Kerr et al. 1994; Obrecht et a1. 1998, Canfield et
al. 2002). Nonetheless, the current programs are still fragmented, incomplete, and
inadequate, often resulting in limited and discontinuous monitoring, especially on
a statewide scale.

Water clarity monitoring in an open sea is often limited by cost and labor.
Adequate monitoring along coastlines and in the open sea is a national concern.
Unfortunately, traditional monitoring programs have never been able to achieve

the necessary quality of a single date shore to shore program.

26

2.3.1 Factors aflecting lake color

Inland waters are visually complex due to the components of
phytoplankton, dissolved organic material (DOM), and suspended sediments.
Suspended sediment, from organic and inorganic materials, backseatters
electromagnetic radiation strongly. These floating particles, in moderate to high
concentrations, generally overwhelm the reflectance signal due to phytoplankton
pigment or DOM absorption (Bukata et al. 1985). Even though the estimation of
algal pigment concentrations, such as chlorophyll a, is problematic in turbid lake
waters, relatively broad-band sensors as the Landsat Multispectral Scanner (M88)
or Thematic Mapper (TM) were shown to be capable of estimating trophic status
of lakes using SDT measurement in the previous studies of Lathrop (1992) and

Kloiber et al. (2002).

2.3 .2 Factors aflecting ocean color

Ocean color, as with that of lakes, is determined by the interaction of
incident light with substances or particles present in the water. The most
significant elements in open sea are free-floating photosynthetic organisms, such
as phytoplankton, and inorganic particulates. Phytoplankton contain chlorophyll,
which absorbs light at blue and red wavelengths and reflects in the green.
Inorganic matter can also reflect and absorb light, which reduces the clarity, or
light transmission, of water. Substances dissolved in water can also affect its color

(Weier, 2002). In the open ocean, light reflection is mainly caused by particulates

27

suspended in the water, and absorption is primarily due to the chlorophyll

pigments present in phytoplankton.

2.3.3 Monitoring programs in Michigan

The joint monitoring program between Michigan Department of
Environmental Quality (MDEQ) Land and Water Management Division, MDEQ
Surface Water Quality Division, and the United States Geological Survey (USGS)
has formed a surface water quality investigator program referred to as the Lake
Water Quality Assessment (LWQA) Monitoring Program in Michigan. During
the 1990’s, funding for surface-water quality monitoring was greatly reduced
(Bednarz et al. 2001); however, the agencies later proposed the LWQA
Monitoring Program relying heavily on the goals outlined in the 1997 MDEQ
report entitled “A Strategic Environmental Quality Monitoring Program for
Michigan’s Surface Waters” and the needs of the State to report on inland lake
quality status and trends. The program consists of a lake water quality assessment
(LWQA) and a volunteer lake trophic status monitoring component, known as the

Cooperative Lakes Monitoring Program (CLMP).

2.3.3.1 The Lake Water Quality Assessment (LWQA) Monitoring Program
The goals of LWQA Monitoring Program are to assess the current status
and conditions of individual waters of the State and determine whether water

quality standards are being met, to measure temporal and spatial trends in the

28

quality of Michigan surface water, and to provide data to support MDEQ water-
quality protection programs and evaluate their effectiveness, as well as, detecting
new and emerging water-quality problems.

LWQA Monitoring Program has an objective to provide current data on
trophic status, water quality variability and trends for Michigan’s inland lakes
through field observations in the five-year rotational cycle (Figure 2.1) (Bednarz et
al. 2001).

The state of Michigan has approximately 11,000 lakes with nearly 3,500
over 25 acres (101,171 m2) in size (Figure 2.2). Although a goal of this plan is to
gain knowledge of the water quality conditions and trophic status of all of
Michigan’s lakes, cuts in financial and staff resources has forced a reduction in the
number of lakes considered for field survey and water-quality monitoring to insure
maximum utilization. Only lakes with public boat access are considered for
selection into the monitoring plan (Bednarz et al. 2001).

The initial targets for monitoring are 730 lakes with public access. These
lakes were historically listed as Michigan’s significant public lakes (Bednarz et al.
2001). Each year, approximately 50 — 70 randomly selected lakes will be
monitored at spring turnover and late summer stratification. All of the 730 lakes
will be monitored and assessed within 14 years. During the first 5-year basin
monitoring cycle, 40 percent of the lakes listed are expected to be visited. After

that, 35 percent during the second, and 25 percent during the third monitoring

29

cycle will be assessed. This scaling approach will reduce costly field data

collection efforts while maintaining quality (Bednarz et a1. 2001).

 

/ 15—YEAR TROPHIC STATUS MONITORING PROGRAM
Based on MI 305b 5-year reporting cycles

 

 

 

 

 

 

 

 

 

- r .4, a}; .
I LAKES _ ‘; 1 £5,
SAMPLE YEAR -, T7351 ‘ .
I 2001 . _- . .
i 2002 - ' ' ' 2.1.1.: ‘ ~ ‘1 ”A
I 2005 - =‘. ‘ :32: if

 

 

 

Figure 2.1 Five-year rotational cycles of field observations under LWQA

monitoring program (Bednarz et al. 2001).

30

 

Michigan Lakes
Larger than 25 Acres

.I,‘ ' .
1.. .' .
“A. s “ \ \ ‘
\ - .
L |~ . ’
.i i _. .. . .. .
\‘\'~_‘ “U. i» ( 7 I 1
M1 ‘ . . r 2
‘\£: I k I .
.,.I g ‘ .
~. 9

T0? . xi
1 \ k I
“'5 that, " "2 s
‘- ‘|;.‘- ‘i "v
.. a ,1

A—
(a
Z I
r r ,. . r J
5
. I \
1 1-
'1, u. 1
. _ . ’- -
a . a.
f w ‘ 4
: , _ ~ ~’
‘\ ‘~
, __ _ _ 3‘1 /.
I 4 ‘ ‘7.
~ . . .. 5:1. ’2
u- ‘ E l
.. . ..- _ .
\ a
, .
’ «.— ‘
Y, ' . .
v r ", -: v K-'.‘ I :3,
j ’ A ' .
t‘ ., r . I
'r r
—————————— L.---——-—-l— J

 

 

 

Figure 2.2 Lakes over 101,171 m2 (25 acres) in size (Bednarz et al. 2001).

31

Baseline WQA surveys will be conducted during spring turnover and late
summer stratification. The deepest basin of each lake will be monitored for
trophic state indicators, baseline water chemistry, and physical properties. For
lakes with distinct multiple deep basins, additional basins will be monitored as

appropriate (Bednarz et al. 2001).

2.3.3.2 The Cooperative Lakes Monitoring Program (CLMP)

The Cooperative Lakes Monitoring Program (CLMP) was developed by the
Department of Environmental Quality (DEQ), in partnership with the Michigan
Lake and Stream Associations, Inc. (ML&SA). The program was designed to be a
cost-effective method for citizens to monitor water quality and to document
changes in lake quality over time (Bednarz et al. 2001).

The CLMP provides volunteers with sampling methods, training,
workshops, technical support, quality control, and laboratory assistance to monitor
indicators of lake productivity. As human populations increase, and more people
use the lakes and surrounding watersheds, the potential for pollution problems
dramatically increases. Consistent information, including water quality data,
levels of use, etc., are important for determining water quality changes and trends
and for developing a management plan for protection.

Problems most commonly reported by lake residents, such as algal blooms
and excessive plant growth, are caused by water quality parameters that lead to

increased lake productivity. Limnologists have developed a variety of numerical

32

indices to express lake productivity on a continuous numerical scale. A widely
used indicator is the Carlson Trophic State Index (TSI) (Bednarz et al. 2001),
including water clarity as measured with a Secchi disk, chlorophyll a from algal
pigment, and total phosphorus. The focus of CLMP is the volunteer-based
monitoring of these primary trophic state parameters (Bednarz et al. 2001).

The CLMP measures SDT in up to 300 lakes and total phosphorus and
chlorophyll a in approximately 200 lakes (Bednarz et al. 2001). Data collected
under this program will be used to educate and involve citizens in lake quality

management issues, and nutrient enrichment and eutrophication trends monitoring.

2.4. Remote sensing technologies for water quality monitoring

The era of high quality observations of the earth’s surface began in 1972
with the launch of ERTS 1, later renamed Landsat-1 (Price, 1987). In subsequent
years Landsats 2—5 have acquired similar data, representing a continuous archive
of 20 years of earth observations. These data have been used extensively for
monitoring changes on the earth’s surface (Price, 1987). Not only have the studies
of land surface benefited from the satellite remote sensing technologies, the
satellites also provide useful tools to integrate limnological data collected from
traditional in—situ measurements (Giardino et al. 2001).

While conventional monitoring of water bodies has to deal with several
limitations, application of existing satellite imagery has the potential to provide

such information at relatively low cost on both spatial and temporal scales.

33

Landsat Thematic Mapper (TM) has been used in several studies as a tool
for water clarity monitoring (Khorram and Cheshire, 1985; Lathrop, 1992; Kloiber
et al. 2002). Covering a large area in a single image, it provides quick and
relatively inexpensive information for a substantial number of lakes.

Landsat satellite data characteristics have many advantages for water
monitoring. The synoptic coverage of 185 km x 167 km allows a large area
monitoring. The spatial resolution of Landsat, 30 x 30 m in the optical bands, is
sufficient to monitor all but the smallest lakes and still provides information on in-
lake variability (Oppenheiner, 1997). The multispectral quality of the data allows
measurement of several water characteristics, while the moderate frequency of
image acquisition allows for change-detection monitoring every 16 days. Finally,
decreasing costs for Landsat-5 Thematic Mapper (TM) and Landsat-7 Enhanced
Thematic Mapper Plus (ETM+) data provide the opportunity for cost-effective

regional assessments.

2.4.1 Landsat sensor cross-calibration (Landsat-5 TM and Landsat-7 E T M +)

The launch of Landsat-7 on April 15, 1999 positioned the spacecraft in an
orbit very close to that of the Landsat-5 spacecraft. The mean altitude of Landsat-
7 was 699 km, 6 km below the mean altitude of Landsat-5 at 705 km above the
ground. The sun-synchronous polar orbit operations of the two spacecraft in this
“underfly” or tandem configuration were almost exactly the same, with a slowly

varying temporal offset on the order of 10 to 30 minutes (Teillet et a1. 2001). This

34

unusual but valuable circumstance was purposely designed to evaluate the data
consistency between the Landsat-7 ETM+ and the Landsat-5 TM. The repeat
coverage cycle of Landsat-7 is now different from that of Landsat-5 by 8 days.

The Landsat-5/Landsat-7 underfly cross-calibration experiment of
Vogelmann et al. (2001) used TM and ETM+ data sets from Niobrara, NE;
Western Michigan; and Zambia, Africa. The researchers developed empirical
relationships between Landsat-5 and Landsat-7 data to facilitate monitoring
investigations (Vogelmann et al. 2001). Mean digital number (DN) values were
extracted from 20 homogeneous targets from each of three pairs of Landsat-7 and
Landsat—5 images, then, L5-L7 regression equations for each band were
developed. Radiometric normalization was used to convert Landsat-7 DN’s and
Landsat-5 DN’s and then to radiance and compare with radiance estimates using
L5 and L7 prelaunch coefficients. Other comparisons, i.e. statistical, NDVI, and
classification, were processed. The results showed that band by band r2 values
(L5 vs L7) (60 Targets) are above 0.9960 (Vogelmann et al. 2001).

The study established coefficients for converting L7 to L5 DN Values as

shown in Table 2.3.

35

Table 2.3 Empirically derived slope and intercept values enabling radiometric
conversion of Landsat-7 ETM+ DN values to Landsat-5 TM DN values

(Vogelmann et al. 2001)

 

 

 

 

 

 

 

Band Slope Intercept R2
1 1.060 -4.208 0.9960
2 0.563 -2.576 0.9977
3 0.650 -2.496 0.9981
4 0.701 -4.800 0.9981
5 1.016 -6.959 0.9983
7 0.767 -5.760 0.9980

 

 

 

 

 

 

Relationships were based upon comparisons between Landsat-5 and
Landsat-7 near-simultaneous data acquisitions on June 2, 1999 at Niobrara,
Nebraska. Landsat-7 gain levels were high for all bands used.

The study concluded that Landsat-7’s variability with homogenous targets,
is lower than Landsat—5 except for Band3. Radiometric normalization of Landsat-
5 to Landsat-7 data provides very similar products (images, NDVI, classification

products).

2. 4.2 Use of satellite data in water monitoring applications
Satellite-based techniques for lake water quality analysis have some
similarities to applications in the marine coastal zone and pelagic environments.

The common approach is based on the use of optical bands, ranging fi'om blue to

36

near infrared, to explore the relationships between the water surface radiance and
the water bio-physical parameters, i.e. SDT, concentrations of chlorophyll a, and
suspended particulates, both mineral and organic (Zilioli and Brivio, 1997).
Compared to other natural surfaces, such as soils and vegetation, the
fraction of light reflected from lake water is very small. Water-leaving radiances
are normally less than 10% of the total radiance measured at the sensor (Gordon,
1987). Typically, in clear water, this contribution is maximal in the blue (A z 440
nm), medium in the green (A z 550 nm) and negligible in the near infrared (A 2
750 nm). Since the true water-leaving radiance is merely a small part of the signal
recorded by the satellite, accurate radiometric normalization of the sensor is

critical (Gordon, 1987).

2.5 Radiometric and atmospheric effects

The retrieval of physical parameters useful for lake monitoring cannot be
successfully achieved without taking into account the sensor characteristics and
atmospheric effects (Brivio et al. 2001). Therefore, an appropriate procedure is
needed in order to take full advantage of the usefulness of the time series data and
to allow a quantitative interpretation of remotely sensed data (Brivio et al. 2001).

The raw satellite images consist of 8-bit raster containing digital numbers
(DNs) between 0 and 255. These digital numbers are controlled by a combination

of (1) sensor parameters (gain and offset), (2) illumination and viewing geometry

37

of the scene, (3) absorption and scattering by the atmosphere, (4) reflectance from
the surfaces under consideration, and (5) solar irradiance outside the atmosphere
(Mackay, 1998).

In the ideal case of no atmospheric interference, a fraction of the incoming
solar photons would be directly reflected back to space. The radiance measured at
satellite would then depend directly upon the actual ground properties (Tanre’ et
al. 1990). In reality, the signal is disturbed by the atmosphere. The missing
photons have either been lost through absorption processes or changed their
propagation direction through scattering processes by molecules and aerosols
(Chavez, 1988).

Radiance reaching the satellite sensor has three components (Gordon,
1987): the water leaving radiance, the contribution from atmospheric scattering of
radiance to air particles (Rayleigh scattering) and the scattering due to particulates
in the atmosphere (Aerosol scattering). Atmospheric correction procedures
estimate the contribution due to Rayleigh and Aerosol scattering and subtracting
that contribution from the observed DN (Pattiaratchi et al. 1994). Mackay et a1.
(1998) found that the derivation of the aerosol optical thickness at a reference
wavelength of 550 nm was strongly affected the absolute radiance.

Although Olmanson (2001) and Nelson (2002) indicated that atmospheric
correction or normalization of the Landsat data was not necessary for the method
similarly to the method use in this study, the application of atmospheric

corrections or, at least, some level of radiometric normalizations is useful to

38

accurately predict water quality parameters as suggested by several other studies

(Gordon, 1987; Ma, 1997; Brivio et al. 2001; Volgelmann et al. 2001).

2. 5. 1 Radiometric normalization

Landsat-5 TM has been operational since March 1984. More than 15 years
after the Landsat-5 launch, postlaunch calibration coefficients are not currently up-
dated due to the poor function of the lamp-system necessary to perform the
onboard sensor calibrations. However, several sets of calibration coefficients,

such as gain and offset (Table 2.4), were provided during the satellite lifetime.

Table 2.4 Calibration coefficients (gain and offset) of Landsat-5 TM data

 

 

TM band Gain Offset
1 0.602 -1.520
2 1.175 -2.840
3 0.806 -1.170
4 0.815 -1.510
5 0.108 -0.370
6 0.055 1.238

 

 

 

 

 

These coefficients can be used to produce radiance values in Wrn'2 sr'l um'1

(Brivio et al. 2001). Calibration coefficients are used in the following linear

39

relationship to convert digital signal value or digital number (DN), in a given
spectral band, into physical units of at-satellite radiance L00»):
L00») = DNOI.) x gainOt) + offsetOc) (2 — 5)
The relationship can be used to correct the data using gains and offsets. It
converts the data from DN to radiance values (Wm‘2 sr'l pm”), which represent

standard physical units (Chavez, 1988).

The radiometric normalization technique to convert DN to radiance is
processed in this study to normalize the images of various date of acquisition from
multiple sensors (TM and ETM+) and different geographic locations (Michigan

and Thailand).

References

Bednarz, R., Rheaume, S., and Minnerick, R., Aichele, S., Jodoin, R. (2001). The
Michigan Department of Environmental Quality’s Lake Water Quality
Assessment Monitoring Program for Michigan’s Inland Lakes (Draft).
Michigan Department of Environmental Quality Land and Water Management
Division and United States Geological Survey Water Resources Division,
Michigan District: 1 - 53

Brivio, P. A., Giardino, C., and Zilioli, E. (2001). Validation of satellite data for
quality assurance in lake monitoring applications. The science of the Total
Environment, 268, 3 — l8.

Bukata, R. P., Jerome J. H., and Burton J. E. (1988). Relationships among Secchi
disk depth, beam attenuation coefficient, and irradiance attenuation coefficient

for Great Lakes waters. Journal of Great Lakes Research, 14(3), 347-355.

40

 

Canfield, D. E., Jr., Brown, C. D., Bachmann, R. W., and Hoyer, M. V. (2002).
Volunteer lake monitoring: testing the reliability of data collected by the
Florida LAKEWATCH program. Lake and Reservoir Management, 18, 1-9.

Carlson, R. E. (1977). A trophic state index for lakes. Limnology and
oceanography, 22 (2), 361 — 368.

Chavez, P. S., Jr. (1988). An improved dark-object subtraction technique for
atmospheric scattering correction of multispectral data. Remote Sensing of
Environment, 24, 459 - 479.

Giardino, C., Pepe, M., Brivio, P. A., Ghezzi, P., and Zilioli, E. (2001). Detecting
chlorophyll, secchi disk depth and surface temperature in a sub-alpine lake
using Landsat imagery. The Science of the Total Environment, 268, 19 — 29.

Gordon, H. R. (1987). Calibration requirements and methodology for remote
sensors viewing the oceans in the visible. Remote Sensing of Environment, 22,
103 - 126.

Heiskary, S., Lindbloom, J ., and Wilson, C. B. (1994). Detecting water quality
trends with citizen volunteer data. Journal of Lake and Reservoir
Management, 9(1), 4-9.

International Cooperative Study of the Gulf of Thailand. “About the Gulf of
Thailand.” IOC- WESTPAC and SEA START RC. Jun 2001. 25 Aug 2002,
<httg//www.start.or.th/gg/&twww/index.htm>

Kerr, M., Ely, E., Lee, V., and Mayio, A. (1994). A profile of volunteer
environmental monitoring: National survey results. Journal of Lake and
Reservoir Management, 9(1), 1-4.

Khorram, S. and Cheshire H. M. (1991). Water quality mapping of Augusta Bay,
Italy from Landsat-TM data. International Journal of Remote Sensing, 12(4),
803-808.

Kloiber, S. M., Brezonik, P. L., Olmanson, L. G., and Bauer, M. E. (2002). A
procedure for regional lake water clarity assessment using Landsat

mutispectral data. Remote Sensing of Environment 82(1), 38-47.

41

11‘.

1:

lo

11.

.\1

Ni

Lathrop, R. G. (1992). Landsat Thematic Mapper monitoring of turbid inland
water quality. Photogrammetric Engineering and Remote Sensing, 58(4), 465 -
470.

Lee, G. F ., Jones-Lee, A. F ., and Rast, W. (1995). Secchi depth as a water quality
parameter. Report of F red Lee & Associates, El Macero, CA. (15 pp.).

Lorenzen, M. W. (1980). Use of chlorophyll — Secchi disk relationships.
Limnology and oceanography, 25 (2), 371 — 372.

Mackay, G. and Steven M. D. (1998). An atmospheric correction procedure for the
ATSR-2 visible and near-infrared land surface data. International Journal of
Remote Sensing, 19(15), 2949 — 2968.

Michigan Farmland and Agriculture Development Task Force (1994). Policy
recommendations and options for the future growth of Michigan agriculture: a
report to Governor John Dngler/ Michigan Farmland and Agriculture
Development Task Force. Lansing, Mich. : The Task Force, 1994.

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.

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

Nelson, SA (2002). The use of remote sensing in freshwater wetland and lake
studies. PH.D. Dissertation. Michigan State University, 134 pp.

Obrecht, D. V., Milanick, M., Perkins B. D., Ready, D., and Jones, J. R. (1998).
Evaluation of data generated from lake samples collected by volunteers.
Journal of Lake and Reservoir Management, 14(1), 21-27.

Oglesby, R. T. (1977). Relationships of fish yield to lake phytoplankton standing
crop, production, and morphedaphic factors. Journal of Fisheries Research
Board of Canada, 34, 2271-2279.

Olmanson, L. G., S. M. Kloiber, M. E. Bauer, and P. L. Brezonik (2001). Image
processing protocol for regional assessments of lake water quality. University

of Minnesota Water Resources Center Public Report Series number 14. Water

42

Resources Center and Remote Sensing Laboratory, University of Minnesota,
St. Paul, MN, 55108. (15 pp.).

Oppenheiner, C. (1997). Remote sensing of the colour and temperature of volcanic
lakes. International Journal of Remote Sensing, 18, 5 — 37.

Pattiaratchi, C. , Lavery, P., Wyllie, A., and Hick, P. (1994). Estimates of water
quality in costal waters using multi-date Landsat Thematic Mapper data.
International Journal of Remote Sensing, 15(8), 1571 -1584.

Price, J. C. (1987). Calibration of satellite radiometers and the comparison of
vegetation indices. Remote Sensing of Environment, 21, 15 — 27.

Tanre', D., Deroo, C., Duhaut, P., Herman, M., Morcrette, J .J ., Perbos, J ., and
Deschamps, D.Y. (1990). Description of a computer code to simulate the
satellite signal in the solar spectrum: the SS code. International Journal of
Remote Sensing, 1 1(4), 659 — 668.

Teillet, P. M., Barker, J. L., Markham, B.L., Irish, R. R., F edosejevs. 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.

The National Statistical Office Thailand (2002). Statistical Standards Branch,
Statistical Policy and Coordination Division, National Statistical Office Larn
Luang, Bangkok, Thailand.

Veatch, J. O. (1941). Agricultural land classification and land types of Michigan.
Michigan State University. Agricultural Experiment Station. Special bulletin;
no. 231 (First revision).

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

Vogelmann, J. E., Helder, D., Morfitt, R., Choate, M. J ., Merchant, J. W., and
Bulley, H. (2001). Effects of Landsat 5 Thematic Mapper and Landsat 7

Enhanced Thematic Mapper Plus radiometric and geometric calibrations and

43

corrections on landscape characterization. Remote Sensing of Environment, 78,
55 — 70.

Wallin, M. L., and Hakanson, L. ( 1992). Morphometry and sedimentation as
regulating factors for nutrient recycling and trophic level in coastal waters.
Hydrobiologia, 235, 33-45.

Weier, John. “Testing the Waters”. Earth Observatory, NASA. Mar 2002. 13 May
2002,
<http ://earthobservatory.nasa.gov/Studv/WaterQualitv/water_quality2 .html>

World Resource Institute. “Farming Fish: The Aquaculture Boom”. NE (Suite
800), Washington, DC 20002. 20 Aug 2002,
<http://www.wri.org/trends/fishfarm.html>

Zilioli, E., and Brivio, P. A. (1997). The satellite derived optical information for
the comparative assessment of lacustrine water quality. The Science of the

Total Environment, 196, 229 — 245.

44

CHAPTER III

METHODOLOGY

3.1 Ground observed water clarity data
3.1.1 Field observation of Michigan

‘kFor this research, the ground-observation data set was comprised of 123
Secchi disc transparency depth (SDT) values within a single satellite path (path21)
across 3 scenes in the lower peninsula of Michigari (Figure 3.1). The area of the
sample lakes ranges from 0.05 — 41.2 km2 with the Secchi depth of 0.61 — 7.62 m
(Appendix A). The data were obtained from two sampling programs: (1) the
Michigan Department of Environmental Quality’s (MDEQ) Lake Water Quality
Assessment (LWQA) Monitoring Program and (2) the Michigan Citizens Lake
Monitoring Program (CLMP) (Table 3.1). (Only lakes that were larger than
101,171 m2 in surface area and were sampled between July - August 2001 were
used in the study as this was found to be the best index period for remote sensing
of water clarity in the nearby state of Minnesota (Kloiber et al. 2000). The data set
was further refined by using only the lakes that were sampled within at 7 days of

the satellite imagery.)

45

 

4".
‘5
(‘ ‘7 I
k\ 1 ..
. — “j
2, 2 1
32
I V: " _«‘
4 'm- "A\f\.\,
‘ o “‘ —v f- I“ y
'. r ,‘ ‘ y
' 1

 

 

 

 

Figure 3.1 Inland lake field sample locations in Landsat path 21

46

Table 3.1. Information for 123 ground observation data from CLMP and MDEQ

 

 

 

Sampling Number of Secchi (m)
Organization Lakes Mean Range
CLMP 94 3.27 0.61 - 7.62
MDEQ 29 2.70 1.07 - 4.88
Total 123 3.13 0.61 — 7.62

 

 

 

 

 

 

3.1.2 Field observation of Thailand

Sea-truth data were collected in the Bight of Bangkok, the Northern part of
the Gulf of Thailand (Figure 3.2). Thirty samples were collected from each
sampling date, at least 1 km apart from each other. Every field sample was
collected no less than 5 km from the shore to remove shoreline effects. Water
sampling was done within :5 days of the satellite overpass between July —
November 2001. SDT was measured using a 0.20 m diameter, black and white
quadrated disk. Water samples were filtered (using GF/C Whatrnans filter) for
chlorophyll a and suspended sediment determination. Two replicated samples
were used at each site. The filters were stored on ice and analyzed for chlorophyll
a concentration by the acetone extraction method (Strickland and Parsons, 1972).
Each sample site was geographically located using (Garmin ETREX) Global
Positioning System (GPS) technique in order to assure exact matching with digital

image pixels.

47

 

P132 P131 P130

  
 

The Bight of
Bangkok

 

 

 

Figure 3.2 The Bight of Bangkok, Thailand

 

3.2 Satellite data

(Six Landsat images in path 21 row 29, 30 and 31 (three Landsat-5 TM and
three Landsat-7 ETM+; Appendix B) were processed for Michigan and four
images (two Landsat 5 TM and two Landsat 7 ETM+) in path 129 row 51 were
processed for the Bight of Bangkok, Thailand) Information of the satellite data is
showed in Table 3.2.

Table 3.2 Information of the satellite data used in the study

 

 

Image scene Date Source of data
Michigan
ETM+ path 21 row 29 08/25/01 ERos"
ETM+ path 21 row 30 08/25/01 EROS
ETM+ path 21 row 31 08/25/01 EROS
TM path 21 row 29 09/05/01 USGS”
TM path 21 row 30 09/05/01 USGS
TM path 21 row 31 07/04/01 USGS
Thailand
ETM+ path 129 row 51 09/02/01 EROS
ETM+ path 129 row 51 01/08/02 EROS
TM path 129 row 51 07/24/01 GISTDA."
TM path 129 row 51 08/25/01 GISTDA

 

 

 

 

 

t

Earth Resource Observation System (EROS) Data Center, U.S.A

" US. Geological Survey, U.S.A
... Geo-Informatics and Space Technology Development Agency (GISTDA), Thailand

49

\e

001

11;

3. 2.1 Image pre-processing and radiometric normalization

In order to benefit fully from the Landsat data record, image processing is
required to ensure that the data are consistent and not significantly affected by
artifacts of the measurement system (Teillet et a1. 2001).

Image processing was done using ERDAS Imagine software version 8.5 on
Network PC. All Landsat-5 TM and Landsat-7 ETM+ images were geometrically
corrected to ensure proper alignment and scale, with registration to the Universal
Transverse Mercator (UTM) projection system using WGS 84 North datum zone
16 for Michigan and zone 47 for Thailand.

The first images of each path row were geometrically corrected using the
“Geometric Correction” module in ERDAS program using the polynomial model
with reference points collected from the keyboard only. Geographic reference
points were indicated in the metadata file included with the image. Other images
of the same paths and rows were co-registered with the first image by selecting
reference points from an existing image viewer.

Radiometric normalization is very important and is strongly recommended
in other studies (Vogelmann et al. 2001; Brivio et al. 2001). Some studies also
suggested that when attempting to physically relate image data to quantitative
ground measurements, especially for water resource evaluation, conversion of
digital number (DN) values to absolute radiance values for each spectral band is
critical (Gordon, 1987; Ma, 1997). The calibration coefficients (Table 3.3) are

used in the following linear relationship to convert DN values, in a given spectra]

50

1.111:

For
\

\l'h

band, into physical units of satellite radiance (L0). There is a slight difference in

the mathematical equation used to convert DN into radiance for TM and ETM+

images as showed below:

Ear—7M

L0=(DNxG)+B (3—1)
Where: L0 = at-satellite spectral radiance

DN = digital number

B = intercept of response function (channel Offset)

G — slope of response function (channel Gain)

For E T M +
L0 = [(LMAX — LMIN) / 255] x DN + LMIN (3 — 2)
Where: L0 = at-satellite spectral radiance
LMAX = upper limit of radiance the Landsat-7 sensor receives
LMIN = lower limit of radiance the Landsat-7 sensor receives

51

Table 3.3 Calibration coefficients of Landsat-5 TM and Landsat—7 ETM+

 

 

 

Band TM ETM+
Gain Bias LMAX LMIN
1 0.602 -1.520 191.6 -6.2
2 1.175 -2.840 196.5 -6.4
3 0.806 -1.170 152.9 -5.0
4 0.815 -1.510 241.1 -5.1
5 0.108 -0.370 31.06 -1.0

 

 

 

 

 

 

 

Conversion of DN to radiance was performed in ERDAS Imagine

“Modeler” module. A flow diagram of this process is shown in Figure 3.3.

 

 

 

 

Radiance Radiance
image image

TM ETM+

 

 

 

Figure 3.3 Flow diagram of conversion process from DN to radiance

52

3.2.2 Water-only image

Terrestrial areas and other unneeded data were removed in this step because
only the water pixels will be processed to predict the clarity/ trophic states. An
advantage of this process is that it reduces the file size of the images as well as the
processing time. I performed an unsupervised classification to differentiate water
classes for the image. Each image was classified into 15 classes using ten
iterations with the convergence threshold of 0.950. In this step, water class or
classes were easily identified because of a unique spectral-radiometric response of
water, which makes it significantly different from terrestrial features. The water-
only classes were used as a mask to remove terrestrial areas from the DN and
radiance images. The masking process was also performed in the “Modeler”
module using conditional function. Processing diagram and the result of this

process, the water only image, is shown in Figure 3.4 - 3.5.

DN/Radiance
image
Water-only>
image

B Keep A if B— — Water, or change to “0” otherwise

Unsupervised
classified
image

 

 

 

 

 

 

 

 

 

 

Figure 3.4 Flow diagram of water-only image

53

 

- Water
:2] Urban
_ Forest
:3 Soil

_ Crops

 

 

 

 

Figure 3.5 Example of water-only image (a) image, (b) unsupervised classified

image, (c) water-only image

The next step was to process an unsupervised classification of the water
only image to detect water pixels that were affected by vegetation, shoreline or
sand bottom. Figure 3.6 shows the spectral signatures of ten classes for a water-

only image. Pixels that had been affected by vegetation, bottom or terrestrial

54

features were colored in red, differently from water clarities in blues and greens.
The classified water only map from this step was used in conjunction with
bathymetric maps to select the representative sample (Area Of Interest: AOI) areas

from the ground measure sample.

 

 

Water pixels
Water pixels
Effected pixels
Effected pixels
Effected pixels

-
ELI!
_
—
_

 

 

 

Figure 3.6 Example of unsupervised classification of the water-only image to

detect affected pixels

3.2.3 Signature acquisition
Limnologists and water monitoring agencies commonly assume that a

single site in the center of the lake provides a representative sample of the entire

55

lake. Satellite imagery provides data across the entire lake surface, including such
areas as shallow bays and littoral areas with emergent vegetation that may have
water quality that is significantly different from the deeper open water area.
Therefore, spectral data from a satellite image must be carefully extracted to
include only the pixels from areas that have similar water quality to the ground
measurements (Kloiber et al. 2002).

I selected Area Of Interest (AOI) locations for each lake, or each GPS
location in the Bight of Bangkok, corresponded with the field measurements.
Using the “A01 tool” in ERDAS Imagine, a polygon was drawn around a cluster
of pixels that represented a pelagic zone (water depth > 5m) of the lake (Figure
3.7). Bathymetric maps and unsupervised classification maps of the water-only
image were used to help determine the proper AOIs. For best results, the size of
AOIs should contain at least 8 pixels as suggested by Lillesand (1983). My AOIs
ranged from 15 pixels for small lakes to 327 pixels for larger lakes, with a median
of 58. A013 for the Bight of Bangkok were 15 pixels for all sample points.

After the A01 location had been selected, the spectral signatures were
acquired by using the mean DN or radiance value within the A01. Extraction of
spectral signatures was performed in the “Signature Editor” module. I activated
the A01 and selected “Create New Signature(s) from AOI”. The name or ID of
each lake was typed in at this step. Next, I opened the “Column Properties” and

selected “Mean” value to identify the statistical value in the output. When all the

56

signatures of water pixels with the AOIs were acquired, the data were imported

into Microsoft Excel Spreadsheet.

 

 

 

 

 

Figure 3.7 Example of A01 selection

3.2.4 Multiple regression

Some previous studies have used nonlinear power models (Y = aXb) to
address the curvilinear behavior of the satellite-ground observation relationship
(Lathrop, 1992; Cox et a1. 1998). Although a power model provided a strong
correlation, residuals from it were not normally distributed. On the other hand, a
semi-log equation was found to meet the model assumptions, as found by
Pattiaratchi et al. (1994) and Kloiber et a1. (2000 and 2002). Consequently, I

chose the regression model using a natural-log transformation of SDT developed

57

by Kloiber et al. (2002) to use in this study. The Kloiber model has been used on
a smaller scale, but has not yet been applied at a broader geographic scale or in
conjunction with a complex system of ground measurements and images from
multiple satellite sensors.

Based on results from Kloiber et al. (2002), the semi-log relationship
between the mean brightness values of Landsat TM bandl and band 3 for the
A015 and the log of SDT produced the best correlation between ground observed
data and satellite data. The TMlzTM3 ratio was found to be the most significant
factor and the TMl variable improved the regression relationship slightly.
Therefore, I used Band]: Band3 ratio, along with Band] for my regression model.

After importing the satellite signature and the ground observation data into
a Microsoft Excel spreadsheet, I performed a natural-log transformation on ground
observed SDT and calculated a satellite radiance B 1: B3 ratio for each image (both
DN and radiance). Then, I performed a multiple regression analysis with ln(SDT)
as the dependent variable and B1:B3 ratio along with Bl as independent variables
using SYSTAT (SPSS Software, Inc., 2001). Statistical outliers were evaluated
and removed from the data set. An example of the output from SYSTAT is shown
in Figure 3.8.

The regression model used in this study was

ln(SDT) = a(B1:B3)+b(Bl)+c (3—3)

Where: a, b, and c = coefficients from regression analysis

58

 

Dep Var: LNSDT N: 93 Multiple R: 0.660 Squared multiple R: 0.436
Adjusted squared multiple R: 0.424 Standard error of estimate: 0.314

Effect Coefficient Std Error Std Coef Tolerance t P(2 Tail)
CONSTANT -2.459 0.695 0.000 -3.537 0.001
B13 1.849 0.222 0.661 0.995 8.331 0.000
B1 -0.009 0.009 -0.083 0.995 -1.051 0.296

Analysis of Variance

Source Sum-of-Squares df Mean-Square F-ratio P
Regression 6.870 2 3.435 34.793 0.000
Residual 8.885 90 0.099

*“ WARNING ***
Case 3 has large leverage (Leverage = 0.179)

Durbin-Watson D Statistic 1.940
First Order Autocorrelation 0.021

Plot of Residuals against Predicted Values

 

1.0 T T I

0.5 ~ —

RESIDUAL
C
o
l
i
't
i
i

:5
(It
I
l

 

 

 

' 0.0 0.5 1.0 1.5 2.0
ESTIMATE

 

 

 

Figure 3.8 Example of an output from SYSTAT

After the coefficients for each satellite image were received, the multiple

regression equation was applied to the water-only image using the ERDAS

59

 

Imagine Modeler. A flow diagram to apply this equation is shown in Figure 3.9.

(The regression model image (ln(SDT)) was then converted to Trophic State Index
(TSI) using the same method (Figure 3.10). The relationship between ln(SDT)
and TSI developed by Carlson (1977) is given in the following equation.

TSI(SDT) = 60— 14.41 ln(SDT) ) (3 — 4)

DN/Radiance Eq. Regression
image (3-3) model image

 

 

 

 

 

 

 

Figure 3.9 Flow diagram of conversion from DN/radiance image to regression

model image

Regression EQ- TSI image
model image (3'4)

 

 

 

 

 

 

 

Figure 3.10 Flow diagram of conversion from regression model image to TSI

60

3.2.5 Lake water clarity/ trophic condition map

l Maps of lake water clarity/ trophic condition were created in two levels,
pixel level and lake level. ) Once the TSI(SDT) image is produced, every pixel of
the image is assigned a TSI value. The three TSI images in path 21 were
mosaiced together to create a pixel-level map. I changed the color display of the
TSI value in “Raster Attribute” to match with the legend as indicated in the
method protocol (Kloiber et al. 2002). An example of a pixel-level water clarity/
trophic condition map is presented in Figure 3.11.

[The lake-level water clarity/ trophic condition map was created using ESRI
GIS software.) The map displays TSI values of lakes larger than 101,171 m2.
First, mean TSI values of each lake were extracted in ERDAS Imagine “Zonal
Statistics to Polygon Attributes” module. 1 used the lake GIS coverage as the
vector layer and TSI image as the raster layer. Then, I selected “Mean” as a zonal
function. @he process added the selected statistical value, the mean TSI valuein
tlrisrcaserof each lake overlain by the GIS coverage as a new field in the coverage
attribute table. In ESRI GIS, the GIS lake coverage was then assigned a color to

match with the same legend. The lake-level water clarity trophic condition map is

shown in Figure 3.12.

61

 

 

  
  
 
 
 
 
 
   

 

 

_TSI

0-30
- 31-40
- 41-50
I: 51-60
:1 61-70
- 71-80

-81-100

SDT (m)
8

4
2
1
0 5
0.25
0 125

 

 

 

Figure 3.11 Example of pixel-level water clarity/ trophic condition map

 

 

 

 

  

i t

4‘

~
1%" c

 

 

 

 

it a
‘ _TSI SDT(m)

.~ :1-3238 2

a, 41-50 2
I 6}. E5130 1

, - 61-70 0.5

71-80 0.25
‘ |;:|81-100 0.125

 

 

 

Figure 3.12 Example of lake-level water clarity/ trophic condition map

62

 

 

References

Brivio, P. A., Giardino, C., and Zilioli, E. (2001). Validation of satellite data for
quality assurance in lake monitoring applications. The science of the Total
Environment, 268, 3 — 18.

Chavez, P. 8., Jr. (1988). An improved dark-object subtraction technique for
atmospheric scattering correction of multispectral data. Remote Sensing of
Environment, 24, 459 — 479.

Cox, R. M., Forsythe, R. D., Vaughan, G. E., and Olmstead, L. L. (1998).
Assessing water quality in the Catawba River reservoirs using Landsat
Thematic Mapper satellite data. Lake and Reservoir Management, 14, 405-416.

Gordon, H. R. (1987). Calibration requirements and methodology for remote
sensors viewing the oceans in the visible. Remote Sensing of Environment, 22,
103 — 126.

Kloiber, S. M., Brezonik, P. L., Olmanson, L. G., and Bauer, M. E. (2002). A
procedure for regional lake water clarity assessment using Landsat
mutispectral data. Remote Sensing of Environment 82(1), 38-47.

Kloiber, S.M., Anderle, T.H., Brezonik, P.L., Olmanson, L., Bauer, M.E. &
Brown, DA. (2000). Trophic state assessment of lakes in the Twin Cities
(Minnesota, USA) region by satellite imagery. Archive Hydrobiologie Special
Issues Advances in Limnology, 55,137-151.

Lathrop, R. G. (1992). Landsat Thematic Mapper monitoring of turbid inland
water quality. Photogrammetric Engineering and Remote Sensing, 58(4), 465 -
470.

Lillesand, T. M., Johnson, W.L., Deuell, R. L., Linstrom, O. M., and Meisner, D.
E. (1983). Use of Landsat data to predict the trophic state of Minnesota lakes.
Photogrammetric Engineering and Remote Sensing, 49(2), 219 - 229.

63

Ma, K. M. (1997). Water quality assessment of the Chilung river using Landsat
Thematic Mapper and Airborne Multispectral Scanner Images: Taipei, Taiwan.
MS. Thesis. Michigan State University, 80 pp.

Pattiaratchi, C. , Lavery, P., Wyllie, A., and Hick, P. (1994). Estimates of water
quality in costal waters using multi-date Landsat Thematic Mapper data.
International Journal of Remote Sensing, 15(8), 1571 -1584.

Strickland, J. D. H. and Parsons, T. R. (1972). A practical handbook of seawater
analysis. Fisheries Research board of Canada, 167.

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.

Vogelmann, J. E., Helder, D., Morfitt, R., Choate, M. J ., Merchant, J. W., and
Bulley, H. (2001). Effects of Landsat 5 Thematic Mapper and Landsat 7
Enhanced Thematic Mapper Plus radiometric and geometric calibrations and
corrections on landscape characterization. Remote Sensing of Environment, 78,

55 — 70.

64

CHAPTER IV

MICHIGAN LAKES WATER MONITORING

4.1 Introduction

Eutrophication, i.e. nutrient pollution impacts from human activities, has
been of national concern for decades. Yet, an adequate monitoring program for a
single date, statewide coverage, has never been possible. Satellite-based water
monitoring using Landsat imagery has the potential to overcome this limitation.

Several studies have found a strong correlation between Secchi disc
transparency (SDT) and a number of other water quality variables, such as trophic
status, phosphorus concentrations, chlorophyll concentrations, suspended sediment
concentrations, and fish yields (Oglesby, 1977; Lorenza, 1980; Lee et al., 1995).
This easily measured parameter is considered a good tool for water quality
monitoring worldwide. SDT has also been found to have potential for use in
satellite-based water quality monitoring (Lathrop, 1992; Cox et al. 1998; Kloiber
et al. 2000 and 2001).

Landsat-5 Thematic Mapper (TM) imagery has been used in several lake
studies as a cost and labor efficient method for sampling water clarity. Fewer
studies have used Landsat-7 Enhanced Thematic Mapper Plus (ETM+). Since

Landsat-7 ETM+ is the latest in the Landsat series and is known to have

65

outstanding data quality and system performance, lake monitoring approaches
using Landsat-7 ETM+ should be explored.

Kloiber et a1 (2002) developed a method to monitor water clarity/ trophic
states using ground observed SDT and Landsat spectral signatures for the state of
Minnesota. However, this method was developed for use with only Landsat-5 TM
at a city-level and across a narrow range of SDT values. My objective in this
study is to test the accuracy and the applicability of this approach when using
multiple images and sensors across a wide diversity of lakes, as an alternative for
water quality monitoring in the state of Michigan. To test the efficacy of this
satellite-based approach in assessing statewide water clarity and trophic status of
Michigan inland lakes, 1) I compared the efficiency of satellite image DN and
radiance values in predicting water quality, and 2) I tested whether the
methodology developed for Landsat-5 TM could be applied to Lansat-7 ETM+

images.

4.2 Results

4. 2. 1 Comparison between DN and radiance images

Previous studies suggested the minimum conversion of DN to radiance for
lake studies because the fraction of light reflected from water is very small
(Gordon 1987; Ma, 1997; Brivio et al. 2001; Volgelmann et al. 2001). I compared
the DN and radiance values of the same geographic location from both Landsat-5

TM and Landsat-7 ETM+ bandl and band3. I randomly sampled DN and radiance

66

value from 16 pixels (Table 4.1). The result from simple correlation between TM
and ETM+ indicated a better goodness of fit for radiance value with the r2 of 0.79
(bandl) and 0.58 (band3; p < 0.001) for DN and 0.84 (bandl and band3; p <
0.001) for radiance. Not only the r2 of band3 was significantly improved, the
fitted line was also very close to the perfectly fit (1:1) line. {This result suggested
that TM and ETM+ values are more comparable after the radiometric

normalization] After this test, I convert the DN image to radiance image for all

SCCI’IBS.

Table 4.1 Preliminary efficacy test between digital number and radiance

 

DN Radiance
ETM+ TM ETM+ TM
Bandl Band3 Bandl Band3 Bandl Band3 Bandl Band3

 

 

 

73 28 57 15 50 12 32 10
87 69 70 50 61 37 43 39
62 32 50 18 41 14 29 13
57 27 49 15 38 11 28 ll
60 26 47 15 40 11 26 ll
59 26 48 17 39 11 27 12
58 29 50 18 38 12 28 13
58 26 49 14 38 11 27 10
67 32 50 19 45 15 28 14
79 36 69 26 55 17 4O 20
61 31 54 18 41 14 31 13
64 28 62 18 43 12 36 13
87 45 73 31 61 23 43 23
70 32 66 25 48 15 38 19
76 43 69 28 53 22 4O 21
63 30 58 22 42 14 32 25

 

 

 

 

 

 

 

 

 

 

67

Landsat-7 E T M +

From 123 sample points of lake SDT, 85 lake spectral signatures were
extracted from the imagery. 38 lakes were omitted due to the extensive area of
cloud and cloud shadow within Landsat-7 image on August 25, 2001 (Appendix
B). After a statistical outlier was removed from the data set, the number was
reduced to 84. Multiple regression of ground observed SDT and satellite spectral
bandl and band3 produced an r2 of 0.365 (p < 0.001) for the DN image and 0.558

(p < 0.001) for the radiance image (Figure 4.1- 4.2).

68

 

 

Bl3=Bl:B3

 

 

 

r2 = 0.365 (p < 0.001)

LN(SDT) = 1.630(BlzB3) + 0.008(Bl) — 3.097
Figure 4.1 Statistical analysis of 84 lakes using data from Landsat-7 ETM+ DN

image

69

 

 

Bl3=Bl:B3

 

 

 

r2 = 0.558 (p < 0.001)

LN(SDT) = 0.949(Bl:B3) + 0.001(131) — 2.322

Figure 4.2 Statistical analysis of 84 lakes using data from Landsat-7 ETM+

radiance image

70

Landsat-5 TM

Because of a very cloudy period in summer 2001, single date satellite
images for the entire path21 could not be obtained. Satellite image rows 29 & 30
were acquired on Sep 5 2001, and row 31 was from Aug 4 2001. The images were
approximately one month apart. Appendix B showed an obvious difference
between the two dates, which may be caused by atmospheric condition. Thus,
images for two dates were processed separately.

From 123 samples of lake SDT, only 45 samples were used in the Landsat-
5 study because SDT for several lakes had not been sampled in the field close to
the satellite acquisition date. Within the 45 sample dataset used for Landsat-5
TM, 34 samples were located in rows 29 and 30, and 10 in row 31. After one
outlier was removed, my multiple regression of 33 SDT and satellite spectral
bandl and band3 produced an r2 of 0.691 (p < 0.001) for DN and 0.804 (p <
0.001) for radiance (Figure 4.3-4.4). The r2 for the image row 31 was 0.353 (p <

0.001) for DN and 0.520 (p < 0.001) for radiance (Figure 4.5- 4.6).

71

 

LNSDT

 

 

Bl3=Bl:B3

r2 = 0.691 (p < 0.001)
LN(SDT) = 1.161(B1:B3) - 0.027(131) — 1.066

Figure 4.3 Statistical analysis of 33 lakes using data from Landsat-5 TM row 29
& 30 DN image

 

 

72

 

o Q
o
o u
.O c

o
0'.
O
c o
.0 .v
. a
g o
o c
o o
a o
o o
o. o-
. p
o I
a

 

Bl3=Bl:B3

 

 

 

r2 = 0.804 (p < 0.001)

LN(SDT) = 1.462(B1 :133) - 0.066(Bl) — 0.542

Figure 4.4 Statistical analysis of 33 lakes using data from Landsat-5 TM row 29

& 30 radiance image

73

 

2.0

 

 

 

Bl3=Bl:B3

 

 

 

r2 = 0.353 (p < 0.001)

LN(SDT) = 0.405(B1:B3)+ 0.012(31) — 1.716

Figure 4.5 Statistical analysis of 12 lakes using data from Landsat-5 TM row 31

DN image

74

 

2,0

 

 

 

Bl3=Bl:B3

 

r2 = 0.520 (p < 0.001)

LN(SDT) = 0.649(B12B3) - 0.008(Bl) — 1.170

Figure 4.6 Statistical analysis of 12 lakes using data from Landsat-5 TM row 31

radiance image

75

LThe results from both the DN and radiance images indicated better r2 for
radiance images (table 4.1). This result proved the first sub-hypothesis of this
study: Image processing into the standard radiance values has a potential to

produce more accurate results than DN values.)

Table 4.2 Summary of the multiple regression results for Michigan lakes

 

 

 

 

 

 

 

 

Data set Number of DN Radiance
sample R-Square R-Square
Lakes in Landsat-7 ETM+ 84 0.365 0.558
Lakes in Landsat-5 TM row 29&30 33 0.691 0.804
row 31 12 0.353 0.520
* p < 0.001

In the following steps, I will compare the efficiency of Landsat-5 and

Landsat-7 using only radiance images.

4. 2.2 Comparison between Landsat—5 TM and Landsat-7 E T M +

The model of 85 lake Landsat-7 ETM+ resulted in an r2 of 0.558 while the
model of 45 lake Landsat-5 TM produced the r2 of 0.804 and 0.520 for row 29 &
30 and row 31, respectively. Results at this point suggests that Landsat-5 TM is

much more effective than Landsat ETM+. However, after evaluating the two sets

2“

of data used for each sensor, I found that the ground data were substantially

76

different. To examine why the model of Landsat-7 ETM+ explained less variation
than Landsat-5 TM, I selected only the lakes that were sampled in both Landsat-7
and Landsat-5 overpasses to process the multiple regression again. My sub-
sample included 43 lakes, 32 lakes in image row 29 & 30 and 11 lakes in row 31.
Comparing r2 of the same 43 lakes, the r2 of Landsat-7 improved to 0.681 (p <
0.001), slightly better than the r 2 of Landsat-5 (whole path 21) of 0.565 (p <

0.001; Figure 4.7-4.8).

77

 

 

Bl3=Bl:B3

 

 

 

r2 = 0.681 (p < 0.001)

LN(SDT) = 1.018(B1:B3)+ 0.009(B1) — 2.928

Figure 4.7 Statistical analysis of 43 sub-sampled lakes using data from Landsat-7

ETM+ radiance image

78

 

 

qs>\ Bl3=Bl:B3

 

 

 

r2 = 0.565 (p < 0.001)

LN(SDT) = 0.796(Bl:B3) - 0.075(131) + 1.352

Figure 4.8 Statistical analysis of 43 sub-sampled lakes using data from Landsat-5

TM radiance image

79

According to the difference in image acquisition date of the Landsat-5 TM
images, I re-processed the regression analysis for row 29 & 30 and row 31
separately, based on image acquisition date. The 32 lakes, from 43 sub-sampled,
in row 29 & 30 gave the r2 of 0.711 (p < 0.001) for Landsat-7 and 0.808 (p <
0.001) for Landsat-5 (Figure 4.9-4.10). The 11 lake samples in row 31 produced
an r2 of 0.445 (p < 0.001) for Landsat-7 and 0.546 (p < 0.001) for Landsat-5
(Figure 4.11-4.12). Table 4.2 summarized the multiple regression results of the

fill] data set and the sub-sample data set.

80

 

 

Bl3=Bl:B3

 

 

 

r2 = 0.711 (p < 0.001)

LN(SDT) = 1.055(B1 :33) + 0.002(131) — 2.708

Figure 4.9 Statistical analysis for 32 lakes using data from Landsat-7 ETM+ row

29 & 30 radiance image

81

 

 

Bl3=Bl:B3

 

 

 

r2 = 0.808 (p < 0.001)

LN(SDT) = 1.460(B 1 :133) - 0.064(B1)— 0.583

Figure 4.10 Statistical analysis for 32 lakes using data from Landsat-5 TM row 29

& 3O radiance image

82

 

 

Bl3=Bl:B3

 

 

 

r2 = 0.445 (p < 0.001)

LN(SDT) = 0.801(B1:B3) + 0.021(B1)- 2.674

Figure 4.11 Statistical analysis for 11 lakes using data from Landsat-7 ETM+ row

31 radiance image

83

 

2.0

 

Bl3=Bl:B3

 

 

 

r2 = 0.546 (p < 0.001)

LN(SDT) = 0.607(B1:B3) —— 0.015(Bl) — 0.667

Figure 4.12 Statistical analysis for 11 lakes using data from Landsat-5 TM Path

21 row 31 radiance image

84

Table 4.3 Summary of the multiple regression results from sub-sample lakes

 

 

 

lrnage scenes Number of Radiance
samples R-square
Full dataset
ETM+ row 29, 30, 31 84 0.558
TM row 29&30 32 0.804
row 31 12 0.520

 

Sub-sample dataset

 

 

 

 

 

 

 

ETM+ row 29, 30, 31 43 0.681
TM row 29, 30, 31 43 0.565
ETM+ row 29&3O 32 0.711
TM row 29&30 32 0.808
ETM+ row 31 11 0.445
TM row 31 11 0.546
* p < 0.001

Landsat-7 ETM+ and Landsat-5 TM were able to predict SDT value at a
similar efficiency. The r2 of both sensors in any cases were not substantially
different. However, using the sub-sample, the Landsat-7 model tend to have a
slightly higher r2 than Landsat-5.

Since the r2 of 84 lakes full dataset of Landsat-7 was smaller than the 43

lakes sub-sample from the same sensor, a question remained as to what caused the

85

r2 to be lower in the full dataset. For better understanding of what factor or factors
caused the difference in predictability, I studied the ground observation data used
in all regression models more closely.

The characteristics of the lakes used in two datasets are compared in terms
of spatial distribution, SDT distribution, and date of data collection in the next

step.

4.2.3 Analysis of ground observation data

4.2.3.1 Geographic distribution of the datasets

The distribution of lakes, the filll and sub-sample, is shown in Figure 4.13.
A Moran’s I test indicated that geographic distribution of the two datasets was not

significantly different (Moran’s I coefficient = 0 for both cases).

86

 

 

O Lakes in both sensors
O Lakes in Landsat-7 only

k

   
  

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Figure 4.13 Geographic location of 43 lakes used in both sensors and 41 lakes

used only in Landsat-7 full dataset

87

 

   

 

 

 

 

4.2.3.2. Comparisons of Secchi disc transparency distribution and temporal
distribution of the dataset

The characteristics of SDT distribution of the two datasets were also found
to be fairly similar (Figure 4.14 - 4.17), with the SDT distributed between 0.61-
7.62 m. The majority of lakes had SDT of 3-4 m in each dataset. Temporal
distribution of the field sample acquisition was not significantly different either, as
tested by the statistical t-test. Although the sub-sample dataset excludes 2 samples
collected 7 days prior to the image acquisition, most of the samples were taken

within 3 days of the image acquisition in both cases.

S YS T A T paired samples t—test output of the temporal distribution

Paired samples t test on LK84 vs LK43 with 42 cases

Mean LK84 = -2.500
Mean LK43 = -1.905
Mean Difference = -0.595 95.00% CI = -1.072 to -0.1 18
SD Difference = 1.531 t = -2.519
df = 41 Prob = 0.016

88

 

 

G
O

 

Q
o
1
l
l

N

O
1
I

 

Number of data collected
A
O
l
l
l

  

 

 

O
l

T T

-7-6-5-4-3-2-101234 567
Days from aqulsltlon date

 

 

ETM+ (84 samples)

count

 

1 2 3 4 5 6 7
Socchl depth (m)

 

 

 

 

 

 

Figure 4.14 Distribution of Michigan lake samples by date of acquisition and SDT

distribution of 84 lakes used in ETM+ model

89

 

 

 

 

 

 

 

 

TM
30
220.-_ 2 a _ fl 2
i
go— - ~ fl _____2
E
3
0“1 r T T T T T Flt r T
-7-6-5-4-3-2-1012 3 4 5 6 7
Days from aqulsltlon date

 

 

 

 

TM (45 samples)

 

 

Number of data collected

 

 

 

1 2 3 4 5 6 7 8 9
Secchi depth (m)

 

 

 

 

 

 

Figure 4.15 Distribution of Michigan lake samples by date of acquisition and SDT

distribution of 45 lakes in TM model

90

 

 

ETM+ (43 sanples)

Number of data collected
M
O

 

 

 

 

10
0
-7-6-5-4-3-2-1012 3 4 5 6 7
Daysfromtheaqulsltlon
ETM+ (43 sanples)

20
315. 22- — —~~_
312. -_____ 2—
s 8< — 22
E 4- i
z

 

 

 

Secehl depth (m)

 

 

 

 

 

 

Figure 4.16 Distribution of Michigan lake samples by date of acquisition and SDT

distribution of 43 lakes used in ETM+ model

91

 

 

TM (43 samples)

 

Number of data collected
'0
O
l
l

 

 

 

 

 

 

10_ 2 7 22 2.27. 27 7.2L
0‘11 l I OIIVM I
-7-6-5-4H-3-2H-112W34567
Days from the aqulsltlon
TM (433arrples)
20

 

 

Number of data collected

 

 

 

Secchl depth (m)

 

 

 

 

 

 

Figure 4.17 Distribution of Michigan lake samples by date of acquisition and SDT

distribution of 43 lakes used in TM model

92

Since the physical characteristics of the full and the sub-sample datasets
were very similar, the difference in r2 may be caused by the quality of data in each
dataset. I further compared lake datasets from the two monitoring programs
MDEQ and CLMP to determine how each dataset would affect the regression

results.

4.2.4 Comparison of ground observation data between the MDEQ dataset and
CLMP dataset

There were 14 MDEQ lakes and 70 CLMP lakes used in the 84 full dataset.
The sub-sample dataset included 6 MDEQ and 37 CLMP lakes. Although the
proportion of the two data sources comprising the Landsat-5 and Landsat-7
datasets were not largely different, I chose to investigate whether data from the
two sources have the potential to cause the differences in r2 between full and sub-
sampled datasets. This was done under the assumption that the MDEQ data are
more reliable as they were collected by skilled-personnel. The full dataset
included more CLMP (less reliable) data so that the r2 of the dataset was lower. I
regressed 14 MDEQ lakes and 70 CLMP lakes used in the full dataset (84 lakes)
separately. The result indicated an r2 of 0.568 (p < 0.001) for CLMP lakes and
0.580 (p < 0.001) for the MDEQ lakes (Figure 4.18-4.19). The results showed an
insignificant difference in reliability between the two data sources. Quality of the

data sources was very similar and did not cause the difference in r2 results.

93

 

 

Bl3=Bl:B3

 

 

 

r2 = 0.580 (p < 0.001)

LN(SDT) = 0.739(BlzB3) + 0.038(Bl) — 3.122

Figure 4.18 Statistical analysis of 14 MDEQ lakes for ETM+ radiance

94

 

LNSDT

 

 

Bl3=Bl:B3

 

r2 = 0.568 (p < 0.001)

LN(SDT) = 0.972(Bl:B3) - 0.004(Bl) — 2.189
Figure 4.19 Statistical analysis of 70 CLMP lakes for ETM+ radiance

95

 

The characteristics of lake sample data from the two datasets (84 filll
dataset and 43 sub-sample dataset), in terms of geographic distribution, temporal
distribution, and the distribution of SDT were very similar. The reliability of data
from MDEQ and CLMP were undoubtedly comparable. The differences of these
datasets that may affected on predictability of the models were that 1) all of the 43
sub-sample data were sampled in a regular basis whereas the 84 data included
many of the irregularly sampling, and 2) the 84 sub-sample dataset included 3
samples that were taken 6-7 days prior to the image acquisition, which may have

weaken the correlation between satellite signature and ground observation data.

4.3 Conclusion
4.3.] Main hypothesis conclusion

RSatellite-based monitoring can be used to enhance inland lake water clarity/
trophic condition monitoring on a regional scale in the state of Michigan. This
approach can be used to investigate a large number of lakes using a model
developed from a small number of reliable in-situ data measurements.)

In this study, I tested the accuracy and applicability of the method developed
for a small scale (city-wide) inland lakes monitoring to a large scale (statewide)
monitoring in Michigan. The results demonstrated that I can apply a standard
regression model for SDT prediction to an extensive number and widely
distributed SDT of lakes within a large geographic region, encompassing multiple

Landsat sensors. These results extend the studies of previous researchers

96

(Khorram and Cheshire, 1985; Lathrop, 1992; Kloiber, Brezonik, and Olmanson,
2000; Dewider and Khedr, 2001) who established the statistical models for
narrower spatial and SDT distribution with a single sensor and single image. Of
all the literature reviewed, only Kloiber et al. (2000) examined this relationship on
a relatively large scale, using 47 lakes within an area of one-third of a Landsat
image. My study had much more inter-lake variation, yet my r2 values of the sub-
sample dataset were not considerably lower than those in the Kloiber study.
Previous single-lake regression studies using similar Landsat bands vary little
from my own (Lathrop, 1992; Pattiaratchi et al., 1994; Cox et al., 1998). This
study also results suggest that large numbers of ground-truth data may not be
necessary, yet good quality samples play an important role in influencing the
strength and reliability of regression models of Landsat and lake SDT. Regularly
sampled lake datasets provide the best prediction potential. {Satellite technology
has considerable potential as a cost-effective means to supplement existing
(ground-based) monitoring programs on the status and trends of water quality in

the state of Michigan. )

4. 3.2 Sub-hypothesis conclusions
l.[The radiance values exhibit stronger correlations with Secchi disc
measurements than the raw DN values.)

The results of this study have proved that the minimal radiometric

correction to convert DN to radiance significantly improved the strength of the

97

relationship between field measurement SDT and satellite spectral signatures
(Table 4.2). Although the radiometric normalization improved the strength of the
relationship between satellite data and ground observed SDT, it is noted that
images in path 21 row 31 (Figure 3.1) are still always produce low r2. This
indicates the potentially significant effects of atmospheric scattering. As shown in
Figure B-l, the southern part of ETM+ image was very cloudy, and the TM image
was notably hazy, on the acquisition date. To address this problem, atmospheric
correction may help improve the goodness of fit of the model as it subtracts the
contribution due to Rayleigh and Aerosol scattering from the images (Pattiaratchi

et a1. 1994).

2. K Both Landsat-5 TM and Landsat-7 ETM+ can be used in this
methodology and produce similar results. )

The results have shown a slight difference in prediction efficacy of the two
sensors in that Landsat-7 ETM+ produced lower r2 than Landsat-5 TM in most
cases. This may be explained by two assumptions 1) Landsat-7 has been
concluded to have higher variability with homogenous targets than Landsat-5 in
band3 (Volgelmann et al. 2001), and 2) atmospheric scattering may play an
important role because Landsat-7 images were much cloudier than the Landsat-5

images (Appendix B).

98

4.3.3 Factors affecting predictability of the models
To address the factors that have the potential to produce the best reliability

in satellite-based water quality monitoring, each data type is covered separately.

In-situ data

1. Lake SDT measurements used in this study range between 0.61 — 7.62 m,
with the majority between 3 — 4 m. This range included lakes from hyper-
eutrophic to oligotrophic, which is normal in the state of Michigan (Nelson, 2002).
In fact, this lake sample dataset is comparable to the range of SDT (n = 675)
collected by USEPA (Nelson, 2002) and should thus represent the nature
variability of lakes in the State.

2. Lakes that were sampled on a regular basis produced a better regression
result than lakes that were sampled only irregularly (page 96). Regular sampling
could indicate a higher sampling quality due to the experience of the sampler.
People who take field samples regularly may gain better skills and increase the
accuracy of measurements

3. Lakes sampled from MDEQ and CLMP produced did not produce
significantly different results in the regression models (Page 93). Both datasets
are appropriate for future studies.

4. Samples taken within i7 days from the image acquisition give good
results. However, measurements on the date further from the image acquisition

may reduce the confidence between satellite data and ground observed data (page

99

96). Samples measured on the same date to the image acquisition better represent
the trophic state and clarity of the water of the date and may help increase r2 of the

regression model.

Satellite data

1. Radiometric conversion of raw DN to radiance helps reduce variance of
residuals and thus improves the goodness of fit in the multiple regression models.

2. The cloudier images (Landsat-7 ETM+) appeared to produce lower r2
than the clearer images (Landsat-5 TM; Appendix B).

3. Although atmospheric effects can vary, even within one image scene,
images of the same path acquired from the same date can partially reduce the
atmospheric effects. However, if the images used in the study area are taken from
different dates, atmospheric correction processing is strongly recommended to
increase the strength of the relationships between ground measurements and
satellite signatures. The Landsat-5 TM images used provide an example of

atmosphere effects caused by two different acquisition dates (Appendix B).

References

Brivio, P. A., Giardino, C., and Zilioli, E. (2001). Validation of satellite data for
quality assurance in lake monitoring applications. The science of the Total

Environment, 268, 3 — 18.

100

Cox, R. M., Forsythe, R. D., Vaughan, G. E., and Olmstead, L. L. (1998).
Assessing water quality in the Catawba River reservoirs using Landsat
Thematic Mapper satellite data. Lake and Reservoir Management, 14, 405-416.

Dewidar, Kh. and Khedr, A. (2001). Water quality assessment with simultaneous
Landsat-5 TM at Manzala Lagoon, Egypt. Hydrobiologia, 457, 49 — 58.

Gordon, H. R. (1987). Calibration requirements and methodology for remote
sensors viewing the oceans in the visible. Remote Sensing of Environment, 22,
103 — 126.

Khorram, S. and Cheshire H. M. (1991). Water quality mapping of Augusta Bay,
Italy from Landsat-TM data. International Journal of Remote Sensing, 12(4),
803-808.

Kloiber, S. M., Brezonik, P. L., Olmanson, L. G., & Bauer, M. E. (2002). A
procedure for regional lake water clarity assessment using Landsat
mutispectral data. Remote Sensing of Environment 82(1), 38-47.

Lathrop, R. G. (1992). Landsat Thematic Mapper monitoring of turbid inland
water quality. Photogrammetric Engineering and Remote Sensing, 58(4), 465 -
470.

Lee, G. F., Jones-Lee, A. F ., and Rast, W. (1995). Secchi depth as a water quality
parameter. Report of F red Lee & Associates, El Macero, CA. (15 pp.).

Lorenzen, M. W. (1980). Use of chlorophyll — Secchi disk relationships.
Limnology and oceanography, 25 (2), 371 — 372.

Ma, K. M. (1997). Water quality assessment of the Chilung liver using Landsat
Thematic Mapper and Airborne Multispectral Scanner Images: Taipei, Taiwan.
MS. Thesis. Michigan State University, 80 pp.

Nelson, S. A. C. (2002). The use of remote sensing in freshwater wetland and lake
studies. PH.D. Dissertation. Michigan State University, 134 pp.

Oglesby, R. T. (1977). Relationships of fish yield to lake phytoplankton standing
crop, production, and morphedaphic factors. Journal of Fisheries Research
Board of Canada, 34, 2271-2279.

101

Pattiaratchi, C., Lavery, P., Wyllie, A., and Hick, P. (1994). Estimates of water
quality in costal waters using multi-date Landsat Thematic Mapper data.
International Journal of Remote Sensing, 15(8), 1571 -1584.

Vogelmann, J. E., Helder, D., Morfitt, R., Choate, M. J., Merchant, J. W., and
Bulley, H. (2001). Effects of Landsat 5 Thematic Mapper and Landsat 7
Enhanced Thematic Mapper Plus radiometric and geometric calibrations and
corrections on landscape characterization. Remote Sensing of Environment, 78,

55 — 70.

102

CHAPTER V

THE BIGHT OF BANGKOK WATER MONITORING

5.1 Introduction

The population of Thailand has increased over the past three decades.
Thailand is one of the world’s major food producers. Almost half of the 67
million people in Thailand make their living in agriculture. Extensive chemical
and biological fertilizers have been used in crop production and aqua-farming.
Annual algal blooms in the Gulf of Thailand are common due to substantial
amounts of nutrients flowing into the Gulf through streams and rivers.

Land-Based Pollution near the gulf of Thailand is derived mainly from
agricultural areas, aquacultural farms, industrial estates, and residential units
(International Cooperative Study of the Gulf of Thailand, 2001). Serious concern
over the environmental impacts of aqua-farming operation has increased,
especially regarding the intensive production system used in the farming of shrimp
and other high-value species. These systems use substantial amounts of food and
chemical nutrients as inputs. Discharges are heavily contaminated with algae and
chemical agents, and when flushed into coastal or river waters can lead to harmful
algal blooms and oxygen depletion.

Although the nutrient pollution problem is seen as a serious issue by

monitoring agencies, financial and labor shortfalls limit comprehensive

103

monitoring programs. Thailand recently acquired its own ground-receiving base
for Landsat satellite data. Thus, Thailand is poised to take advantage of the water
monitoring opportunities presented by remote sensing technologies to replace the
current, inefficient methods used for water monitoring in the huge area of the Gulf
of Thailand.

In this study, I applied the water clarity/ trophic state monitoring approach
developed for inland lake monitoring in the Upper Midwest Region of the United
State to the upper part of the Gulf of Thailand, the Bight of Bangkok. The method
used the relationship between ground observed Secchi disc transparency (SDT)
and Landsat signatures from bandl and band3 to predict trophic status and water

clarity for the entire Bight.

5.2 Results
5. 2. 1 Comparisons between DN and radiance images
Landsat- 7 E T M+

One Landsat-7 ETM+ image acquired on Sep 2, 2001 was processed. The
regression model based on 30 ground measurements of SDT collected on Sep 4
and satellite signatures bandl and band3 produced an r2 of 0.786 (p < 0.001) for

DN image and 0.863 (p < 0.001) for radiance image (Figure 5.1-5.2).

104

 

LNSDT

 

 

Bl3=Bl:B3

 

t2 = 0.786 (p < 0.001)
LN(SDT) = 3.390(B1 :133) + 0.143(31) — 15.052

Figure 5.1 Statistical analysis of 30 samples using data from Landsat-7 ETM+

09/02/01 DN image

 

 

LNSDT

 

 

Bl3=Bl:B3

 

r2 = 0.863 (p < 0.001)

LN(SDT) = 3.014(B1:B3)+ 0.062(B1)— 10.516

Figure 5.2 Statistical analysis of 30 samples using data from Landsat-7 ETM+

09/02/01 radiance image

106

 

Landsat-5 TM

Two Landsat-5 TM images acquired on Jul 24 and Aug 25, 2001 were
processed. The regression model from Jul 24 with 13 SDT samples on Jul 19 had
an r2 of 0.899 (p < 0.001) for both DN and radiance images (Figure 5.3-5.4). The
Aug 25 satellite signatures were regressed with 30 SDT samples from Aug 20.
The regression model had an r2 of 0.779 (p < 0.001) for DN and 0.836 (p < 0.001)

for radiance (Figure 5.5-5.6).

107

 

k
D
9
’1

 

 

Bl3=Bl:B3

 

t3 = 0.899 (p < 0.001)
LN(SDT) = 2.329(BlzB3) + 0.005(B1) — 7.830

Figure 5.3 Statistical analysis of 13 samples using data from Landsat-5 TM

07/24/01 DN image

 

 

LNSDT

 

 

Bl3=Bl:B3

 

13 = 0.899 (p < 0.001)
LN(SDT) = 2.930(B1:B3) + 0.013(B1)— 7.740
Figure 5.4 Statistical analysis of 13 samples using data from Landsat-5 TM

07/24/01 radiance image

 

 

LNSDT

 

 

Bl3=Bl:B3

 

08/25/01 DN image

r2 = 0.779 (p < 0.001)

LN(SDT) = 3.046(Bl:B3) + 0.071(B1) — 17.550
Figure 5.5 Statistical analysis of 30 samples using data from Landsat-5 TM

110

 

 

LNSDT

 

 

Bl3=Bl:B3

 

,2: 0.836 (p < 0.001)

LN(SDT) = 4.034(BlzB3) + 0.130(131) — 18.211

Figure 5.6 Statistical analysis of 30 samples using data from Landsat-5 TM

08/25/01 radiance image

 

After all three images from the period of study were processed, I performed
an experiment to apply the regression model produced from Sep 2, 2001 image
(shallow SDT season) to a Landsat-7 ETM+ image acquired on Jan 8, 2002
(expected deep SDT season). The objective of the experiment was to validate the
accuracy of the approach. An ideal validation was to check the result of predicted
TSI with an exact ground measurement location. Unfortunately, cloud-free
images were very difficult to obtain throughout the end of the year. The cloud-
free image closest to the study period was Jan 8, 2002. Under the assumption that
if the model was capable of predicting SDT in all seasons, the resulting map from
Jan 8, 2002 was expected to present low TSI, or deep SDT in the January image.
There was no statistical report for the Jan 8 image because the model had been
built upon Sep 2 data. However, the TSI map showed low TSI values for the most
of the map. Figures 5.7-5. 10 show four output TSI maps from Jul 2001- Jan 2002.

Results from the regression models show that correlation between ground
measurement SDT and satellite spectral signature of radiance images are generally
better than DN images. Although the differences were not significant, the result is
in agreement with the study of Michigan lakes in the previous chapter. The square
of the correlation coefficient (r2) from the model of the Bangkok Bight is very

high, between 0.836 — 0.899 (Table 5.1).

112

Table 5.1 Summary of the multiple regression results for the Bight of Bangkok

 

 

Date Number of DN Radiance
samples R-square R-square
09/02/01 ETM+ 30 0.786 0.863
07/24/01 TM 13 0.899 0.899
08/25/01 TM 30 0.779 0.836

 

 

 

 

 

 

*p<0.001

5. 2.2 Comparisons between Landsat-5 TM and Landsat-7 E T M +

The TSI maps (Figures 5.7-5.10) illustrate the spatial distribution of water
clarity within the Bight of Bangkok. Regression models produced fairly good r2
of higher than 0.800 for both Landsat-5TM and Landsat-7 ETM+. Trophic State
Index models from both sensors were effective in identifying the “hot spot” areas
near the mouths of the rivers, the industrial estates and the aquaculture area
(Figure 5.11). In addition, the application of model from Sep 2, 2001 to Jan 8,
2002 also showed the clearer water in Jan 8 as expected. The water trophic/

clarity condition are known to be low at the end and the beginning of a year.

113

 

 

 

 

 

 

_TSI SDT (m)
- 0-30 8

31-40 4
! 41-50 2
51-60 1

0.5

61-70
71-80 0.25
-81-100 0.125

     

 

 

 

 

Figure 5.7 TSI map from Landsat-5 TM 07/24/01

114

 

 

 

 

 

 

 

 

 

  
      
 

_TSI SDT(m)
- 0-30 8
3140 4
E4150 2
51-60 1
61-70 0.5
71-80 0.25
-81-100 0.125

 

 

Figure 5.8 TSI map from Landsat-5 TM 08/25/01

115

 

 

 

 

 

 

 

 

 

 

 

   
 
   
 
     

_ TSI

- 0-30
51-60
61-70

531-40
71-80

41-50
-81-100

SDT (m)
8
4
2
1
0.5
0.25
0.125

 

 

Figure 5.9 TSI map from Landsat-7 ETM+ 09/02/01

 

 

 

 

 

 

 

 

 

TSI

- 0-30
3140
4150

1:1 51-60
61-70
71-80

-81-100

SDT (m)
8

4

2

1
0.5
0.25
0.125

 

 

Figure 5.10 TSI map fi'om Landsat-7 ETM+ 01/08/02

117

 

 

 

Chao Praya Bang Pakong
River Aquaculture River

 

 

 

Tourism area

 

Figure 5.11 Example of land use near the Bight of Bangkok
5.2.3 Seasonal changes of Secchi disc transparency, Chlorophyll a, and suspended
sediment

Ground observation data were studied closely for a better understanding of
each water quality parameter. Figure 5.12 presented the approximate locations of
the two sample sites within the Bight of Bangkok. Sample site 1 was close to a

tourism area, Bang Saen, with clearer water (deeper SDT). Sample site 2 was

118

located near the mouth of Bang Pakong river and an aquacultural zone where

water was normally more turbid (shallower SDT).

 

 

 

 

 

Figure 5.12 Ground sample sites

Figure 5.13-5.15 shows the trend of SDT, Chlorophyll, and suspended
sediment (SS) changes through time. SDT (Figure 5.13) was shallow in late July
through August because this period was the beginning of rainy season in Thailand.
Tremendous loads of chemical nutrients and SS flowed into the upper Bight of
Bangkok through rivers. SDT increased from late October through the end of the

year as the water was clearer.

119

Chlorophyll a (Figure 5.14) increased in response to the accumulation of
nutrients at the beginning of the wet season. Chlorophyll concentrations increase
from July through the middle of September, when it reaches the seasonal season
peak, then declines toward the end of the year.

The suspended sediment trend (Figure 5.15) for sample site 2 matched well
with the Chlorophyll a trend. Sample site 1 did not behave in the same way. A
possible reason is because the site was farther away from the mouth of the river
and a large amount of SS had sunk to the bottom of the sea before it reached the

sample area.

120

 

 

0.5

Secchi Depth (m)

 

Secchi Depth Slto1
19Jul Augzo Sep4 Sep20 Oct6 Oct22 Nov4 Nov23

1 1
Y I

 

«i—

J
T

-0-

_ -__.___ —_.

 

 

 

 

 

 

 

 

 

 

 

 

 

Secchi Depth (m)

 

1.5 —’
I1. s1.820.1,900

2 ‘T‘“ ‘ T— T T ‘ " 1.900

2.5
Date
Secchi Depth $1192
19 Jul Aug 20 Sep4 Sep 20 Oct6 Oct22 Nov4 Nov 23
O i i i i i 1 1

 

 

 

 

 

 

Date

 

 

 

 

Figure 5.13 Seasonal changes of Secchi disc transparency in the study area

121

 

 

 

Chlorophyll a Slte‘l

 

 

 

 

 

 

 

4
. 35r——** 2 -———— —— —
i
a 2 __ 22___ _ . ,_ _
5 1.687 11.608
‘

o 1 2.----. 0.187 fl
0.320
0.760
0 0.383' 0.3%21 % % 10.130 r If

 

Jul19 A0920 Sep4 Sep20 OctG Oc122 Nov4 Nov23
Date

 

 

 

 

Chlorophyll a Sitez

 

 

 

 

 

 

 

4
.322 2_ 2 2 22
a 12.362
32“ 2 ----—- .2 2mm ~ ~—*~-— -~— —
5 1 1.300

.714
0.371 0.131
0 l i I r 1 1

AugZO Sep4 Sep20 Oct6 Oct22 Nov4 Nov23
Date

Jul 19

 

 

 

 

 

 

Figure 5.14 Seasonal changes of Chlorophyll a in the study area

122

 

 

Suspended Solid Slte1

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

0.03

$0.021 — ~ L-,- ————— ————— -—————
0.016 0.015
002 I _ 0.013 _ _ 1 .
0 01 _ 1 711114
' 0.013 0.013 0.010

,3 0.01 ~——-- — ——- - —— ——— — ——

0.00 1 1 r r 1 i

Jul 19 Aug 20 Sep4 Sep20 Oct6 Nov4 Nov 23
Date
Suspended Solid 81192

0.06
g0.051- — m- -- — — — ~—
13 0.04 -
E 0.03 ~
8. 0.02 -
3
In 0.01 ~

0.00 i 1 1 1 1 1

Jul 19 Aug 20 Sep 4 Sep 20 Oct6 Nov4 Nov 23
Date

 

 

 

Figure 5.15 Seasonal changes of suspended sediments in the study area

123

 

 

 

 

5.2.4 Correlation between Secchi disc transparency, Chlorophyll a, and
suspended sediment

The approach for water clarity/ tr0phic condition monitoring I used in the
study was originally developed for freshwater lake monitoring (Kloiber et al.
2002). Typically, turbidity in a lake was caused by algal biomass; therefore, SDT
generally represented trophic status of the water (Khorram and Cheshire, 1991).
However, this assumption may not hold valid in a volatile environment such as the
Bight of Bangkok. 1 used SYSTAT (SPSS Software, Inc., 2001) to analyze
statistical correlations between three water quality parameters collected from the

field sites; SDT, Chlorophyll a, and SS.

Table 5.2 Correlations among Secchi disc transparency, Chlorophyll a, and

suspended sediment

 

 

Sample date SDT/Chlor0phyll a SDT/SS SS/Chlorophyll a
Jul 19 0.359 0.243 0.075
Aug 20 -0539 -0.865 0.727
Sep 4 0.629 -0.800 0365
Sep 20 -0501 -0775 0.240

 

 

 

 

 

 

The results indicated more interaction between SDT and SS than between

SDT and Chlorophyll a.

124

5. 2.5 Changes in water clarity/ trophic status

Maps in Figure 5.16 were produced by subtracting TSI values for each
pixel of the second date image from the first date image. The maps showed how
TSI changes, spatially, over time. These thematic maps, if processed continuously
every 8 or 16 days as the overpass cycle of, or between, Landsat-5 and Landsat-7
satellites, could be useful for monitoring agencies to predict where eutrophication
is likely to occur. The change during September to January was negative, which
means that TSI decreased. These results matched trends of SDT, chlorophyll a,

and SS from ground observation.

125

 

   

Jul - Aug Aug - Sep

 

- TSI Change
0-5
6-10
11-15
16-20
21-25
26-30
31-35

1 Iflifll

 

 

 

 

Sep - Jan

 

 

 

Figure 5.16 TSI change maps

126

5.2.6 Integrated T SI value

In order to conclude whether the model output actually produces TSI, I
summed up the number of pixels within each TSI class to evaluate how the
magnitude of the peak of TSI changed by season and to use as a statistical proof
(Table 5.3; Figure5.17). Number of pixels decreased in the lower TSI classes
from Jul — Sep, but the number of pixels increased in the higher TSI classes. For

January of the next year, the TSI characteristics changed back to a level similar to

July as they are both clear seasons.

Table 5.3 Integrated TSI value

 

 

 

 

Number of pixel
C1355 TM image TM image ETM+ image ETM+ image
07/24/01 08/25/01 09/02/01 01/08/02
1 — 20 2529905 749726 279046 2486250
21 - 40 4162156 3382172 2871148 5619375
41 - 60 3446050 4903477 7194680 2604375
61 — 80 354691 1657466 578264 354375
81 — 100 75645 191634 43706 50625
> 100 0 15129 0 0

 

 

 

 

 

127

 

 

Integrated TSI

8.000.000

7.000.000 200‘
2001

6.000.000 2001

5,000,000 I 8-Jan 2002
4,000,000
3,000,000
2,000,000

Total number of pixels

1 ,000,000

 

0

1-20 21-40 41-60 61-80 81-100 >100
TSI

 

 

 

Figure 5.17 Integrated TSI trend

5.3 Conclusion
5.3.] Main Hypothesis conclusion

(Satellite-based monitoring can be used to enhance tropical marine water
clarity/ trophic condition monitoring for the entire Bight of Bangkok. This
approach can be used to investigate a large semi-open sea area across widely
distributed characteristics of water using a model developed from a small number
of in-situ measurements. l

Satellite-based monitoring on the Bight of Bangkok generated a broad

geographic view with fine spatial resolution. Change detection and hot-spot

128

monitoring of the whole area is achievable within a very short time, allowing the
possibility to identify a problem before it causes extensive damage to the sensitive

environment.

5.3.2 Sub-hypothesis conclusions
l1. Radiance images also produced a better r2 than the DN image for the
Bight of Bangkok.)

The differences of r2 results are not substantial because water in the bight is
very turbid in some area, especially near the sample sites, which thus reflect strong

signal to satellite sensors. Thereby, r2 of DN images were already high (Table

5.1).

2.11Both TM and ETM+ sensors can be used, and produced a similar result
for SDT predictionl

The two sensors were capable of capturing a tight correlation between the

satellite and in-situ data.

3. This standard regression model for water quality monitoring developed
for freshwater inland lakes is applicable to the open sea of the upper Gulf of
Thailand;

Statistical r2 results indicated a very high capability r2 > 0.830 (p < 0.001)

to predict SDT from Landsat imagery. The standard regression model developed

129

on this study can be used effectively to determine the clarity and trophic condition
of water in the Bight of Bangkok, even though the environmental characteristics of
the Bight of Bangkok and Michigan inland lakes are considerably different. Algae
cells are the main cause of lake turbidity in the Upper Great Lakes Region but
both algae and sediment are the primary sources of turbidity in the Bight of
Bangkok. Both algae and sediment cause undesirable water conditions in
Thailand. Suspended sediments and cloudy water affects aesthetic value and
impacts aquatic organisms. As this study attempted to predict SDT, water clarity
and trophic state, the suspended sediment data proves usefill in understanding the

status of the volatile aquatic environment found in the Bight of Bangkok.

5. 3. 3 General conclusions

1. Changes in TSI through time followed the expected trend. TSI was very
low at the beginning of the rainy season (Jun — Jul). From late July through
September, TSI increased with the eventual appearance of an algal bloom.
Following this, TSI declined to the normal clear stage in late November.
Integrated TSI has verified the ability of this model to capture this natural trend.

2. The standard regression model was capable of predicting the trophic
condition in January from satellite imagery even though field observations had not
been collected. The prediction agreed with the Chlorophyll trend and the

integrated TSI in that January trophic status should be similar to that of July. This

130

 

experiment also indicated that the standard regression model was sensitive to TSI
changes through time.

3. The models were valid in capturing the “hot spots” such as industrial
estates area, aquaculture zone, and the tourism areas. This should prove useful in
monitoring sources of eutrophication problem (Figure 5.7-5.10).

4. The models used in this study captured the interaction between satellite
signatures and SDT very well. However, there are more interactions between SDT
and SS than between SDT and Chlorophyll in the Bangkok Bight. Therefore, the
signatures of chlorophyll may be weakened by suspended sediments in the water.
Hence, until the method is improved, caution should be used when translating the
TSI into algal productivity in the water.

5. In this study, only one Landsat image was used; therefore, atmospheric
effects had only minimal influence on the prediction. However, atmospheric
correction may have the potential to improve the r2. My suggestion is that, if the
model is to be used across a wider geographic region, atmospheric correction
should be performed.

6. If the model is to be used for the purpose of quantitative comparison
across time, atmospheric correction is recommended. Atmospheric scattering has
the most impact at a wavelength of 550 nm, Landsat bandl (Mackay et al. 1998).
Since the spectral signature of bandl is used in the regression model, atmospheric
correction may have the potential to reduce the errors between sensor and ground

surface reflectance.

l3l

References

International Cooperative Study of the Gulf of Thailand. “About the Gulf of
Thailand.” IOC- WEST PA C and SEA START RC. Jun 2001. 25 Aug 2002,
<http://www.start.or.th/got/gotwww/index.htm>

Khorram, S. and Cheshire H. M. (1991). Water quality mapping of Augusta Bay,
Italy from Landsat-TM data. International Journal of Remote Sensing, 12(4),
803-808.

Kloiber, S. M., Brezonik, P. L., Olmanson, L. G., and Bauer, M. E. (2002). A
procedure for regional lake water clarity assessment using Landsat
mutispectral data. Remote Sensing of Environment 82( 1), 38-47.

Mackay, G. and Steven M. D. (1998). An atmospheric correction procedure for the
ATSR-2 visible and near-infrared land surface data. International Journal of

Remote Sensing, 19(15), 2949 — 2968.

132

CHAPTER VI

RESEARCH CHALLENGES AND FUTURE NEEDS

During the research undertaken for this thesis, I have encountered several
challenges which lead to suggestions for further research to improve our capacity
to monitor water quality using satellite remote sensing data.

1. Cloud-free images were very difficult to obtain during the study period,
in both case studies, Michigan and Thailand. This is a disadvantage of Landsat
satellites, indicating that the potential for satellite remote sensing as an
independent monitoring tool in areas with a high probability of cloud cover could
be limited. It is apparent that in these {areas of high cloud cover, satellite remote
sensing may not be able to replace current monitoring techniques. Yet it does
provide a supplementary tool that can reduce the cost of over-all monitoring
programs.)

2.( Other satellite sensors that have a more frequent image acquisition
should be explored in future research as they have the potential to solve the cloud

:1) if.- ; 'l""'rl/I rah/4,1
problem.) However, there is a trade off between spatial and temporal resolution,

A

 

for instance, a daily temporal resolution satellite as SPOT Vegetation has a very
coarse pixel resolution of 1 km. New sensors that are able to penetrate clouds yet
still acquire a water signature may be another solution. New models are worth

exploring to develop new approaches for water quality monitoring.

133

 

3. Atmospheric correction can be useful. Unfortunately, some data needed

for processing the atmospheric model (5s) were not obtainable during the study.

I The advent of new sensors such as MODIS, which have detectors in atmospheric

spectral regions may provide direct measurements of atmospheric scattering and
absorption. /

{4. To use remote sensing for statewide assessments of water clarity, there is
still a requirement for statewide field sampling that coincides with satellite over-
flights.) Field data collected simultaneously with satellite image acquisition can
produce a better result-in prediction. Also, even though this study had 123 ground
sample data, only a portion of these were of use in the model because many lakes
were not sampled regularly and or reliably.

5 Most of the ground observation data used in the study did not have
geographical reference location./. Although my sample locations (AOIs) were
selected based. on a standard theory that water samples are collected from the
deepest basins of the water bodies, there is a possibility that the theories were not
actually followed by the samplers in this situation. Consequently, there is some
uncertainty in locatinngth‘e pixel corresponding to the in-situ location.

6. I experienced {difficulties in creating a single regression relationship from
images from multiple dates. / Further research is necessary to explore the effect of
path-to-path atmospheric variations.)

7. The water column contains a mixture of dissolved organics, inorganic

suspended sediments, and chlorophyll, which mask and interfere with the spectral

134

identification of the chlorophyll alone. [Further study is necessarily to develop the
regression model that can effectively predict trophic condition of waters where

non-algal substances play an important part in turbidity.)

135

APPENDICES

136

 

APPENDIX A

Table A-1 Detailed information for the 123 ground observation data from MDEQ

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

and CLMP
Lake Name County Sampler Date Secchi Date Secchi2
Sample1 (m) Sample2 (m)
Arbutus Lake Grand Trave CLMP 26-Aug 4.88 2-Sep 5.79
Arnold Lake Clare CLMP 264% 4.88 2-Sep 4.88
Austin Lake CLMP 264% 4.42
Avalon Lake Montmorency CLMP 26-Aug 5.18 2-Sep 6.71
Baldwin Lake Montcalm CLMP 26-Aug 5.18 2-Sep 3.35
Barlow Lake Barry CLMP 26-Aug 2.74
Base Line Lake Livingston CLMP 26-Au 4.42 5-AUL 4.11
Bass Lake Kent CLMP 26% 2.74 2-Sep 2.59
Bear Lake Kalkaska CLMP 26-Aug 7.62 2-Sep 8.38
Beaver Lake Alpena CLMP 264% 3.91 2-Sep 4.09
Bergess CLMP 26-AUL 2.60
Big Bradford Lake Otsego 26-Aug 5.49 29-Jul 5.79
Big Lake Osceola MDEQ 20-Au 2.51
fig Lake Osceola CLMP 26-Aug 3.35 2-Sep 3.35
Bills Lake Newalgo CLMP 26-Aug 2.74 2-Sep 2.74
Birch Lake Cass CLMP 26-Aug 5.18 5&9 3.66
Blanch Lake MDEQ 21-Au 1.83
Blue Lake Mecosta MDEQ 21-Agg 3.35
Blue Lake Mecosta CLMP 26-AuL 3.96 2-Sep 4.57
Bostwick Lake Kent CLMP 26-Au 1.76
Brooks Lake Newaygo CLMP 26-Aug 0.91
Burkhart Lake Washtenaw CLMP 26-Aug 4.60
Byram Lake Genesee CLMP 26-Aug 3.35
Camelot Lake Isabella CLMP 26-Aug 2.90
Campau Lake Kent CLMP 26-Aug 2.44
Cedar Lake Alcona\losc CLMP 26-Aug 2.98 2-Sep_ 2.98
Cedar Lake Van Buren CLMP 26-Aul 4.42 5-Aug 3.35
Center Lake MDEQ 27-Au 2.59
Christiana Lake Cass CLMP 26-Au 2.59 5qu 1.98
Clear_Lake Berrien CLMP ZG-fl 4.27
Clear_Lake Jackson CLMP 26-Aug 2.74
Clifford Lake Montcalm CLMP 26-Aug 3.66 2-Sep 3.66
Coldwater Lake Branch CLMP 26-AUL 1.83 5-AuL 1.07
Coon Lake Livinflm CLMP 26-AUL 2.26
CoreLLake St. Joseph CLMP 26-AAL 3.51 5-AUL 2.59
Cowan Lake Kent CLMP 26-Au 1.98
Cranberry Lake MDEQ 21-fl 2.59
Crockery Lake Ottawa CLMP 26-Aul 1.47 2-Sep 1.98
Crooked_Big Lake Alcona CLMP 26-Aug 4.11 5-Aug 5.03
Croton Dam Pond MDEQ 28-Au1 1.83
Cfistal Lake Hillsdale CLMP 26-Aug 4.72

 

 

 

 

 

137

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Lake Name County Sampler Date Secchi Date Secchi2

Sample1 (m) Sample2 (In)

Cub Lake Kalkaska CLMP 26-Aug 6.71 2-Sep 6.10

Derby Lake Montcalm CLMP 26-Aug 6.40 2-Sep 5.79

Devils Lake Lenawee CLMP 9-Sep 3.66

Diamond Lake Cass CLMP 26-Aug 3.05 Sflg 1.68

Donnell Lake Cass CLMP 2-Sep 2.74 5-Algg 1.68

Eagle Lake Allegan 8 V CLMP 26-Aug 4.42 5-Aug 4.57

East Twin Lake Montmorency CLMP 26-Aug 3.81 2-Sep 2.59

Evans Lake Lenawee CLMP 26M 5.79

Fair Lake Barry CLMP 26qu 4.33

Fish Lake Van Buren CLMP 26qu 2.59 5-Aug 3.35

Freska Lake Kent CLMP 26%ch 3.26

George Lake Clare CLMP 26-AUL 2.74 2-Sep 2.13

Goose Lake MDEQ 22-Aug 3.05

Grass Lake MDEQ 2%qu 1.83

Hamilton Lake Dickinson CLMP 26-Aug 4.57 2-Sep 4.27

Hardy Darn Pond MDEQ 27-Au 3.66

Harper Lake Lake CLMP 26-AUL 4.62 2-Sep 4.78

Hess Lake Newaygo CLMP 26.0% 0.61 2-Sep 0.46

Hicks Lake MDEQ 22-Aug 1.07

Higgins Lake Roscommon MDEQ 24-Aug 4.98

Higgins Lake Roscommon CLMP 2-Sep 6.40

Horsehead Lake Mecosta CLMP 26-Aug 2.51 8-Jan

Hubbard Lake Alcona CLMP 26-Aug 5.49 2-Sep 5.94

Hutchins Lake Allegan CLMP 26-quk 3.38 9-Sep 3.28

Indian Lake MDEQ 15-Ag 1.52

Jehnsen Lake MDEQ 22-Au 2.13

Jordan Lake Barry CLMP 26-Aug 1.27 2-Sep 1.07

Juno Lake Cass CLMP 26-Aug 2.59 5-Aug 1.83

Kettle Lake Kent CLMP 26-Au 3.05 2-Sep 2.74

Klinger Lake St. Joseph CLMP 26-Aug 2.90 5-Am 2.74

KP Lake Crawford CLMP 26-AUL 3.35 2-Sep 3.35

Lake Lansing lngham CLMP 26-Aug 1.65 2-Sep 1.52

Lake Manrethe Crawford CLMP 26-Aug 3.05 2-Sep 3.05

Lake Mecosta Mecosta MDEQ 21-AU 3.12

Lake Mecosta Mecosta CLMP 9-Sep 4.27

Lake of the Woods Van Buren CLMP 26-Aug 3.10 5-ALLL 3.51

Lake Sapphire Missaukee MDEQ 27-Aug 1.83

Lake Sapphire Missaukee CLMP 26-Aug 2.29 2-Sep 2.29

Leisure Lake Shiawassee CLMP 26-Aug 3.10

Lilly Lake Clare CLMP 26-Aug 3.20

Little Pine Island Lake Kent CLMP 26-Aul 2.13

Little Twin Lake Kalkaska CLMP 26-Aug 2.62

Little Whitefish Lake MDEQ 29-Aug 3.96

Londo Lake MDEQ 14-Aug 2.21

Long Lake Iosco CLMP 9-Sep 3.18

Long Lake Iosco MDEQ 28-fllg 4.88

Loon Lake MDEQ 14-Aug 4.27

Mary Lake Dickinson CLMP 26-AUL 4.88 2-Sep 4.88

Mud Lake MDEQ 28-AglL 1.07

Murray Lake Kent CLMP 26-Aug 2.90 9-Sep 2.74

North Lake Alcona CLMP 26qu 4.57 2-Sep 5.49

Osterhout Lake Allegan CLMP 26-Aug 2.13 5-AUL 2.44

 

138

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Lake Name County Sampler Date Secchi Date Secchi2

Sample1 (m) Sample2 (m)

Painter Lake Cass CLMP 26-Aug 1.68

Pardee Lake Liviggston CLMP 26-Aug 3.90

Paw Paw Lake Berrien CLMP 26-Aug 2.71 5-AILL 1.65

Perch Hillsdale CLMP 26qu 2.13

Pleasant Lake St. Joseph CLMP 26-Aug 4.88 2-Sep 4.57

Portage Lake MDEQ 27-Aug 2.74

Reeds Lake Kent CLMP 26-Aug 1.49 2-Sep 1.25

Robinson Lake Newaygo CLMP 26-Au 2.74 2-Sep 3.05

Round Lake Clinton CLMP 26qu 3.20

Round Lake Mecosta CLMP 26-Aug 3.51 2-Sep 3.51

Sage Lake Ogemaw CLMP 26-AUL 3.51 2-Sep 3.81

Sand Lake MDEQ 21-Aug 2.74

Shingle Lake Clare MDEQ 27-Aug 3.05

Shingle Lake Clare CLMP 26-Aug 3.66 2-Sep 3.05

Silver Lake Livingston MDEQ 26-AUL 3.96

Spider Lake Grand Trave CLMP 26-Aug 3.96 9-Sep 4.57

Stone Ledge Lake Wexford CLMP 26-Aug 3.35

Sunrise Lake MDEQ 27-Aug 2.44

Townline Lake MDEQ 21-Aug 3.35

Van Etten Lake Iosco CLMP 26-Aug 0.91 2-Sep 0.91

Vaughn Lake Alcona CLMP 26-Aug 1.98

Viking Lake Otsego CLMP 26-A_ug 3.05

Vineyard Lake Jackson CLMP 26-AUL 3.35 5-Aug 2.29

Wells Lake Osceola CLMP 26-Agg 3.66 2-Sep 3.20

West Londo Lake Iosco MDEQ 14-ALLtL 2.44

West Londo Lake Iosco CLMP 26-A;ug 2.49 2-Sep 3.07

West Twin Lake Montmorency CLMP 26-AugL 3.96

Whitefish Lake MDEQ 25% 4.11

Winfield Lake MDEQ 29-AUL 2.13

Woods Lake Kalamazoo CLMP 26-Aug 2.44

 

139

 

 

APPENDIX B

Landsat-7 ETM+ (left) path 21 row 29, 30, and 31 from 08/25/01 and
Landsat-5 (right) TM path 21 row 29&30 from 09/05/01 and row 31 from

07/04/01.

 

 

 

Figure B-l Pictures of ETM+ and TM images used in the study

140

APPENDIX C

Data used in the regression model

Table C-l Data of 84 lakes used in Landsat-7 ETM+ DN model

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Name Band1 Band3 LN(SDT)
Arnold Lake 58.524 25.714 1.584
Austin Lake 58.244 25.610 1.486
Avalon Lake 73.529 28.244 1.645
Baldwin Lake 58.188 25.250 1.297
Barlow Lake 62.853 26.721 1.009
Bass Lake 60.250 26.607 1.009
Bear Lake 61.382 24.311 2.031
Beaver Lake 67.705 28.641 1.364
Bergess Lake 57.923 25.885 0.956
Big Lake 59.421 26.609 1.210
Bills Lake 66.324 27.824 1.009
Blue Lake 60.125 26.008 1.377
Bostwick Lake 63.540 30.067 0.568
Brooks Lake 62.117 31.312 -0.089
Burkhart Lake 55.500 24.833 1.527
Byram Lake 60.378 28.014 1.210
Camelot Lake 59.833 32.167 1.063
Campau Lake 60.194 27.463 0.891
Cedar Lake 62.940 26.120 1.486
Center Lake 60.621 26.931 1.009
Center Lake SW 58.674 28.696 0.956
Clear Lake_Berrien 63.276 26.828 1.451
Clear Lake_Jackson 58.797 26.234 1.009
Clifford Lake 59.784 26.440 1.297
Coldwater Lake 70.943 35.146 0.604
Coon Lake 59.370 29.439 0.758
Cowan Lake 58.211 26.737 0.684
Crockery Lake 61.234 28.447 0.387
Crooked_Big Lake 64.108 27.559 1.415
Crystal Lake 59.421 26.711 1.553
Derby Lake 60.237 25.695 1.856
Diamond Lake 74.301 32.551 1.114
Eflle Lake 60.290 25.275 1.486
East Twin Lake 63.563 28.881 1.338
Evans Lake 58.800 26.550 1.756
Fair Lake 57.789 24.752 1.465
Fish Lake 59.750 26.500 0.952
Freska Lake 59.625 25.667 1.182

 

141

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Name Band1 Band3 LN(SDT)
George Lake 59.213 26.766 1.009
Grass Lake 58.733 29.433 0.588
Hardy Dam Pond 59.287 26.350 1.308
Hess Lake 52.684 25.028 -0.495
Hicks Lake 62.476 33.452 0.095

_liiggins Lake 72.549 27.984 1.756
Horsehead Lake 61.896 28.493 0.922
Hubbard Lake 68.754 30.859 1.702
Hutchins Lake 63.543 27.743 1.217
Indian Lake 60.162 26.858 1.297
Jordan Lake 61.181 29.181 0.240
Juno Lake 67.128 30.994 0.952
Kettle Lake 60.688 26.500 1.114
Klinger Lake 70.424 31.390 1.063
KP Lake 59.163 26.041 1.210
Lake Lansing 61.047 30.953 0.498
Lake of the Woods 61.848 26.428 1.131
Leisure Lake 62.634 29.195 1.131
Lilly Lake 58.337 26.221 1.163
Little Pine Island Lake 62.476 29.119 0.729
Little Twin Lake 60.500 26.433 0.964
Mud Lake 58.130 30.435 0.095
Murray Lake 61.242 26.736 1.063
Osterhout Lake 60.806 26.478 0.758
Painter Lake 66.727 30.673 0.517
Pardee Lake 57.583 27.194 1.361
Paw Paw Lake 62.353 26.848 1.063
Perch Lake 58.257 26.771 0.758
Pleasant Lake 61.151 29.698 0.271
Portage Lake 60.710 30.376 0.993
Reeds Lake 65.250 31.350 0.401
Round Lake 65.490 28.204 1.163
Round Lake_Clinton 59.529 27.912 1.042
Round LakLMecosta 58.462 25.354 1.254
Sage Lake 65.076 32.038 1.254
Sand Lake 60.774 25.161 0.993
Silver Lake 63.138 30.770 1.377
Stone_Ledngake 59.190 26.381 1.210
Sunrise Lake 59.091 26.091 0.875
Townline Lake 58.606 25.788 1.224
Van Etten Lake 70.996 38.619 -0.089
Viking Lake 63.069 31.207 1.114
Vineyard Lake 49.067 16.515 1.210
Wells Lake 60.045 26.682 1.297
West Twin Lake 64.763 29.123 1.377
Winfield Lake 60.100 27.367 0.742
Woods Lake 61.750 28.125 0.891

 

 

142

 

‘.- “~' e_-'_-
H.153-

 

Table C-2 Data of 84 lakes used in Landsat-7 ETM+ radiance model

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Name Band1 Band3 LN(SDT)
Arnold Lake 38.711 10.342 1.584
Austin Lake 38.583 10.472 1.486
Avalon Lake 50.379 12.024 1.645
Baldwin Lake 39.475 10.675 1.297
Barlow Lake 41.986 11.068 1.009
Bass Lake 40.104 11.122 1.009
Bear Lake 40.971 9.576 2.031
Beaver Lake 45.682 12.064 1.364
Bergess Lake 38.321 10.679 0.956
gig Lake 39.415 11.008 1.210
Bills Lake 44.697 11.636 1.009
Blue Lake 39.951 10.656 1.377
Bostwick Lake 42.467 13.042 0.568
Brooks Lake 41.415 13.939 -0.089
Burkhart Lake 37.000 9.833 1.527
Blram Lake 40.238 11.746 1.210
Camelot Lake 39.700 14.050 1.063
Campau Lake 39.944 11.408 0.891
Cedar Lake 42.039 10.718 1.486
Center Lake 40.100 11.133 0.935
Center Lake SW 38.855 12.188 0.956
Clear Lake_Berrien 42.414 11.121 1.451
Clear Lake_Jackson 38.971 10.824 1.009
Clifford Lake 39.657 10.900 1.297
Coldwater Lake 48.289 16.317 0.604
Coon Lake 38.833 11.786 0.809
Cowan Lake 38.583 11.139 0.684
Crockery Lake 40.898 12.286 0.387
Crooked_BilLake 42.859 1 1.489 1 .415
Crystal Lake 39.125 10.575 1.553
Derby Lake 40.098 10.541 1.856
Diamond Lake 50.102 14.663 1.114
Eagle Lake 40.100 10.200 1.486
East Twin Lake 42.785 12.558 1.338
Evans Lake 38.946 10.946 1.756
Fair Lake 38.139 9.806 1.465
Fish Lake 39.773 11.000 0.952
Freska Lake 39.565 10.391 1.182
George Lake 39.258 11.169 1.009
Grass Lake 38.894 12.675 0.588
Hardy Dam Pond 39.311 10.867 1.308
Hess Lake 45.915 22.562 -0.495
Hicks Lake 41.810 15.286 0.095
Higgins Lake 49.865 11.916 1.756
Horsehead Lake 41.169 12.046 0.922

 

143

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Name Band1 Band3 LN(SDT)
Hubbard Lake 46.450 13.375 1.702
Hutchins Lake 42.517 11.622 1.217
Indian Lake 39.980 11.094 1.297
Jordan Lake 40.810 12.441 0.240
Juno Lake 45.211 13.600 0.952
Kettle Lake 40.500 10.786 1.114
Klinger Lake 47.916 13.848 1.063
KP Lake 39.240 10.880 1.210
Lake Lansing 40.764 13.688 0.498
Lake of the Woods 41.217 10.957 1.131
Leisure Lake 42.125 12.650 1.131
Lilly Lake 38.576 10.753 1.163
Little Pine Island Lake 41.610 12.463 0.758
Little Twin Lake 41.133 11.767 0.964
Mud Lake 38.348 13.174 0.095
Murray Lake 40.920 11.080 1.063
Osterhout Lake 40.683 10.857 0.758
Painter Lake 44.871 13.306 0.517
Pardee Lake 38.182 11.515 1.361
Paw Paw Lake 41.594 11.152 1.063
Perch Lake 38.556 10.917 0.758
Pleasant Lake 40.851 13.106 0.271
Portage Lake 40.372 13.032 0.993
Reeds Lake 43.858 13.967 0.401
Round Lake 44.143 11.816 1.163
Round Lake_Clinton 39.667 11.727 1.042
Round Lake_Mecosta 38.773 10.303 1.254
qug Lake 43.877 14.487 1.254
Sand Lake 40.438 10.063 0.993
Silver Lake 42.382 13.640 1.377
Stone_Legge Lake 39.333 10.714 1.210
Sunrise Lake 39.031 10.594 0.875
Townline Lake 38.710 10.387 1.224
Van Etten Lake 48.286 18.313 -0.089
Viking Lake 42.167 13.933 1.114
Vineyard Lake 41.040 11.972 1.210
Wells Lake 39.682 10.955 1.297
West Twin Lake 43.369 12.342 1.377
Winfield Lake 40.000 11.233 0.742
Woods Lake 41.222 11.889 0.891

 

 

 

 

144

 

Table C-3 Data of 33 lakes used in Landsat-5 TM row 29 and 30 DN model

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Name Band1 Band3 LN(SDT)
Arnold Lake 49.527 15.782 1.584
Avalon Lake 56.549 15.774 1.903
Baldwin Lake 51.957 16.872 1.210
Bass Lake 52.315 15.978 0.952
Bear Lake 51.564 14.470 2.126
Beaver Lake 52.010 16.569 1.409
_B_ig Lake 51.930 17.561 1.210
Bills Lake 58.771 18.457 1.009
Blue Lake 52.500 16.992 1.520
Clifford Lake 52.296 17.024 1.297
Crockery Lake 53.826 17.739 0.684
Derby Lake 53.578 16.822 1.756
East twin Lake 50.164 16.684 0.952
George Lake 50.288 17.561 0.758
Hardy Dam Pond 52.007 16.523 1.297
Hess Lake 61.324 34.744 -0.783
Higgins Lake 58.869 14.545 1.856
Hubbard Lake 51.401 15.450 1.782
Hutchins Lake 53.591 17.497 1.187
Jordan Lake 54.088 20.637 0.065
KP Lake 49.788 15.247 1.210
Kettle Lake 50.943 16.142 1.009
Lake Lansing 50.353 18.141 0.421
Murray Lake 52.596 17.020 1.009
Pleasant Lake 49.800 16.867 1 .520
Portage Lake 48.856 16.308 1.009
Reeds Lake 55.905 20.211 0.223
Round Lake_Mecosta 51.270 16.649 1.254
Sage—Lake 47.476 14.427 1 .338
Sunrise Lake 50.028 15.722 0.891
Van Etten Lake 55.104 24.343 0089
Wells Lake 52.500 17.711 1.163
Winfield Lake 52.697 18.485 0.758

 

 

145

-.‘.J‘_" u_- _-

I

W... v:

 

 

Table C-4 Data of 33 lakes used in Landsat-5 TM row 29 and 30 radiance model

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Name Band1 Band3 LN(SDT)
Arnold Lake 27.500 10.620 1.584
Avalon Lake 32.040 10.848 1.903
Baldwin Lake 29.327 11.857 1.210
Bass Lake 29.737 11.585 0.952
Bear Lake 29.100 9.959 2.126
Beaver Lake 29.504 11.524 1.409
Big Lake 29.299 12.569 1.210
Bills Lake 33.622 13.324 1.009
Blue Lake 29.815 11.919 1.520
Clifford Lake 29.682 12.068 1.297
Crockery Lake 30.455 12.691 0.684
Derby Lake 30.147 11.721 1.756
East twin Lake 28.277 11.923 0.952
George Lake 28.237 12.619 0.758
Hardy Dam Pond 29.465 11.435 1.297
Hess Lake 34.903 26.428 -0.783
flggins Lake 33.435 9.919 1.856
Hubbard Lake 28.984 10.499 1.782
Hutchins Lake 30.416 12.474 1.187
Jordan Lake 30.771 14.729 0.065
KP Lake 27.852 10.444 1.210
Kettle Lake 29.059 11.353 1.009
Lake Lansing 28.216 13.281 0.421
Murray Lake 29.842 12.050 1.009
Pleasant Lake 28.019 12.135 1.520
Portage Lake 27.941 12.922 1.009
Reeds Lake 31.748 14.244 0.223
Round Lake_Mecosta 29.000 11.729 1.254
Sage Lake 26.494 9.927 1.338
Sunrise Lake 27.857 11.143 0.891
Van Etten Lake 31.711 18.193 0089
Wells Lake 29.542 12.417 1.163
Winfield Lake 29.657 13.343 0.758

 

146

 

 

Table C-5 Data of 12 lakes used in Landsat-5 TM row 31 DN model

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Name Band1 Band3 LN(SDT)
Christiana Lake 71.228 15.924 0.684
Coldwater Lake 87.092 24.865 0.065
Crooked_BLq Lake 71.854 14.427 1.615
Diamond Lake 86.070 18.797 0.517
Eagle Lake 70.532 13.671 1.520
Fish Lake 70.267 14.567 1.210
Juno Lake 69.669 14.263 0.604
Klinger Lake 85.611 19.482 1.009
Lake of the Woods 69.461 14.117 1.254
Osterhout Lake 71 .789 15.092 0.891
Paw Paw Lake 74.179 17.723 0.498
Vineyard Lake 78.870 21.438 0.827

 

Table C-6 Data of 12 lakes used in Landsat-5 TM row 31 radiance model

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Name Band1 Band3 LN(SDT)
Christiana Lake 40.941 10.906 0.684
Coldwater Lake 50.463 18.225 0.065
Crooked_Big Lake 41.320 10.040 1.615
Diamond Lake 50.394 13.875 0.517
Eagle Lake 40.603 9.628 1.520
Fish Lake 40.280 10.080 1.210
Juno Lake 40.108 9.985 0.604
Klinger Lake 49.618 14.014 1.009
Lake of the Woods 39.908 9.948 1.254
Osterhout Lake 41.312 10.247 0.891
Paw Paw Lake 42.768 12.658 0.498
Vineyard Lake 45.529 15.447 0.827

 

147

 

 

Table C-7 Data of 43 lakes sampled in both sensors from ETM+ radiance model

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Name Band1 Band3 LN(SDT)
Arnold Lake 38.711 10.342 1.584
Avalon Lake 50.379 12.024 1.645
Baldwin Lake 39.475 10.675 1.297
Bass Lake 40.104 11.122 1.009
Bear Lake 40.971 9.576 2.031
Beaver Lake 45.682 12.064 1.364
Big Lake 39.415 11.008 1.210
Bills Lake 44.697 11.636 1.009
Blue Lake 39.951 10.656 1.377
Clifford Lake 39.657 10.900 1.297
Coldwater Lake 48.289 16.317 0.604
Crooked_BM.ake 42.859 1 1 .489 1 .415
Derby Lake 40.098 10.541 1.856
Diamond Lake 50.102 14.663 1.114
Eggle Lake 40.100 10.200 1.486
East Twin Lake 42.785 12.558 1.338
Fish Lake 39.773 11.000 0.952
George Lake 39.258 11.169 1.009
Hardy Darn Pond 39.311 10.867 1.308
Hess Lake 45.915 22.562 -0.495
Higgins Lake 49.865 11.916 1.756
Hubbard Lake 46.450 13.375 1.702
Hutchins Lake 42.517 11.622 1.217
Jordan Lake 40.810 12.441 0.240
Juno Lake 45.211 13.600 0.952
Kettle Lake 40.500 10.786 1.114
Klinger Lake 47.916 13.848 1.063
KP Lake 39.240 10.880 1.210
Lake Lansing 40.764 13.688 0.498
Lake of the Woods 41.217 10.957 1.131
Murray Lake 40.920 11.080 1.063
Osterhout Lake 40.683 10.857 0.758
Paw Paw Lake 41.594 11.152 1.063
Pleasant Lake 40.851 13.106 0.271
Portage Lake 40.372 13.032 0.993
Reeds Lake 43.858 13.967 0.401
Round Lake_Mecosta 38.773 10.303 1.254
Sage Lake 43.877 14.487 1.254
Sunrise Lake 39.031 10.594 0.875
Van Etten Lake 48.286 18.313 -0.089
Vineyard Lake 41.040 11.972 1.210
Wells Lake 39.682 10.955 1.297
Winfield Lake 40.000 11.233 0.742

 

148

 

 

Table C-8 Data of 43 lakes sampled in both sensors from TM radiance model

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Name Band1 Band3 LN(SDT)
Arnold Lake 27.500 10.620 1.584
Avalon Lake 32.040 10.848 1.903
Baldwin Lake 29.327 11.857 1.210
Bass Lake 29.737 11.585 0.952
Bear Lake 29.100 9.959 2.126
Beaver Lake 29.504 11.524 1.409
Big Lake 29.299 12.569 1.210
Bills Lake 33.622 13.324 1.009
Blue Lake 29.815 11.919 1.520
Clifford Lake 29.682 12.068 1.297
Coldwater Lake 50.463 18.225 0.065
Crooked_BiLLake 41.320 10.040 1.615
Derby Lake 30.147 11.721 1.756
Diamond Lake 50.394 13.875 0.517
Eflle Lake 40.603 9.628 1.520
East Twin Lake 28.277 11.923 0.952
Fish Lake 40.280 10.080 1.210
George Lake 28.237 12.619 0.758
Hardy Dam Pond 29.465 11.435 1.297
Hess Lake 34.903 26.428 -0.783
Higgins Lake 33.435 9.919 1.856
Hubbard Lake 28.984 10.499 1.782
Hutchins Lake 30.416 12.474 1.187
Jordan Lake 30.771 14.729 0.065
Juno Lake 40.108 9.985 0.604
Kettle Lake 29.059 11.353 1.009
Klinger Lake 49.618 14.014 1.009
KP Lake 27.852 10.444 1.210
Lake Lansigq 28.216 13.281 0.421
Lake of the Woods 39.908 9.948 1.254
Murray Lake 29.842 12.050 1.009
Osterhout Lake 41.312 10.247 0.891
Paw Paw Lake 42.768 12.658 0.498
Pleasant Lake 28.019 12.135 1.520
Portage Lake 27.941 12.922 1.009
Reeds Lake 31.748 14.244 0.223
Round Lake_Mecosta 29.000 11.729 1.254
Sage Lake 26.494 9.927 1.338
Sunrise Lake 27.857 11.143 0.891
Van Etten Lake 31.711 18.193 -0.089
Vineyard Lake 45.529 15.447 0.827
Wells Lake 29.542 12.417 1.163
Winfield Lake 29.657 13.343 0.758

 

149

 

radiance model

Table C-9 Data of 32 lakes sampled in both sensors from ETM+ row 29 and 30

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Name Band1 Band3 LN(SDT)
Arnold Lake 38.711 10.342 1.584
Avalon Lake 50.379 12.024 1.645
Baldwin Lake 39.475 10.675 1.297
Bass Lake 40.104 11.122 1.009
Bear Lake 40.971 9.576 2.031
Beaver Lake 45.682 12.064 1.364
fig Lake 39.415 11.008 1.210
Bills Lake 44.697 11.636 1.009
Blue Lake 39.951 10.656 1.377
Clifford Lake 39.657 10.900 1.297
Derby Lake 40.098 10.541 1.856
East Twin Lake 42.785 12.558 1.338
George Lake 39.258 11.169 1.009
Hardy Dam Pond 39.311 10.867 1.308
Hess Lake 45.915 22.562 -0.495
Higgins Lake 49.865 11.916 1.756
Hubbard Lake 46.450 13.375 1.702
Hutchins Lake 42.517 11.622 1 .217
Jordan Lake 40.810 12.441 0.240
Kettle Lake 40.500 10.786 1.114
KP Lake 39.240 10.880 1.210
Lake Lansing 40.764 13.688 0.498
Murray Lake 40.920 11.080 1.063
Pleasant Lake 40.851 13.106 0.271
Portage Lake 40.372 13.032 0.993
Reeds Lake 43.858 13.967 0.401
Round Lake_Mecosta 38.773 10.303 1.254
Sage Lake 43.877 14.487 1.254
Sunrise Lake 39.031 10.594 0.875
Van Etten Lake 48.286 18.313 0089
Wells Lake 39.682 10.955 1.297
Winfield Lake 40.000 11.233 0.742

 

150

 

radiance model

Table C-10 Data of 32 lakes sampled in both sensors from TM row 29 and 30

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Name Band1 Band3 LN(SDT)
Arnold Lake 27.500 10.620 1.584
Avalon Lake 32.040 10.848 1.903
Baldwin Lake 29.327 11.857 1.210
Bass Lake 29.737 11.585 0.952
Bear Lake 29.100 9.959 2.126
Beaver Lake 29.504 11.524 1.409
figLake 29.299 12.569 1.210
Bills Lake 33.622 13.324 1.009
Blue Lake 29.815 11.919 1.520
Clifford Lake 29.682 12.068 1.297
Derby Lake 30.147 11.721 1.756
East Twin Lake 28.277 11.923 0.952
George Lake 28.237 12.619 0.758
Hardy Dam Pond 29.465 11.435 1.297
Hess Lake 34.903 26.428 -0.783
Higgins Lake 33.435 9.919 1.856
Hubbard Lake 28.984 10.499 1.782
Hutchins Lake 30.416 12.474 1.187
Jordan Lake 30.771 14.729 0.065
Kettle Lake 29.059 11.353 1.009
KP Lake 27.852 10.444 1.210
Lake Lansing 28.216 13.281 0.421
Murray Lake 29.842 12.050 1.009
Pleasant Lake 28.019 12.135 1.520
Portage Lake 27.941 12.922 1.009
Reeds Lake 31.748 14.244 0.223
Round Lakg_Mecosta 29.000 11.729 1.254
Sage Lake 26.494 9.927 1.338
Sunrise Lake 27.857 11.143 0.891
Van Etten Lake 31.711 18.193 -0.089
Wells Lake 29.542 12.417 1.163
Winfield Lake 29.657 13.343 0.758

 

151

 

Table C-ll Data of 11 lakes sampled in both sensors from ETM+ row 31 radiance

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

model
Name Band1 Band3 LN(SDT)
Coldwater Lake 48.289 16.317 0.604
Crooked_Big Lake 42.859 11.489 1.415
Diamond Lake 50.102 14.663 1.114
Eagle Lake 40.100 10.200 1.486
Fish Lake 39.773 11.000 0.952
Juno Lake 45.211 13.600 0.952
Klinger Lake 47.916 13.848 1.063
Lake of the Woods 41.217 10.957 1.131
Osterhout Lake 40.683 10.857 0.758
Paw Paw Lake 41.594 11.152 1.063
Vineyard Lake 41.040 11.972 1.210

 

Table C-12 Data of 11 lakes sampled in both sensors from TM row 31 radiance

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

model
Name Band1 Band3 LN(SDT)
Coldwater Lake 50.463 18.225 0.065
Crooked_Big Lake 41.320 10.040 1.615
Diamond Lake 50.394 13.875 0.517
Eagle Lake 40.603 9.628 1.520
Fish Lake 40.280 10.080 1.210
Juno Lake 40.108 9.985 0.604
Klinger Lake 49.618 14.014 1.009
Lake of the Woods 39.908 9.948 1.254
Osterhout Lake 41.312 10.247 0.891
Paw Paw Lake 42.768 12.658 0.498
Vineyard Lake 45.529 15.447 0.827

 

152