JPL PUBLICATION 78-100 EPA-600/4-79-005
This document has not been
submitted to NTIS, therefore it
should be retained.
Trophic Classification of Selected
Colorado Lakes
Richard J. Blackwell
Jet Propulsion Laboratory
Dale H. P. Boland
U.S. Environmental Protection Agency
January 1979
Prepared for
National Aeronautics and
Space Administration
and
U.S. Environmental Protection Agency
Jet Propulsion Laboratory
California Institute of Technology
Pasadena, California
-------
JPL PUBLICATION 78-100 EPA-600/4-79-005
TROPHIC CLASSIFICATION OF
SELECTED COLORADO LAKES
Lake Classification Through the Amalgamation of
Contact-Sensed Data and Digitally Processed
Multispectral Scanner Data Acquired by Satellite
and Aircraft
Richard J. Blackwell Dale H. P. Boland
Earth Resources Application Group Water and Land Quality Branch
Jet Propulsion Laboratory Environmental Monitoring and
California Institute of Technology Support Laboratory - Las Vegas
Pasadena, California
National Aeronautics and U.S. Environmental Protection Agency
Space Administration Office of Research and Development
Washington, D.C. 205^6 Las Vegas, Nevada 89114
January 1979
-------
The research described in this report was carried out by the Jet Propulsion
Laboratory, California Institute of Technology, together with the U. S.
Environmental Protection Agency and was jointly sponsored by the National
Aeronautics and Space Administration and the U. S. Environmental Protection
Agency.
-------
ABSTRACT
Multispectral scanner data, acquired over several Colorado lakes
using Landsat-1 and aircraft, were used in conjunction with contact-
sensed water quality data to determine the feasibility of assessing lacus-
trine trophic levels. A trophic state index was developed using contact-
sensed data for several trophic indicators (chlorophyll a., inverse of
Secchi disc transparency, conductivity, total phosphorous, total organic
nitrogen, algal assay yield). Relationships between the digitally pro-
cessed multispectral scanner data, several trophic indicators, and the
trophic index were examined using a supervised multispectral classification
technique and regression techniques. Statistically significant correlations
exist between spectral bands, several of the trophic indicators (chlorophyll
a, Secchi disc transparency, total organic nitrogen) , and the trophic state
index. Color-coded photomaps were generated which depict the spectral
aspects of trophic state. Multispectral scanner data acquired from satel-
lite and aircraft platforms can be used to advantage in lake monitoring
and survey programs when amalgamated with contact-sensed data.
111
-------
CONTENTS
1. INTRODUCTION 1-1
A. THE NATIONAL EUTROPHICAT10N SURVEY^ 1-1
B. REQUIREMENTS OF THE FEDERAL WATER POLLUTION
CONTROL ACT 1-3
C. PROJECT HISTORY AND BACKGROUND 1-5
D. PROJECT OBJECTIVES 1-7
E. STUDY AREA AND LAKES 1-7
II. BACKGROUND 2-1
A. LACUSTRINE CONCEPTS 2-1
B. OPTICAL PROPERTIES OF PURE AND NATURAL WATERS 2-5
C. REMOTE SENSING SYSTEMS 2-7
D. PERIPHERAL EFFECTS 2-19
E. REMOTE SENSING OF COLORADO LAKES 2-25
F. TROPHIC INDICATORS AND A MULTIVARIATE
TROPHIC INDEX 2-31
G. MULTISPECTRAL CLASSIFICATION 2-36
III. METHODS 3-1
A. DATA ACQUISITION 3-1
B. MULTISPECTRAL DATA PREPROCESSING 3-2
C. LAKE EXTRACTION METHODS 3-8
D. WATER SAMPLE SITE LOCATIONS 3-13
E. PIXEL DENSITY AT SAMPLE SITES 3-13
F. LAKE SURFACE AREA DETERMINATION 3-14
G. TROPHIC INDICATOR SELECTION AND MULTIVARIATE
INDICES DEVELOPMENT 3-17
H. ANALYSES OF TROPHIC INDICATOR, TROPHIC STATE INDEX,
AND REMOTELY SENSED DATA RELATIONSHIPS 3-21
-------
IV.
RESULTS AND DISCUSSION 4-1
A. LAKE SURFACE AREA 4-1
B. PRINCIPAL COMPONENT TROPHIC ORDINATION OF LAKES AND
SAMPLING SITES INDICATORS 4-4
C. CORRELATION AND REGRESSION ANALYSIS RESULTS 4-18
D. REGRESSION MODELS FOR THE ESTIMATION OF TROPHIC
INDICATORS AND MULTIVARIATE TROPHIC INDICATORS 4-35
E. THEMATIC MAPPING RESULTS 4-43
F. POINTS OF CONCERN 4-59
V. CONCLUSIONS AND RECOMMENDATIONS 5-1
A. CONCLUSIONS 5-1
B. RECOMMENDATIONS 5-1
VI. GLOSSARY 6-1
REFERENCES R-1
APPENDIXES
A. PHYSICAL-CHEMICAL DATA FOR THE COLORADO
STUDY LAKES A-1
B. PHOTOGRAPHIC FLIGHT LOG OF NASA AIRCRAFT COVERAGE
OF COLORADO LAKES, AUGUST 25, 1975 B-1
C. REGRESSION MODEL PREDICTED, RESIDUAL, AND
ASSOCIATED OBSERVED VALUES C-1
Figures
1-1. Number of Lakes Sampled by the National
Eutrophication Survey 1-4
1-2. Landsat-1 Coverage of Colorado NES-Sampled Lakes
for August 22-24, 1975 1-11
1-3. Color Composite Image of Landsat-1 Scene 5126-16474
Showing Several of the Colorado Test Lakes 1-12
VI
-------
1-4. Color Composite Image of Landsat-1 Scene 5127-16532
Showing Several of the Colorado Test Lakes ----------- 1-13
1-5. Color Composite Image of Landsat-1 Scene 5127-16534
Showing Several of the Colorado Test Lakes ----------- 1-14
2-1. Hypothetical Productivity Growth 'Curve of a
Hydrosere ---------------- . ---------------------------- 2-3
2-2. Reflection Characteristics of Filtered and Unfiltered
Water Samples from Two Wisconsin Lakes in the
Area of Madison -------------------------------------- 2-8
2-3. The Landsat Space Observatory ------------------------ 2-10
2-4. Landsat-1 Ground Coverage Pattern -------------------- 2-12
2-5. Schematic Diagram of the Landsat-1 MSS Scanning
Arrangement ------------------------------------------ 2-15
2-6. Ground Scan Pattern for a Single MSS Detector -------- 2-16
2-7. Generalized Spectral Reflectance Curve --------------- 2-17
2-8. Generalized Output of the Landsat MSS in Response
to the Spectral Distribution Illustrated in
Figure 2-7 ------------------------------------------- 2-17
2-9. Generalized Output of the Aircraft-Borne MMS in
Response to the Spectral Distribution Illustrated
in Figure 2-10 --------------------------------------- 2-21
2-10. Generalized Spectral Reflectance Curve for a Single
Picture Element of a Hypothetical Lake --------------- 2-22
2-11. Some Components and Interactions of Light with a
Hypothetical Lake and the Atmosphere ----------------- 2-23
2-12. IR2 Image of Landsat Scene 5126-16474,
August 23, 1975 -------------------------------------- 2-26
2-13. IR1 Image of Landsat Scene 5126-16474,
August 23, 1975 -------------------------------------- 2-27
2-14. Red Image of Landsat Scene 5126-16474,
August 23, 1975 -------------------------------------- 2-28
2-15. CRN Image of Landsat Scene 5126-16474,
August 23, 1975 -------------------------------------- 2-29
2-16. Geometrical Interpretation of the Principal Components
for a Hypothetical Bivariate System ------------------ 2-35
vii
-------
3-1. Geometric Corrections Typically Applied to
Multispectral Scanner Data 3-5
3-2. Aircraft-Acquired MMS Imagery Before Geometric
Corrections 3-6
3-3- Aircraft-Acquired MMS Imagery After Geometric
Corrections Have Been Made 3-7
3-4. GRN, RED, IR1, and IR2 Images of a Landsat Scene
5127-16532 Subscene — 3-9
3-5. GRN, RED, IR1, and IR2 Images of a Landsat Scene
5127-1653^ Subscene After the Application of the
Binary Mask Generated Using the IR2 DN Range of
0-2& as Representing Water 3-10
3-6. Landsat-1 MSS IR2 Concatenation of Nine
Colorado Lakes 3-11
3-7. MMS Channel 10 Concatenation of Five
Colorado Lakes 3-12
3-8. Landsat Pixel Size in Relation to the U.S. Standard
One-Mile Section 3-15
4-1. Proportion Estimation Diagram 4-4
4-2. Trophic Classification Maps of Five Colorado Lakes
Based on Landsat-1 MSS Data and a Multivariate
Trophic Index for 13 Sampling Sites (PC1-13) 4-45
4-3- Trophic Classification Maps of Five Colorado Lakes
Based on MMS Data and a Multivariate Trophic Index
for 13 Sampling Sites (PC1-13) 4-46
4-4. Trophic Classification of Nine Colorado Lakes Based
on a Pooled Multivariate Index (PC1-2?) 4-50
4-5. Trophic Classification of Nine Colorado Lakes Based
on a Nine-Class Multivariate Trophic Index 4-53
4-6. Chlorophyll a. Classification of Nine Colorado
Lakes 4-55
4-7. Inverse Secchi Depth Classification Maps for
Nine Colorado Lakes 4-58
Vlll
-------
Tab!es
1-1. Colorado Lakes Included in the Study 1-8
2-1. Landsat-1 Orbital Parameters _ 2-11
2-2. Landsat-1 MSS Characteristics 2-14
2-3. MMS System Specifications 2-20
2-4. Trophic Indicators 2-32
3-1. Acquisition Dates for Colorado Lake Data 3-1
3-2. Availability of Remotely Sensed Data for 12 Colorado
Lakes and 32 Lake Sampling Sites 3-3
3-3. Area of Colorado Lakes As Determined From NASA
Aerial Photographs 3-16
3-4. Acronyms Used for Trophic Indicators 3-17
3-5. Trophic Indicator Values for 12 Colorado Lakes
for August 1975 Sampling Period 3-19
3-6. Landsat MSS Functions Investigated in Statistical
Stepwise Regression Analysis 3-25
3-7. Landsat MSS Mean Values and Standard Deviations for
Nine Colorado Lakes (Destriped "Whole" Lake Data) 3-26
3-8. Landsat MSS Band Means for 27 Sampling Sites
in 9 Colorado Lakes (Destriped Data) 3-27
3-9. MMS Channel Means, Standard Deviations, and Pixel
Counts for Five Colorado Lakes 3-28
3-10. MMS Channel Means and Standard Deviations for 13
(11_by-11 Pixel Array) Sampling Sites in 5
Colorado Lakes 3-29
3-11. Pearson Product-Moment Correlation Coefficients
Generated From MMS Channels for 13 Sites Located
in 5 Colorado Lakes 3-30
4-1. Surface Area Estimates for Colorado Lakes Using
Three Types of Sensors 4-2
4-2. Normalized Eigenvectors and Eigenvalues Extracted
from 11 Colorado Lakes' Six Trophic Indicator
Data Correlation Coefficient Matrix 4-6
IX
-------
4-3. Product-Moment Correlation Coefficients for 6
Trophic Indicators and the Principal Components
Extracted from 11 Colorado Lakes' Data Correlation
Matrix 4-7
4-4. Trophic Ranking of 11 Colorado Lakes Derived From
Principal Components Analysis of Six Trophic
Indicators 4-8
4-5. Rankings of 11 Colorado Lakes as Derived from Two
Trophic Indices and Ordered by the PC1-11 Index 4-8
4-6. Normalized Eigenvectors and Eigenvalues Extracted
from 27 Colorado Lake Sampling Sites' 6 Trophic
Indicator Data Correlation Coefficient Matrix 4-10
4-7. Product-Moment Correlation Coefficients for 6 Trophic
Indicators and the Principal Components Extracted from
the 27 Colorado Lake Sampling Sites, Data Correlation
Coefficients Matrix 4-11
4-8. Trophic Ranking of 27 Colorado Lake Sampling Sites
Derived From Principal Components Analysis of 6 Trophic
Indicators 4-12
4-9. Normalized Eigenvectors and Eigenvalues Extracted
from 13 Colorado Lake Sampling Sites' 6 Trophic
Indicator Data Product-Moment Correlation Matrix 4-13
4-10. Product-Moment Correlation Coefficients for 6 Trophic
Indicators and the Principal Components Extracted
from 13 Colorado Lake Sampling Sites Data Correlation
Coefficient Matrix 4-14
4-11. Trophic State Index (PC1-13) Ranking Generated for
13 Colorado Lake Sampling Sites Using Principal
Components Analysis of 6 Natural Log-Transformed
Trophic Indicators 4-15
4-12. Comparison of Trophic State Index (PC1) Values for
13 Sampling Sites 4-16
4-13. Landsat MSS Interband Product-Moment Correlation
Coefficients for Three Sets of Observations 4-19
4-14. MMS Interchannel Product-Moment Correlation
Coefficients Based on Data from 13 Sites 4-21
4-15. Normalized Eigenvectors and Eigenvalues Extracted
from 13 Colorado Lake Sampling Sites, Eight-Channel MMS
Data Covariance Matrix 4-22
-------
4-16. New MMS Variables and Associated Data Generated
Through Principal Components Analysis of MMS
Channels 1-4, 8-9 Data for 13 Sites in 5 Colorado
Lakes r ^-23
4-17. Pearson Product-Moment Correlation Coefficients of
Eight Channel MMS"Data for 13 Colorado Lake Sampling
Sites and Associated Principal Components 4-24
4-18. Product-Moment Correlation Coefficients Between
"New" MMS Channels Developed from Principal Component
Analysis of MMS Data from 13 Sites 4-25
4-19. Pearson Product-Moment Correlation Coefficients
Between Landsat MSS Bands and MMS Channels for 13
Sampling Sites on 5 Colorado Lakes 4-26
4-20. Pearson Product-Moment Correlation Coefficients
Between Landsat MSS Bands and MMS Principal Component-
Derived Variables for 13 Sampling Sites on 5
Colorado Lakes 4-27
4-21. Pearson Product-Moment Correlation Coefficients for
9 Colorado Lakes (August 1975 Contact-Sensed and
Destriped MSS Data) 4-28
4-22. Pearson Product-Moment Correlation Coefficients
Generated from Landsat MSS and Trophic Indicator
Data for 27 Sites Located in 9 Colorado Lakes 4-30
4-23- Pearson Product-Moment Correlation Coefficients
Between Landsat MSS Bands and Trophic Indicators
for 13 Sampling Sites on 5 Colorado Lakes 4-32
4-24. Pearson Product-Moment Correlation Coefficients
Generated from MMS and Trophic Indicator Data for
13 Sites Located in 5 Colorado Lakes 4-33
4-25. Pearson Product-Moment Correlation Coefficients
Generated from Principal Component-Derived MMS
Variable and Trophic Indicator Data for 13 Sampling
Sites in 5 Colorado Lakes 4-34
4-26. Regression Models Developed from Contact,
MSS, and MMS Data 4-36
4-27. Landsat MSS and Bendix MMS Training Site Classifi-
cation Accuracies Expressed as a Percentage of Each
Site's Pixel Count 4-48
4-28. Landsat MSS and Bendix MMS Lake Classification
Results Expressed as a Percentage of Each Lake's
Pixel Count 4-49
XI
-------
4-29- Classification Analysis of Nine Colorado Lakes
Based on a Pooled Multivariate Index (PC1-27) 4-51
4-30. Classification Analysis of Nine Colorado Lakes
Based on a Nine-Class Multivariate Trophic Index 4-54
4-31. Analysis of Chlorophyll a. Classification of Nine
Colorado Lakes 4-56
4-32. Analysis of Inverse Secchi Depth Classification of
Nine Colorado Lakes 4-56
4-33. Changes in DN Values for Two Colorado Lakes Over
Two Days 4-60
xii
-------
ACKNOWLEDGMENT
The contributions to this document of J. D. Addington, A. Y. Smith,
and A. L. Mendoza are gratefully acknowledged.
Xlll
-------
SECTION I
INTRODUCTION
Liranologists and other scientists concerned with eutrophication
have frequently become entangled in the semantics associated with the
word "eutrophication.1' Originally the term was limited to the concept
of changes in the nutrient levels in lakes, but has now broadened in
meaning to include the consequences of nutrient enrichment.
Eutrophication of surface waters is a major contemporary water
quality management problem. Many of man's activities accelerate naturally
occurring eutrophication. Municipal sewage and industrial waste disposal
activities as well as land use practices often impose relatively large
nutrient loadings on lakes and rivers. In some cases, this enrichment
results in algae blooms and other symptoms of eutrophication. The
consequences of man-induced eutrophication often make the water body
less attractive to potential users, or completely unusable. More impor-
tantly, at least when a long-range viewpoint is adopted, eutrophication
shortens the time period of natural lake succession.
A. THE NATIONAL EUTROPHICATION SURVEY
In December 1971 the U. S. Environmental Protection Agency (EPA)
announced initiation of the National Eutrophication Survey (NES) , an
intensive survey to identify water bodies in the United States which
have potential or actual eutrophication problems due to phosphorus
from municipal sources and assess the degree of this problem. Estab-
lishment of the survey project was announced jointly by the Surgeon
General of the United States, the Administrator of EPA, the Commissioner
of the Food and Drug Administration, and the Chairman of the Council
on Environmental Quality. They further announced that nitrilotriacetic
acid should not be used as a phosphate substitute because of unresolved
questions concerning long-term effects on human health and the environment.
The results of the Eutrophication Survey have been integrated into
an EPA control program. Through construction grants to state and local
governments, this program will seek to improve municipal waste treatment
facilities so as to reduce phosphate levels to the extent necessary to
protect water quality.
It should be noted that the survey centered upon phosphorus and
nitrogen, the nutrients most often implicated as a cause of eutrophication
in freshwater and, occasionally, in salt or brackish water. While the
process and rate of eutrophication within any particular water body can
also be related to other substances including elements such as iron,
manganese, and molybdenum, nitrogen and phosphorus, in particular, play
key roles.
1-1
-------
For each lake considered, the work of the Eutrophication Survey
consisted of three actions, undertaken in response to three fundamental
questions:
Question
(1) What is the water body's
trophic condition and what
are the nutrient levels at
the present time?
(2) Which nutrients control the
growth of aquatic plant life
and can they be controlled?
(3) What is the extent of nutrient
loading from sources supplying
the water body and, specifically,
how much is from municipal sewage
treatment plants?
Action
Assess lake's condition.
Determine limiting
nutrient(s).
Estimate the nutrient
loading to the water body.
Responsibility for conducting the survey was assigned jointly
to the EPA's Environmental Monitoring and Support Laboratory in Las
Vegas, Nevada (EMSL-Las Vegas), and Corvallis Environmental Research
Laboratory, Oregon (CERL)^. The EPA was supported by National Guard
volunteers, who collected samples from tributary streams (a project
sanctioned by the Department of Defense), and by local sewage treatment
plant operators, who collected effluent samples.
During the first "sampling season," from March to November 1972, about
235 lakes in Minnesota, Wisconsin, Michigan, New York, and the New England
states were sampled, as were some 1100 stream sites and 230 waste effluents.
In October 1972, the Congress enacted the Federal Water Pollution
Control Act Amendments (PL 92-500), which assigned to the states respon-
sibility for classifying lakes within their boundaries as to degree of
eutrophication, defining the cause and nature of lake pollution, and
devising procedures for eutrophication control. In response to the new
Federal legislation, NES program objectives were recast to better match
EPA strategies. The NES sampling program, which had previously been
limited to lakes directly subject to pollution from point sources, was
broadened to include lakes subject to only non-point-source pollution.
In 1973, about 250 lakes and their tributaries were sampled.
These lakes are located in 17 states east of the Mississippi River and
south of the states in which lakes had been sampled the previous year.
1EMSL-Las Vegas was formerly named the National Environmental Research
Center-Las Vegas (NERC-Las Vegas), CERL was named the National Environ-
mental Research Center-Corvallis (NERC-Corvallis) .
1-2
-------
About 180 lakes in the 10 Great Plains States located between the
Rocky Mountains and the Mississippi River were sampled in 1974. During
1975, the last NES field year, some 155 lakes located in the remaining
11 Rocky Mountain and Far Western States were sampled. A grand total
of some 820 lakes was sampled during the project (Figure 1-1).
To conduct the sampling, the EPA used three Bell UH-1H "Huey"
helicopters on loan from the U.S. Army. Two of the helicopters were
deployed at field locations at a given time,'while the third was out of
service, on a rotating basis, for routine maintenance. Each of the
helicopter sampling teams from EMSL-Las Vegas included a pilot, a
limnologist, and a technician. The helicopter would land on a particular
lake, and the crew would lower an electronic sensor package into the water
to measure temperature, conductivity, turbidity, pH (whether acidic or
basic), and dissolved oxygen continuously as the sensor package descended.
Later, as the sensor package was raised, individual water samples were
pumped from selected depths for more extensive laboratory analysis.
Sightings of prominent landmarks were made for position information.
When all of the measurements and samples had been gathered at a given
point the sensor package was stowed back aboard the aircraft and the
helicopter moved to a new site on the same, or another, lake.
At the end of the day, the helicopter returned to a temporary
field base in the area, where the samples were unloaded and taken to
a mobile field laboratory manned by chemists. There the samples were
analyzed for chlorophyll a. and dissolved oxygen, after which they were
filtered and packed for shipment. Nutrient analyses and algae identi-
fications were made at EMSL-Las Vegas; algal assays and heavy metal
determinations were performed at CERL. The tributary stream samples
collected by National Guardsmen and the treatment-plant-effluent samples
submitted by plant operators were sent to CERL for chemical analyses.
Detailed descriptions of the lake sampling procedures and the water
sample analysis techniques are given in NES Working Paper Number 1
(U.S. EPA 197*0 and NES Working Paper Number 175 (U.S. EPA 1975).
B. REQUIREMENTS OF THE FEDERAL WATER POLLUTION CONTROL
ACT (PUBLIC LAW 92-500)
In October 1972 Congress enacted the "Federal Water Pollution
Control Act Amendments." This act assigns to the states responsibility
for classifying lakes within their boundaries as to degree of eutrophi-
cation or aging, defining the cause and nature of lake pollution, and
devising procedures for control.
The act itself, a comprehensive document of 89 pages, outlines in
great detail the goals and policies of the act. It further defines com-
prehensive programs for water pollution control, research, investigations,
training, and the dissemination of information. Section 104-a-5 in parti-
cular directs the EPA administrator to establish, equip, and maintain
a water quality surveillance system for monitoring purposes. The same
citation further suggests that this water quality surveillance be
1-3
-------
NATIONAL EUTROPHICATION SURVEY
NUMBER OF LAKES & YEAR SAMPLED
'72 - 231
'75 - 154
'73 - 249
GRAND TOTAL-813
Figure 1-1. Number of Lakes Sampled by the National
Eutrophication Survey
-------
conducted by utilizing wherever practicable the resources of NASA,
NOAA, USGS, and the Coast Guard.
Several sections within the act, particularly 106 and 314, speci-
fically require each state to identify and classify all publicly owned
freshwater lakes within its boundaries. These requirements are noteworthy
in that, in order for a state to gain assistance in the form of grants for
pollution control programs or for lake restoration, it must conduct a
classification of its lakes.
A lake classification system which utilizes a combination of remote
and contact-sensed data, if developed to meet EPA standards of classifi-
cation, offers a potentially cost-effective method whereby states can
comply with PL 92-500 without a heavy investment at the state level
to train and equip technical personnel to conduct lake classifications
by water analysis.
C. PROJECT HISTORY AND BACKGROUND
When Landsat-1 (ERTS-1) was placed into orbit in July 1972, it
became apparent that an opportunity existed to apply remote-sensing
techniques to the problem of trophic classification of lakes. In 1974 a
research proposal was accepted by NASA Office of Technology Utilization
to examine the feasibility of lake classification using Landsat data
in combination with contact-sensed data. A joint effort by the Image
Processing Laboratory at JPL, the Corvallis Environmental Research
Laboratory, Corvallis, Oregon and the Environmental Monitoring and
Support Laboratory, Las Vegas, Nevada, was then undertaken to examine
this potential application.
Landsat digital data in the form of computer-compatible tapes (CCT's)
were obtained for lakes in New York, Wisconsin, and Minnesota. Water
sample data from EPA's STORET (STOrage and RETrieval) information storage
system were acquired for the lakes under consideration. The Landsat
data and the STORET data were examined to obtain the best match possible
in terms of water sample data and Landsat coverage data.
Two basic approaches for classification were undertaken during
this period. The first was to develop a method and procedure to produce
a numerical trophic index from the water quality measurements made
by the EPA water sampling teams. The second approach was to relate
the trophic index to the Landsat multispectral scanner (MSS) data and
achieve classification and/or to use the MSS data to estimate either
the numerical trophic index directly or one or more of the water quality
parameters used to generate the trophic index.
The results from this joint activity, including many of the techniques
and methods for data analysis, are found in Boland (1976), Boland and
Blackwell (1975), Blackwell and Boland (1975), and Blackwell and Boland
(1974). Boland (1976:1) drew the following conclusions:
(1) The Landsat-1 MSS is an effective tool for lake enumeration
and for the estimation of lake surface area.
1-5
-------
(2) Estimates of lake Secchi disc transparency and chlorophyll a.
levels having practical significance can be achieved through
the incorporation of lake MSS color ratios in regression
models. Each trophic indicator has a model which is unique to
the specific date of Landsat-1 coverage.
(3) MSS color ratios can be used to estimate lake position on a
multivariate trophic scale. However, each date of Landsat-1
coverage has its unique model. The models for different
dates vary greatly in their predictive capabilities. The
standard error of the models tends to decrease as the growing
season progresses and the manifestations of eutrophication
become more evident.
(4) While information relating to lacustrine trophic state can be
extracted from Landsat-1 MSS imagery, either by visual
inspection or through microdensitometry and optical density
slicing, the maximum benefits in water-based studies can be
derived only through the use of the digital data contained
on the computer-compatible tapes in conjunction with automatic
image processing techniques.
(5) Although Landsat-1 provides 18-day repetitive coverage,
systematic times-series are difficult, if not impossible, to
obtain because of excessive cloud cover on many dates of
satellite coverage.
(6) The Landsat-1 MSS has utility as a supplemental data source in
lake survey and monitoring programs. Its value is most
apparent in situations involving large lakes and/or large
numbers of lakes."
While the aforementioned project did enjoy some success, it is
evident that several of the conclusions reflect and articulate areas
of concern. Several deficiencies have been noted in the past study.
These shortcomings, related more to procedural activity and project
planning than to the underlying research effort, include
(1) The use of contact-sensed data acquired nonconcurrently with
the MSS data. With one exception, the contact-sensed data
were collected several days prior to (or after) Landsat
passage.
(2) The unavailability of intermediate-level aircraft-acquired
multispectral scanner data with which to determine the
segments of the spectrum more suited for classification
purposes.
(3) The unavailability of aerial photography for the determination
of lake surface area.
(4) The use, by the regression models and classification maps, of
average trophic indicator and MSS band values, computed
1-6
-------
separately for each lake, as opposed to values determined at
each sampling site in each lake.
(5) The fact that the thematic photomaps generated depicted only the
spatial aspects of a trophic index developed using annual mean
trophic indicator values, computed separately for each lake.
(6) The failure to generate thematic maps depicting the spatial
aspects of some of the more obvious trophic indicators such as
Secchi depth transparency or chlorophyll .a.
The activity described in this report reflects efforts made to
overcome the more obvious shortcomings of the previous study.
D. PROJECT OBJECTIVES
The general objective of this investigation is the further elucida-
tion of the role of Landsat multispectral scanner data, when used with
contact-sensed data, in lake monitoring and survey programs. Although
the focus of attention is directed toward Landsat, data acquired from
an aircraft-borne modular multispectral scanner (MMS) were also examined.
Specific objectives include an evaluation of the scanners' capabilities to
(1) Estimate lake surface area.
(2) Estimate lacustrine trophic state.
(3) Estimate several trophic indicators including Secchi disc
transparency and chlorophyll a..
(4) Aid in the development of lake thematic photomaps which depict
specific lake trophic indicators and trophic state.
E. STUDY AREA AND LAKES
The selection of the study area and lakes was largely dictated by
weather conditions and the sampling schedule laid out by Water and Land
Quality Branch personnel at Las Vegas, Nevada. Initially, lakes in
several western states (.e.g.., California, Utah, Nevada) were proposed as
test lakes. A combination of factors, including the availability of NASA
aircraft, weather, and EPA logistical and sampling constraints led to the
selection of lakes in Colorado as the test lakes (Table 1-1). The lakes
are described below. The surface area and water depth figures are taken
from the literature and do not necessarily reflect the situation existing
at the time of the Landsat flyover. Most of 'the "lakes" are actually
artificial reservoirs which experience large fluctuations in surface area
and depth during the summer.
1-7
-------
Table 1-1. Colorado Lakes Included in the Study
Lake/Reservoir
Name
Barker H.
Barr L.
Blue Mesa H.
Cherry Creek fl.
Cucharaa H.
Dillon R.
Grand L.
Green Mt. fl.
Holbrook L.
Meredith H.
Milton fl.
Shadow Mt. R.
STORE!,,
Number
0001--
0802--
0603 —
0804--
Od06—
Ott07 —
OBOo1--
0(309--
0610--
0811 —
0813--
County
Boulder
Adams
Junnison
Arapahoe
Summit
Grand
Summit
Otero
Crowley
Weld
Grand
Latitude /Longitude
39-58-00/105-28-54
39-57-32/104-45-17
38-27-10/107-20-15
39-39-11/104-51-20
37—44-55/1 04 -35-5 5
39_37_20/106-03-5«
40-H-10/105-48-08
39-52-00/106-20-00
30-03-37/103-36-06
36-06-48/103-44-50
40-14-20/104-38-15
36-40-50/107-40-22
USGS Quadrangles
Nederland 7.5'
Tungsten 7 .5'
Brighton 7.5 '
Mile High L. 7.5'
Little Soap Park 7.5
Sapinero 7.5'
Carpenter Ridge 7.5'
Big Mesa 7.5'
Mclntosh 7 .5'
Fitzsimons 7.5'
Parker 7.5'
Frisco 7.5'
Dillon 7.5'
Grand L. 7.5'
Mt. Powell 15.0'
Cheraw 7.5'
Sugar City 7.5'
Milton R. 7.5'
Shadow Mt. 7.5'
Grand L. 7.5'
1.
Barker Reservoir (0801)
Barker Reservoir, located in Boulder County, is maintained by the
Public Service Company of Colorado. It is situated in Sections 17 and 18,
T1S, R72W, and Section 13, T1S, R73W. The lake was formed by constructing
a 53-meter high concrete power dam across South Boulder Creek. It is open
to public fishing, but boating is prohibited. The surface area measures
154 hectares and it has a maximum depth greater than 31 meters. The water
depth is known to fluctuate greatly.
2.
Barr Lake (0802)
Barr Lake is an irrigation reservoir located in Adams County
approximately 24 km northeast of Denver. At its maximum extent it
covers some 757 ha to a maximum depth of 12 m. The conservation pool
has a mean area of 324 ha and a maximum depth of 2.4 m. The lake provides
a diversity of habitats and is noted for its abundance of wildlife.
The uniqueness of the lake has been recognized by the Colorado General
Assembly and it has purchased, through the Colorado Division of Parks
and Outdoor Recreation, a perpetual recreational easement from the
irrigation company. Detailed plans have been formulated to develop
v,he area, but at the same time to preserve the different habitats.
1-f
-------
3. Blue Mesa Reservoir (0803)
Blue Mesa Reservoir is located in Gunnison County. It is the largest
of the Curecanti National Recreation Area reservoirs located on the Gunnison
River. The reservoir, first filled in October 1965, has a maximum depth of
104 m and a surface area of 3,662 ha, making it the largest reservoir in
the state of Colorado.
4. Cherry Creek Reservoir (0804)
Cherry Creek Reservoir is located in Arapahoe County on the south-
eastern fringe of the Denver metropolitan area. It has a surface area of
356 ha and a maximum depth of 12.2 m. The lake, formed by damming Cherry
Creek, is used for recreational fishing, water skiing, and pleasure boating.
It is stocked with both rainbow trout and warm water species of fish.
5. Cucharas Reservoir (0805)
Cucharas Reservoir is located on the Cucharas River in Huerfano
County. The water level fluctuates greatly; little water was found in the
reservoir at the time of EPA sampling (August 22, 1975). The sampling
crews reported it as being dry on September 7, 1975.
6. Dillon Reservoir (0806)
Dillon Reservoir is located in Summit County on the Blue River.
It covers an area of 1,276 ha and has a maximum depth of 61 m. Water
from this reservoir flows into Green Mountain Reservoir.
7. Grand Lake (0807)
Grand Lake is a natural lake located at the headwaters of the
north fork of the Colorado River in Grand County. It is part of the
Colorado-Big Thompson Project, which also includes five storage reservoirs
and five regulation reservoirs. The lake has a surface area of 205
ha, a maximum depth of 81 m, and a mean depth of 41 m. It is connected
to Shadow Mountain Reservoir by means of a water level channel. The
two water bodies are operated as single entity. Water level fluctuations
are limited to 0.3 m by law. See Shadow Mountain Reservoir (0813).
8. Green Mountain Reservoir (0808)
Green Mountain Reservoir is located on the Blue River in Summit
County and is part of the Colorado-Big Thompson Project. It was formed by
the construction of an earth and rock fill dam 94 m in height and 351 m
in length. It has a surface area of 820 ha and a maximum depth of 76 m.
The U.S. Bureau of Reclamation operates the reservoir for stabilizing
purposes as well as for the generation of hydroelectric power. Its
depth fluctuates as much as 27 m.
1-9
-------
9. Holbrook Lake (0809)
Holbrook Lake is located in Otero County in southeastern Colorado.
It lies about 8 km northeast of Swink in the Arkansas Valley. It has
a maximum surface area of 272 ha and a maximum depth of 7 m. The surface
area and depth vary greatly because the lake serves as a source of
irrigation water. The conservation pool has a surface area of about
24 ha and a maximum depth of approximately 1 m.
Although 90% of the game fish caught by anglers are planted
rainbow trout, the annual reduction in water level, high summer water
temperatures, and high alkalinity impose a great stress on the reservoir's
game and pan fish. The natural habitats conducive to fish survival
are adequate when the reservoir is at its maximum pool level, but are
grossly inadequate when the irrigation drawdown takes place. Artificial
habitat has been constructed from rocks, old tires, logs, brush, and
straw to supply the fish with places for spawning, foraging, and hiding.
With the start of June the reservoir is dominated by water skiers and
boating enthusiasts.
10. Lake Meredith Reservoir (0810)
Lake Meredith Reservoir is situated in Crowley County near Sugar
City. It has a surface area of 1,303 ha and a maximum depth of 4.6 m.
The major stream feeding the reservoir is Lake Meredith Reservoir Inlet,
11. Milton Reservoir (0811)
Milton Reservoir is located in Weld County. It has a maximum
surface area of 841 ha and a maximum depth of 12.2 m. The minimum
conservation pool has a depth of about 6 m.
12. Shadow Mountain Reservoir (0813)
Shadow Mountain Reservoir, located in Grand County, has a surface
area of 548 ha and a maximum depth of 11 m. It is part of the Colorado-
Big Thompson Project and under the jurisdiction of the U.S. Bureau
of Reclamation. Shadow Mountain Reservoir and Grand Lake (0807) are
operated as one unit by the Bureau. The natural flow of water is from
Grand Lake to Shadow Mountain Reservoir and then to Lake Granby, which
serves as a major storage facility. When the need arises, Lake Granby
water is pumped back to the Shadow Mountain Reservoir where it seeks
an equilibrium level with Grand Lake. Water can leave Grand Lake via
an outlet at its west end and pass under the continental divide through
the Adams Tunnel, emerging on the east slope of the Rocky Mountains.
The geographic distribution and Landsat coverage of the study
lakes are depicted in Figure 1-2. The water bodies, as viewed from
Landsat orbital altitudes, are shown in Figures 1-3 through 1-5.
1-10
-------
37 N
109 W
37 N
108 W
107 W 106
NEW MEXICO
Figure 1-2. Landsat-1 Coverage of NES-Sampled Lakes for August 22-
24, 1975. Scene 5125-16422 and the unnumbered scene were
not processed because of excessive cloud cover
1-11
-------
Figure 1-3. Color Composite Image of Landsat-1 Scene 5126-164?^
Showing Several of the Colorado Test Lakes. The
multispectral scanner data were recorded on August
23, 1975 by Landsat-1. Several of the lakes under
study in this report are labeled
1-12
-------
.
Figure 1-4. Color Composite Image of Landsat-1 Scene 5127-16532
Showing Several of the Colorado Test Lakes. The
multispectral scanner data were recorded on August
24, 1975 by Landsat-1. Several of the lakes under
study in this report are labeled
1-13
-------
Figure 1-5. Color Composite Image of Landsat-1 Scene 5127-
16531* Showing Several of the Colorado Test Lakes.
The multispectral scanner data were recorded on
August 24, 1975 by Landsat-1. Several of the
lakes under study in this report are labeled
1-14
-------
SECTION II
BACKGROUND
In this section a number of concepts relating to lakes, the processes
of eutrophication and succession, peripheral effects, and remote sensing
are discussed. In addition, an overview of two remote sensing systems is
presented. Detailed discussion on specific points are found in the
publications cited.
A. LACUSTRINE CONCEPTS
There is far from universal agreement as to what constitutes a lake.
Veatch and Humphrys (1966) suggested that to give the word "lake" a pre-
cise, limited meaning would probably be an exercise in futility because
the word has been in use for a long time and been given a diversity
of applications. The word is used as a synonym for pond, reservoir,
and sea. It has been applied to bodies of fresh water and saline water;
to standing water and widenings in rivers; to bodies of water measuring
less than a hectare and to those gauged in hundreds of thousands of
hectares; to naturally occurring water bodies and man-made reservoirs;
to water-filled or partially filled basins; and to basins void of water.
"Lake" is generally more prestigious than other common names (.e.g.. , pond,
slough, reservoir) and is preferred by promoters of waterbased tourist
and recreational businesses and commercial developers of shoreline pro-
perty (Veatch and Humphrys 1966). Nevertheless, numerous attempts
have been made to define and delimit the members of lentic series (i.e.-,
lake, pond, marsh, and their intergrades).
Forel defined a lake as a body of standing water occupying a basin
and lacking continuity with the sea, and a pond as a lake of slight depth
(toelch 1952). Welch (1952: 15-16) defined a lake as a "...body of standing
water completely isolated from the sea and having an area of open,
relatively deep water sufficiently large to produce somewhere on its
periphery a barren, wave-swept shore." He employed the term "pond"
"...for that class of very small, very shallow bodies of standing water
in which quiet water and extensive occupancy by higher aquatic plants
are common characteristics"...and suggested that all larger bodies of
standing water be referred to as lakes. Zumberge (1952) defined a lake
as an inland basin filled with water. Harding (19^2) described lakes
as "...bodies of water filling depressions in the earth's surface." In
this report no deliberate effort will be made to carefully distinguish
a lake from another lentic body on the basis of a definition. For
example, an artificial reservoir may be called a lake or, at times, a
water body.
1. Lake Succession
Lakes, although giving the impression of permanence when measured
on the scale of the human life span, are transitory features of the
earth's surface. All lakes, regardless of their origin, pass through
2-1
-------
the process of ecological succession which ultimately results in a terres-
trial environment. The ephemeral nature of lakes is a consequence of two
fundamental processes, the downcutting of the outlet and, more importantly,
the deposition of allochthonous (arising in another biotope) and autoch-
thonous (arising in the biotope under consideration) materials in the basin.
Generally speaking, most lakes commence the successional process as
bodies possessing relatively low concentrations of nutrients and, generally,
low levels of productivity.^ The importation and deposition of materials
(.e.g.. , sediment) from the shoreline and the surrounding watershed gradually
decreases the lake depth. The addition of allochthonous materials normally
enriches the water and thereby stimulates the production of organic materials.
Autochthonous materials increase the sedimentation rate, thus accelerating
succession. Marked floral and faunal changes occur. Algal blooms become
more common along with submergent and, eventually, emergent aquatic macro-
phytes. Desirable game fish may be replaced by less desirable species,
the so-called "rough fish." When the successional process is not disrupted
by major geologic or climatic changes, a lake eventually becomes a marsh
or swamp which, in turn, terminates as dry land.
Lindeman (1942) stressed the productivity aspects in relation to lake
succession. Figure 2-1 represents the probable successional productivity
relationships for a hypothetical hydrosere (an ecological sere originating
in an aquatic habitat) developing from a moderately deep lake located in a
fertile humid continental region. Productivity is initially low, a conse-
quence of low nutrient levels, but increases rapidly as nutrients become
more available. The length of time required for completion of the succes-
sional process is a function of several factors including lake basin morpho-
logy, climate, and the rate of influx and nutrient value of allochthonous
materials. It is readily apparent that allochthonous nutrients can drasti-
cally increase lake productivity and thereby shorten the life span of a lake.
2. Eutrophication
The word eutrophication is often used to denote the process whereby
a pristine water body (.e.g.. , lake) is transformed into one characterized
by dense algal scums and obnoxious odors. Thick beds of aquatic macrophytes
may become common. However, the word has been applied differently, accord-
ing to the respective interests of its users. Weber (1907) used the German
adjectival form of eutrophication "nahrstoffreichere" (eutrophe) to describe
the high concentration of elements requisite for initiating the floral se-
quence in German peat bogs (Hutchinson 1973: 269). The leaching of nutrients
from the developing bog resulted in a condition of "mitterlreiche" (mesotrophe)
and eventually "nahrstoffearme" (oligotrophe). Naumann (1919) applied the
words oligotrophic (under-fed), mesotrophic, and eutrophic (well-fed) to
describe the nutrient levels (calcium, phosphorus, combined nitrogen) of
water contained in springs, streams, lakes, and bogs (Hutchinson 1973: 269).
^Edmondson (1974) suggested that the idea that all lakes are born oligotro-
phic and gradually become eutrophic as they age, is an old misconception.
2-2
-------
EUTROPHY
MAT
OLIGOTROPHY
SENESCENCE
TIME
Figure 2-1. Hypothetical Productivity Growth-Curve of a
Hydrosere. Lindeman (19^2: 413) describes the curve
as representing a hydrosere "...developing from a
moderately deep lake in a fertile cold temperate
climatic condition." It must be kept in mind that
this is a generalized curve and not all lakes will
follow it in total. For example, lakes that are
light-limited because of suspended inorganic materials
may never experience the initial dramatic increase
in productivity. Adapted from Lindeman (1942: 413)
2-3
-------
Naumann (1931) defined eutrophication as the increase of nutritive sub-
stances, especially phosphorus and nitrogen, in a lake. Hasler (194?)
broadly interpreted eutrophication as the "Enrichment of water, be it
intentional or unintentional...'1 Fruh, et ai. (1966: 1237) defined the
word as the "enhancement of nutrients in natural water..." while Edmondson
(1974) suggested that many limnologists seem to use the term to describe
"...an increase in the rate of nutrient input..." Hasler and Ingersoll
(1968: 9) suggested that eutrophication is the "...process of enrichment
and aging undergone by bodies of fresh water..." Vollenweider (1968)
summarized the eutrophication of waters as meaning "...their enrichment
in nutrients and the ensuring progressive deterioration of their quality,
especially lakes, due to the luxuriant growth of plants with its repercussions
on the overall metabolism of the water affected..." A search of the liter-
rature on eutrophication indicates that the meaning of the term, originally
limited to the concept of changing nutrient levels, has been gradually
expanded to include the consequences of nutrient enrichment.
Eutrophication occurs both naturally and as a result of man's
activities (cultural or anthropogenic eutrophication). Many of man's
practices relating to land use and to the disposition of municipal sewage
and industrial wastes impose relatively large nutrient loadings on many
lakes and rivers. In many cases, the enrichment results in algal blooms
and other symptoms of eutrophication. The consequences of man-induced
eutrophication often make the water body less attractive to potential
users. More importantly, at least when a long-range viewpoint is adopted,
eutrophication accelerates lake succession, thus shortening the time
period before a lake loses its identity.
A comment regarding eutrophication is in order. The term, in the
popular press and the mind of the layman, is being equated with a "bad" or
highly undesirable situation. Certainly, when the enrichment levels reach
extremes and undesirable manifestations occur (e_.g.. , algal blooms, fish
kills, obnoxious odors), the water body loses much of its value as a natural
resource. However, enrichment of natural waters can result in increased
primary productivity, leading to a larger biomass of consumers; eutrophic
water bodies often provide excellent fishing opportunities.
The lakes of Colorado are undergoing the successional process and
eutrophication. In general, the successional and enrichment processes are
slow-moving in lentic bodies located in high mountain areas which are far
removed from human activities. On the other hand, lakes and reservoirs
which are located near metropolitan centers and in regions of intense
agricultural activity are subjected to processes which can have serious
consequences (i.e.., degradation of water quality). The water quality in
Colorado's lakes and streams is appreciably affected by the dissolved and
suspended matter carried by runoff from the land surface.
Agricultural runoff and soil erosion are two nonpoint sources which
affect the water quality of Colorado's lakes and streams. Other major non-
point sources (of a more localized nature) which affect water quality include
active and abandoned coal mines, intensive livestock and specialized agri-
cultural operations, and storm drainage from urbanized area and construction
sites. Agricultural runoff and runoff from ordinary precipitation events
contain many contaminants (.e.g.., organic materials which are oxygen-demanding,
2-4
-------
minerals derived from the soil or applied by man, fecal coliforms, pesticides,
herbicides, fertilizers, and other chemicals) from ground surface and ground
cover which have accumulated through natural processes and nonintensive land
husbandry. When rainfall of sufficient intensity occurs, soil erosion
results. The severity and frequency of soil erosion are functions of many
factors including intensity of immediate rainfall, prior climatic conditions,
soil cover, soil texture, topography, and antecedent human activities. The
eroded soil contributes both dissolved and suspended matter to the flowing
waters. The suspended materials contributed by agricultural runoff and
erosion are deposited in stream and lake beds. The deposited soil can
bury aquatic life, create an oxygen demand, and release nutrients and
chemicals to the flowing stream or overlying lake water. The influx of
nutrients to a lake, assuming they are not deposited on the bottom and
overlain by other materials, tend to make the water body more eutrophic.
The accumulation of materials on the lake or reservoir bottom decreases
the water depth and moves the water body closer to the point in time
when its identity as a lake or reservoir is lost.
B. OPTICAL PROPERTIES OF PURE3 AND NATURAL WATERS
It is readily apparent, even to the casual observer, particularly
if looking downward from an aircraft, that lakes differ in color and
brightness. Many investigations have been undertaken to develop a
comprehension of the processes which result in the observed phenomena.
Although a detailed discussion of the interaction of electromagnetic
energy with the components of the hydrosphere and atmosphere is outside
the scope of this report, a brief survey is essential to gain some
understanding of the principles which both permit and yet constrain
the use of remote sensing techniques in lake classificatory work.
The interaction between electromagnetic energy and chemically pure
water has been studied by numerous investigators (e..£. , Ewan 189*1, Sawyer
1931, Collins 1925, James and Birge 1938, Hulburt 1945, Raman 1922, Dawson
and Hulburt 1937). The transmission of electromagnetic energy through a
material medium is always accompanied by the loss of some radiant energy
by absorption. Some of the energy is transformed into other forms (.e.g..,
heat, chemical) or to some longer wavelength of radiation (James and
Birge 1938). Pure water is very transparent to violet, blue, and green
light. In the infrared region, the extinction coefficient is high with
a complementary low degree of transmission (see Hutchinson 1957: 381-383).
The absorption spectral characteristics of pure water can be modified
greatly through the addition Of dissolved and particulate materials.
The absorption spectra of natural water (e.g.. , lake and ocean)
have been studied in detail by Jerlov (1968), Duntley (1963), Atkins
and Poole (1952), Birge and Juday (1929, 1930, 1932), and Juday and
Birge (1933), to mention a few. Hutchinson (1957) has summarized the
more important attempts to elucidate the interactions of light with
natural waters, particularly in regard to lakes.
Water which is totally free of dissolved and particulate substances.
2-5
-------
An electromagnetic wave impinging on the surface of a lake decom-
poses into two waves, one of which is refracted, proceeding into the
aquatic medium, and a second wave which is reflected back to the atmo-
sphere (Jerlov 1968). The wave entering the water is refracted as
it passes through the air-water interface according to Snell's law,
which may be expressed as:
n = sin (i)/sin(r)
where (i) is the angle of incidence, (r) is the angle of refraction,
and (n) is the refractive index, which for water is approximately 1.33.
Most of the electromagnetic energy entering a lake is attenuated
through the process of absorption. Although only a small percentage
(less than 3 percent, Davis 1941) of the incident energy is backscattered
from the lake water volume, this light (volume reflectance) is the focus
of interest in remote sensing of water quality investigations. Its
spectral characteristics have been shaped by the materials found in the
lake's waters (dissolved and suspended materials, plankton, aquatic
macrophytes, and air bubbles) and, in some cases, by bottom effects.
The attenuation of electromagnetic radiation in lake waters is a
consequence of the relatively unselective effect of suspended particulate
materials and the highly selective effect of dissolved coloring matter,
usually of organic origin, on the electromagnetic spectrum. The dissolved
matter absorbs strongly in the violet and blue wavelengths, moderately
in the middle wavelengths (.e.g. , green), and much less strongly at longer
wavelengths (Hutchinson 1957: 423). When the dissolved materials are
present in small quantities, the water will be most transmissive in the
green wavelengths. Lake waters with large amounts of dissolved substance
are more transmissive in the orange and red wavelengths than in shorter
wavelengths. However, the transmission of the red and orange light is
still greater in pure water than in water containing particulate and/or
dissolved materials. As water transparency diminishes, the detectable
electromagnetic energy will be of progressively longer wavelength,
at increasingly shallower depths (Hutchinson 1957: 424).
The color of a lake is the color of the electromagnetic energy
backscattered from the lake body to the sensor. Lake color ranges from
the blue of pure water through greenish blue, bluish green, pure green,
yellowish green, greenish yellow, yellow, yellow brown, and clear brown
(Hutchinson 1957: 415). Lake color need not be, and is usually not, the
same as the lake water. Lakes which are blue in color lack appreciable
quantities of humic materials and colored materials in suspension (.e.g. ,
phytoplankton). The bluer the lake color, the smaller the amount of
free-floating organisms contained in the water (Ruttner 1963: 21).
Waters with a high plankton content possess a characteristic yellow-
^Welch (1952: 84) defines water color as "...those hues which are inherent
within the water itself, resulting from coloidal substances or substances
in solution" (i.e.., true color).
2-6
-------
green to yellow color. The characteristic color may not be apparent
owing to masking by other materials (.e.g.. , suspended sediments) . Ruttner
(1963: 104) suggested that
A lake with very transparent and dark blue, blue-green,
or green water is always oligotrophic. On the other hand,
eutrophic lakes always have a relatively low transparency
and are yellow-green to yellow-brown in color; but the
determination of these optical properties alone will not
establish the productivity type, for the turbidity can be of
inorganic origin, and the color can come from humic substances.
Seston color, attributable to the reflection spectra of suspensoids of
microscopic or submicroscopic size, is often observed in highly productive
lakes. Lakes containing large quantities of suspended inorganic matter
(je.g.. , silt) may acquire a characteristic seston color. However, in
most cases the color is related to large concentrations of phytoplanktonic
organisms (Hutchinson 1957: 417).
Scherz e_t aj^. (1969) have investigated the total reflectance (surface
reflectance plus volume reflectance) curves of pure water and natural waters
under laboratory conditions using a spectrophotometer. They reported that
the addition of dissolved oxygen, nitrogen gases, and salts (.e.g.. , NaCl,
Na2SOi}, Na^POipt^O) had no apparent effect on the reflection curve. How-
ever, water from lakes in the Madison (Wisconsin) area had reflectance
curves that differed both from the distilled water curve and from each
other. These differences can be attributed to the presence of different
algal organisms, since filtration of the lake waters produced similar
reflectance curves, though different from that of pure water (Figure 2-2).
The color of natural waters is the end result of optical processes
that are both numerous and complex. It is relatively easy to detect
differences in color within a lake and also among a population of lakes.
It is, however, more difficult to attach physical, chemical, or biological
significance to the color, particularly when quantitative estimates are
desired. The difficulty is compounded in waters having more than one
class of particulates present, which is normally the case in natural water
(McCluney 1976: 3-4).
The degree of success in sensing and interpreting the significance
of color is partially a function of the sensor type employed for the
collection of spectral data. We will now turn our attention briefly
to the two remote sensing systems used in this project.
C. REMOTE SENSING SYSTEMS
Most attempts to classify or ordinate lakes employ contact sensing
techniques coupled with the observations of the field crew to document
the characteristics of the water bodies. The major constraints of most
classification systems are the necessity of elaborate field data, diffi-
culties in obtaining data for all lakes within a comparable time period
or under similar circumstances, and lack of sufficient or appropriate
2-7
-------
600 800
WAVELENGTH, nrr
1,000 1,200
Figure 2-2. Reflection Characteristics of Filtered and
Unfiltered Water Samples from Two Wisconsin Lakes
in the Area of Madison. Adapted from Scherz el
al. (1969)
2-t
-------
sample locations to characterize the entire lake. A good historical data
base for most of the lakes in the United States is either not available
or is not suited to the development of an overall lake classification
system. Efforts to characterize large numbers of lakes employing only
contact-sensed data and field observations have limited usefulness since
the data base was not intensively collected within a short time period
and some of the data relied upon subjective observations of field personnel.
It appears that satellite-borne sensors, such as the multispectral scanner
carried by Landsat-1 and Landsat-2, and aircraft-borne sensors (_§_.£. ,
modular multispectral scanner, MMS) have the capability to collect data
of value for lake classification and monitoring activities.
The Landsat space observatories are attractive because they provide (1)
repetitive coverage, (2) a synoptic view, and (3) a permanent record. The
Landsat capabilities offer a unique opportunity to obtain a data base which
could group the lakes into categories according to their spectral responses
and also provide the opportunity to study relationships between certain
trophic indicators and the spectral data with an eye toward the development
of predictive models. Landsat provides what may be an economically viable
technique for collecting data for the entire area of each lake within a
reasonable time period. In addition, the sensible physical and optical
properties of the lake waters are measured objectively. In about 25 seconds
the Landsat multispectral scanner collects data in four bands of the spec-
trum for an area of the earth covering about 3^,225 square kilometers. In
regions of the earth where lakes are very abundant, a typical Landsat scene
may contain several hundred to more than a thousand inland water bodies.
With two satellites in operation, and assuming cloud-free conditions,
repetitive coverage is provided on a 9-day basis. Clearly, the satellite
offers certain advantages over conventional contact-sensing techniques.
While presenting a much more restricted synoptic view than Landsat,
and having their own unique shortcomings, aircraft-borne modular multi-
spectral scanner (MMS) systems are also attractive data sources for lake
monitoring and classification programs. They are capable of providing
greater spatial and spectral resolution than the Landsat MSS. In addition,
the opportunity exists, at least in theory, to collect data more frequently
because the aircraft can be scheduled to take advantage of cloud-free days.
Landsat is locked into a predetermined 18-day (ignoring side overlap) data
acquisition cycle; it cannot be moved into a particular area to take advan-
tage of a cloud-free day. Rather its acquisition of lake data is of a
more fortuitous nature. A more detailed examination of the two systems
is presented below.
1. Landsat-1 System
Landsat-1 is one of three satellites NASA has orbited to survey and
monitor the earth's natural resources (Figure 2-3). Both it and a companion,
Landsat-2, were operational at the time the water truth data were collected
from the Colorado Lakes. Landsat-1, however, provided coverage closer to
the dates of water truth data acquisition than Landsat-2 and was therefore
selected as the source of satellite-acquired multispectral scanner data.
2-9
-------
SOLAR ARRAY
ORBIT ADJUST TANK
DATA COLLECTION ANTENNA
ATTITUDE CONTROL SUBSYSTEM
RETURN BEAM
VIDICON CAMERAS (3)
WIDEBAND
RECORDER
ELECTRONICS
WIDEBAND ANTENNA
ATTITUDE MEASUREMENT SENSOR
MULTISPECTRAL SCANNER
S-BAND ANTENNA
Figure 2-3. The Landsat Space Observatory
2-10
-------
a. Landsat-1 Orbit Parameters and Earth Coverage. Landsat-
1 was placed into a nominal sun-synchronous near-polar orbit by a Delta
launch vehicle on July 23, 1972 (Freden 1973). Orbital parameters
are listed in Table 2-1.
The earth coverage pattern is shown in Figure 2-4 for two orbits on
two consecutive days. Orbital parameters result in a 1.43-deg westward
migration of the daily coverage swath, equivalent to a distance of 159 km
at the equator. The westward progression of the satellite revolutions
continues, exposing all of the area between orbit N and orbit N+1 to the
satellite sensors by day M. This constitutes one complete coverage cycle
and consists of 251 revolutions. The cycle takes exactly 18 days and
results in total global coverage between 81°N and 81°S latitude. Fourteen
orbits (i.e.., revolutions) are completed during each of 17 days in a cycle
with 13 revolutions during one day (NASA 1972). Approximately 188 scenes
are acquired on an average day (Nordberg 1972).
b. Landsat-1 Instrumentation. The Landsat-1 payload consists of
a return beam vidicon (RBV) camera subsystem, a multispectral scanner sub-
system (MSS), and a data collection system (DCS). The RBV and MSS are
designed to provide multispectral imagery of the earth beneath the obser-
vatory (i.e.., satellite). A malfunction occurred on August 6, 1972 (orbit
198), in the RBV power switching circuit, and the RBV cameras were turned
off as only one sensor system can be used in conjunction with the one
functioning video tape recorder. The second recorder aboard the observa-
tory malfunctioned between orbits 148 and 181 (Freden 1973) necessitating
data transmisison in a real time mode. The DCS serves to relay environmental
information from geographically remote ground-based sensors to Landsat
ground stations for processing and delivery to users. The RBV and
DCS aspects of the satellite need not concern us.
Table 2-1. Landsat-1 Orbital Parameters3
Orbit Parameter
Actual Orbit
Semi-major axis, km
Inclination, deg
Period, min
Eccentricity
Time at descending node (a.m.)
(southbound equatorial crossing)
Coverage cycle duration, days
Distance between adjacent ground
tracks (at equator), km
7285.82
99.114
103.267
0.0006
9:42
18
(251 revolutions)
159.38
aAdapted from Data Users Handbook (NASA 1972)
2-11
-------
ORBIT N -f 1,DAY M + 1
ORBIT N + 1, DAY M
ORBIT N, DAY M + 1
ORBIT N, DAY M
'// /' NOTE: ORBIT N, DAY M + 1
OCCURS 14 REVOLUTIONS
AFTER ORBIT N, DAY M
t
EQUATOR = 159 km *-K
Figure 2-4. Landsat-1 Ground Coverage Pattern. Adapted from Pat;
Users Handbook (NASA 1972)
2-12
-------
The MSS is a line-scanning radiometer which collects data by
creating images of the earth's surface in four spectral bands simulta-
neously through the same optical system. The instrument operates in
the solar-reflected spectral band region from 500 to 1,100 nanometers.
Scanner characteristics are listed in Table 2-2. The MSS scans crosstrack
swaths 185 km in width, simultaneously imaging six scan lines for each of
the four bands. The object plane is scanned by an oscillating flat mirror
positioned between the scene and a double reflector telescope-type of
optical chain. An 11.5 deg (Horan, Schwartz, and Love 1974) cross-track
field of view is produced by the mirror oscillating ±2.89 deg about its
nominal position (Figure 2.5).
A nominal orbital velocity of 6.47 km/s, neglecting observatory
perturbations and earth rotation effects, produces the requisite along-
track scan. The subsatellite point moves 474 m along the track during
the 73.42-ms active scan-retrace cycle which is itself a consequence
of the 13-62-Hz mirror oscillation rate. The track distance of 474 m
synchronizes with the 474 m along-track field of view of each set of
six detectors. The line scanned by the first detector in one cycle
of the active scan is in juxtaposition to the line scanned by the sixth
detector of the previous scan (Figure 2-6).
Twenty-four glass optical fibers, arranged in a 4-by-6 matrix
in the focused area of the telescope, intercept the light from the
earth scene. Light impinging on the square input end of each optical
fiber is conducted to an individual detector through an optical filter
unique to the respective spectral band under consideration. Photomultiplier
tubes (PMT) serve as detectors for Bands 4 through 6; Band 7 uses silicon
photodiodes. A video signal is produced at the scanner electronics
output as the image of a line across the swath is swept across the
fiber during active scan. The signal is sampled at 9-95- s intervals,
which correspond to a 56-m cross-track motion of the instantaneous
field of view. The sampled signal is digitized and arranged into a
serial digit data stream for transmission to ground stations. Individual
signals are derived from light passing through each fiber, resulting
in 24 channels of output.
The MSS, as found on Landsat-1 and Landsat-2, is a low-resolution
device, both spatially and spectrally speaking. As can be seen in Table
2-2, three of the bands [green (CRN), red (RED), infrared-one (IR1)] are
100 nm in width with the infrared-two (IR2) band covering 300 nm. Figure
2-7 illustrates a generalized spectral reflectance curve for a single
picture element (pixel - the MSS spatial resolution unit with nominal
measurements of 57-by-79 m) of a hypothetical lake. The width of the MSS
bands disallows the recording of the fine details in the curve. The MSS
output more closely resembles Figure 2-8. Responses are given as values
for the various wavelengths bands (.e.g..> 500 to 600 nm, CRN) instead of
specific values for the entire spectral range. This procedure crudely
defines an entire range of wavelength responses as four single readings.
2-13
-------
Table 2-2. Landsat-1 MSS Characteristics
Item
Characteristics
Telescope optics
Scanning method
Scan (swath) width
Scan duty cycle
Instantaneous field of view (IFOV)
Number of bands
Number of lines (detectors)
scanned per band
Limiting ground resolution from
917 km altitude
Spectral band wavelength:
NDPF band code
Band 4 (green)
Band 5 (red)
Band 6 (near-infrared one)
Band 7 (near-infrared two)
Sensor response:
Detector
Nominal input for 4-V scanner
output (lO'^W cnr2sr-1)
Scanner and multiplexer weight
Signal channels
Telemetry channels
Command capability
Scanner size
22 cm (aperture diameter), f/3.6
Ritchey-Chretien
flat mirror oscillating ±2.9
deg at 13.62 Hz
11.5 deg (185 km at 917 km altitude)
44$
86 yirad
four
six
80 m
500-600 nm
600-700 nm
700-800 nm
800-1,100 nm
Band 4 Band 5 Band 6 Band 7
PMTa PMT PMT Photodiode
24.8 20.0 17-6 46.0
50 kg
24
97
72
Approximately 36 x 38 x 89 cm
aPhotomultiplier tube.
2-14
-------
SCANNER
-OPTICS
6 DETECTORS
PER BAND:
24 TOTAL,
+ 2 FOR BAND 5
(ERTS B)
SCAN MIRROR
(OSCILLATES
NOMINALLY
±2.89°)
NOTE: ACTIVE SCAN IS
WEST TO EAST
6 LINES/SCAN/BAND
Figure 2-5. Schematic Diagram of the Landsat-1 MSS Scanning Arrangement,
Adapted from Data Users Handbook (NASA 1972)
2-15
-------
SPACECRAFT
VELOCITY VECTOR
z
u
t/i
|.
I
O
a:
i
LINE 1
2
3
4
5
6
LINE 1
2
3
4
5
COMPOSITE
TOTAL AREA SCAN
FOR ANY BAND
FORMED BY
REPEATED 6 LINE
PER BAND SWEEPS
PER ACTIVE
MIRROR CYCLE
Figure 2-6. Ground Scan Pattern for a Single MSS Detector,
Adapted from Data Users Handbook (NASA 1972)
2-16
-------
500 600
700 800 900
WAVELENGTH, nm
1000 1100
Figure 2-7. Generalized Spectral Reflectance Curve for
a Single Picture Element (Pixel) of a
Hypothetical Lake
Q
z
o
on
LANDSAT fv
GRN
RED
IR 1
IR2
1 1
500 600 700 800 900 1000 HOC
WAVELENGTH, nm
Figure 2-8. Generalized Output of the Landsat MSS in
Response to the Spectral Distribution
Illustrated in Figure 2-7
2-17
-------
As mentioned above, the nominal MSS pixel measures 57 by 70 m,
thereby covering an area of 0.3933 ha. Through the use of resampling
techniques it is possible to adjust the pixel size (.e.g.. , an 80-by-80-m
pixel corresponding to 0.6M ha was employed in this study). It must be
kept in mind that the MSS gathers energy over the area of its nominal
pixel. Many measurements made using contact techniques are of the point
type, a direct contrast to those acquired by the Landsat MSS. It is
commonly recognized that some Landsat MSS pixels contain a mixture of
water and land features. This normally occurs along the water-land inter-
face or in situations where the water body is much smaller than the pixel
or, conversely, where an island is much smaller than the pixel. The pixel
size also tends to give small water bodies or those with very irregular
shorelines a "blocky" appearance. A visual examination of imagery gen-
erated from the Landsat MSS will usually detect a pattern of stripes
running orthogonal to the satellite's path. This is a consequence of
an imbalance among the MSS's 24 detectors. The problem is particularly
noticeable when work is in the digital domain (i.e.. , work is with the
CCT's).
It should be kept in mind that, although the Landsat MSS was de-
signed with the earth's resources in mind, it was not developed specifi-
cally for water. In fact, if anything, its design favors terrestrial
features. Most of the incident solar energy entering a water body is
attenuated through absorption. The volume reflectance of a water body
is generally less than 3% of the incident light. Thus the energy reaching
the MSS from water bodies is relatively small in magnitude compared to
that received from most land features. While it is possible to increase
the MSS's gain in the CRN and RED bands, this is not normally done.
2. Aircraft-Borne Modular Multispectral Scanner System (MMS)
The modular multispectral scanner system is an airborne, 11-channel
scanning multiband radiometer designed to perform quantitative measurements
of the electromagnetic radiation reflected or emitted from a ground
scene. Table 2-3 lists the characteristics of this system.
The scanner is mounted in a bay on the underside of the aircraft
and views the ground through an opening in the aircraft skin. A rotating
wedge mirror is used to scan the ground below the aircraft perpendicular
to the direction of flight. This allows a telescope assembly to focus
an image of successive small areas of the ground onto an aperture called
the primary aperture. Light passing through this aperture is split
into 11 spectral bands (i.e.., channels) by a spectrometer consisting
of a combination of spectral filters and a diffraction grating. The
images of the scene on the ground, observed in each wavelength band,
are then focused onto separate detectors and are converted into electrical
signals. Sequential line-by-line scans are generated by virtue of
the forward motion of the aircraft. In order to compensate for changes
in the velocity/altitude (V/H) ratio of the aircraft and maintain contiguous
scanning along the flight path, the speed of the scan mirror may be
varied from 10-100 scans/second by the sensor operator.
2-18
-------
The instantaneous field-of-view (IFOV) provided by the field
defining apertures and the telescope is 2.5 mrad. Basic registration
of 0.1 IFOV is maintained between the thermal and the visible/near
IR channels because of the use of a common optical system up to the
beam splitter. The visible/near IR channels are inherently registered
because the optical input to all those channels comes from the single
spectrometer slit, and the registration of the thermal channel is maintained
by a mechanical positioning of the thermal detector.
As evident from Table 2-3, the MMS channels (CH) or bands are
considerably narrower than those of the Landsat MSS. This is very
advantageous because it permits a more detailed examination of the
spectral curve for areas or objects of interest. Figure 2-9 depicts
the generalized output of the MMS in response to the spectral distri-
bution illustrated in Figure 2-10. In this particular comparison,
picture element (pixel) sizes are the same and atmospheric effects
are assumed to be equal. You will note that the MMS has greater resolving
power of the spectral distribution than does the Landsat MSS.
D. PERIPHERAL EFFECTS
The character of the electromagnetic energy impinging on both
the MMS and Landsat-1 MSS has been shaped through interactions with
numerous environmental phenomena (Figure 2-10). Some of the interac-
tions are highly desirable because they alter the character of the
light, which may then be interpreted in terms of some parameter of
interest (.e.g.., Secchi depth). Other interactions of light energy
with the environment may be detrimental to a particular study. It
goes without saying that what may be a vitally important interaction
in one study may be devastating in another.
The earth's atmosphere has a pronounced effect on the solar spectrum
and on lake water color as sensed from aircraft and satellite altitudes.
Atmospheric conditions (.e.g.. , degree of cloudiness; presence of fog,
smoke, and dust; amount of water vapor) affect the degree of insolation
attenuation. Weather conditions strongly affect the distribution of
energy between sunlight and skylight (Piech and Walker 1971: 186-187),
contributing a degree of uncertainty in water quality assessment through
remotely sensed color measurements, hulstrom (1973: 370-376) has point-
ed out the adverse impact that cloud bright spots can have on remote
sensing techniques which utilize reflected energy.
The degree of scattering and absorption imposed on the return
signal from water bodies is related to atmospheric transmittance and
can result in changes in lake color when sensed at aircraft high flight
and satellite altitudes. The attenuated return signal is also contami-
nated by electromagnetic radiation from the air column (path radiance).
Rogers and Peacock (1973) have reported that solar and atmospheric
parameters have a serious adverse impact on the radiometric fidelity
of Landsat-1 data. Path radiance was found to account for 50% or more
of the signal received by the MSS when viewing water and some land
masses. Even at aircraft altitudes, the atmosphere can have a substantial
2-19
-------
Table 2-3. MMS System Specifications
Center
Channel Wavelength
No. Ac, pro
1
2
3
4
5
6
7
8
9
10
11
Scan
V/H
IFOV
FOV
Roll
0.410
0.465
0.515
0.560
0.600
0.640
0.680
0.720
0.910
1.015
-
Speed
compensation
Spectral
Bandwidth
A, fj.m
0.06
0.05
0.05
0.04
0.04
0.04
0.04
0.04
0.10
0.09
8-14 (thermal
10 to 100 rps
continuously
0.025 to 0.25
2 . 5 mrad
100 deg
± 10 deg
Detector
Type
Silicon
Silicon
Silicon
Silicon
Silicon
Silicon
Silicon
Silicon
Silicon
Silicon
) Mercury
Cadmium
Telluride
variable from
rad/s
Detectors
Frequency response
% obscuration
Focal length
Primary mirror diameter
Second mirror diameter
Registration
Scene video resolution
elements per scan
Silicon PIN diodes (channels 1
through 10) HgCdTe (channel 11)
DC to 200 kHz
18% of primary mirror area
15 in. (.381 m)
4 in. (.0762 m) (3-6 in. unobscured)
1.7 in. (0.04318 m)
0.1 IFOV for all channels including
thermal channel
802
2-20
-------
MMS CHANNEL NUMBER
Q
_i
Z
o
01
s
i
u
a:
1
80
2
—
4C
3
—
>0
4
—
S
5
-
JO
6
6<
7
—
)0
8
—
7
40 8
9 10
1
i
l|
1
!
1
II
1
SO 960 l.C
THERMAL
60
440 540 620 700
WAVELENGTH, nm
-
Figure 2-9- Generalized Output of the Aircraft-Borne MMS in Response
to the Spectral Distribution Illustrated in Figure 2-10.
Compare Figure 2-9 with Figure 2-8
2-21
-------
MMS CHANNEL NUMBER
Figure 2-10.
THERMAL
380 490 580 660 740
440 540 620 700
860 960 1.06
WAVELENGTH, nrt
970
Generalized Spectral Reflectance Curve for a Single
Picture Element (Pixel) of a Hypothetical Lake. The
MMS Spectral Windows Are Superimposed on the Curve
2-22
-------
IV)
FLOATING AND EMERGENT MACROPHYTES
Figure 2-11. Some Components and Interactions of Light
with Hypothetical Lake and Atmosphere
-------
impact on the character of electromagnetic energy reflected from the
earth's surface. The magnitude of the adverse atmospheric effects
can be reduced, though not completely eliminated, by using imagery
or digital data collected on clear, cloudless days. This is the approach
used in this investigation.
The Landsat-1 spacecraft passes over the same point on the earth
at essentially the same local time every 18 days. However, even though
the time of passover will remain essentially the same throughout the
year, solar elevation angle changes cause variations in the lighting
conditions under which the MSS data are obtained. The changes are
due primarily to the north or south seasonal motion of the sun (NASA
1972). Changes in solar elevation angle produce changes in the average
scene irradiance as seen by the sensor from space. The change in irradi-
ance is influenced both by the change in the intrinsic reflectance
of the ground scene and by the change in atmospheric backscatter (path
radiance). The actual effect of changing solar elevation angle on a
given scene is very dependent on the scene itself (NASA 1972). For
example, the intrinsic reflectance of sand is significantly more sensi-
tive to changing solar elevation angle than are most types of vegetation
(NASA 1972). The effects of changing solar elevation angle are of
particular importance when scenes taken under significantly different
angles are compared. The use of color ratios in lieu of raw data values
may be of value in reducing the magnitude of the solar-angle-induced
effects by normalizing the brightness components. The approach is
given some consideration in this study.
A portion of the radiation impinging on the lake surface will
be reflected. The percentage of surface-reflected energy is a strong
function of the angle of incidence. The light reflected from the water-
atmosphere interface is composed of diffuse light from the sky (skylight)
and specularly reflected sunlight. Specular reflection areas contained
in a scene are of little value in most water studies, the possible
exception being the determination of surface roughness. The specularly
reflected radiation exceeds, by several orders of magnitude, the reflect-
ed energy emanating from beneath the water surface (Curran 1972: 1857).
Specular reflection has not been demonstrated as being a major problem
in water-related projects employing Landsat MSS data. This is probably
largely because the MSS has a relatively small scan width or field
of view (FOV) of 11.5 deg. On the other hand, the MMS has a FOV of
100 deg, which makes it very susceptible to specular reflection.
Surface reflected skylight, containing no water quality color
information, can compose from 10% of the return signal on a clear day
to 50$ on a cloudy day (Piech and Walker 1971). The surface-related
skylight not only increases the apparent reflectance from the water
body (volume reflectance), but also affects the shape of the reflectance
curve. Surface roughness is known to have an effect on the percentages
of light reflected and refracted at the interface (Jerlov 1968).
However, the effect of surface is negligible in estimating total radia-
tion entering a water body when the solar elevation angle is greater
than 15 deg (Hutchinson 1957: 375).
2-24
-------
The lake bottom characteristics (color and composition) will also
affect the intensity and/or the spectrum of the volume reflectance in
settings where water transparency permits the reflection of a significant
amount of radiation from the bottom materials. In studies involving
the estimation of water depth or the mapping of bottom features, it is
essential that the lake bottom be "seen" by the sensor. Bottom effects
are capitalized upon and put to a beneficial use. However, in this inves-
tigation, bottom effects are considered to be an undesirable peripheral
effect. A sensor with the capabilities of the Landsat MSS or the MMS
is not able to "see" much deeper into a lake than Secchi depth. The
Secchi transparency of the study lakes is, in most cases, relatively
small when compared to the mean depth of each lake. The assumption
is made, as a first approximation, that the bottom effect is relatively
insignificant when considering each of the selected lakes as an entity.
It is evident that many factors influence the intensity and spectral
characteristics of the electromagnetic radiation which is collected by the
sensor. Absolute quantification of remotely sensed phenomena requires
that all of the adverse effects be accounted for in the return signal.
Failure to account for all of the variation introduced by the detrimental
effects might be criticized as being simplistic or naive. However, given
the present state of the art along with manpower, time, and monetary
constraints a complete accounting is not possible.
E. REMOTE SENSING OF COLORADO LAKES
In this subsection an overview is presented by taking a "first
look" at several Colorado lakes and reservoirs. Initially, the focus
of attention will be directed toward several Landsat MSS images of
Scene 5127-16534.
A visual examination of Landsat MSS imagery indicates that gray tone
differences can be detected in the population of Colorado lakes. Figures
2-12 through 2-15 represent respectively, the IR2, IR1, RED, and CRN gray
tone images of LANDSAT Scene 5126-16W. The IR2 image (Figure 2-12)
clearly demonstrates a great contrast between water bodies and terrestrial
features. Water is an excellent absorber of radiation wavelengths com-
prising the IR2 band and, hence, water bodies appear black. Figure 2-13,
the scene's IR1 counterpart, exhibits a similar contrast between water
and land. A careful examination of the water bodies suggests surface or
near-surface phenomena in some lakes. Gray tone differences both within
specific water bodies and among members of the lake population are most
pronounced in the RED image (Figure 2-14). In this band, lakes with ex-
tremely turbid water often meld with the terrain features, a consequence
of similar gray tone values. A vivid example of this occurring in
Wisconsin lakes is presented in Boland (1976: 12-17). Though less obvious
to the eye, gray tone differences are also noted among water bodies in the
GRN band image (Figure 2-15).
When viewing LANDSAT scenes such as the black and white standard
photographs produced by the EROS Data Center, it should be kept in
mind that no special effort is made to enhance water bodies and related
2-25
-------
Figure 2-12.
IR2 Image of Landsat Scene 5126-16W, August 23, 1975.
The IR2 band (also called Band 7) is excellent for
discriminating water from terrain
2-26
-------
Figure 2-13.
IR1 image of Landsat Scene 5126-16474, August 23, 1975.
This band (also called Band 6) is excellent for discrim-
inating water from terrain. Surface or near-surface
phenomena are evident in some of the water bodies
2-27
-------
Figure 2-14.
RED image of Landsat Scene 5126-16W (August 23, 1975)
The variations in water body gray tones suggest differ-
ences in water quality
2-28
-------
Figure 2-15.
CRN Image of Landsat Scene 5126-16474 (August 23, 1975).
While lacking in the contrast evident in the RED, IR1,
and 1R2 images of the scene, gray tone differences are
still apparent among the water bodies. Compare with Figures
2-12, 2-13, and 2-14
2-29
-------
phenomena. Indeed, a loss of spectral information occurs when the
MSS digital data are transformed into photographic products. Specif-
ically, the products have a relatively small density range compared
to the sensitivity range of the MSS. This results in a scale compression
when the MSS data are transformed into a film image on an electron
beam recorder. In addition, the range of energy returns from water
bodies is small and concentrated at the lower end of the MSS intensity
scale. Scale compression coupled with the small range of digital number
(DN) values adds to the difficulty of determining trophic state index
and indicator values through visual and densitometric evaluation of
"standard" EROS black and white photographs.
As can be seen from Figures 2-12 through 2-15, it is possible to
detect spectral differences for Colorado lakes using Landsat imagery
coupled with photointerpretive techniques. Spectral differences, ex-
hibited as gray tone variations, are also evident in imagery generated
from MMS data (examples of single-channel MMS imagery will not be shown
here). The real challenge is one of relating the spectral variations
to chemical, biological and physical phenomena measurable through contact-
sensing techniques and/or acquired through ground level observation.
As indicated earlier, the quantity and spectral composition of
radiation directed upward across the water-atmosphere interface is,
in part, a function of the dissolved substances and particulate materi-
als in the water. While water itself is capable of scattering and
absorbing light, the major portion of the scattering is due to materi-
als in the water. Scattering due to dissolved color is highly selec-
tive, while suspended solids tend to affect volume reflectance in a
rather nonselective fashion. It then follows that increases in suspend-
ed particulate materials in lake water will tend to increase the reflec-
tance in the Landsat bands and MMS channels. It should be noted that
some natural waters will, at least for a portion of the spectrum, exhi-
bit a lower volume reflectance than that of pure water. Humic waters
have this characteristic as demonstrated by Rogers (1977: 3-85). None
of the Colorado water bodies in this study fall in the humic category.
It has been well documented that the MSS and MMS are incapable of
directly detecting substances such as nutrients (.e.g.. , phosphorus) in
water. This does not mean, however, that it is impossible to get some
estimate of such substances. Phosphorus, for example, is known to be a
key element in primary productivity, stimulating the production of bio-
mass. Differences in nutrient levels are often directly related to the
magnitude of the manifestations of eutrophication (.e.g.. , turbidity,
chlorophyll a., algal blooms) . Such phenomena are sensible to the MSS and
MMS. Again, it should be kept in mind that the energy return from natural
water bodies is generally low compared to that of land features. Thus,
all of the water quality related information is contained in a relatively
small range of DN levels for each band or channel for the Colorado lakes.
This precludes developing trophic indicator estimates having the
accuracies and precisions of the contact-sensed data.
This project is based on the premise that the volume reflectances
of water bodies represent distinct characterizations of their optical
2-30
-------
properties which are then interpretable in terms of parameters considered
important in assessing the trophic state. This concept assumes:
(1) Waters with similar optical properties will yield similar
spectral responses.
(2) Under identical light conditions, the volume reflectance
as measured in all bands and channels will generally be
lowest for clear water lakes. The inverse is also assumed.
(3) Detritus, phytoplankton, suspended solids and most other
natural large particulates are Mie scatterers and, there-
fore, scatter approximately uniformly over the spectrum
sensed by the sensors.
As the quantity of scattering materials increases, there
is a relatively uniform increase in the reflectance curve
(Piech and Walker 1971: 195). In other words, the reflec-
tance curve will become higher and flatter as the water
becomes more turbid.
(4) Substances (.e.g., phosphorus) which are not directly detect-
able to the sensors can be sensed indirectly through their
effects on parameters sensible to the sensors.
(5) Shifts in dominant color reflectance from the blue range
toward the red-brown range reflect increases in lake produc-
tivity or associated increases in dissolved color or inorganic
turbidity.
It should be recognized that the contact-sensed data for this
project were collected as part of an ongoing national survey of lakes
and reservoirs; little or no thought was given to the possibility of
the data being used in a satellite-related project. Thus, some highly
desirable parameters (.e.g.., total suspended solids, organic particulates,
inorganic particulates) were not measured during the time of satellite
flyover. In some cases the location of the sampling stations, selected
prior to the planning of this project, was less than nominal when viewed
through the "eyes" of the sensors.
F. TROPHIC INDICATORS AND A MULTIVARIATE TROPHIC INDEX
1. Trophic Indicators and Trophic State
Limnologists and other individuals concerned with lakes have used
the term "trophic state" (i.e.., degree of eutrophy) to describe two
different lake characteristics, nutrient status and productivity.
Thus, trophic state is a hybrid concept as suggested by Margalef (1958).
A multiplicity of classificatory schemes has been devised to group
and rank lakes. Examples of some approaches to lake typology are found
in: Lueschow, et al. (1970), Rawson (1956, 1960), Margalef (1958),
Hansen (1962), Jarnefelt (1958), Larking and Northcote (1958), Moyle
(19^5, 1946), Pennak (1958), Round (1958), Whipple (1898, Winner (1972),
2-31
-------
Zafar (1959), Beeton (1965), Donaldson (1969), and Gerd (1957). Hutchinson
(1957, 1967) has reviewed many of the attempts to arrange lakes into orderly
systems. Sheldon (1972) provides a particularly enlightening discussion
on quantitative approaches to lake classification.
Perhaps no single area concerning eutrophication and trophic classi-
fication receives more attention than the selection of parameters used to
characterize it. There are numerous indicators of trophic state, each
with its merits and shortcomings. Reviews have been written on the subject
by Fruh, et al. (1966), Stewart and Rohlich (19&7), Vollenweider (1968),
Taylor JLt al- (in preparation), and Hooper (1969). Wezernak and Polycyn
(1972) have compiled a tabulation of indicators commonly used to assess
eutrophication (Table 2-M). There is no universal agreement as to which
(or how many) parameters adequately define the trophic state of a lake.
The number and types of parameters used to reflect levels of eutrophication
clearly indicate that no single parameter currently serves as a universal
measure of trophic state. The use of a single indicator has the virtue
of simplicity but may produce misleading rankings or groupings because
lakes are normally too complex to be adequately gauged on such a simpli-
fied basis. On the other hand, the use of a large number of indicators
may result in a problem of character redundancy.
Because of its multidimensional nature, lacustrine trophic state
is amenable to analysis by multivariate statistical techniques. Multi-
variate techniques minimize the personal bias often present when data
Table 2-M. Trophic Indicators
Standard crop of algae and aquatic plants3
Amount of suspended solids3
Volume of algae3
Chlorophyll levels3
Number of algal blooms3
Transparency3
Plant regression3
Photosynthesis
Primary production
Aquatic plant nutrient content
Hypolimnetic oxygen concentrations
Sediment composition
Dissolved solids
Conductivity
Nutrient concentrations
Cation ratio (Na + K) / (Mg + Ca)
3Indicator may be remotely sensed using operational or near-operational
sensors.
2-32
-------
are examined for groups and rankings are developed. They are of parti-
cular value in situations where large numbers of objects or parameters
are to be classified. Principal components analysis is one such technique
meriting consideration as an approach to the problem of trophic index
development.
2. Principal Components Ordination
Principal components analysis may be used to reduce the dimensional-
ity of a multivariate system, such as water trophic indicators, by
representing the original attributes as functions of the attributes
(Boland 1976). This approach was also used by Shannon (1970) when he
undertook the establishment of trophic state relationships for lakes
in Florida. Wezernak, Tanis, and Bajza (1976) replicated this approach
as well.
The objective of the principal components approach is to combine
all of the various water quality measurements into a single numerical
expression. In undertaking the principal components analysis approach
it is required that the initial parameters selected (i.e.. , the six
trophic indicators) exhibit intercorrelations. The resulting index,
the first principal component (PC1), thus represents the maximum total
variation of any of the components.
The computation of principal components can be undertaken using
either a covariance matrix (S) or a p x p matrix of Pearson product-
moment correlation coefficients (r). Use of the r-matrix is indicated
when the variates are measured in different units (.e.g.. , grams and
meters) . Computation of the r-matrix principal components involves
the extraction of its eigenvalues (characteristic or latent roots) and
eigenvectors (characteristic or latent vectors). The eigenvalues are
a set of nonzero, positive scalar quantities. The sum of the r-matrix
eigenvalues is the matrix trace and is equal to the number of dimensions
in the original system (i.e.. , the number of variates p). The rank
of the matrix is equal to p.
Normalized eigenvectors give the A-space coordinates of an orthogonal
set of axes known as the principal axes. The normalized eigenvectors
are commonly designated as principal components.
The first principal component of the observations of the p-variates
X-|,...,Xp is the linear compound
Y-| = an X1 + .. . + ap1 Xp
whose coefficients (a-|-|) are the elements of the eigenvector associated
with the largest eigenvalue of the r-matrix (Morrison 1967). The vari-
ance of the first principal component is associated with its eigenvalue.
The jth principal component of the p-variate system is the linear compound
X +...+ apjXp
2-33
-------
whose coefficients are the elements of the eigenvector associated with
the jth eigenvalue extracted from the r-matrix. The jth eigenvalue is
a measure of the variance of the jth principal component.
The proportion of the total sample variance in the cloud of dimen-
sionless standard scores attributable to any component is found by div-
iding its eigenvalue by p. The first principal component has the innate
property of explaining the greatest proportion of the sample variance,
and each successive component explains progressively smaller amounts of
the toal sample variance. Frequently, as a consequence of the decreasing
order of variance, k < p dimensions will adequately summarize the varia-
bility of the original variates X1f...,Xp. The first three components
generally account for most of the variation permitting the ordination
of the subjects in 1-D, 2-D, and 3-D space. All of the dispersion in
the data can be accounted for by using p dimensions, but this negates
the analysis objective.
The principal components of N p-variate observations may be defined
geometrically (Morrison 196?) as "...the new variates specified by the
axes of a rigid rotation of the original response coordinate system into
an orientation corresponding to the directions of maximum variance in the
sample scatter configuration." The normalized eigenvectors give the di-
rections of the new orthogonal axes and the eigenvalues determine the
lengths (.i.e.., variance) of their respective axes. The coordinate system
is expressed in standard units (zero means, unit variances) when the com-
ponents are extracted from the r-matrix. Figure 2-16 is a hypothetical
bivariate example of the geometric meaning of principal components.
Principal components may be interpreted geometrically as the variates
corresponding to the orthogonal principal axes of observation scatter in
A-space. The elements of the first normalized eigenvector (i.e.., coeffi-
cients of the first principal component) define the axis which passes
through the direction of maximum variance in the scatter of observations.
The associated eigenvalue corresponds to the length of the first principal
axis and estimates the dispersion along it. The second principal component
corresponds to the second principal axis, the length of which represents
the maximum variance in that direction. In our example the first component
accounts for most of the dispersion in the data swarm and the original
2-dimensional system could be summarized in one dimension with little loss
of information. The new variate value (PC1) for each lake is obtained
by evaluating the first component
Y-, = aX-, + bX2
The PC1 for each lake in 1-D A-space is its coordinate on the first com-
ponent axis and is shown diagrammatically by projecting each observation
to the principal axis.
Detailed descriptions of the theoretical and computational aspects
of principal components are found in Hotelling (1933a, 1933b, 1936),
Anderson (1957), and Morrison (1967).
-------
>
h-
U
Q
2
Q-
£
I
FIRST PRINCIPAL AXIS
CORRESPONDING TO THE
FIRST PRINCIPAL
COMPONENT:
Y, =aX] +bX2
SECOND PRINCIPAL AXIS
CORRESPONDING TO THE
SECOND COMPONENT:
Y =cX - dX
CHLOROPHYLL a (X)
Figure 2-16.
Geometrical Interpretation of the Principal Components
for a Hypothetical Bivariate System (modified from
Brezonik and Shannon 1971)
2-35
-------
G. MULTISPECTRAL CLASSIFICATION
The automated classification of remotely sensed water quality data
is undertaken to determine whether trophic patterns, as determined by
contact-sensed data, are discernible in multispectral data and to what
extent these patterns can be mapped and correlated in different water
bodies. Many algorithms exist which can be applied to multispectral data
for purposes of classification. The one of interest in this study is the
Bayesian maximum likelihood algorithm.
Like most spectral classification schemes, the Bayesian algorithm
assumes gaussian distribution of its classes. Training area statistics
are computed for each of the possible water quality types. These statis-
tics, which consist of a mean multispectral signature and a covariance
matrix, essentially instruct the classifier what it is to search for in
spectral space.
Given these statistics, the classifier decides to which possible
class a picture element belongs, based upon the probability of such an
occurrence. All a priori probabilities are assumed equal. This procedure
has the advantage of minimizing the probability of misclassification,
hence the name "maximum likelihood."
Assume n multispectral channels of data. Mathematically, each
picture element can be considered as an n-vector X = (x-| ,X2-. .xn) , where
Xj is the DN from channel J. For each training area p, compute the mean
vector X = (xjp,X2p.. .xnp) and the covariance matrix (n by n), Kp whose i,
jth element is the covariance between channels i and j. The decision for
class assignment is made by finding the class with the largest probability
density function at the point X.
The probability density function for class is given by
1
p = exp{-1/2(X - Xp)T K~1 (X - Xp)}
(2TT)n/2|Kpl1/2
where |Kp| = det (Kp). In order to save computer time, it is not P
that is computed, but rather loge (Pp). This is appropriate since
loge is a monotonic increasing function and since one is only concerned
with the P that gives rise to the largest P . Therefore one finds
max (Qp) where
Qp = loge(Pp) = Cp -1/2 (X - X )T K -1(X - Xp)
and
Cp = -1/2(n loge[2Tr] + loge |Kp|)
Each picture element is assigned to one of the possible classes in
this manner.
2-36
-------
The Bayesian maximum likelihood algorithm is considered to be
an expensive classification algorithm in terms of computer time neces-
sary. However, since only those picture elements which have been prede-
termined to be water are to be classified, and because of its sensitivity
to subtle differences in spectral signatures, the Bayesian maximum
likelihood algorithm continues to be used effectively in water quality
classification studies.
2-37
-------
SECTION III
METHODS
A.
DATA ACQUISITION
1. Contact-Sensed Water Quality Data
The water quality data were collected by helicopter-borne limnolo-
gists between August 22 and 25, 1975 inclusive (Table 3-1). The parameters
measured and techniques employed are described in U.S. EPA (1975). With
the exception of the algal assay yields, the data are reproduced in
Appendix A. The parameters commanding attention in this investigation
include chlorophyll a. (CHLA), conductivity (COND), inverse of Secchi
depth transparency (ISEC), total phosphorus (TPHOS), total organic
nitrogen (TON), and algal assay yield (AAY).
2.
Remotely Sensed Water Data
NASA-Houston provided an Orion P-3A aircraft equipped with a Bendix
11-channel modular multispectral scanner (MMS) and an aerial mapping
camera outfitted with a 15.25-cm Zeiss lens and Kodak type S0397 film.
The details of the August 25, 1975, flyover are found in Table 3-1 and
Appendix B.
Table 3-1. Acquisition Dates for Colorado Lake Data
Lake/Reservoir
Name
Date Number
Water of
Sampled Sampling
(1975) Stations
Data of Landsat
Aircraft Overpass
Flight Date
(1975) (1975)
Data Quality
Remarks
Barker R.
Barr L.
Blue Mesa R.
Cherry Creek R.
Cucharas R.
Dillon R.
Grand L.
Green Mt. R.
Aug
Aug
Aug
Aug
Aug
Aug
Aug
Aug
26
26
25
22
22
25
26
25
2
2
6
3
1
4
2
3
Aug
Aug
Aug
Aug
Aug
Aug
Aug
Aug
25
25
25
25
25
25
25
25
Aug
Aug
Aug
Aug
Aug
Aug
Aug
Aug
23
23
24
23
22
24
24
24
Aircraft MMS
sun glint
Aircraft MMS
sun glint
Cloud cover;
voir almost
Aircraft MMS
reser
dry
sun glint
3-1
-------
Table 3-1. Acquisition Dates for Colorado Lake Data
(Continuation 1)
Lake /Reservoir
Name
Holbrook L.
Meredith R.
Milton R.
Shadow Mt. R.
Date
Water
Sampled
(1975)
Aug 22
Aug 22
Aug 26
Aug 2?
Number
of
Sampling
Stations
1
3
2
3
Data of
Aircraft
Flight
(1975)
Aug 25
Aug 25
Aug 25
Aug 25
Landsat
Overpass
Date
(1975)
Aug 22
Aug 22
Aug 23
Aug 24
Data Quality
Remarks
Cloud cover on
Landsat
Cloud cover on
Landsat
Aircraft MMS
sun glint
Landsat-1 passed over the study area on August 22, 23, and 24. The
Landsat-1 coverage of August 22 was not processed because of excessive cloud
cover. Landsat-1 scenes of August 23 (5126-16474) and August 24 (5127-16532,
5127-16534) were selected for processing.
Of the 12 lakes under consideration, three (Cucharas, Holbrook, Meredith)
were eliminated because of cloud cover on the Landsat scenes. Of the nine
remaining lakes, four (Barker, Blue Mesa, Green Mountain, Shadow Mountain)
were dropped from the MMS analyses because of sun glint. For multispectral
analysis purposes, Landsat multispectral scanner (MSS) data were available
for nine lakes and MMS data for five of the same nine lakes (see Table 3-2).
Each of the Landsat computer compatible tapes (CCT's) contained MSS
data for all of the spectral bands (GRN, RED, IR1, IR2). Data acquisition
with the MMS was less successful, with readable data being available for
nine of the 11 channels. Only "noise" was found for Channels 5 and 6.
B.
MULTISPECTRAL DATA PREPROCESSING
Prior to attempting classifications of any sort, certain corrections
and processing are applied to both the Landsat and aircraft data. These
preprocessing functions are applied not only to correct the imagery for
cosmetic purposes but also for geometric reasons. The cosmetic processing
relates to correcting for MSS line dropouts, slipped or missing lines, and
other obvious defects in the MSS imagery. Similar types of corrections
are also applied to the aircraft MMS data (Figure 3-1).
In terms of geometric corrections, the Landsat CCT's are not in a
format compatible with the processing approaches used in the JPL Image
Processing Laboratory (IPL). The CCT's, as received from the EROS data
3-2
-------
center, have the data from the four MSS bands interleaved. The IPL soft-
ware program, VERTSLOG, unravels these interleaved data and creates a
separate image for each band. The program also applies the various geo-
metric corrections to the Landsat data. These corrections include scan
mirror velocity, panorama correction and resampling the data to create
an approximately 80-m IFOV (instantaneous field of view).
Landsat-1 MSS imagery is plagued by a striping problem, a consequence
of an imbalance in sensors detectors. The MSS data were also preprocessed
to reduce the magnitude of the striping problem, also known as "sixth
line" banding.
Similarly, the aircraft MMS data must be corrected so that the
imagery has a 1:1 aspect ratio. The most obvious distortions seen in this
imagery are square fields elongated into rectangles or circular irrigation
patches which appear as ovals of high ellipticity. The cause of this
appearance is related to a nonsynchronization of aircraft speed with the
MMS scan mirror sweep rate. Figures 3-2 and 3-3 show circular fields
before and after the aspect ratio has been adjusted.
Table 3-2. Availability of Remotely Sensed Data for 12
Colorado Lakes and 32 Lake Sampling Sites
Lake /Reservoir
Name
Barker R.
Barr L.
Blue Mesa R.
Cherry Creek R.
Sampling
Lake STORET Site STORET
Number Number
0801
080101
080102
0802
080201
080202
0803
080301
080302
080303
080304
080305
080306
0804
080401
080402
080403
Zeiss
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
MSS
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
MMS
X
X
X
X
X
X
X
3-3
-------
Table 3-2. Availability of Remotely Sensed Data for 12
Colorado Lakes and 32 Lake Sampling Sites
(Continuation 1)
Lake/Reservoir
Name
Cucharas R.
Dillon R.
Grand L.
Green Mt. R.
Holbrook L.
Lake Meredith R.
Milton R.
Shadow Mt. R.
Lake STORET
Number
0805
0806
0807
0808
0809
0810
0811
0813
Sampling
Site STORET
Number
080501
080502
080601
080602
080603
080604
080701
080702
080801
080802
080803
080901
081001
081002
081003
081101
081102
081301
081302
081303
Zeiss
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
MSS
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
MMS
X
X
X
X
X
X
X
X
X
X
X
3-4
-------
PARAMETER
SCALE ADJUSTMENT -
ALTITUDE VARIA-
TIONS WITHIN A
FRAME
9 CORRECTIONS
WITHIN THE
FRAME
ATTITUDE
VARIATIONS
WITHIN A FRAME
ROLL
9 CORRECTIONS
WITHIN A FRAME
PITCH
9 CORRECTIONS
YAW
9 CORRECTIONS
WITHIN A FRAME
IMAGE SKEW
CAUSED BY EARTH
ROTATION
(FUNCTION OF
LATITUDE)
AVERAGE VELOCITY
CHANGE FROM
NOMINAL
IMAGE SKEW
CAUSED BY FINITE
SCAN TIME
GEOMETRIC
FOOTPRINT
rv \~~i
1 / 1 \
/
/| S/C \
/ i \
AX— H h-
AX— •>] N-
\! V
v^
i 1 s/c ! 1
AY
,— a
i 1 f
I
"^
\Z'~&(>fy
AY
-*) |^AX
'/ !/
\
l\ vc /!
! AY
__5^c__jL
t
SCAN
-4— »• DIV.
t
S/C AX
MAGNITUDE OF
CORRECTION -
METERS ON
THE GROUND1
AX ~~ 4 Ah
9.26X 10 ^-
FRAME)
AX*hA0D
K
AY - h A 6 p
AX- 2
4.63X 104 (A9^)
AY*
9.26X I04A0^
AX~
8.88 X 104 HER
AY*
S.SSXIO4^
^ n
(AT TOP AND
BOTTOM
OF FRAME)
AX* 218 METERS
PARAMETER
SCAN MIRROR
VELOCITY
CORRECTION
PERSPECTIVE
CORRECTIONS
SWEEP TO
FRAME IN DIRECTION
OF SCAN
FRAME CORRECTION
IN DIRECTION OF
SPACECRAFT TRAVEL -
(EBR CORRECTION)
ALIGNMENT
CORRECTIONS -
RELATIVE
ALIGNMENT OF MSS
TO AMS
ROLL
PITCH
YAW
GEOMETRIC
FOOTPRINT
BAND 1 |
| BAND 2 |
BANDS |
| BAND 4 |
* »-| — AX
_J • — AX
— J U-AX
25 ACTUAL PATH
1 I
§ 4,X /^PATH FOR
O i/ CONSTANT
g // .SCAN RATE
<° T.MP TS
TIME •>
v/^V<^SCAN
/T~\. KATE
/ ! V.AT\
/ / EARTH \
-•jf^^AX
SCAN
EBR~^J DIR.
7 f
/ TF^S/C;
o [ w
— »4 -«-AY
1 1
1
1
1
L 1
AX H h—
-|
AX
L
t
r^x~^ti
r/-^^^7r|— I
/ y>^~^
N^^ AA0/
MAGNITUDE OF
CORRECTION -
METERS ON
THE GROUND
AX =
112 METERS
AXMAX^
395 METERS
AXMAX =
115 METERS
AYMAX =
64 METERS
AX = hAA.
K
AY = hAA
P
AX*
8.88 X 104AAi/>
AY*
9.26 X 10 AAi//
Figure 3-1. Geometric Corrections Typically Applied to
Multispectral Scanner Data
3-5
-------
n*
Cr«"<
4 tr 11
,1. «\ ...
0,540 - o.5*onicMifr
MI" -ION :i" CITE
cue s
0 LINC
13
S«M rot 9. JPU/IPL
Figure 3-2. Aircraft-Acquired MMS Imagery Before Geometric
Corrections. The circular fields have been
distorted into highly elliptical figures. The
water body is Milton Reservoir.
See Figure 3-3
3-6
-------
MM: CHWfCL 4 IF II 0.540 0.580 CLSC 8
._ ,, HI;SIDN 317 SITE 0 LINE 13 - PEWflR - *CLttSf>
STRETCH 70-105
Figure 3-3- Aircraft-Acquired MMS Imagery After Geometric
Corrections Have Been Made
3-7
-------
C. LAKE EXTRACTION METHODS
The primary thrust of this task is water quality monitoring and lake
classification. At this time, the project is not concerned with land use
or land use practices as they relate to water quality. The image processing
techniques used were those designed to extract and manipulate MSS and MMS
pixels representing surface water. The extraction procedure, explained
in detail by Blackwell and Boland (1975), is outlined as follows using
Landsat MSS for illustrative purposes. The approach uses the digital
data contained on CCT's.
The MSS data for a Landsat scene are rescaled to 8 bits of precision
using an IBM 360-65 and associated software and peripherals. The range of
the new digital number (DN) scale is from 0 to 255 or a total of 256 DN
levels for each of the four Landsat MSS bands. A hard copy image is then
generated from the rescaled IR2 (Band 7) data. Using the newly generated
photograph, a candidate lake is selected from the scene and a polygon
is constructed around it. The polygon's coordinates are input to the
computer system and four images are generated of the newly defined Landsat
subscene, one for each of the four MSS bands. Each image, representing
both water and surrounding terrain, and its histogram of DN values, is
concatenated into a single photograph along with the three images
representing the remaining bands (Figure 3-*O •
Through inspection and after comparative testing it has been determined
that the IR2 DN value of 28 provides optimum segregation of water and land
features. A binary mask is developed from the IR2 extracted lake image.
With this method, IR2 data values between 0 and 28 are set equal to 1 and
all other IR2 DN values (29 to 255) are set equal to 0. The binary mask,
in which water pixel values equal 1 and non-water pixel values equal 0,
is then used to eliminate all but water-related features in the subscene.
Multiplication of each MSS band subscene image (4 (GRN), 5 (RED), 6 (IR1),
7 (IR2)) pixel by its IR2 binary mask counterpart produces an image for
each band. If processed correctly, the images will represent only pixels
containing water-related information.
Figure 3-5 is a concatenation of the four subscene images after multi-
plication by the counterpart IR2 binary mask. Some final editing may be
required to eliminate rivers, streams, and other water-related features not
considered to be part of the lake proper. Once this has been accomplished,
listings are generated of pixel counts, DN histograms, mean DN values for
each band for the entire water body along with their associated standard
deviations. At this point, the lake (or reservoir) is treated as a whole
and mean DN values are for the overall water body.
Each of the nine lakes for which Landsat MSS data were available was
processed in this manner. After final editing, the nine lakes were con-
catenated into a single image for further processing. Figure 3-6 is an
example of the process output. The aircraft 11-channel MMS data were
processed in a similar manner. Channel 10 data were used in the generation
of the binary mask with a DN value of 60 being selected as the water cut-
off point (i.e.. , pixels with DN values of 60 or less were considered as
representing water). Figure 3-7 is an IR2 concatenation of the five lakes
for which MMS data were available.
3-8
-------
UJ
I
Figure 3-
CRN, RED, IR1, and IR2 Images of a Landsat Scene 5127-16532 Subscene. The histograms depict the
DN distributions for the subscene including both the water body and terrain. In the IR2 band, most
of the water-related information is found in the DN range of 0-28. In most cases, IR2 DN values
greater than 28 are related to terrestrial and atmospheric phenomena. The water body is Dillon
Reservoir. Note the small cloud just southeast of the Blue River Arm. Its shadow is cast to the
northwest, falling in part on the water, with most falling on the peninsula separating Frisco and
Giberson Bays from the Blue River Arm
-------
uo
o
Figure 3-5. CRN, RED, IR1, and IR2 Images of a Landsat Scene 5127-16534 Subscene After the Application of
the Binary Mask Generated Using the IR2 DN Range of 0-28 as Representing Water. See Figure 3-4.
The histograms depict the DN distributions for the pixels comprising Dillon Reservoir. At this
stage, the computer was unable to separate the shadow from the reservoir and it was necessary to
manually override the computer, eliminating the cloud shadow through the use of an editing program
-------
fHi
S130J
11 i i . 11 nli..»I. i! i J . > i. I IK 11 > it i In., 11 ml i... I 11 111 11 i J . < j I. i it'< - . 11 > 111.
LANDSAT CLASSIFICATION SAMPLE SITES
TEXTAD
BOX
THU DCT 149 JPL/IPL
Figure 3-6. Landsat-1 MSS IR2 Concatenation of Nine Colorado
Lakes. The boxlike figures define the MSS sampling
sites and encompass the lake sampling stations
3-11
-------
Figure 3-7. MMS Channel 10 Concatenation of Five Colorado
Lakes. The boxlike figures define the MMS
sampling sites and encompass the lake sampling
stations
3-12
-------
D. WATER SAMPLE SITE LOCATIONS
Unlike remote-sensing problems related to land features, the problem
of contact sampling site location and stability of the nature of the material
or substance being sampled is an order of magnitude more difficult with
water. Even when costly electronic positioning equipment such as Loran
is used, there is a high degree of ambiguity inherent in ascertaining
sampling site locations on large lakes. The field crews located the
position of the helicopter (i.e.. , the sampling site) by sighting on
prominent land or cultural features with the helicopter compass and
then estimating its distance from the shore and/or features. These
locational data were recorded in the field notes and subsequently entered
into the STORET data system along with the trophic indicator data.
Color photographic prints of the 12 lakes were made from the color
transparencies supplied by NASA's Johnson Space Center. The aerial
transparencies were taken from absolute altitudes ranging from 4600 to
6400 m using a Zeiss camera equipped with a 15.25-cm lens. Using the
location data from STORET and sampling site locations marked on topographic
maps by the field crew, each lake sampling site was pinpointed on the
color prints by EPA personnel.
Since the Zeiss camera imagery also included sufficient cultural and
landform features it was possible to establish visual ground control tie
points in the extracted Landsat imagery as well as the aircraft 11-channel
multispectral scanner imagery. With the aid of a Bendix datagrid coordinate
digitizer it was possible to digitize the tie-point locations in the photo-
graphs as well as the sample site locations. With these data coordinates,
a geometric transformation was made to map these locations onto the Landsat
and aircraft multispectral scanner images. These locations are seen as
the small rectangles within each lake in Figures 3-6 and 3-7.
Procedurely, the process consists of mapping a blank image containing
tie-point locations and sample site positions on a lake to the same set
of tie-points on the LANDSAT and aircraft imagery.
E. PIXEL DENSITY AT SAMPLE SITES
Once the sample site locations had been established for each lake,
a decision was necessary relative to the matrix size of pixels to be used
for the spectral analysis. In the previous task for the Wisconsin study
(Boland 1976), all of the pixels for each lake had been used as the spectral
training site. The selection of the appropriate matrix size for this
application was guided in large part by the shape and size of some of the
lakes. Some lakes, such as Blue Mesa, had very narrow sections which
were only 4 to 6 pixels in width. With one or two exceptions the matrix
size selected was 5-by-5, 4-by-4, or 3-by-3, depending on the lake and
the sample location. The Landsat imagery being used was resampled to
produce pixels measuring approximately 80 by 80 m. More specifically,
the resampling produced square Landsat pixels representing an earth
surface distance of 79.98 m per pixel edge. A resampled pixel represents
3-13
-------
an area of 6,396.8 m2 (68,854.6 ft2) or 0.6396 ha (1.58 acres). Therefore,
the following pixel matrix sizes will represent
5-by-5 matrix = 15.99 ha = 39.54 acres
4-by-4 matrix = 10.23 ha = 25.30 acres
3-by-3 matrix = 5.76 ha = 14.23 acres
on the earth's surface. Figure 3-8 illustrates the areas these matrix
sizes encompass relative to a standard U.S. one-mile-square section.
The aircraft modular multiband scanner (MMS) which was used to
acquire intermediate altitude multispectral data of the same lakes produced,
after geometry corrections, imagery with an average pixel resolution equal
to 15 m. The pixel density or matrix size used to obtain MMS training site
statistics was 11-by-11. This results in a ground area surface coverage of
2.72 ha (6.727 acres) per H-by-11 MMS matrix sample site.
F. LAKE SURFACE AREA DETERMINATION
The surface area of many of the study lakes fluctuates greatly as a
consequence of evaporation, and more importantly, drawdown, a result of
irrigation and hydroelectric power demands. The use of area figures
derived from reports and topographic sheets can only serve as "rough"
reference values and are of little use in evaluating the area prediction
capabilities of remote sensors in this study. Grand Lake and Shadow
Mountain Reservoir are possible exceptions, because water level fluctua-
tions are limited to 0.3 m by law.
An effort was therefore undertaken to determine the lakes' area
using the well-accepted practice of taking the relevant measurements from
vertical aerial photographs. The photos in this case were the 9-by-9 inch
color prints supplied by JPL from the NASA overflight of August 25, 1975.
The area of each lake was measured on its respective prints using a
Numonics Model 253 electronic planimeter. The area of a lake's photo-
graphic image was measured four times, an average was computed, and then
converted to hectares using the appropriate conversion factors. The
results are displayed in Table 3-3- The computed values will serve as
"target" figures for the Landsat MSS and MMS. Aerial photography was
available for the 12 NES lakes and the calculations were made for each.
Landsat MSS-derived surface area estimates were made for 9 of the 12
lakes. The estimate for a specific lake was made by multiplying the sum
of lake MSS pixels by the appropriate conversion factor. In other words,
Area ^a) = Z pixels x 0.6396 ha/pixel
MMS-derived surface area estimates were made for 5 of the 12 lakes. The
estimate was made by multiplying the sum of the lake MMS pixels by the
appropriate conversion factor.
Area (ha) = £ pixels x 0.0225 ha/pixel
An average pixel size of 225 m2 was used for the MMS calculations.
3-14
-------
STANDARD U.S. SECTION:! SQUARE MILE
1/4 SECTION
160 ACRES
1/16
SECTION
40 ACRES
! 10
IACRE
; 1 PIXEL ^
J(80X 80 METERS)^?
I
5280
Figure 3-8.
1 PIXEL = 1.58 ACRES = 0.64 HECTARES
3X3 - PIXEL MATRIX = 14.23 ACRES = 5.76 HECTARES
4 X 4 - PIXEL MATRIX = 25.30 ACRES = 10.23 HECTARES
5 X 5 - PIXEL MATRIX = 39.54 ACRES = 15.99 HECTARES
Landsat Pixel Size in Relation to
the U.S. Standard One-Mile Section
3-15
-------
Table 3-3. Area of Colorado Lakes As Determined B'rom NASA Aerial Photographs
oo
I
Lake/Reservoir
Name
Barker R.
Barr L.
Blue Mesa R.
Cherry Creek R.
Cucharas R.
Dillon R.
Grand L.
Green Mt. R.
Holbrook L.
Meredith R.
Milton R.
Shadow Mt. R.
County
Boulder
Adams
Gunnison
Arapahoe
Huerfano
Summit
Grand
Summit
Otero
Crowley
Weld
Grand
STORE!
Number
0801
0802
0803
0804
0805
0806
0807
0808
0809
0810
0811
0813
Pool
Elevation ,
m
2
1
2
1
1
2
2
2
1
1
1
2
,495
,553
,292
,691
,750
,804
,550
,423
,269
,297
,463
,550
Absolute
Aircraft
Elevation, a
m
5
5
5
5
5
5
5
5
6
6
5
5
,670
,760
,740
,640
,880
,670
,700
,820
,400
,370
,850
,670
Photograph
Scale
1:37
1:37
1:37
1:37
1:38
1:30
1:37
1:38
1:42
1:41
1:38
1:37
,200
,800
,667
,000
,600
,200
,400
,200
,000
,800
,400
,200
Lake Area on
Photograph,
?
cm^
5
32
217
25
1
96
14
52
4
55
26
36
.87
.12
.62
.74
.90
.60
.88
.23
.86
.72
.22
.94
Lake Surface
Area,
ha
81
459
308ab
352
26
1337
208
762
86
974
386
511
aBased on aircraft radar altimeter.
DSeveral small areas of the lake fell outside the camera's field of view.
-------
G.
TROPHIC INDICATOR SELECTION AND MULTIVARIATE INDICES DEVELOPMENT
1.
Tropic Indicator Selection
The NES, in the selection of water quality parameters, had to
weigh parameter usefulness, length of acquisition time, and complexity
of data reduction as well as other factors against resources, total
number of lakes, and the element of time. Similarly, this particular
feasibility study was constrained in the selection of trophic state
indicators, with the following being selected for incorporation into
a trophic state index (Table 3-4):
Table 3-H. Acronyms Used for Trophic Indicators
Trophic Indicator
Units
Acronym
1.
2.
3.
4.
5.
6.
Chlorophyll a.
Secchi disc transparency, inverse
Secchi disc transparency
Total phosphorus
Total organic nitrogen
Conductivity
Algal assay control yield3
|ig/l
m, 1/m
mg/1
mg/1
(j.mhos
mg/1
[dry weight]
CHLA
SECCHI, ISEC
TPHOS
TON
COND
AAY
aControl samples of lakes water are spiked with various concentrations
of phosphorus, nitrogen, and phosphorus plus nitrogen. Test cells of
Selenastrum Capricornutum are injected into flasks with the controlled
nutrients and allowed to incubate. The maximum growth attained is then
quantified in terms of dry weight (mg/1).
In addition to being incorporated into a multivariate trophic index,
each of the above parameters was used individually as the dependent
parameter variable in attempts to develop regression models employing
sensor bands or channels as independent variables. The trophic indicators
had been used with some success in previous studies (.§_.£., Boland, 1976).
2.
Multivariate Trophic Index Development
The indicators selected for principal components analysis are those
previously listed: conductivity (COND, ^mhos), chlorophyll a (CHLA, |jg/l),
algal assay yield (AAY, dry weight in mg) and Secchi disc transparency
(Secchi, m). The inverse of Secchi disk transparency was employed
(ISEC, m~1) so that all indicator values would increase as the trophic
state increases. Since the raw data seen in Table 3-5 are skewed,
each trophic indicator value was natural log (LN) transformed to give
a distribution more closely approximating a normal one. The transformed
3-17
-------
trophic indicator data are identified as follows: LNCHLA, LNISEC,
LNCOND, LNTPHOS, LNTON, LNAAY. The data matrix was further standardized
(zero mean, unit variance) by attributes using the relationship
j - Xi
where z^.; is the standardized values for attribute i of observation j
(I.e.., lake), Xj_j is the LN-transformed value of observation j, and x^ and
s^ are the mean and standard deviation of attribute i, respectively. Eigen-
vectors and eigenvalues were then extracted from the associated correlation
matrix. Mext the first normalized eigenvector (principal component) was
evaluated for each.
As described in some detail below, three multivariate trophic indices
were developed for the Colorado lakes using the above approach. These indi-
ces are identified as PC1-11, PC1-13, and PC1-27. The PC1-11 was developed
using "whole lake" parameter values; the others employed parameter values
for a specific number of sampling sites (13 and 27), each treated as a sepa-
rate observation or entity. While the original intention was to develop a
"whole lake" index based on 12 lakes and a site index based on a total of 38
sampling sites, a combination of missing contacted-sensed and remotely sensed
data resulted in the development of the three indices reported here.
a. Eleven-Lake Index (PC1-11). The principal component-derived
ranking of Colorado Lakes was accomplished using the mean trophic indicator
values calculated from data collected for 11 lakes during August 22-26, 1975
(Table 3-5). As the data set for Lake Meredith was not complete at the time
the analysis was run, it was excluded. The analysis was undertaken using
the Statistical Interactive Programming System (SIPS) on a Control Data
Corporation CDC 3300 digital computer at Oregon State University. The
procedure used was that reported by Boland (1975, 1976) with one important
exception. The original analysis, performed on 100 NES-sampled lakes,
used trophic indicator values averaged over a sampling year consisting
of three sampling rounds (spring, summer, fall); this time only the
summer sampling round data were used. In both cases, the lakes were
treated to generate a trophic scale on which each lake would have a
number (i.e.., trophic index value (PC1)) defining its position.
b. Twenty-Seven Site Index (PC1-27). In addition to treating each
of the 11 lakes as an entity, the 27 sampling sites located on the 9
Colorado lakes for which Landsat coverage was available were also ordinated
using principal components analysis. The same six trophic indicators were
used but, with the exception of AAY, the data were averaged at each station
where multidepth sampling occurred. The AAY values were determined from
composite water samples, each representing all or several lake sampling sites,
Therefore, the AAY values used for the 27 site principal components analysis
were averages for whole lakes or, as in the case of Blue Mesa Reservoir,
representative of two sets of of several sampling sites each.
3-18
-------
Table 3-5. Trophic Indicator Values for 12 Colorado
Lakes for August 1975 Sampling Period
Lake/Reservoir STORET
Name Number
Barker R.
Barr L.
Blue Mesa R.
Cherry
Creek R.
Cucharas R.
Dillon R.
Grand L.
Green Mt. R.
Holbrook L.
0801xxa
080101
080102
0802xx
080201
080202
0803xx
080301
080302
080303
080304
080305
080306
0804xx
080401
080402
080403
0805xx
080501
0806 xx
080601
080602
080603
080604
0807xx
080701
080702
0808xx
080801
080802
080803
0809xx
080901
CHLA,
Hg/1
3-7
3-7
3.7
51.7
74.4
29.0
4.9
6.0
4.1
4.6
4.2
5.2
5.4
48.7
9.8
124.6
11.6
27.4
27.4
2.3
2.2
2.5
2.1
2.4
5.5
5.5
5.4
8.3
7.9
7.1
9.8
146.9
146.9
ISEC,
m~^
0.461
0.410
0.525
2.187
3.281
1.640
0.490
0.547
0.787
0.437
0.410
0.394
0.525
1.373
1.094
1.514
1.640
3.937
3.937
0. 120
0.219
0.193
0.066
0.156
0.398
0.410
0.386
0.471
0.410
0.492
0.525
5.624
5.624
SECCHI,
m
2.
2.
1.
0.
0.
0.
2.
1.
1.
2.
2.
2.
1.
0.
0.
0.
0.
0.
0.
8.
4.
5.
15.
6.
2.
2.
2.
2.
2.
2.
1.
0.
0.
17
44
91
46
31
61
04
83
27
29
44
54
905
73
91
66
610
25
25
33
57
18
24
40
52
44
59
13
44
03
91
17
17
TPHOS,
mg/1
0.015
0.016
0.014
0.747
0.761
0.733
0.022
0.059
0.020
0.019
0.022
0.025
0.020
0.054
0.041
0.089
0.043
0.263
0.263
0.009
0.008
0.009
0.011
0.006
0.011
0.012
0.010
0.010
0.009
0.010
0.013
0.367
0.367
TON,
mg/1
0.
0.
0.
1.
1.
1.
0.
0.
0.
0.
0.
0.
0.
0.
0.
1.
0.
1.
1.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
2.
2.
180
180
180
623
890
357
277
380
180
180
197
143
150
816
533
130
710
48
48
190
200
180
180
158
116
180
197
237
180
180
380
96
96
AAY,
mg/1
0.
186.
0.
0.
0.
0.
0.
0.
0.
3.
1.
1.
0.
0.
0.
63.
5
3
7
4
4
4
9
9
9
2
9
9
3
2
3
3
COND,
(amhos
30
29
32
595
597
593
152
123
132
160
167
180
180
600
637
586
571
849
849
92
89
93
91
89
7
8
5
107
109
105
109
2368
2368
3-19
-------
Table 3-5. Trophic Indicator Values for 12 Colorado
Lakes for August 1975 Sampling Period
(Continuation 1)
Lake/Reservoir STORET
Name Number
Meredith R.
Milton R.
Shadow Mt. R.
08lOxx
081001
081002
081003
08l1xx
081101
081102
081 3xx
081301
081302
081303
CHLA,
Hg/1
146
151
151
138
12
8
16
6
8
6
3
.9
.3
.3
.0
.2
.3
.0
.2
.1
.5
.9
ISEC,
m~1
3.
3.
3.
3.
0.
0.
0.
0.
0.
0.
0.
698
937
579
579
787
656
984
562
656
525
525
SECCHI,
m
0.27
0.25
0.28
0.28
1.27
1.52
1.02
1.78
1.52
1.91
1.91
, TPHOS,
rag/1
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
357
429
406
236
720
714
728
025
049
021
018
TON,
mg/1
4.
4.
4.
3.
1.
1.
1.
0.
0.
0.
0.
21
62
52
48
092
050
155
320
330
347
330
AAY , COND ,
mg/1 (amhos
40.2 7095
7096
7094
7095
7.2 1295
1304
1282
0.5 24
24
25
24
aSTORET numbers with xx, e.g., 0801xx, represent average values.
3-20
-------
c. Thirteen-Site Index (PC1-13). MMS coverage was available for
only 5 of the 12 study lakes. A trophic ranking was developed for the
13 sampling sites on the lakes (Barr, Cherry Creek, Grand, Dillon, and
Milton) using the procedure described in the preceding paragraph.
H. ANALYSES OF TROPHIC INDICATOR, TROPHIC STATE INDEX,
AND REMOTELY SENSED DATA RELATIONSHIPS
1. Introduction of Analysis Methods
As stated earlier, the specific objectives of the study include an
evaluation of the Landsat MSS's capabilities, when used in conjunction
with contact-sensed data, to (a) estimate lacustrine trophic state, (b)
estimate several trophic state indicators including Secchi transparency
and chlorophyll a., and (c) to aid in the development of lake thematic
photomaps which depict trophic indicator magnitudes and trophic state
as defined by a numeric index.
As the project was originally conceived, contact and remotely
sensed data (aerial photography, MMS, MSS) would be obtained from 12
lakes including 33 sampling sites. Unfortunately, cloud cover and
sun glint reduced the number of lakes and sampling stations available
for analysis purposes (Table 3-2). In addition, the MMS did not produce
usable data in channels 5 and 6.
Two basic approaches were employed to evaluate the feasibility of
using Landsat MSS and Bendix MMS digital data in lake classification and
monitoring programs:
(1) Correlation/regression approach.
(2) Baysian maximum likelihood-derived thematic mapping approach.
2. Correlation/Regression Method
Correlation and regression analysis is one approach used to deter-
mine the feasibility of using Landsat MSS and MMS data for the estimation
of trophic indicator and trophic index magnitudes. Passive remote sensors,
such as the Landsat MSS, are not capable of sensing all of the lake trophic
indicators of interest to limnologists. For example, the MSS is not able
to sense directly nutrients such as phosphorus and nitrogen. However, phos-
phorus and nitrogen can and often do stimulate the production of algae and
macrophytes which can have a measurable effect on the water body's volume
reflectance. Another trophic indicator, conductivity, also cannot be directly
sensed by the multispectral scanners but may still correlate with the remotely
sensed data because of secondary effects.
a. Dependent Variables. Correlation and regression analyses were
made to determine statistical relationships between the remotely sensed
data, both MMS and MSS, and the following contact-sensed parameters:
3-21
-------
(1) Chlorophyll a. (CHLA, LNCHLA).
(2) Inverse of Secchi depth (ISEC, LNISEC).
(3) Secchi depth (SEC, LNSEC).
(4) Total phosphorus (TPHOS, LNTPHOS).
(5) Total organic nitrogen (TON, LNTON).
(6) Algal assay yield (AAY, LNAAY).
(7) Conductivity (COND, LNCOND).
(8) Multivariate trophic state indices (PC1-11, PC1-13, PC1-27).
The above-listed trophic indicators and indices are treated as dependent
variables and the MSS bands and MMS channels as independent variables.
b. Model Development and Adequacy Criteria. Most of the statis-
tical analyses were made using the Oregon State University Statistical
Interactive Programming System (SIPS) on a CDC 3300. The regression
analyses were made using the backward selection procedure. Many different
regression models were developed. Model selection was made using a
combination of several guidelines:
(1) Multiple correlation coefficient: the larger the better.
(2) Mean residual square: the smaller the better.
(3) Individual regression coefficients: significant at 0.05
level of probability.
(4) Calculated F-value for regression: significant at 0.05
level of probability.
(5) Number of independent variables: avoid many because the
number of observations is small.
The number of observations available for analysis purposes was
relatively small (n = 9, 13, 27), thereby making model saturation an
acute problem. For example, it is possible to obtain a large multiple
correlation coefficient (R^) simply by adding more independent variables
to the regression model. In the case of Landsat MSS data the investigator
has four basic values to work with (i.e.., CRN, RED, IR1, IR2) in his
regression analysis. However, the situation rapidly becomes complex
when band ratios and other functions are used. Excluding the thermal
channel, the MMS produced usable data in eight channels, and when the
possible functions are considered, the picture becomes even more com-
plex. With exception of some of the correlation analyses, the modeling
effort on SIPS was limited to using the basic Landsat MSS and MMS values.
Table 3-6 lists Landsat MSS functions considered and investigated by
personnel at the Jet Propulsion Laboratory.
3-22
-------
c. Analyses as Applied to Five Groups of Lakes and Sample Sites.
The fragmentary character of the remotely sensed data, a consequence of
cloud cover and sun glint, precluded the planned analyses of (a) 12 "whole"
lakes and (b) 32 sampling sites. Instead, the modeling efforts were applied
to the contact and remotely sensed data as segregated into five sets on the
basis of remotely sensed data availablity. The analysis groups are
(1) Landsat MSS nine "whole" lake set.
(2) Landsat MSS 27-site set.
(3) Landsat MSS 13-site set.
(4) MMS 13-site set.
(5) Modified MMS 13-site set.
Each of the above approaches will be discussed below. The results of the
correlation analyses and regression modeling efforts are reported in
Section IV, Results and Discussion.
1) Landsat MSS Nine "Whole" Lake Set. This effort utilized
Landsat MSS DN lake means for each of nine lakes along with their mean
trophic indicator values. The trophic indicator means are found in
Table 3-5 and the Landsat data are in Table 3-7. The trophic index
values (PC1-11), generated through the principal components analysis
of 11 Colorado lakes, are given in Table 4-4.
2) Landsat MSS 27-Site Set. Twenty-seven sampling sites
were included in this modeling effort. The sites are located on the
nine lakes for which Landsat MSS coverage was available. The trophic
indication means are found in Table 3-5 and the Landsat data are in
Table 3-8. The trophic index values (PC1-27) generated through the
principal components analysis of water truth from the 27 sites, are
given in Table 4-8.
3) Landsat MSS 13-Site Set. Although MSS data are available for
the 27 sampling sites, MMS data are available for only 13 sites on a total
of 5 lakes (Barr, Cherry Creek, Dillon, Grand, Milton). In an effort to
compare Landsat MSS data with Bendix MMS data, correlation and regression
analyses were performed using MSS data from the same 13 sites for which MMS
data are available. The contact-sensed data used in the correlation and
regression analyses are found in Table 3-5 and the Landsat MSS data in
Table 3-8. The trophic state index (PC1-13) values, produced from a
principal components analysis of the contact-sensed data for the 13 sites,
are given in Table 4-11.
4) MMS 13-Site Set. MMS data are available for 5 of the 12 lakes
(Barr, Cherry Creek, Dillon, Grand, Milton). With the exception of surface
area determination, no "whole" lake analyses were conducted on the 5 lakes
3-23
-------
using the MMS data in Table 3-9. The number of observations (N = 5)
was considered to be too small to be of much value. Instead, an effort
was made to analyze the MMS data for the 13 sites.
Correlation and regression analyses were run using the water truth
in Table 3-5, the MMS channel values in Table 3-10, and the trophic index
values in Table 4-11. The regression modeling was accomplished using two
different approaches:
(1) Eight-channel approach: The modeling was initiated using
eight of the nine available MMS channels as independent
variables. The thermal channel was excluded. The full model
was developed and then the backstep procedure was employed.
(2) Four-channel approach: Four MMS channels approximating the
four Landsat MSS bands were selected for analysis. The MMS
channels (4, 5, 8, 9) were then used as independent variables
with the trophic indicators and trophic index serving as
dependent variables.
5) Modified MMS 13-site Set. A correlation analysis of the
MMS data for the 13 sampling sites indicated that, with exception of
Channel 10, strong correlations exist between channels (Table 3-11).
Mueller ( 1972) demonstrated the use of principal components analysis
as a means of reducing the dimensionality of the data and at the same
time generating new variables that are orthogonal. With this in mind
the MMS data for the 13 sites were processed on SIPS using the multi-
variate subsystem principal components analysis program. The program
outputs included the MMS-related variables: MMSPC1, ... MMSPC8. The
data relating to the new variables are given in Table 4-16.
As in the "MSS 13-Site Set", two approaches were employed in the
development of regression models for the prediction of trophic indicator
and index magnitudes:
(1) Eight-variable approach: The modeling was initiated
using the eight newly defined MMS variables (MMSPC1
MMSPC8) as independent variables. The full regression model
was developed and then the backstep procedure was employed.
(2) Four-variable approach: The first four principal component-
derived MMS variables (MMSPC1, MMSPC2, MMSPC3, MMSPC4),
representing 99-54? of the variation in the MMS data were
then used as independent variables with the trophic indicators
and trophic index serving as dependent variables.
The contact-sensed data and trophic index values used to develop
regression models through the above two approaches are found in Tables
3-5 and 3-7, respectively.
3-24
-------
Table 3-6. Landsat MSS Functions Investigated in
Statistical Stepwise Regression Analysis,
MSS data included means for 9 lakes and
27 sampling sites
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
18.
19.
20.
21.
22.
23.
24.
25.
26.
27.
28.
29.
30.
31-
32.
33.
34.
35.
36.
37.
38.
39.
40.
GRN
RED
IR1
IR2
GRN /RED
GRN/IR1
GRN/IR2
RED/IR1
RED/IR 2
IR1/IR2
LN (GRN )
LN(RED)
LN(IR1)
LN(IR2)
GRN + RED/IR 1
GRN - RED/IR 1
RED - IR1/RED
IR1 - IR2/IR1
LN(GRN)/AVG
LN(RED)/AVG
LN( IRD/AVG
LN(IR2)/AVG
GRN - RED/IR 1
RED - IR1/RED
GRN - RED/IR 1
LN(GRN)/AVG x
LN(GRN)/AVG x
LN(GRN)/AVG x
LN(RED)/AVG x
LN(RED)/AVG x
LN (IRD/AVG x
LN(GRN)/AVG x
LN(GRN)/AVG x
LN(GRN)/AVG x
LN(RED)/AVG x
(LN(RED)/AVG x
GRN /SUM
RED/SUM
IR1/SUM
IR2/SUM
+ IR2
+ IR2
+ IR1
+ IR2
+ IR2 - RED - IR1/RED + IR1
+ IR1 - IR1 - IR2/IR1 + IR2
+ IR2 - IR1 - IR2/IR1 + IR2
LN(RED)/AVG
LN( IRD/AVG
LN(IR2)/AVG
LN (IRD/AVG
LN(IR2)/AVG
LN(IR2)/AVG
LN(RED)/AVG x LN(IR1)/AVG
LN(IR2)/AVG x LN(RED)/AVG
LN(IR2)/AVG x LN( IRD/AVG
LN (IRD/AVG x LN(IR2)/AVG
LN( IRD/AVG x LN(IR2)/AVG x LN(GRN)/AVG
3-25
-------
Table 3-7. Landsat MSS Mean Values and Standard Deviations for
Nine Colorado Lakes (Destriped "Whole" Lake Data)
Lake/ Reservoir
Name
Barker R.
Barr L.
Blue Mesa R.
Cherry Creek R.
Cucharas R.
Dillon R.
Grand L.
Green Mt. R.
Holbrook L.
Lake Meredith
R.
Milton R.
Shadow Mt. R.
STORET LANDSAT
Number Date
(1975)
0801 Aug 23
0802 Aug 23
0803 Aug 24
0804 Aug 23
0805
0806 Aug 24
0807 Aug 24
0808 Aug 24
0809
0810
0811 Aug 23
0813 Aug 24
Band Means and
(Standard Deviations)
GRN
34.99
(2.88)
46.99
(3.54)
36.52
(3.93)
46.59
(3.55)
RED
19.40
(3.88)
29.32
(5.63)
19.82
(5.42)
29.58
(6.14)
IR1
14.24
(6.01)
22.41
(6.46)
15.61
(14.99)
16.02
(6.61)
IR2
6.79
(6.45)
7.58
(5.87)
5.54
(5.66)
6.09
(6.01)
Number
of
Pixels
119
647
5,082
517
(Not available)
34.01
(4.03)
31.89
(2.47)
40.35
(3.34)
19.02
(5.07)
18.09
(3.12)
23.85
(4.19)
14.34
(6.13)
13.40
(7.04)
14.93
(5.65)
7.26
(6.21)
6.61
(6.62)
6.97
(5.78)
2,114
328
1,267
(Not available)
(Not available)
42.03
(5.51)
33.87
(2.02)
23-47
(7.22)
19.50
(2.52)
14.41
(7.19)
13-50
(5.23)
5.04
(5.02)
5.96
(5.99)
584
862
3-26
-------
Table 3-8. Landsat MSS Band Means for 27 Sampling Sites
in 9 Colorado Lakes (Destriped Data)
Lake/Reservoir
Name
Barker R.
Barr L.a
Blue Mesa R.
Cherry Creek R.a
Dillon R.
Grand L.a
Green Mt. R.
Milton R.a
Shadow Mt. R.
Sampling
Site
080101
080102
080201
080202
080301
080302
080303
080304
080305
080306
080401
080402
080403
080601
080602
080603
080604
080701
080702
080801
080802
080803
081101
081102
081301
081302
081303
GRN
33.44
34.81
44.40
48.36
37.22
38.12
34.80
35.56
34.40
34.20
45.08
47.76
47.40
34.44
31.48
33.56
30.56
30.48
31.36
42.19
40.12
39.36
39.84
42.16
35.28
33.28
33.84
Landsat
RED
17.89
19.25
24.56
31.16
19.56
20.44
17.24
18.94
16.52
17.00
24.80
29.32
33-04
17.12
15.96
19.36
14.64
16.56
16.36
23.81
24.12
22.76
19.12
23.96
20.76
18.32
18.80
Band Means
IR1
11.78
12.31
21.12
17.16
11.44
12.20
9.28
19.56
8.56
10.28
12.28
15.24
17.16
11.16
10.64
13.84
9.60
10.32
9.92
15.81
14.28
14.12
11.16
12.52
13.12
10.20
11.40
IR2
2.89
4.31
6.08
4.00
6.00
3.48
3.68
5.06
3.80
4.64
4.44
3.80
5.20
4.20
3.92
5.64
3.52
3.40
4.28
7.44
5.28
5.68
3.68
3.64
5.32
2.80
3.72
aLandsat MSS data for the 13 sampling sites associated with these 5
lakes were used in 13-site analyses.
3-27
-------
Table 3-9. MMS Channel Means, Standard Deviations, and Pixel Counts for Five Colorado Lakes3
CO
Lake/Reservoir
Name
Dillon ri.
Grand L.
Cherry Creek R.
Barr L.
Milton R.
Lake
Number 1
0«06 111
(5
0807 115
(6
0804 123
(6
0802 124
(3
0811 126
(5
MMS Channel Means and (Standard Deviations)
.00
.00)
.40
.78)
.50
.02)
.0
.52)
.10
.36)
2
110.30
(6.87)
105.50
(8.04)
123.20
(8.47)
115.90
(4.69)
120.30
(7.02)
3
67.76
(5.70)
70.62
(5.16)
87.82
(6.39)
81 .43
(3.51)
83-70
(5.35)
4 5
53-68
(5.01)
56.30
(3.56)
71.96
(5.09)
71.08
(1.95)
68.34
(4.67)
6 7
41.66
(3-81)
44.51
(2.92)
- 53-98
(4.79)
50.01
(3.16)
48.62
(4.53)
8
47.76
(4.21)
51.17
(3.24)
57.46
(5.34)
62.65
(3.93)
54.64
(6.27)
9
37 . 57
(4.71)
41.84
(3-98)
43.89
(5.90)
48.69
(6.96)
42.61
(6.66)
10
51.05
(4.59)
54.80
(4.16)
54.05
(5.42)
52.80
(5.32)
53-32
(4.36)
11
23. 8«
(7.56)
44.50
(8.46)
103.0
(9.44)
106.6
7.76
114.8
(9.27)
Number
of
Pixels
56,537
9,773
17,590
19,247
17,118
aThe channel mean for a lake was determined by summing the band DN values for all of the lake's pixels and tnen dividing by the
total number of pixels.
-------
Table 3-10. MMS Channel Means and Standard Deviations for 13 (11 x 11 Pixel Array)
Sampling Sites in 5 Colorado Lakes
Lake/Reservoir
Name
Milton fi.
Milton R.
Barr L.
Barr L.
Cherry Creek R.
Cherry Creek R.
Cherry Creek H.
Grand L.
Grand L.
Dillon R.
Dillon R.
Dillon R.
Dillon R.
Sampling
Site
STORE!
Number
081101
061102
080201
080202
080402
080401
080403
080702
080701
080603
080604
080602
080601
MMS Channel Means and (Standard Deviations)
1
122. 96a
(2.37)b
128. 17
(3-13)
122.75
(2.62)
125.90
(2.67)
131.84
(3.34)
122.26
(3-82)
122.74
(2.91)
119.16
(3-88)
117.22
(5.17)
111.75
(3-15)
108.18
(2.88)
113-76
(2.64)
113-03
(2.87)
2
118.49
(1.82)
123.19
(1 .64)
112.68
(1.52)
120.22
(1.44)
133-53
(2.90)
119.00
(1.96)
122.20
(1.71)
110.07
(2.95)
108.97
(6.24)
99.04
(2.23)
94.75
(2.41)
106.69
(1.64)
103-70
(1.57)
3
82.88
(1.15)
86.34
(1 .07)
78.98
(0.92)
84.66
(0.96)
95.07
(1 .79)
83-99
(1.27)
87.63
(1.21)
73.26
(2.03)
72.96
(4.05)
66.00
(1.50)
63.00
(1.65)
72.46
(1.08)
71.88
(1.07)
4
67.
(1.
69.
(0.
69.
(0.
71 .
(0.
77.
(1.
68.
(1.
72.
(1 .
57-
(1.
57.
(2.
52.
(1 .
49.
(1.
5
79
05)
65
99)
83
85)
71
72)
08
46)
49
21)
16
00)
69
46)
71
.89)
07
.13)
.54
.21)
56.60
(0.91)
57.70
(0.97)
6 7
46.89
(0.85)
47.78
(0.78)
47.82
(0.73)
52.41
(0.83)
56.98
(1 .09)
50.22
(1.01)
54.40
45.60
(1.24)
45.57
(2.29)
40.90
(0.93)
38.60
(0.90)
44.11
(0.79)
41.41
(0.77)
8
52.31
(1.09)
51 .75
(0.76)
59.84
(0.62)
60.79
(0.77)
60.34
(1.27)
54.08
(1.24)
56.45
(1.08)
51 .92
( 1 . 30 )
52.07
(2.59)
47.30
(0.93)
44. 12
(1.07)
50.05
(0.84)
46.68
(0.89)
9
40.81
(1.00)
39-35
(0.81)
45.73
(0.85)
43.93
(0.75)
47.20
(1.35)
41 .93
(1.31)
40.26
(0.89)
41 .96
(1.58)
42.25
(2.8)
36.84
(1.12)
33.56
(1.14)
39.88
(1 .01)
35.29
(0.81)
10
52.50
(1. 12)
51 .10
(1.03)
52.25
(0.99)
48.64
(0.58)
58.09
(1 .65)
53-50
(1.01)
49.26
(0.73)
54.56
(1.23)
54.67
(2.53)
50.80
(1 .21)
47.04
(1 .00)
53.16
(0.72)
48.04
(0.66)
11
121 .26
(3-52)
110. 14
(2.44)
109.01
(3.06)
101 .74
(3-94)
94.50
(3.12)
107.38
(3-91)
102.26
(3.63)
38.85
(2.70)
44. H
(6.76)
22.62
(2.22)
19.92
( 2 . 20 )
24.38
(4.69)
22.22
(2.20)
aAverage pixel value for (11 x 11) training site.
b( ) - standard deviation.
-------
Table 3-11. Pearson Product-Moment Correlation Coefficients Generated From MMS Channels for
13 Sites Located in 5 Colorado Lakesa
1
2
3
4
7
OJ Q
1 0
t_0
o
9
10
11
MMS Chan
(370-413
(440-490
(495-535
(540-580
(660-700
(700-740
(760-860
(970-1060
MMS Channel
nM
123^
nm) 1.000 0.970 0.954 0.946
nm) 1.000 0.992 0.952
nm) 1.000 0.974
nm) 1.000
nm)
nm)
nm)
nm)
7 8
0.892 0.838
0.941 0.801
0.943 0.803
0.929 0.877
1.000 0.892
1.000
9
0.788
0.743
0.707
0.748
0.791
0.920
1.000
0
0
0
0
0
0
0
1
10
.474
.489
.407
.322
.436
.406
.706
.000
(8000-13000 nm)
0
0
0
0
0
0
0
0
1
11
.854
.824
.849
.904
.757
-747
.621
.197
.000
adf = n-2 = 13 - 2 = 11 5% level = 0.553,
level = 0.684.
-------
3. Bayesian Maximum Likelihood-Derived Thematic Mapping Method
Several standard procedures are followed in the development of color
classification maps. The initial phase is the application of the Bayesian
maximum likelihood algorithm to the multispectral data to achieve classi-
fication. Once a satisfactory classification has been achieved, a color
image of the resulting thematic classification map is constructed to
illustrate the trophic pattern identified and mapped by the Bayesian
classifier.
a. Operational Aspects of Classification. The Bayesian maximum
likelihood algorithm was chosen for use in this water quality classification
effort because of its sensitivity to subtle differences in spectral signa-
tures. Multispectral signatures of water quality classes typically have
a great deal of overlap, especially in Landsat data. There is no easily
determined dividing line between one class and another in terms of the
multispectral data. Rather, there is a range of possible ambiguity unless
the classification algorithm is very sensitive to data differences.
The operational aspects used to develop a number of spectral classes
which are relatable to specific trophic indicators and multivariate trophic
indices are as follows:
(1) Selection of training sites. The initial step is the selec-
tion of training sites which represent different phenomena
of interest or a range of values for a particular parameter
(.e.g.. > trophic indicator, multivariate trophic index). A
training site may consist of just a few pixels or many thou-
sands. It could, for example, consist of an entire lake or
just a small portion of it. In this study, both whole lakes
and specific portions of lakes were used as training sites.
The training sites, consisting of subsections of the lakes,
encompass the ground truth sampling sites (Figures 3-6 and
3-7). While it is recognized that the ground truth sites
are geographic points and that an entire site falls into
a single pixel, the Bayesian maximum likelihood algorithm
operates on statistical parameters (means, covariance) to
generate probability density functions. Thus, more than
one pixel is required for each training area. Generally,
25 to 40 pixels are selected, although on occasions fewer
are used. The use of more than one pixel dampens the noise
produced by the sixth line banding. Ideally, a training
site is homogeneous and the resulting spectral curves are
unimodal.
(2) Analysis of training site statistics. After the selection
of the training sites have been completed, descriptive
statistics are generated for each site; the statistics
describe each site in terms of its spectral properties
as measured by the Landsat MSS and/or the Bendix MMS. If
each site demonstrates a unique spectral signature and the
classification site accuracies are acceptable, the next
step is to proceed with the generation of the thematic
mapping product. If the signature demonstrates a marked
3-31
-------
overlapping or the classification site accuracies are too
low, an effort is made to pool training sites (on the basis
of spectral similarity) and/or select new sites. In this
project, other than using the entire lake as a training
site, site selection was limited to the NES sampling sites.
(3) Generation of the thematic photomaps. Once the decision
has been made to extend the signatures to the entire water
body or group of water bodies, the Bayesian maximum likeli-
hood algorithm is applied to all of the data. This process
is accomplished through a video information communication
and retrieval application program which also outputs a
classification map as an end product. The map is constructed
as the program progresses through the classification of
the water quality data. After a pixel has been assigned
to a particular class, it is also assigned a corresponding
DN in the output map which signifies the class to which
it belongs. Thus a pixel assigned to class 4 is represented
by DN=Jj, class 5 by DN=5, and so on. When the classification
is completed on all the input data, a corresponding map
delineating the trophic patterns as classified remains.
This map is perhaps the single most important visual tool
available to the water quality analyst to aid in the evalua-
tion of the accuracy and significance of a classification.
t>. Color Thematic Maps. As colors are more easily discerned
than gray levels by the human eye, classification maps are
normally reproduced in color for interpretation. Colors
are created by mixtures of the three primary colors of
blue, green and red. A color is chosen to correspond to
each class represented, with special attention given to
the ease with which these colors can be distinguished from
one another. A wide range in colors is preferable, as
colors from the same family closely resembling one another
are easily confused when in close proximity. In lake class-
ifications an attempt is also made to assign colors of blue
and green to classes lying at the oligotrophic end of the
relative trophic scale. This is done more for esthetic
purposes and should not be construed as suggesting that
classes represented by blues denote clear or pristine water.
Once the desired colors have been selected, three separate
positive images of the classification map are created.
Each is individually contrast-enhanced in such a way as
to produce a given hue when exposed through an individual
primary filter. Each image is then registered to the other
and is exposed separately through its corresponding filter
of red, green, or blue. The result is a color image repre-
senting the classification map in as many different colors
as there are classes.
3-32
-------
SECTION IV
RESULTS AND DISCUSSION
This section is devoted to the presentation of the analysis results
and to discussions of pertinent aspects. The order of presentation is
essentially the same as that used in Section III, Methods.
A. LAKE SURFACE AREA
The results of the lake surface area calculations as determined
using data from three types of sensors (Zeiss camera, Landsat MSS, Bendix
MRS) are given in Table 4-1. The estimates made from the Zeiss camera
photographs are treated as the "ground truth." Although neither of the
scanners was able to give an estimate that corresponded exactly with the
Zeiss estimates, they did provide estimates that approximate the true
values.
Three-sensor coverage was available for five of the lakes. Total
surface area estimates for the five lakes are as follows:
Zeiss camera 2,742 ha
Landsat MSS 2,677 ha
Bendix MMS 2,706 ha
Landsat underestimated the total area of the five lakes by 2.3756.
The Bendix MMS gave slightly better results, underestimating the area by
1-31 %.
Complete Landsat MSS and Bendix MMS coverage was available for eight
of the nine lakes; Zeiss coverage for Blue Mesa was not complete. For the
eight lakes, Landsat overshot the Zeiss estimates in four cases and
underestimated in the remaining four cases. Total surface area estimates
for the eight lakes are:
Zeiss 4096 ha
Landsat MSS 4114 ha
The Landsat MSS, on the basis of eight lakes, overestimated the total
surface area by 0.40J.
Although the sample sizes are small (N = 5, N = 8), it is apparent
that both Landsat and the Bendix MMS can provide estimates of lake surface
area which are of practical value.
The approach used during this project to separate water pixels
from pixels associated with other features including terrain and clouds
is very simple. However, the approach, consisting of establishing
an IR2 DN threshold value of 28 for Landsat (DN values from 0 to 28
as assumed to represent water) and a CH10 value of 60 for the MMS (DN
values of 0-59 are assumed to represent water), can lead to problems.
For example, if you examine the subscene of Landsat Scene 5127-16532
4-1
-------
Table 4-1. Surface Area Estimates for Colorado Lakes
Using Three Types of Sensors
Sensor Estimates
j_ian.^/ 11^0^-1 v \j j.i
Name
Barker R.
Barr L.
Blue Mesa R.
Cherry Creek R.
Cucharas R.
Dillon R.
Grand L .
Green Mt. R.
Holbrook, L.
Lake Meredith R.
Milton R.
Shadow Mt. R.
kj i WII.LJ j.
Mumber
0801
0802
0803
0801
0805
0806
0807
0808
0809
0810
0811
0813
Zeiss Camera
ha
81
459
3088a
352
28
1337
208
762
86
974
386
511
Landsat MSS
ha
76
413
3247
330
(clouds)
1351
210
810
(clouds)
( clouds)
373
551
Bendix MMS
ha
N/Ab
433
N/A
396
N/A
1272
220
N/A
N/A
N/A
385
N/A
aSeveral small areas of the lake fell outside the camera's field of
view.
bN/A = not available.
(Figure 3-4) you will note a small cloud just southeast of the Blue
River Arm of Dillon Rerservoir. The cloud's shadow is cast to the
northwest, falling in part on the water, with the remainder falling
on the peninsula separating Frisco and Giberson Bays from the Blue
River Arm. Using the Landsat 0 to 28 IR2 DN range as representing
water, the computer was unable to separate the shadow from the water
(i.e_. , it included the cloud shadow as part of the water mass). Yet,
if you further examine the cloud's shadow in the false-color image,
you will note that it is not black like the water of the Blue Arm,
but has a slightly bluish hue (the hue is very apparent on the image
original). This suggests that the land portion of the cloud is separable
if more than one spectral band is used in the water detection algorithm.
The pixels which straddle the water-land interface present another
problem of substantial proportions when using a single-band water detection
algorithm. These pixels, sometimes called ''mixed" pixels, encompass
both water and land and therefore provide a signal consisting of a
mixture of spectral information. This can have an adverse impact on
determination of the area of water bodies.
4-2
-------
1. A Multiband Water Detection Algorithm
In this and past EPA/JPL Landsat lake classification efforts,
the detection of pixels whose instantaneous field of view (IFOV) is
that of water was done in the manner of thresholding the IR2 band.
The low reflectance of water in this spectral range conveniently produced
a bi-modal distribution of DNs: one peak for water, another peak for
non-water. This technique works quite well except in cases where the
IFOV of the scanner includes both water and non-water areas (e.g.,
the shoreline of a lake and cloud shadow on the water-land interface).
In this situation, the problem becomes one of trying to estimate the
proportion of each material in the IFOV.
Horowitz (1971) and Work and Gilmer (1976) have investigated
the proportion estimation problem and have obtained encouraging results.
Work and Gilmer estimated the proportions of water, bare soils, and
green vegetation using Landsat bands 5 and 7. This technique requires
an estimate of the spectral signature for pure water, pure bare soil,
and pure vegetation. Although the spectral signature of water is fairly
easy to estimate, that for soil and vegetation is more difficult.
The many variables involved—e..£. , differences in type of soil, variations
in type and thickness of vegetative cover—cause considerable error
when estimation is attempted by a completely automatic processor.
An alternate approach meriting consideration treats the mixture
classes as consisting of only water and non-water. Only bands 5 and
7 are used in the detection process, since it has been found that bands
4 and 6 offer little in additional information. The estimation of
the spectral signature for water and non-water is made over a region
within and immediately surrounding the water body.
The spectral signatures (mean DN values) were estimated by an
iterative procedure. First the 2-dimensional space (band 5 vs band 7)
is partitioned into two regions in which the populations of water and
non-water typically cluster and the mean is then recomputed for those
DNs which fall within the neighborhood of the initial mean. This process
is continued until a convergent mean has been found for each region.
The proportion estimation implemented uses a technique proposed
by McCloy (1977). In Figure 4-1 W is the mean for water, U is the
mean for non-water, and P is the DN for any given pixel. P1 is the
projection of P onto the line segment WU. If /WU/ is the length of the
line segment WU and /WP'/ is the length of line segment WP1 then the
proportion estimate q for water is
where CKq<.1.
If P' does not fall between W and U then it is given the position
of the closest point, W or U. A decision threshold is set for q at
which the pixel is defined to be water or non-water.
4-3
-------
The necessary algorithm for the multiband approach described above
was developed during the course of this project, but was not employed
on Colorado lakes.
B. PRINCIPAL COMPONENT TROPHIC ORDINATION OF LAKES AND SAMPLING SITES
INDICATORS
The principal components analysis of the six trophic indicators was
undertaken to reduce the dimensionality of the data. The six natural
logarithm-transformed indicators include LNChLA, LNISEC, LNCOND, LNTPHOS,
LNAAY , and LtfTON. The analysis was performed once for each of three sets
of data points. The results of each analysis will be discussed
separately.
BAND 5
Figure 4-1. Proportion Estimation Diagram
4-4
-------
1. Principal Component-Derived Trophic Ranking of Colorado Lakes
At the time the analysis was made, complete trophic indicator data
sets were available for 11 of the 12 Colorado lakes (data for Lake
Meredith were lacking). Although Landsat data were available for only
nine lakes, the decision was made to generate the "whole" lake trophic
ranking using as many contact-sensed Colorado lakes as possible.
The normalized eigenvectors and eigenvalues are found in Table 4-2.
Although the principal components analysis is of value in reducing the
dimensionality of a multivariate system, it is sometimes difficult to
interpret the new variates in terms of subject matter identities. Some
indication of a principal component's meaning may be ascertained by
an examination of the algebraic sign and magnitude of its coefficients.
The coefficients of the first component (Table 4-2) are nearly
equal in magnitude, suggesting that it represents a general measure
of trophic state, accounting for approximately 86% of the variation
in the data. Correlations between the new variate and the LN-transformed
trophic indicators are found in Table 4-3- The first component correlates
most strongly with LNTON (r = 0.989) and least with LNISEC (r = 0.894)
(Table 4-3).
The second component (Table 4-2) explains about 6% of the variation
in the trophic indicator data. It correlates best with LNCOND (r = 0.367)
(Table 4-3).
The third component (Table 4-2) accounts for about 4$ of the
variation in the data. Together, the first three components explain
about 97% of the variation in the trophic indicator data (Table 4-2).
The first component (Table 4-2) was evaluated for each of the
lakes included in the analysis. This resulted in the generation of
a numeric value for each of the lakes. The index (PC1-11) number defines
the lake's position on a trophic scale. Table 4-4 displays the resulting
PC1-11 values and rank position of the lakes. As the values increase,
the trophic state increases. It is these values that were used in
the nine "whole" lake correlation and regression analyses.
Although the details will not be presented in this report, once the
Lake Meredith trophic indicator data set was complete, the principal com-
ponents analysis was rerun using 12 lakes instead of the 11. The trophic
index generated correlated strongly with that produced above (r = 0.999).
Again, it should be noted that the above principal components
analysis used August sampling round data which is unlike that of Boland
(1975), where the "whole" lake ranking was generated using annual means
of the indicator data.
4-5
-------
Table 4-2. Normalized Eigenvectors and Eigenvalues Extracted from 11 Colorado Lakes' Six
Trophic Indicator Data Correlation Coefficient Matrix
Eigenvector
Number LNCHLA
1 0.409
2 -0.497
3 -0.059
4 -0.437
5 -0.614
6 -0.120
LNCOND
0.392
0.457
-0.687
-0.189
0.156
-0.325
LNISEC
0.400
-0.599
-0.078
0.435
0.490
-0.214
LNTPHOS
0.408
0.379
0.349
0.570
-0.426
-0.248
LNTON
0.434
0.073
-0.151
0.111
-0.005
-0.878
LNAAY
0.405
0.191
0.612
-0.497
0.420
-0.046
Eigenvalue
5.187
0.374
0.245
0.151
0.028
0.015
6.000
Variance,
86.45
6.23
4.08
2.52
0.47
0.25
Cumulative
Variance,
86.45
92.68
96.77
99.28
99.75
100.00
-------
Table 4-3. Product-Moment Correlation Coefficients for 6 Trophic
Indicators and the Principal Components Extracted from the
11 Colorado Lakes' Data Correlation Coefficient Matrix3
Principal Component
Indicator
LNCHLA
LNISEC
LNCOND
LNTPHOS
LNTON
LNAAY
0.931
0.894
0.910
0.928
0.989
0.923
-0.304
0.279
-0.367
0.232
0.044
0.117
-0.029
-0.340
-0.039
0.173
-0.075
0.303
-0.169
-0.073
-0.169
0.221
0.043
-0.193
-0.103
-0.026
-0.082
-0.071
-0.001
0.070
-0.015
-0.040
-0.026
-0.031
0.108
-0.006
aAt 9 degrees of freedom, the 0.05 level of significance is 0.602;
the 0.01 level is 0.735.
The question arises, How well does the PC1-11 trophic ranking
compare with those developed using other means? This question can be
answered, in part, by comparing the resultant ranking with that developed
by CERL for the same Colorado lakes. The CERL approach, described in
detail in U.S. EPA (1974), employs a percentile ranking procedure. In
this procedure, for each of the unweighted parameters used, the percentage
of the Colorado lakes (N = 13) exceeding lake X in that parameter (e.g.,
chlorophyll a.) was determined. The final ranking value or index number
is simply the sum of the percentile ranks for each of the parameters
used. The parameters incorporated into the CERL index include median
total phosphorus, median inorganic nitrogen, mean chlorophyll a., median
dissolved ortho-phosphorus, mean Secchi depth, and dissolved oxygen
minimum. The values for mean Secchi depth and dissolved oxygen minimum
are subtracted from the values of 500 and 15, respectively, so that
all of the parameters contribute positively to the ranking. The means,
medians, and minimum measures were calculated using data collected
from all of the sampling rounds. A comparison of the two indices is
shown in Table 4-5. When comparing the two indices,
4-7
-------
Table 4-4. Irophic Ranking of 11 Colorado Lakes Derived From
Principal Components Analysis of Six Trophic Indicators3
Lake/Reservoir
Name
Grand L.
Dillon R.
Barker R.
Green Mt. R.
Shadow Mt. R.
Blue Mesa R.
Cherry Creek R.
Milton R.
Cucharas R.
Barr L.
Hoi brook L.
STORET
Number
0807
0806
0801
0808
0813
0803
0804
0811
0805
0802
0809
Rank
1
2
3
4
5
6
7
8
9
10
11
PC 1-11 Value
-2.59
-2.52
-1.95
-1.49
-1.41
-1.17
1.11
1.50
1.89
2.90
3.74
aCucharas Reservoir and Holbrook Reservoir trophic state index
values were not used in Landsat MSS and Bendix MMS analyses be-
cause of cloud cover problems. The remaining 9 lakes' index values
were used in the Landsat MSS 9-lake analyses.
Table 4-5. Rankings of 11 Colorado Lakes as Derived from Two Trophic
Indices and Ordered by the PC1-11 Index
Lake/
Reservoir
Name
Grand L.
Dillon R.
Barker R.
Green R.
Shadow Mountain R.
Blue Mesa R.
Cherry Creek R.
Milton R.
Cucharas R.
Barr
Holbrook
PC1-11
Values
-2.59
-2.52
-1.95
-1 .49
-1.41
-1.17
-1.11
1.50
1.89
2.90
3.74
Rank
1
2
3
4
5
6
7
8
9
10
11
CERL
Trophic
Index
Value
453
521
358
479
433
354
291
183
157
104
183
Rank
3
1
5
2
4
6
7
8.5
9
10
8.5
CERL
Trophic
Classification
Mesotrophic
Oligotrophic
Mesotrophic
Oligotrophic
Mesotrophic
Mesotrophic
Eutrophic
Eutrophic
Eutrophic
Eutrophic
Eutrophic
4-8
-------
it should be kept in mind that they do not incorporate all of the same
parameters and, in addition, the PC 1-11 used only fall sampling round
data. However, the agreement is very good between the two indices,
the product-moment correlation coefficient (r) and rank correlation
coefficients (Spearman's rho and Kendall's tau) being -0.922, -0.898,
and -0.764, respectively. This close agreement does not provide direct
evidence that the PC1-11 (and the CERL trophic index for that matter)
adequately represents the trophic state of the water bodies. However,
they do appear to be in general agreement with what has been observed
in the field. Incidentally, the negative correlations are a consequence
of the manner in which the two scales were constructed. In the case
of the PC 1-11 index, eutrophic lakes are located toward the positive
end of the scale and oligotrophic lakes toward the negative end. Lakes
with low rank sum values in the CERL index are considered to be eutrophic;
those with high values are located toward the oligotrophic end of the
scale.
2. Principal Coraponent-Derived Trophic Ranking of 27 Sampling Sites
on 9 Colorado Lakes
The preceding principal components analysis produced a trophic
state index (PC1-11) which defines a lake's position on a numeric scale.
The lake was treated as a whole unit. However, 27 contact-sensed data
sites were established on the 9 lakes for which Landsat coverage is
available. An analysis was conducted to rank each of the 27 sampling
sites on a trophic scale. The normalized eigenvectors and eigenvalues
are found in Table 4-6.
The coefficients of the first component (Table 4-6), with the
exception of LNCOND, are of about the same magnitude and compare favorably
with those of the first component derived from the 1 1-lake analysis
(Table 4-2). About 18% of the data variation is explained by the first
component (Table 4-6). With the exception of LNCOND, the first component
exhibits high correlations with the trophic indicators (Table 4-7). The
second component explains about 10$ of the data variation (Table 4-6) and
correlates best with LNCOND (Table 4-7). About 5% of the variation is
explained by the third component, and correlations with the trophic
indicators are less obvious. Together, the first three components are
associated with approximately 93? of the variation.
The first component (Table 4-6) was evaluated for each of the 27
sampling sites included in the analysis. This resulted in a numeric value
for each of the sites. The index (PC1-27) number defines a sampling
site's position on a trophic scale. Table 4-8 lists the PC1-27 values and
site rank. It is these values that were used in the Landsat 27-site
trophic index analysis.
4-9
-------
-tr
I
Table 4-6. Normalized Eigenvectors and Eigenvalues Extracted from 27 Colorado Lake Sampling
Sites' 6 Trophic Indicator Data Correlation Coefficient Matrix3
Eigenvector
Number LNCHLA
1 0.405
2 -0.425
3 -0.375
4 -0.544
5 0.327
6 -0.335
LNISEC
0.396
-0.484
-0.299
0.653
-0.295
0.080
LNCOND
0.347
0.735
-0.576
0.031
-0.086
-0.003
LNTPHOS
0
0
0
0
-0
-0
.428
.165
.524
.040
.313
.648
LNTON
0.437
-0.040
0.243
-0.437
-0.390
0.636
LNAAY
0.430
0.129
0.328
0.290
0.741
0.240
Variance ,
Eigenvalue %
4.702 78.37
0.617 10.28
0.303 5.05
0.188 3.13
0.121 2.02
0.070 1 . 17
6.001
Cumulative
Variance,
78
88
93
96
98
100
• 37
.65
• 37
.83
.85
.02
aThe principal components analysis was performed using a r-matrix of correlation coefficients for six
trophic state indictors. The natural log-transformed data were collected from 27 sampling stations
in 9 Colorado lakes between August 22-26, 1975.
-------
Table 4-7. Product-Moment Correlation Coefficients for 6 Trophic
Indicators and the Principal Components Extracted from
the 27 Colorado Lake Sampling Sites Data Correlation
Coefficients Matrix3
Indicator
Principal Component
LNCHLA
LNISEC
LNCOND
LNTPHOS
LNTON
LNAAY
0.877
0.858
0.752
0.927
0.948
0.933
-0.334
-0.380
0.577
0.130
-0.032
0. 101
-0.207
-0.165
-0.317
0.286
0. 134
0. 181
-0.236
0.283
0.014
0.017
-0.190
0.126
0.114
-0.103
-0.030
-0.109
-0.136
0.258
-0.088
0.021
-0.001
-0.171
0.168
0.064
*At 11 degrees of freedom, the 0.05 level of significance is 0.553;
the 0.01 level is 0.684.
3. Principal Component-Derived Trophic Ranking of 13
Sampling Sites on 5 Colorado Lakes
MMS data were available for only 5 lakes containing 13 NES sampling
sites. Because the 5 lakes are a very small sample, a trophic ranking
was developed for the 13 sites. The resulting normalized eigenvectors
and eigenvalues are found in Table 4-9.
With the exception of LNCOND, the coefficients of the first component
are of the same general magnitude. The first component relates to about
83 % of the data variation (Table 4-9). It correlates strongly with all
of the trophic indicators except LNCOND (Table 4-10). The second component
explains about 10$ of the variation and correlates most strongly with
LNCOND (Table 4-10). Approximately 4$ of the data variation is associated
with the third component; correlations with the trophic indicators
are low. The first three components explain about 96% of the variation.
The trophic index values (PC1-13) for the 13 sites are found in
Table 4-11. The index values are used in the Landsat MSS and MMS trophic
index analyses. Rather than generating the PC 1-13 values for the 13 sites,
the 13-site PC1 values could have been selected from the list of PC1-27
values (Table 4-8). A comparison was made of the PC1-13 and PC1-27 indices
(Table 4-12). The product-moment correlation coefficient between the two
indices is 0.9999. An arbitrary decision was made to use the PC1-13 values
in MSS-MMS analyses of the 13 sites.
4-11
-------
Table 4-8. Trophic Ranking of 27 Colorado Lake Sampling
Sites Derived From Principal Components
Analysis of 6 Trophic Indicators3
Rank
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
Lake/
Reservoir Name
Dillon R.
Dillon R.
Grand L.
Dillon R.
Dillon R.
Grand L.
Barker R.
Barker R.
Green Mt. R.
Green Mt. R.
Shadow Mt. R.
Blue Mesa R.
Blue Mesa R.
Blue Mesa R.
Blue Mesa R.
Shadow Mt. R.
Blue Mesa R.
Green Mt. R.
Shadow Mt. R.
blue Mesa R.
Cherry Creek. R
Cherry Creek. R
Milton R.
Milton R.
Cherry Creek R.
Barr L.
Barr L.
Sampling Site .
STORET Number
080603
080604
080702
080602
080601
080701
080101
080102
080801
080802
081303
080303
080305
080302
080304
081302
080306
080803
081301
080301
080401
080403
081101
081102
080402
080202
080201
PC1-27
-2.51
-2.28
-2.01
-1.96
-1.93
-1.85
-1.41
-1.30
-1.06
-0.99
-0.94
-0.88
-0.71
-0.66
-0.65
-0.64
-0.59
-0.33
-0.23
0.02
1.54
1.96
2.78
3.31
3.39
4.49
5.43
Indicators: LNCHLA, LNCOND, LNISEC, LNTPHOS, LNTON, LNAAY,
The data incorporated into the analysis were
collected August 22-26, 1975.
4-12
-------
Table 4-9. Normalized Eigenvectors and Eigenvalues Extracted from 13 Colorado Lake Sampling
Sites' 6 Trophic Indicator Data Product-Moment Correlation Matrix3
Eigenvector
Number
1
2
3
.pr
i 4
UJ
5
6
LNCHLA
0.394
-0.511
-0.339
-0.581
-0.282
-0.226
LNISEC
0.403
-0.447
-0.193
0.670
0.354
-0.162
LNCOND
0.358
0.700
-0.578
0.072
-0.064
-0.196
LNTPHOS
0.419
0.207
0.541
-0.314
0.504
-0.370
LNTON
0.446
0.032
-0.160
-0.140
-0.182
-0.864
LNAAY
0.423
0.072
0.470
0.300
-0.710
0.015
Eigenvalue
4.979
0.579
0.230
0.122
0.086
cum
6.001
Variance ,
82.98
9.65
3.83
2.03
1.43
0.08
Cumulative
Variance,
82.98
92.26
96.47
98.50
99.93
100.02
aThe principal components analysis were performed using a r-matrix of correlation coefficients for six
trophic state indicators (LNCHLA, LNISEC, LNCOND, LNTPHOS, LNTON, LNAAY). The natural log-transformed
data were collected from 13 sampling stations in five Colorado lakes between August 22-26, 1975 by NES.
-------
Table 4-10. Product-Moment Correlation Coefficients for
6 Trophic Indicators and the Principal
Components Extracted from 13 Colorado Lake
Sampling Sites Data Correlation Coefficient
Matrix
Indicator
LNChLA
LNISLC
LNCOND
LNTPhOS
LNTON
LNAAY
Principal Components
0
0
0
0
0
0
1
.880
.900
.799
.935
• 995
.944
2
-0.
-0.
0.
0.
0.
0.
389
3^0
532
157
024
054
-0
-0
-0
0
-0
0
3
.163
.093
.277
.260
.008
.226
4
-0.
-0.
0.
-0.
-0.
0.
203
234
025
110
049
105
5
-0.083
-0. 104
-0.019
0.147
0.053
-0.208
6
-0.
-0.
-0.
-0.
0.
-0.
015
111
014
025
059
001
4-14
-------
Table 4-11. Trophic State Index (PC1-13) Ranking Generated
for 13 Colorado Lake Sampling Sites Using
Principal Components Analysis of 6 Natural
Log-Transformed Trophic Indicators3
Lake/ Sampling Site
Reservoir Name STORET Number PC 1-13 Rank
Barr L.
Cherry Creek A.
Dillon R.
Grand L.
Milton R.
080201
080202
080401
080402
080403
080601
080602
080603
080604
080701
080702
081101
081102
3-47
2.78
0.57
1.94
0.89
-2.05
-2.07
-2.46
-2.31
-2.07
-2.18
1.55
1.93
13
12
7
11
8
6
4
1
2
5
3
9
10
aPrincipal components analysis was performed using a r-matrix of
correlation coefficients for six trophic indicators (LNCHLA, LNISEC,
LNCOND, LNTPHOS, LNTON, LNAAY). The natural log-transformed data
were collected from 13 sampling stations in 5 Colorado lakes between
August 22 and 26, 1975.
4-15
-------
Table 4-12. Comparison of Trophic State Index (PC1)
Values for 13 Sampling Sites
Lake/
Reservoir Name
barr L.
Cherry Creek ft.
Dillon H.
Grand L.
Milton R.
Range
Mean
Standard deviation
Sampling Site
STORET Number
080201
080202
080401
080402
080403
080601
080602
080603
080604
080701
080702
0801101
0801102
13-Site PC1
(PC1-13)
3.47
2.78
0.57
1.94
0.89
-2.05
-2.07
-2.46
-2.31
-2.07
-2. 18
1.55
1.93
5.93
0.00
2.23
27-Site PC1
(PC1-27)
5.43
4.49
1.54
3.39
1.96
-1.93
-1 .96
-2.51
-2.28
-1.85
-2.01
2.78
3.31
7.94
0.80
2.95
Correlation PC1-13, PC1-27
0.9999
4-16
-------
4. Some Aspects of the Multivariate Indices
The three multivariate trophic indices (PC1-11, PC1-27, and PC1-13)
have much intuitive appeal. However, it should be kept in mind that they
were developed from a relatively small number of sampling sites. While
the assumption is made that the sites and the data collected at them are
truly representative of the water bodies, this may not be the case. It
may well be that more sites should have been sampled by the helicopter-
borne field teams. It is not known what changes might have taken place
in the rankings had the same number of sites been selected but at different
locations in the water bodies.
As they now stand, the indices do not include any parameters directly
relating to macrophyte distributions, much less macrophyte biomass. While
such parameters would be desirable, they were not measured other than
to note whether or not macrophytes were present.
The algal assay yield parameter included as one of the six trophic
indicators in the indices development was measured in the laboratory
under control conditions. Some credence must be given to criticisms
suggesting that the inclusion of that parameter is inappropriate because
of its laboratory derivation. The parameter was dropped from the principal
components analysis scheme and three indices were developed using the
remaining five trophic indicators: CHLA, COND, TPHOS, ISEC, and TPHOS.
The correlations between the self-pairing members of the two sets (the
three trophic indices derived from six indicators and the three indices
derived from the five indicators) were highly significant (greater
than 0.95), suggesting that little or no information was lost by dropping
the algal assay yield parameter. The decision was made, however, to
use the indices derived from the six indicators in the development
of regression models and thematic classification products. The decision
was arbitrary.
Up to this point, little attention has been given to the light
inhibition effects that inorganic materials (e.g., silts, clays) can
have on algae and submerged aquatic macrophytes. If available
in substantial quantities, inorganic materials result in a reduction
of the algal productivity of a lake through the reduction of light
levels below those needed by the autotrophs to manufacture food.
Thus, a water body may have very high nutrient levels but not experience
algal blooms and/or major problems with macrophytes. As currently
constructed, the index might result in some water bodies being
incorrectly classified or ranked. It may be more fitting to segregate
water bodies into two or more groups according to the primary contributor
of turbidity (inorganic suspended sediments, volatile suspended
sediments) and then constructing an index for each group.
The concept of a trophic index is usually thought of in terms of
"whole" water bodies. In this project, trophic rankings were devised
for both the lakes and reservoirs as whole bodies, and also for individual
sampling sites. Each sampling site is treated as if it were an entity
with its own unique values, which make it separable from the other
members of the set. The CERL index, discussed in one of the preceding
subsections, was not used on a sampling site basis, and therefore a
-------
parallel correlation analysis cannot be made. Once again, the principal
component-derived indices do not account for the presence of aquatic
macrophytes; this may result in some disparity between the indices and
remotely sensed data because floating and emergent macrophytes can
directly affect the character of the spectral curve(s).
C. CORRELATION AND REGRESSION ANALYSIS RESULTS
As indicated in the methods section, the fragmentary nature of
the data set led to the organization of the data into five basic data
sets for analysis purposes. The sets are:
(1) Landsat MSS nine "whole" lake set
(2) Landsat MSS 27-site set.
(3) Landsat MSS 13-site set.
(4) MMS 13-site set.
(5) Modified MMS 13-site set.
Correlations were calculated between each sensor's bands or channels,
between the two sensors, and between the multispectral data and the
trophic indicators and their associated multivariate trophic indices.
Regression models for the estimation of trophic indicator and multivariate
trophic indices' values were developed from each of the five sets of
observations and, in the case of the modified MMS 13-site set, for
two categories. It is recognized that some of the parameters estimated
from the regression models are not directly sensible by the sensors
and dependent on secondary effects.
Data sets (3), (4), and (5) are particularly interesting because
they permit a direct comparison of the Landsat MSS and the MMS regression
model results. Unfortunately, good MMS data were available from only
five of the lakes, thus preventing comparisons both at the nine "whole"
lake level and at the 27-site level.
In regression analysis, the term "independent variables" is often
used to describe the parameters that will be used to predict another
parameter, the dependent variable. As discussed by Draper and Smith
(1966:4), the "independent" should not be interpreted too literally
because two or more independent variables may vary together in some
fashion. In general, this is not desirable because it, among other
things, "...restricts the information on the separate roles of the
variables—but it may be unavoidable" (Draper and Smith 1966: 4).
Table 4-13 displays the correlations between the Landsat MSS
bands for three sets of observations. A very high correlation exists
between the CRN and RED bands. This suggests a very limited amount
of information is available to discriminate between their separate
roles. The correlations decrease as the band separations increase;
overall, IR2 correlates least with any of the bands. The MMS
4-18
-------
Table 4-13. Landsat MSS Interband Product-Moment
Correlation Coefficients for Three Sets
of Observations
GRN RED IR1
Set 1 (N = 9 "whole" lakes)
GRN 1.000 0.981 0.72?
RED 1.000 0.736
IR1 1.000
IR2
At 9-2 degrees of freedom, 0.05 level = 0.666,
0.01 level = 0.798
Set 2 (N = 13 sites on 5 lakes)
GRN 1.000 0.947 0.750
RED 1.000 0.768
IR1 1.000
IR2
At 13-2 degrees of freedom, 0.05 level = 0.553,
0.01 level = 0.684
Set 3 (N = 27 sites on 9 lakes)
GRN 1.000 0.943 0.660
RED 1.000 0.691
IR1 1.000
IR2
IR2
0.022
0.130
0.438
1.000
0.256
0.300
0.694
1.000
0.320
0.319
0.562
1.000
At 27-2 degrees of freedom, 0.05 level = 0.381,
0.01 level = 0.487
4-19
-------
interchannel correlation coefficients, based on data from 13 sites on five
Colorado water bodies, are listed in Table 4-14. Again, it is noted
that the correlations are highest between the shorter wavelength channels.
A possible exception is CH11, the thermal channel. In general, as
the distance between channels increases, the correlation coefficients
decrease. This phenomenon is not unexpected. In addition, the bands
and channels are relatively broad compared to those found in laboratory
instrumentation designed for purposes of chemical analyses. Because
of their spectral resolution, the sensors tend to average out the finer
aspects of the spectral curves.
Present computer and software technology has made the development
of regression models incorporating large data masses and many variables
feasible; it is intuitively appealing to use "independent" variables
that are not correlated. It is also desirable to avoid the use of
variables that contribute little to the model(s). With this in mind,
and as described in the methods section, a principal components
analysis was made of the MMS channels CH1, CH2, CH3, CH4, CH7, CH8,
CH9, and CH10; the thermal channel, CH11, was excluded from the analysis.
The normalized eigenvectors and eigenvalues are displayed in Table 4-15.
Unlike the principal components analysis used to develop the trophic
indices, the covariance matrix for the MMS channels was used in place
of the correlation matrix because all of the MMS data are measured
in the same units. The new MMS variables and their associated data
values are given in Table 4-16. The coefficients of correlation between
the "old" variables and the "new" are in Table 4-17. The newly created
channels (MMSPC1, ...,MMSPC8) are orthogonal or independent as demonstrated
in Table 4-18; the correlation coefficients are essentially zero, though
not exactly because of "rounding error."
The question arises, "How well do the Landsat bands correlate with
the MMS channels?" Statistical relationships can be demonstrated using
the scanner data collected from the 13 sites on five Colorado lakes
(Table 4-19). Many of the correlations are statistically significant.
For example, the (CH4, CRN) is 0.936, (CH7, RED) is 0.882 and (CH8,
IR1) is 0.763. These values are especially high. Overall, the Landsat
GRN band correlates best with the MMS channels; IR2 has the lowest
correlations with the MMS channels. Based on the premise that the
longer wavelengths are affected less by the atmosphere than the shorter
wavelengths, it was expected that longer wavelength bands and channels
would have larger correlation coefficients than, for example, the GRN
and CH4. The unexpected correlations may be a consequence of weak
signals received by the sensors (in particular Landsat) in the longer
bands.
Table 4-20 presents the correlations between the Landsat bands
and the "new" MMS channels developed through principal component analysis.
The coefficients are large between the first "new" channel (MMSPC1)
and Landsat GRN and RED. The principal component-derived variables
contain a decreasing amount of information in the order MMSPC1, ...,
MMSPC8. Ideally, most of the information would be contained in the
first three or four variables.
4-20
-------
Table 4-14. MMS Interchannel Product-Moment Correlation Coefficients Based on Data from 13 Sites
CH1 CH2 CH3 CH4 CH5
CH1 1.000 0.970 0.954 0.946
CH2 1.000 0.992 0.952
CH3 1.000 0.974
CH4 1.000
f CHS
rv>
— »
CH6
CH7
CH8
CH9
CH10
CH11
CH6 CH7 CH8
0.892 0.838
0.941 0.801
0.943 0.803
0.929 0.877
- - -
- - -
1.000 0.892
1.000
CH9
0.788
0.743
0.707
0.748
-
-
0.791
0.920
1.000
CH10
0.475
0.489
0.407
0.322
-
-
0.436
0.406
0.706
1.000
CH11
0.854
0.824
0.849
0.904
-
-
0.757
0.747
0.621
0.197
1.000
At 11 degrees of freedom. 0.05 level = 0.553, 0.01 level = 0.684
-------
Table 4-15. Normalized Eigenvectors and Eigenvalues Extracted from 13 Colorado Lake Sampling
Site, Eight-Channel MMS Data Covariance Matrix
Eigen-
vector
Number
1
2
3
f 4
fY>
rv>
5
6
7
8
CH1
0.347
0.026
0. 130
-0.672
-0.460
0.401
0.114
0.160
CH2
0.551
-0.208
0.367
0.094
-0.230
-0.437
-0.339
-0.388
CH3
0.474
-0.305
0.048
0.183
0.244
-0.167
0.601
0.444
CH4
0.442
-0.089
-0.455
-0.188
0.585
0.285
-0.336
-0.167
CH7
0.264
0.121
-0.106
0.675
-0.351
0.509
-0.189
0.175
CH8
0.237
0.559
-0.436
0.033
-0. 178
-0.208
0.441
-0.412
CH9
0.158
0.683
0.060
-0.092
0.113
-0.347
-0.364
0.577
CH10
0.073
0.404
0.666
0.078
0.408
0.340
0.189
-0.256
Eigen-
value
378.613
15.165
10.904
3.926
1.644
0.193
0.032
0.008
Variance,
92.235
3.694
2.656
0.956
0.401
0.047
0.007
0.002
99.998
Cumulative
Variance,
92.235
95.929
98.585
99.541
99.942
99.989
99.996
99.998
99.998
-------
Table 4-16. New MMS Variables and Associated Data Generated Through Principal Components Analysis
of MMS Channels 1-4, 8-9 Data for 13 Sites in 5 Colorado Lakes
MMS New Variable
Lake Site
STORET Number
081101
081102
080201
-Cr
1
w 080202
080402
080403
080401
080702
080701
080603
080604
080602
080601
MSSPC1
212.23
218.86
210.80
220.46
239.47
214.65
221.02
197.15
195.82
179.16
170.86
191.29
186.98
MSSPC2
28.00
24.29
37-41
32.56
32.38
29.95
26.97
34.64
35.21
31.69
28.08
32.49
26.13
MMSPC3
42.89
43.76
36.38
35.58
45.54
42.33
37.71
45.09
44.43
41.93
39.81
43.51
39-87
MMSPC4
-35.34
-37.51
-36.43
-34.14
-32.48
-32.47
-29.79
-34.26
-33-15
-33-74
-33.47
-31.88
-33-90
and Associated Data
MMSPC5
-23-73
-26.22
-23.27
-27.42
-24.84
-23-80
-25.53
-26.81
-25-66
-25.55
-25.66
-24.33
-23-86
MMSPC6
19.69
20.27
20.49
19.58
20.44
20.48
20.48
20.30
19-95
20.83
20.40
19.31
19.91
MMSPC7
10.06
10.36
10.28
10.39
10.53
10.17
10.15
10.13
10.09
10.56
10.18
10.50
10.45
MMSPC8
-3.98
-3.84
-3-86
-3.78
-3-81
-3-67
-3-95
-3.84
-3-79
-3.95
-3.79
-3.90
-3-72
-------
Table 4-17. Pearson Product-Moment Correlation Coefficients of Eight Channel MMS Data for 13 Colorado
Lake Sampling Sites and Associated Principal Components
MMS
Channel
1
2
-tr
' 3
j=
4
5
6
7
8
9
10
MMSPC1
0.975
0.990
0.990
0.980
-
-
0.958
0.869
0.791
0.457
MMSPC2
0.015
-0.075
-0.128
-0.039
-
-
0.087
0.410
0.605
0.506
MMSPC3
0.062
0.112
0.017
-0.168
-
-
-0.065
-0.271
0.051
0.708
Principal
MMSPC4
-0.192
0.017
0.039
-0.043
-
-
0.249
0.013
-0.047
0.050
Components
MMSPC5
-0.085
-0.027
0.034
0.086
-
-
-0.084
-0.043
0.037
0.168
MMSPC6
0.025
-0.018
-0.008
0.014
-
-
0.042
-0.017
-0.039
0.048
MMSPC7
0.003
-0.006
0.012
-0.007
-
-
-0.006
0.015
-0.017
0.011
MMSPC8
0.002
-0.003
0.004
-0.002
-
-
0.003
-0.007
0.014
-0.008
-------
Table 4-18. Product-Moment Correlation Coefficients
Between "New" MMS Channels Developed fror
Principal Component Analysis of MMS Data
from 13 Sites
"New"
MMSPC1
MMSPC1
MMSPC1
MMSPC1
MMSPC1
MMSPC1
MMSPC1
MMSPC2
MMSPC2
MMSPC2
MMSPC2
MMSPC2
MMSPC2
MMSPC3
MMSPC3
MMSPC3
MMSPC3
MMSPC3
MMSPC4
MMSPC4
MMSPC4
MMSPC4
MMSPC5
MMSPC5
MMSPC5
MMSPC6
MMSPC6
MMSPC7
Channels
,MMSPC2
,MMSPC3
,MMSPC4
,MMSPC5
,MMSPC6
,MMSPC?
,MMSPC8
,MMSPC3
,MMSPC4
,MMSPC5
,MMSPC6
,MMSPC?
,MMSPC8
,MMSPC4
,MMSPC5
,MMSPC6
,MMSPC7
,MMSPC8
.MMSPC5
,MMSPC6
,MMSPC7
,MMSPC8
,MMSPC6
,MMSPC7
,MMSPC8
,MMSPC7
,MMSPC8
,MMSPC8
-
_
=
=
=
=
=
=
=
=
=
=
=
=
=
=
=
=
=
=
=
=
=
=
=
=
z
=
=
Correlation
Coefficient
1.950902922E-008
1.017656610E-010
-4. 36^313^-008
-8. 158675052E-010
4.664763204E-007
-1.155939702E-009
6.^75880184E-007
6.865259266E-011
-2.176259774E-009
1.954145026E-008
-2.930489124E-008
2.241735482E-007
-1 .215319346E-006
4.225637112E-011
-3-832076098E-006
9.697340292E-008
0.000031555
4.486070722E-006
-1 .090250368E-009
-3.570639032E-010
9.135440976E-009
-9.139718356E-007
9.125326640E-007
-3.150138508E-006
6.604071588E-010
-0.000019889
6.998971632E-007
1.03525188ME-007
4-25
-------
Table 4-19- Pearson Product-Moment Correlation Coefficients Between Landsat MSS Bands
and MMS Channels for 13 Sampling Sites on 5 Colorado Lakesa'b
MMS Channel
CH1
CH2
CH3
CH4
CH5
CH6
CH7
CH8
CH9
CH10
CH11
[370-413 run]
[440-490 nm]
[495-535 nm]
[540-580 nm]
[660-700 nm]
[700-740 nm]
[760-860 nm]
[970-1060 nm]
[8000-13000 nm]
Landsat MSS Band
GRN RED IR1
[500-600 nm] [600-700 nm] [700-800 nm]
0.824
0.827
0.877
0.216
(MMS data missing)
(MMS data missing)
0.864
0.825
0.606
0.083
0.865
0.756
0.777
0.821
0.863
0.882
0.812
0.563
0.035
0.735
0.498
0.426
0.483
0.639
0.568
HJtt
0.561
-0.041
0.568
IR2
[800-1 ,100 nm]
-0.050
-0.105
-0.051
0.089
0.053
0.234
0.131
-0.122
0.108
aThe lakes are Barr, Cherry Creek, Grand, Dillon and Milton,
b5% level = 0.553, 11 d.f. 1% level = 0.684, 11 d.f.
-------
Table 4-20. Pearson Product-Moment Correlation Coefficients
Between Landsat MSS Bands and MMS Principal
Component-Derived Variables for 13 Sampling Sites
on 5 Colorado Lakes
Landsat MSS Band
MMS
Variable
MMSPC1
MMSPC2
MMSPC3
MMSPC4
MMSPC5
MMSPC6
MMSPC7
MMSPC8
CRN
0.877
-0.102
-0.401
0.022
0.053
0.153
0.081
0.037
RED
0.825
-0.069
-0.427
0.204
-0.152
0.201
0.127
-0.035
IR1
0.537
0.297
-0.660
-0.066
0.099
0.276
0.220
-0.187
IR2
-0.010
0.305
-0.492
0.024
0.267
0.544
0.188
-0.312
Up to this point, only correlations between the remotely-sensed
spectral data and transformed spectral data have been examined. Of
particular interest to the limnologist are the correlations between
the contact-sensed data and the multispectral data (i.e., the correlations
between the ground or water truth and the multispectral data). The
correlations for the five sets of observations (i.e.. , the five data
sets discussed previously in this section) are displayed in Tables 4-21
to 4-25, inclusive. Overall, the correlations decline as the set size
increases. In general, the aircraft-borne MMS channels correlate better
with the contact-sensed data than the Landsat bands.
Correlations do not demonstrate cause and effect relationships.
As mentioned earlier, not all of the trophic indicators can be sensed
directly by the remote sensors. Such indicators include conductivity,
total phosphorus, and algal assay yield. However, statistically signi-
ficant correlations may exist between the remotely acquired data and
"nonsensible" indicators, a consequence of secondary effects. If
correlations of a relatively large magnitude exist, predictive models
can be developed.
4-27
-------
Table 4-21. Pearson Product-Moment Correlation Coefficients for 9 Colorado Lakes
(August 1975 Contact-Sensed and Destriped MSS Data)
MSS
Bands
GHN
RED
IR1
-tr
w IR2
GRNRED
GRNIR1
GRN1R2
REDIR1
PC1
0.910
0.665
0.761
-0.115
-0 . 565
0.143
0.743
0.359
CHLA
(LNCHLA)
0.881
(0.916)
0.938
(0.950)
0.799
(0.707)
0.213
( 0 . 020 )
-0.891
(-0.868)
-0.006
(0.177)
0.465
(0.624)
0.384
(0.487)
COND
(LNCOND)
0.688
(-0.861)
0.582
(0.764)
0.327
(0.539)
-0.455
(-0.242)
-0.241
(-0.394)
0.425
(0.337)
0.867
(0.824)
0.437
(0.435)
SEC
(LNSEC)
0.545
(-0.807)
-0.534
(-0.814)
-0.351
(-0.684)
0.337
(0.112)
0.426
(0.688)
-0.215
(-0.081)
-0.578
(-0.623)
-0.343
(-0.330)
IS EC
(LNISEC)
0.849
(0.807)
0.877
(0.814)
0.902
(0.686)
0.205
(-0.110)
-0.781
(-0.689)
-0.163
(0.080)
0.447
(0.622)
0.157
(0.330)
TPHOS
(LNTPHOS)
0.625
(0.723)
0.539
(0.646)
0.649
(0.662)
-0.079
(-0.222)
-0.256
(-0.358)
-0.061
( 0 . 027 )
0.549
(0.703)
0.010
(0.135)
TON
(LNTON)
0.845
(0.888)
0.810
(0.840)
0.822
(0.713)
0.018
(-0.167)
-0.587
(-0.584)
-0.046
(0.147)
0.613
(0.767)
0.168
(0.338)
AAY
(LNAA50
0.578
(0.816)
0.597
(0.783)
0.953
(0.894)
0.497
(0.092)
-0.530
(-0.559)
-0.545
(-0.182)
0.062
(0.532)
-0.281
(0.043)
MSS band values used were the mean values for each of the nine lakes. Contact-sensed data were also mean
values for each lake.
-------
Table 4-21. Pearson Product-Moment Correlation Coefficients for 9 Colorado Lakes
(August 1975 Contact-Sensed and Destriped MSS Data) (Continuation 1)
MSS
Bands
REDIR2
IR1IR2
PCIMSS
-fcr
1
PC 1 CHLA
(LNCHLA)
0.825 0.659
(0.788)
0.870 0.606
(0.668)
0.874 0.937
(0.900)
COND
(LNCOND)
0.831
(0.840)
0.722
(0.771)
0.508
(0.729)
SEC
(LNSEC)
-0.622
(-0.735)
-0.583
(-0.741)
-0.462
(-0.790)
IS EC
(LNISEC)
0.604
(0.734)
0.710
(0.740)
0.934
(0.791)
TPHOS
(LNTPHOS)
0.537
(0.708)
0.723
(0.842)
0.620
(0.681)
TON
(LNTON)
0.693
(0.833)
0.804
(0.848)
0.863
(0.831)
AAY
(LNAAY)
0.182
(0.614)
0.531
(0.813)
0.790
(0.875)
CRN - mean green DN level
RED - mean red DN level
IR1 - mean infrared - one DN level
IR2 - mean infrared - two DN level
GRNRED - green to infrared - one ratio
GRNIR1 - green to infrared - two ratio
REDIR1 - red to infrared - one ratio
REDIR2 - red to infrared - two ratio
IR1IR2 - infrared-one to infrared-two ratio
CHLA - chlorophyll .a (ug/liter)
COND - conductivity (micromhos)
SEC - Secchi disc transparency (meters)
ISEC - inverse of Secchi disc transparency (1/meter)
TPHOS - Total phosphorus (milligrams/liter)
TON - total organic nitrogen (milligrams/liter)
AAY - algal assay control yield (milligram/liter dry wt.)
LN - natural logarithm transformation
PCI - trophic index values for lakes
generated from principal component
ordination analysis of 6 trophic
indicators (LNCHLA, LNCOND, LNISEC,
LNTPHOS, LNTON, LNAAY). August data
used.
PC1MSS - "new" MSS variable generated by
principal component analysis of
the four MSS bands.
-------
Table 4-22. Pearson Product-Moment Correlation Coefficients Generated from Landsat MSS and Trophic
Indicator Data for 2? Sites Located in 9 Colorado Lakes3'b»c
MSS
Bands
GRN
RED
IR1
-f
UJ
o
IR2
GRNRED
GRNIR1
GRNIR2
REDIR1
REDIR2
PC1
0.838
0.731
0.570
0.081
-0.410
-0.039
0.437
0.236
0.553
CHLA
(LNCHLA)
0.590
(0.799)
0.535
(0.740)
0.471
(0.574)
0.057
(0.167)
-0.352
(-0.516)
-0.144
(-0.103)
0.324
(0.3^9)
0.082
(0.230)
0.434
(0.512)
COND
(LNCOND)
0.640
(0.754)
0.485
(0.581)
0.253
(0.415)
-0.085
(0.141)
-0.129
(-0.149)
0.213
(0.126)
0.471
(0.311)
0.333
(0.252)
0.470
(0.351)
SEC
(LNSEC)
-0.458
(-0.770)
-0.379
(-0.710)
-0.174
(-0.509)
0.1011
(-0.057)
0.180
(0.474)
-0.177
(0.024)
-0.376
(-0.409)
-0 . 325
(-0.296)
-0.405
(-0.545)
ISEC
(LNISEC)
0.731
(0.770)
0.679
(0.709)
0.677
(0.511)
0.198
(0.058)
-0.456
(-0.475)
-0.257
(-0.026)
0.236
(0.408)
0.023
(0.293)
0.389
(0.544)
TPHOS
(LNTPHOS)
0.519
(0.657)
0.390
(0.524)
0.386
(0.414)
-0.022
(-0.015)
-0.121
(-0.238)
0.046
(0.024)
0.352
(0.424)
0.030
(0.185)
0.361
(0.472)
TON
(LNTON)
0.730
(0.776)
0.637
(0.707)
0.577
(0.526)
0.076
(0.060)
-0.366
(-0.459)
-0.145
(-0.076)
0.378
(0.425)
0.089
(0.231)
0.483
(0.555)
AAY
(LNAAY)
0.487
(0.725)
0.445
(0.627)
0.582
(0.584)
0.151
(0.039)
-0.280
(-0.323)
-0.325
(-0.124)
0.143
(0.393)
-0.177
(0.084)
0.252
(0.489)
PC1-27-53
0.850
0.743
0.556
0.090
-0.425
0.018
0.438
0.269
0.558
-------
Table 4-22. Pearson Product-Moment Correlation Coefficients Generated from Landsat MSS and Trophic
Indicator Data for 2? Sites Located in 9 Colorado Lakes3>b»c (Continuation 1)
MSS
Bands
IR1IR2
LNGRN
LNRED
LNIR1
LNIR2
PC1
0.502
0.833
0.736
0.557
0.096
CHLA
(LNCHLA)
0.440
(0.441)
0.571
(0.787)
0.531
(0.746)
0.458
(0.564)
0.061
(0.167)
COND
(LNCOND)
0.333
(0.257)
0.642
(0.763)
0.491
(0.583)
0.260
(0.410)
-0.062
(0.168)
SEC
(LNSEC)
-0.299
(-0.469)
-0.463
(-0.768)
-0.387
(-0.718)
-0.162
(-0.497)
0.104
(-0.065)
ISEC
(LNISEC)
0.457
(0.469)
0.719
(0.767)
0.678
(0.718)
0.641
(0.498)
0.207
(0.066)
TPHOS
(LNTPHOS)
0.409
(0.443)
0.518
(0.649)
0.401
(0.538)
0.365
(0.401)
-0.015
(-0.000)
TON
(LNTON)
0.506
(0.490)
0.720
(0.768)
0.640
(0.712)
0.557
(0.529)
0.085
(0.072)
AAY
(LNAAY)
0.422
(0.537)
0.471
(0.714)
0.435
(0.620)
0.534
(0.551)
0.155
(0.058)
PC1-27-53
0.483
0.846
0.751
0.547
0.104
aDestriped MSS data.
b25 degrees of freedom. 0.05 level = 0.381, 0.01 level = 0.487-
cTrophic index for 27 sites developed from principal components analysis of five indicators:
LNISEC, LNCOND, LNTPHOS, and LNTON.
LNCHLA,
-------
Table 4-23. Pearson Product-Moment Correlation Coefficients
Between Landsat MSS Bands and Trophic Indicators
for 13 Sampling Sites on 5 Colorado Lakes
Trophic
Indicator
CHLA
LNCHLA
ISEC
LNISEC
SEC
LNSEC
TPHOS
LNTPHOS
TON
LNTON
AAY
LNAAY
COND
LNCOND
CRN
0.584
0.812
0.763
0.847
-0.589
-0.847
0.506
0.732
0.783
0.890
0.477
0.813
0.650
0.800
Landsat
RED
0.521
0.739
0.700
0.769
-0.488
-0.769
0.351
0.586
0.655
0.765
0.432
0.712
0.477
0.652
Band
IR1
0.592
0.727
0.907
0.692
-0.284
-0.690
0.492
0.601
0.793
0.735
0.755
0.802
0.288
0.534
IR2
0.148
0.162
0.516
0.159
0.245
-0.156
0.067
0.098
0.278
0.198
0.388
0.299
-0.091
0.148
PC1-13
0.893
0.771
0.753
0.196
4-32
-------
Table 4-24. Pearson Product-Moment Correlation Coefficients Generated from MMS and Trophic Indicator Data
for 13 Sites Located in 5 Colorado Lakesa»b»c
MMS Channel
1 (370-413
2 (440-490
3 (495-535
4 (540-560
j_ 7 (660-700
I
UJ
LO
8 (700-740
9 (760-860
10 (970-1060
nm)
nm)
nm)
nm)
nm)
nm)
nm)
nm)
11 (8000-13000 nm)
CHLA
(LNCHLA)
0.652
(0.875)
0.61 1
(0.806)
0.609
(0.808)
0.647
(0.877)
0.632
(0.816)
0.706
(0.910)
0.732
(0.856)
0.527
(0.431)
0.428
(0.764)
COND
(LNCOND)
0.708
(0.578)
0.673
(0.585)
0.694
(0.658)
0.707
(0.710)
0.475
(0.501)
0.403
(0.442)
0.295
(0.223)
0.056
(-0.143)
0.873
(0.780)
SEC
(LNSEC)
-0.717
(-0.840)
-0.722
(-0.803)
-0.726
(-0.827)
-0.750
(-0.910)
-0.685
(-0.826)
-0.673
(-0.888)
-0.618
(-0.766)
-0.259
(-0.239)
-0.709
(-0.875)
ISEC
(LNISEC)
0.625
(0.839)
0.541
(0.802)
0.585
(0.825)
0.740
(0.907)
0.627
(0.825)
0.831
(0.889)
0.695
(0.766)
0.117
(0.239)
0.715
(0.874)
TPHOS
(LNTPHOS)
0.547
(0.764)
0.395
(0.652)
0.411
(0.671)
0.532
(0.767)
0.263
(0.541)
0.481
(0.675
0.366
(0.549)
-0.119
(0.055)
0.706
(0.883)
TON
(LNTON)
0.750
(0.856)
0.634
(0.784)
0.664
(0.816)
0.796
(0.908)
0.596
(0.730)
0.787
(0.812)
0.657
(0.659)
0.101
(0.135)
0.826
(0.932)
AAY
(LNAAY)
0.304
(0.663)
0.155
(0.557)
0. 183
(0.602)
0.382
(0.752)
0.263
(0.570)
0.632
(0.776)
0.481
(0.582)
-0.192
(-0.080)
0.402
(0.813)
TEMP
°C
0.838
0.824
0.867
0.929
0.792
0.773
0.609
0.161
0.953
PC1-13
0.842
0.768
0.801
0.903
0.733
0.835
0.679
0.126
0.923
PC1-13-53
0.872
0.806
0.836
0.924
0.761
0.837
0.693
0.174
0.934
aBarr ft., Cherry Creek R., Grand L., Dillon R., and Milton R.
b11 degrees of freedom. 0.05 level = 0.553; 0.01 level = 0.684.
cTrophio index for 13 sites developed from principal components analysis of five trophic indices: LNCHLA, LNISEC, LNCOND,
LNTPHOS, and LNTON.
-------
Table 4-25. Pearson Product-Moment Correlation Coefficients Generated from Principal Component-Derived
MMS Variable and Trophic Indicator Data for 13 Sampling Sites in 5 Colorado Lakes
(JO
MMS
Variable
MMSPC1
MMSPC2
MMSPC3
MMSPC4
MMSPC5
MMSPC6
MMSPC7
MMSPC8
CHLA
(LNCHLA)
0.656
(0.863)
0.357
(0.294)
0.008
(-0.166)
-0.077
(-0.201)
0.179
(0.044)
0.232
(0.161)
0.336
(0.066)
0.058
(0.046)
COND
(LNCOND)
0.663
(0.612)
-0.422
(-0.424)
-0.047
(-0.359)
-0.447
(-0.183)
0.152
(0.308)
-0.050
(0.034)
-0.168
(0.155)
-0.234
(-0.142)
SEC
(LNSEC)
-0.740
(-0.863)
-0.019
(-0.143)
0.108
(0.330)
0.078
(0.123)
-0.067
(-0.085)
0.265
(-0.029)
0.434
(0.239)
-0.257
(-0.099)
ISEC
(LNISEC)
0.646
(0.862)
0.336
(0.144)
-0.541
(-0.332)
-0.191
(-0.125)
0.205
(0.086)
0.210
(0.030)
-0.041
(-0.238)
-0.028
(0.098)
TPHOS
(LNTPHOS)
0.448
(0.700)
-0.017
(-0.031)
-0.380
(-0.316)
-0.726
(-0.560)
-0.003
(0.039)
-0.218
(-0.108)
-0.075
(-0.067)
-0.214
(-0.192)
TON
(LNTON)
0.712
(0.840)
0.152
(-0.010)
-0.428
(-0.350)
-0.470
(-0.353)
0.132
(0.113)
0.029
(0.034)
0.038
(-0.015)
-0.132
(-0.122)
AAY
(LNAAY)
0.277
(0.648)
0.465
(0.13D
-0.732
(-0.608)
-0.362
(-0.379)
-0.071
(0.053)
-0.131
(-0.053)
0.091
(0.005)
-0.055
(-0.043)
PC1-13
0.831
0.038
-0.390
-0.334
0.108
0.016
-0.025
-0.061
-------
D. REGRESSION MODELS FOR THE ESTIMATION OF TROPHIC INDICATORS
AND MULTIVARIATE TROPHIC INDICATORS
Regression models were developed for the estimation of CHLA,
ISEC, SEC, TPHOS, TON, COND, AAY, and three principal component-derived
trophic state indices, PC1-11, PC1-27, and PC1-13 (Table 4-26). It is
noted that the models are grouped by dependent variable (.£.£.., CHLA)
with each group consisting of one to seven models. The comments column
in Table 4-26 provides information explaining some factors on which
the models are based. This can be illustrated by examining the comments
made regarding the models for ISEC (models 7 to 13, inclusive):
(1) Model 7. The four Landsat bands were treated as independent
variables during the interactive model development phase.
The final model incorporates a single independent, the
Landsat RED band. The MSS values used for modeling purposes
consisted of averages for each band for each of nine
Colorado lakes. The dependent variable values, ISEC
(actually its natural logarithm transformed version), were
means for each of the same nine lakes. Landsat band ratios
were also considered as independent variables.
(2) Model 8. As with Model 7, the four Landsat bands and their
ratios were examined during the multiple regression procedure;
the final model incorporates only the Landsat CRN band.
In this particular case, the values of the dependent and
independent variables were means calculated for each of
27 sites located on 9 Colorado lakes. This is in contrast
to Model 7 in which the same 9 lakes were treated as "whole"
water bodies.
(3) Model 9. The four Landsat bands and their ratios were employed
in the modeling effort. The data values were means for
each of 13 sampling sites on 5 Colorado lakes. These sites
are included in the 27 sites discussed above and are the
same sites used in the development of the models discussed
next (10, 11, 13, and 13).
(4) Model 10. In this case, eight MMS channels (2 of the 11
channels were missing and it was not deemed appropriate
to include the thermal channel, CH11) were treated as
independent variables during the model development phase;
no channel ratios were employed. The data values were
averages for each of 13 sites on 5 Colorado lakes.
(5) Model 11. The same data and sites were employed as in
the developmental phase of model 10. However, the selection
of the independent variables was limited to channels 4, 7,
8, and 9. This was an effort to mimic the bands found
in the Landsat MSS.
4-35
-------
Table 4-26. Regression Models Developed from Contact, MSS and MMS Data
Model Dependent Intercept Independent Variables and Associated
Number Variable Value Coefficients
CHL'A
1 LNCHLA -3.036 +0.233 RED
2 LNCHLA -3-367 +0. 142 CRN
3 LNCHLA 0.728 +0.449 IR1; -1.039 IR2
4 LNCHLA -25.628 +0.358 CH1; -0.220 CH2; +0.207 CH7
5 LNCHLA -9.281 +0.081 CH4; +0.156 CH9
-Cr
1
^ 6 LNCHLA -12.901 +0.059 MMSPC1; +0.001 MSSPC2
ISEC
7 LNISEC • -3.931 +0.150 RED
8 LNISEC -4.817 +0.1 12 CRN
9 LNISEC -5.811 +0.1 35 CRN
10 LNISEC -8.120 +0.119 CH4
11 LNISEC -8.120 +0.119 CH4
Regression,
Residual Calculated
d.f. F-value R2x100
1, 7 64.87 90.26
1, 25 44.07 63.80
2, 10 15.35 75.43
3, 9 20.49 87.23
2, 10 30.71 86.00
2, 10 24.69 83.16
1, 7 13.76 66.28
1, 25 36.34 59.25
1, 11 27.92 71.73
1, 11 52.78 82.75
1, 11 52.78 82.75
Standard
Error
of
Estimate
3-2
21.6
28.7
23.0
18.0
17.4
0.272
0.660
0.629
0.717
0.717
Comments
9 lakes. Landsat
MSS
27 sites. Landsat
MSS
13 sites. Landsat
MSS
13 sites. MMS. 8
channel selection
13 sites. MMS.
Channel selection
limited to 4,7,8,9
13 sites. MMS. 8
principal component-
derived "new"
channels
9 lakes. Landsat
MSS
27 sites. Landsat
MSS
13 sites. Landsat
MSS
13 sites. MMS. 8
channel selection
13 sites. MMS.
12
LNISEC
9.474 +0.51 MMSPC1; -0.115 MMSPC3; -1.519 MMSPC7 3, 9
Channel selection limited
to 4,7,8,9
30.23 90.97 0.620 13 sites. MMS. 8
principal component
derived "new"
channels
-------
Table 4-26. Regression Models Developed from Contact, MSS and MMS Data (Continuation 1)
I
<_0
—a
Regression,
Model ' Dependent Intercept Independent Variables and Associated Residual Calculated
Number Variable Value Coefficients d.f. F-value
I5EC
13 LNISEC -6.165 +0.051 MMSPC1; -0.115 MMSPC3 2, 10 29.11
SEC
LNSEC (Model development not attempted)
LNSEC (Model development not attempted)
LNSEC (Model development not attempted)
LNSEC (Model development not attempted)
14 LNSEC 8.119 -0.119 CH4 1, 11 53.24
15 LNSEC -9.520 -0.051 MMSPC1; +0.114 MMSPC3; +1.527 MMSPC7 3, 9 30.78
16 LNSEC 6.196 -0.051 MMSPC1; +0.114 MMSPC3 2, 10 29.29
TPHOS
17 LNTPHOS -3.746 +0.575 Ifil; -1.301 IR2 2, 6 7.63
18 LNTPHOS -10.053 +0.176 GRN 1, 25 ^8.26
Standard
Error
of
R2x100 Estimate Comments
85.34 0.636 13 sites. MMS.
Selection limited
to first 4 compo-
nents of principal
component trans-
formed 8 channels
9 lakes. Landsat
MSS
27 sites. Landsat
MSS
13 sites. Landsat
MSS
13 sites. MMS
82.88 2.59 13 sites. MMS.
Channel selection limited
to 4,7,8,9
91.12 1.50 13 sites. MMS. 6
principal component-
derived "new"
channels
85.42 2.55 13 sites. MMS.
Selection limited to
first 4 components of
principal component
transformed 8
channels
71.75 0.215 9 lakes. Landsat MSS
41.84 0.237 27 sites. Landsat MSS
-------
Table 4-26. Regression Models Developed from Contact, MSS and MMS Data (Continuation 2)
Model
Number
TPHOS
19
20
21
22
1
CO TON
OO
23
24
25
26
27
28
COND
29
Standard
Regression, Error
Dependent Intercept Independent Variables and Associated Residual Calculated of
Variable Value Coefficients d.f. F-value R2x100 Estimate Comments
LNTPHOS -10.785 +0.201 GRN 1, 11 12.70 53.59 0.323 13 sites. Landsat
MSS
LNTPHOS -16.360 +0.462 CH2; -0.768 CH3; +0.712 CH4 4, 8 16.64 89.27 0.284 13 sites. MMS.
-0.462 CH7 8 channel selection
LNTPHOS -8.947 +0.437 CH4; -0.463 CH7 2, 10 20.53 80.42 0.198 13 sites. MMS
channel selection
limited to 4,7,8,9
LNTPHOS -28.507 +0.071 MMSPC1; -0.190 MMSPC3; -0.559 MMSPC4 3, 9 28.38 90.44 0.171 13 sites. MMS. 8
principal compo-
nent-derived "new"
cnannels
LNTON -6.511 +0.143 GRN 1, 7 25.99 78.78 0.268 9 lakes. Landsat
MSS
LNTON -5.367 +0.1 13 GRN 1 , 25 37.94 60.28 0.317 27 sites. Landsat
MSS
LNTON -5.363 +0.118 GRN 1, 11 41.83 79.18 0.410 13 sites. Landsat
MSS
LNTON -4.793 -0.091 CH3; +0.253 CH4; -0.104 CH7 3, 9 58.23 95.10 0.116 13 sites. MMS. 8
channel selection
LNTON -6.449 +0.167 CH4; -0.190 CH7; +0.076 CH8 3, 9 59.69 95.21 0.091 13 sites. MMS
channel selection
limited to 4,7,8,9
LNTON -10.732 +0.041 MMSPC1; -0.101 MMSPC3; -0.170 MMSPC4 3, 9 59.80 95.22 0.176 13 sites. MMS. 8
principal compo-
nent-derived "new"
channels
LNCOND -11.690 +0.882 GRN; -0.807 RED 2, 6 29.53 90.78 120 9 lakes. Landsat MJ
-------
Table M-26. Regression Models Developed from Contact, MSS and MMS Data (Continuation 3)
Model
Number
COND
30
31
32
33
34
_Cr
uo
AAY
35
36
37
38
39
40
Regression ,
Dependent Intercept Independent Variables and Associated Residual
Variable Value Coefficients d.f.
LNCOND -6.277 +0.493 GRN; -0.360 RED 2, 24
LNCOND -2.618 +0.201 GRN 1, 11
LNCOND 2.251 +0.255 CH4; -0.326 CH9 2, 10
LNCOND 2.251 +0.255 CH4; -0.326 CH9 2, 10
LNCOND -6.412 +0.057 MSSPC1 1, 11
LNAAY -6.285 +0.823 IR1; -0.954 IR2 2, 6
LNAAY -7.481 +0.190 QRN; +0.246 IR1; -0.634 IR2 3, 23
LNAAY -13.288 +0.439 GRN; -0.376 RED; +0.392 IR1 3, 9
LNAAY -12.273 +0.289 CH4; -0.654 CH7; +0.835 CH8 4, 8
-0.460 CH9
LNAAY -12.273 +0.289 CH4; -0.654 CH7; +0.835 CH8 4, 8
-0.460 CH9
LNAAY -12.841 +0.081 MMSPC1; -0.446 MMSPC3; -0.462 MMSPC4 3, 9
Standard
Error
Calculated of
F-value H2x100 Estimate Comments
31.38 72.33 271 27 sites. Landsat
MSS
19.46 63.89 4390 13 sites. Landsat
MSS
12.85 72.00 462 13 sites. MMS.
8 channel selection
12.85 72.00 462 13 sites. MMS
channel selection
limited to 4,7,8,9
6.57 38.40 444 13 sites. MMS. 8
principal compo-
nent-derived "new"
channels
21.37 87.69 15.7 9 lakes. Landsat MSS
13.84 64.35 44.9 27 lakes. Landsat
MSS
15.21 83.53 46.9 13 sites. Landsat
MSS
39.22 95.15 30.1 13 sites. MSS.
d cnannel selection
39.22 95.15 30.1 13 sites. MMS
channel selection
limited to 4,7,8,9
41.32 93.23 24.8 13 sites. MMS.
Eight principal
component-derived
"new" channels
-------
Table 4-26. Regression Models Developed from Contact, MSS and MMS Data (Continuation
Model Dependent Intercept Independent Variables and Associated
Number Variable Value Coefficients
PC1-11
41 PC1-11 -5.359 +0.218 RED; +0.421 IR1; -1.036 IR2
PC1-27
42 PC1-27 -12.414 +0.331 CRN
PC 1-1 3
43 PC1-13 -13.153 +0.539 CRN; -0.442 RED; +0.632 IR1
-1 .332 IR2
44 PC1-13 -12.617 -0.476 CH3; +0.797 CH4
-Cr
O 45 PC1-13 -17.912 +0.480 CH4; -0.595 CH7; +0.305 CH8
Regression,
Residual
d.f.
3, 5
1, 25
4, 8
2, 10
3, 9
Standard
Error
Calculated of
F-value R^xlOO Estimate Comments
21.06 92.67 0.67 9 lakes. Landsat MSS
59.59 70.45 1.19 27 sites. Landsat
MSS
33-22 94.32 0.86 13 sites. Landsat
MSS
68.14 93.16 0.84 13 sites. MMS. 8
channel selection.
60.89 95.30 0.86 13 sites. MMS.
46
PC1-13 -27.289 +0.126 MMSPC1; -0.348 MMSPC3; -0.497 MMSPC4 3, 9
62.93
95.45
Channel selection
limited to 4,7,8,9
13 sites. MMS.
Selection limited
to first 4 component
of principal compo-
nent transformed
8 channels
-------
(6) Model 12. As was noted in the preceding subsection on
interband and interchannel correlations, the correlative
problem is substantial. An attempt was made to circumvent
this situation, in the case of the MMS channels, by generating
"new" independent or orthogonal variables (MMSPC1, ... ,
MMSPC8) through principal components analysis. During the
development of Model 12, these "new" variables were treated
as independent variables. The dependent and independent
variable values were means for each of the 13 sites on
the 5 lakes.
(7) Model 13. The same basic approach was used as in the develop-
ment of Model 12. However, the selection of independent
variables was limited to the first four "new" variables
generated by the principal component analysis of the eight
MMS channels. These components (MMSPC1, ..., MMSPC4) account
for about 99-54/S of the variation in the eight-channel MMS
data available for the 13 sites.
As is evident from Table 4-26, some models were not developed for SEC.
It was determined that to do so would be to needlessly repeat the devel-
opment of the ISEC models, but in an inverse way. The actual, estimated,
and residual values for the dependent variables found in Models 1 to and
including 46 are in Appendix C.
An examination of Table 4-26 indicates that the Landsat-derived
models incorporate only the MSS bands as the independent variables.
The aircraft-borne MMS-derived models utilize either the MMS channels
(excluding CH5, CH6, and CH11) or "new" variables generated through
the principal component analysis of CH1, CH2, CH3, CH4, CH7, CH8, CH9,
and CH10. The development of regression models employing Landsat MSS
bands or MMS channels as independent variables for the estimation of
a variety of water quality parameters is not new. Several investigators
have also used MSS band ratios in attempts to reduce the magnitude
of the atmospheric effects.
Yarger and McCauley (1975), reporting on an investigation to
determine the feasibility of developing quantitative estimates of water
quality in Kansas reservoirs, employed Landsat MSS ratios in their
regression models. They indicate that (1) MSS ratios were very effective
in providing quantitative estimates of suspended solids up to at least 900
ppm; (2) significant correlations were not found between Landsat MSS data
and dissolved solids, potassium, phosphate, and nitrate, at least at the
levels found in the Kansas reservoirs, and the Landsat RED/CRN ratio
correlates weakly with total chlorophyll levels above approximately
8 )Jig/liter.
Rogers et ai. (1976) have developed regression models for prediction
of several water quality indicators in Saginaw Bay (Michigan). Their
models contain the following Landsat MSS band and band ratios as independent
variables:
(1) Temperature (°C) - RED.
(2) Secchi depth (m) - RED.
4-41
-------
(3) Chloride (mg/liter) - CRN/RED.
(4) Conductivity (|amhos/cm) - CRN/RED.
(5) Total Kjeldahl nitrogen (mg/liter) - GRN/RED.
(6) Total phosphorus (mg/liter) - GRN/RED.
(7) Chlorophyll a. (fig/liter) - GRN/RED.
They report the addition of other independent variables into the models
did not improve the regression correlation coefficient significantly.
Boland (1976) has also described the use of Landsat MSS ratios
as independent variables in regression models for prediction of trophic
indicator magnitudes. Attempts to use MSS ratios in the development
of regression models for the Colorado lakes were less successful than
efforts using the MSS bands' raw values.
Ideally, each regression model would have an R2 of 100$ and a
standard error of estimate of 0. An examination of the models in Table
4-26 indicates that all fall short of the ideal. This is not surprising
since in the area of environmental studies there tends to be much variation
in data and the attainment of very high R2 (e_.g. , > 95%) is the exception
rather than the rule. A few general comments can be made about the
models.
(1) As the number of observations increases, the R2 decreases
and the estimate of standard error increases. Model 1-3
and 7-9 are examples of the phenomenon. The estimates
for the nine "whole" lakes are much better than those for
the 27 sites located on the same lakes. The loss in accuracy
may be related to factors such as sensor "noise" and the
relatively small range of DN values compared to those of
the trophic indicators.
(2) The dependent variables in the final models for a specific
parameter (e..g.. , LNCHLA) and developed from a particular
sensor's data (e.g.., Landsat MSS) are sometimes different
and appear related to the sampling sites included in model
development. For example, in Model 1 for CHLA, RED is
the dependent variable and in Model 2, CRN is the dependent
variable; the 27 sites are on the 9 lakes. This suggests
that model development is very sensitive to the sampling
sites selected for calibration purposes.
(3) Some of the models for different parameters employ the
same dependent variable(s) (e.g., Model 1 and Model 7),
differing in their slope, intercept, and associated parameters.
This is not surprising because, in general, the trophic
indicators are correlated.
(4) In general, the models incorporating the principal component
transformed MMS data provide slightly better results as
measured by the R2 and standard error of estimate values
than those developed using the raw MMS data.
4-42
-------
(5) Overall, the MSS and MMS data provide better estimates
of the multivariate indices than of the individual trophic
indicators. This may be a consequence of the relatively
small range of the multivariate index values.
(6) A comparison of the models developed using ground truth from
the 13 sites indicates that, as measured by the R^ and
standard error of estimate values, the aircraft-acquired
MMS data provide better estimates of the trophic indicators
and multivariate index examined here. This outcome may
be a consequence of the MMS's greater spectral and spatial
resolution and/or the lower altitude of the aircraft (i.e.. ,
atmospheric effects of a lesser magnitude) compared to
that of Landsat and its MSS.
(7) The models may yield grossly inaccurate estimations when
applied to water bodies from other regions or when applied
to the same lake populations using MSS data from another
period of time.
While the regression approach is attractive the user is confronted
by several problems including:
(1) The prospect of data non-normality. This occurs frequently
in inland lake water quality studies.
(2) Statistically significant correlations between MSS bands
or MMS channels. This reduces the predictive value of some
bands or channels.
(3) The relatively small range of MSS and MMS DN values compared
to the ranges trophic indicator values often encountered
in surveys involving large numbers of lakes. This makes
it impossible to get accuracies and precisions obtainable
through contact sensing.
The investigator can reduce the magnitude of the first problems
by employing data transformation techniques which will induce a sense
of normality. Mueller (1972) has demonstrated the use of principal
component analysis to achieve orthogonality between variables. It
is unclear from the study reported here as to what advantage was gained
by generating "new" variables through the principal components analysis
of the MMS channels. It appears that some consideration should be
given to the use of class intervals for the trophic indicators because
the user will often be confronted with the problem of the limited dynamic
ranges of the MSS and MMS relative to the ranges of the ground truth
parameters.
E. THEMATIC MAPPING RESULTS
The purpose of the following classification maps is to visually
illustrate the relative lake rankings on a eutrophic-oligotrophic scale,
employing either a multivariate index or a single trophic indicator. The
maps serve as a visual aid in determining (1) the different trophic levels
-------
within a lake, (2) the locations of these different levels of trophy,
and (3) the positional relationships of the different levels of trophy
with each other and in comparison to the other lakes.
In determining the accuracy of a classification map it is important
to be aware of certain factors. The horizontal bands which occasionally
appear in the lake surface and are visually denoted by a continuous
stripe of color are not to be interpreted as trophic features. In this
Landsat imagery these are caused by residual sixth line banding which
was not removed. The extreme sensitivity of the Bayesian Classifier
causes the residual striping to appear due to the algorithm's ability
to detect relatively small differences in data. As such, the stipes
are an unfortunate artifact caused by the sensors and are not lacustrine
features. When interpreting the accuracy of the clasification maps,
one key is to look for uniform shading of colors from blues to greens,
class to class. Uniform shading would tend to uphold the classification
results as would large contiguous areas of color. A sudden jump from
one class to another quite far down the relative trophic scale may
suggest an error in the classification. However, this in not always
the case as discrete water type boundaries may exist.
Classification accuracy can further be examined, in the following
pages under the discussion of individual classification maps, through
an interpretation of the "classification analysis" tables listed with
each map. The tables enumerate the number of classes specified in a
trophic classification and the percentage of pixels which were classified
as belonging to each particular class. Ideally, the analyst would wish
to see 100$ accuracy across the diagonal of the classification analysis
table, but with Landsat data expecially, this is rarely the case.
Six color thematic maps have been chosen to represent the range
of trophic classifications attempted during the course of this project.
Both Landsat-acquired and aircraft-acquired contact-sensed data were
utilized in the classifications presented. All classifications were
based on this sensor-acquired data in relationship to the water quality
measurements taken by EPA personnel.
Individual water quality parameters, such as chlorophyll a, and
inverse Secchi disc, as well as combined parameters derived through
the principal components ordination method were mapped and are illustrated
on the following pages. The 13-site multivariate trophic index results
will be examined first.
1. Trophic Classification of Five Colorado Lakes Based on 13-site
Multivariate Index (PC1-13)
Figures 4-2 and 4-3 depict the spatial aspects of the trophic classi-
fication of 5 Colorado lakes based on Landsat MSS data and aircraft MMS
data respectively. The number of lakes classified was reduced from 9 to
5 due to the unavailability of aircraft sensed data for Barker Reservoir,
Green Mountain Reservoir, Shadow Mountain Reservoir and Blue Mesa Reservoir,
4-44
-------
LANDSAT MSS CLASSIFICATION
PCI WITH 6 TRDPHIC INDICATORS
13 CLASSES
CLASS 1
CLASS £
CLASS 3
CLASS 4
CLASS 5
CLASS 6
CLASS 7
CLASS 8
CLASS 9
CLASS 10
CLASS 11
CLASS 1£
CLASS 13
Figure 4-2. Trophic Classification Maps of Five Colorado Lakes
Based on Landsat-1 MSS Data and a Multivariate Trophic
Index for 13 Sampling Sites (PC1-13)
4-45
-------
AIRCRAFT HMS CLASSIFICATION
PCI WITH 6 TROPHIC INDICATORS
13 CLASSES
OLIGOTROPH1C
Figure 4-3. Trophic Classification Maps of Five Colorado Lakes
Based on MMS Data and a Multivariate Trophic Index
for 13 Sampling Sites (PC1-13)
4-46
-------
Each of the 13 sampling sites served as a trophic class (PC 1-13) training
site for both the Landsat MSS and MMS. In other words, parallel classi-
fication efforts were made using the maximum likelihood algorithm, one
employing MSS data and the other MMS data. The training site classifi-
cation results are compiled in Table 4-27. The computer classified 68
percent of the MSS pixels in Class 1 as having spectral properties
characteristic of Class 1. Ideally it would have classified 100 percent
of the pixels as belonging to Class 1.
Similarly 80/6 of the MMS pixels which served as Class 1 training
sites were classified as Class 1. In general, MMS training site classi-
fication results are better than those of the Landsat MSS. Overall, approxi-
mately 83% of the MMS pixels were recognized as belonging in the proper
class as compared to 63% of the Landsat pixels. Landsat training site
classification accuracies range from 28% for Class 6 to 100? for Class 13.
MMS accuracies range from about 64$ for Class 12 to 99% for Class 9.
Table 4-28 displays the results of extending both Landsat MSS
and MMS site-related trophic class signatures to all of the pixels
in each of the five lakes. Though not identical, MSS and MMS results
tend to follow the same general pattern. Based on the results of the
site classification (Table 4-27), it follows that the MMS gives a more
representative ''picture" of the trophic status of the five lakes.
While the color-coded maps can be of practical values to limnologists,
it must be kept in mind that several of the training sites were relatively
heterogeneous, resulting in low site classification accuracies. Inclusion
of such classes can adversely affect the lake classification results.
Ideally, each training site should be homogenous and spectrally
separable from the other sites. Each site would then be spectrally
unique for its multivariate trophic index value (i.e.. , have a spectral
signature). If the above criteria of homogeneity and separability
are not met, then the investigator should consider selecting different
training sites and/or pooling some of those currently used. Generally
the procedure is one of trial and error, guided by training site statistical
information and the investigator's knowledge about the water bodies.
It appears that some of the PC1-13 classes should be pooled.
2. Trophic Classification of Nine Colorado Lakes Based on a Pooled
Multivariate Index (PC1-27)
The color map, Figure 4-4, depicts the results of the classification
of nine Colorado lakes using Landsat MSS data. The classification was based
on the trophic ranking of 27 sampling sites derived from principal components
ordination. By using principal components analysis a single numerical ex-
pression representing 6 trophic indicators was derived. The 6 trophic
indicators used were chlorophyll a, inverse Secchi disc, conductivity, total
phosphorus, total organic nitrogen and algal assay yield. Table 4-29
lists the PC1 values for these 27 sampling sites as derived from principal
components analysis. For classification, the PC1 values were ranked on
a relative trophic scale and grouped into classes by identifying natural
groupings in the numerical values.
4-4?
-------
Table 4-27. Landsat MSS and Bendix MMS Training Site Classification Accuracies Expressed as a Percentage
of Each Site's Pixel Count
-tr
GO
Sen-
Class sor 1
1
2
3
4
5
6
7
8
9
10
11
12
13
MSS 68.0
MMS 80 . 1
MSS 4 . 0
MMS 14.0
MMS 14.0
MMS
MSS 20 . 0
MMS 5.7
MSS 8 . 0
MMS 2.4
MSS 12.0
MMS
MSS
MMS
MSS
MMS
MSS
MMS
MSS
MMS
MSS
MMS
MSS
MMS
MSS
MMS
Class
_!_ -J- • -5- 6 _j_ 8 _*_
4.0 4.0 16.0 8.0
10. 7 5.7 3.3
52.0 12.0 20.0 12.0
85.9
12.0 52.0 20.0 4.0 12.0
92.5 7.4
12.0 16.0 36.0 16.0
1.6 82.6 9.0 0.8
4.0 88.0
16.5 80.9
16.0 24.0 12.0 8.0 28.0
12.3 85.9 0.8
92.0 4.0
66.9 12.3 2.4
4.0 48.0
10.7 83.4
12.0 8.0 80.0
0.8 99.1
8.0 8.0
17.3 0.8
20.0 4.0
4.1
4.0 12.0
19.0 12.3
12.3 0.8
0.8
4.0
2.4 0.8
4.0 20.0 24.0
4.1
76.0 4.0 4.0
81.8
4.0 40.0 32.0
95.0
8.0 16.0 60.0
1.6 63.6
6.6
14.8
1.6
0.8
100.0
81,0
Percentage values of less than 0.05? are represented by empty matrix cells.
The Landsat MSS training sites samples generally consisted of 25 pixels (5x5 matrix). The MMS training site samples
consisted of 121 pixels (11 x 11 matrix).
-------
Table 4-28. Landsat MSS and Bendix MMS Lake Classification Results Expressed as a Percentage of Each
Lake's Pixel Count
Class
Num-
ber
1
2
3
4
5
6
7
8
9
10
11
f 12
£ 13
PC 1-13
Value
-2.46
-2.31
-2.18
-2.07
-2.07
-2.05
0.57
0.89
1.55
1.93
1.94
2.78
3.47
Site
Storet
080603
080604
080702
080602
080701
080601
080401
080403
081101
081102
080402
080202
080201
Dillon
MSS
35.63
10.01
10.01
11 .17
16.73
5.36
6.19
1.78
0.91
0.38
0.04
0.74
MMS
26.83
28.29
4.82
24.11
11.98
3.63
0.01
0.01
o.ooa
0.26
0.01
0.01
Grand
MSS
26.76
8.30
28.00
6.46
6.46
19.07
4.30
0.30
0.30
MMS
0.05
32.05
0.16
65.95
0.07
0.20
0.15
0.15
1.04
0.07
0.08
Cherry Creek
MSS
30.56
16.63
0.19
11.99
16.63
20.50
3.48
MMS
0.43
0.29
24.17
43.72
3.71
1.82
8.68
6.94
10.21
Milton
MSS
1.37
7.53
6.33
4.11
34.41
25.51
8.04
7.36
5.30
MMS
35.39
2.40
31.60
16.72
0.29
13-58
Barr
MSS
0.30
1.23
13.44
0.77
6.49
14.21
14.83
48.68
MMS
24.69
8.85
7.41
0.01
1.52
22.49
35.00
Percentage values of less than 0.05? are represented by empty matrix cells.
aLess than 0.005J.
-------
. ettf «•, ran
i ,
-*.*» fa -A,**.
PCI 6 IND.
1
a
CLASS 3
CLASS 4
CLASS 5
Figure 4-4. Trophic Classification of Nine Colorado Lakes Based on
a Pooled Multivariate Index (PC 1-2?)
4-50
-------
Table 4-29. Classification Analysis of Nine Colorado Lakes Based on
a Pooled Multivariate Index (PC1-27)
Percent Classified as
*
Class 0
1 0.0
2 0.0
3 0.0
4 0.0
5 0.0
1
76.0
24.0
0.0
7.5
0.0
2
19.0
68.0
0.0
42.6
1.7
3
2.0
0.0
90.2
13-3
13-7
4
3.0
8.0
4.8
34.2
1.7
5
0.0
0.0
4.8
2.2
82.8
Ambiguous
0.0
0.0
0.0
0.0
0.0
•Undefined classification
A total of five classes were selected for this particular classification.
Table 4-29 enumerates the classification accuracy in terms of percent of
pixels actually assigned to a particular class. Ideally, each cell on the
matrix diagonal would contain a value of 100 indicating that all of the
pixels within a particular training site (in this case pooled training sites)
were identified by the classification program as being in the class for which
they provided the calibration data. Class 4 with a classification percentage
of 34.2 is a highly suspect class; 42.6$ of the pixels serving as calibration
types have been classified as belonging to Class 2. The confusion resulting
from the heterogenous nature is evident in Blue Mesa (Figure 4-4), where
Class 2 and Class 4 dominate. Particularly evident in Blue Mesa is the
striping, a consequence of the MSS's sixth line banding. This striping
contributes to the heterogeneity of training sites.
The six training sites on Blue Mesa were pooled with the three
sites on Shadow Mountain Reservoir and one on Green Mountain Reservoir
to form Class 4. It was expected that Blue Mesa would be wholly Class 4.
The Class 2 spatial features of Blue Mesa are markedly linear, suggesting
that they tie in with sixth line banding.
The five-class trophic maps of the nine lakes were examined visually
on a training site by training site basis and, with three exceptions,
excellent results were obtained for the 27 sites. Site 80803 on Green
Mountain Reservoir is classified largely as being Class 3; it was originally
pooled into Class 4. On Dillon Reservoir, site 80601 was pooled into
Class 1; it is classified largely as Class 2. While pooled into Class 4,
site 81302 on Shadow Mountain Reservoir is pictured on being largely
Class 3- Overall, the resulting classifications of the nine reservoirs
appear to be representative of their actual condition. Further work
might lead to pooled classes which are spectrally more distinguishable
and result in more accurate classification products.
4-51
-------
3- Trophic Classification of Nine Colorado Lakes Based on a Nine-Class
Multivariate Trophic Index
Figure 4-5, illustrates the classification results from nine
Colorado lakes again using Landsat MSS data. The classification was based
on the trophic ranking of 27 sampling sites derived from principal com-
ponents ordination. Mean trophic indicator values were used to generate
the PC1 index and 9 classes were chosen to represent the trophic scale.
Six trophic indicators were again used in the principal components
analysis. These were chlorophyll a., inverse Secchi disk, conductivity,
total phosphorus, total organic nitrogen and algal assay yield. Table
4-30 depicts the classification accuracy.
There is a wide range in training site classification accuracy, with
Class 1 having the greatest (80.0?) and Class 6 the lowest (36.7$). Again,
the ideal value would be 100?. Each training class is not recognized as
being homogeneous by the computer program. This suggests the need for
merging some of the classes and/or deleting some of the sampling sites
from the analysis.
As seen in Figure 4-5, the nine water bodies are not homogeneous.
Again, the sixth line striping or banding is apparent, particularly in
Blue Mesa Reservoir. The thematic photo maps suggest that Grand Lake,
Cherry Creek Reservoir, and Milton Reservoir are relatively homogeneous.
It is most unfortunate that the sixth line striping is so pronounced.
4. Chlorophyll .a Classification of Nine Colorado Lakes
Figure 4-6 depicts the results of a classification based on chloro-
phyll .a measurements taken at 27 sampling sites on nine Colorado lakes.
The classification is again based on Landsat data. The field teams
measured chlorophyll a. in jig/liter; these measurements constitute one
of several trophic indicators measured and recorded. Unlike the preceding
two classifications, which are based on the combination of 6 different
trophic indicators, this classification illustrates the distribution of
an individual trophic indicator in each of nine lakes. The measurements
for the 27 sampling sites were ranked on a numerical scale and then
grouped into four classes to represent relative levels of chlorophyll a.
Table 4-31 represents the classification accuracy.
As evidenced from the table, Class 1 with chlorophyll a values
ranging from 2.1 to 4.1 ^jig/liter is the least homogeneous and Class
4 is the most homogeneous. A visual examination of the maps indicates
that they compare favorably with the CHLA values found in Table 3-5.
However, the Class 4 training sites have a very large CHLA range and
perhaps should be divided to provide additional classes.
4-52
-------
OLXGOTRQPHIC
•«<••« tff- nli ',»< i,** ft
EUTiOPHIC
Figure 4-5. Trophic Classification of Nine Colorado Lakes Based on
a Nine Class Multivariate Trophic Index
4-53
-------
4-30. Classification Analysis of Nine Colorado Lakes Based on a Nine-Class Multivariate
Trophic Index
-Cr
Lake
Grand
Dillon
Barker
Green
Shadow
Mountain
Blue Mesa
Cherry
Creek
Milton
Barr
Class
1
2
3
4
5
6
7
8
9
0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
1
80.0
24.0
4.0
0.0
10.6
3.1
0.0
0.0
0.0
2
16.0
42.0
12.0
0.0
5.3
12.7
0.0
2.0
0.0
3
4.0
19.0
52.0
1.5
32.0
16.7
0.0
2.0
0.0
4
0.0
2.0
0.0
72.7
1.3
11.1
0.0
4.0
2.0
5
0.0
8.0
28.0
1.5
46.6
8.7
0.0
0.0
0.0
6
0.0
4.0
4.0
1.5
4.0
36.7
0.0
2.0
0.0
7
0.0
0.0
0.0
0.0
0.0
0.0
73.3
8.0
26.0
8
0.0
1.0
0.0
15.1
0.0
9.5
16.0
78.0
4.0
9
0.0
0.0
0.0
7.5
0.0
0.7
10.6
4.0
68.0
Ambiguous
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
-------
Figure 4-6. Chlorphyll & Classification of Nine Colorado Lakes
4-55
-------
Table 1-31• Analysis of Chlorophyll a. Classification of Nine
Colorado Lakes
Percent Classified as
*
Class 0123
1 0.0 42.2 33-1 21.7
2 0.0 18.9 77.3 1.4
3 0.0 19.0 16.0 53.0
4 0.0 0.0 3-5 15.0
4 Ambiguous
2.8 0.0
2.1 0.0
12.0 0.0
81.5 0.0
*Undefined classification
Table 4-32. Analysis of Inverse Secchi Depth Classification
of Nine Colorado Lakes
Percent Classified as
*
Class 0123
1 0.0 73.6 3.5 21.6
2 0.0 33-7 24.8 26.7
3 0.0 31.1 1.8 59.6.
4 0.0 0.0 5.1 12.0
4
1.1
14.6
7.3
82.8
Ambiguous
0.0
0.0
0.0
0.0
*
Undefined classification
4-56
-------
5. Inverse Secchi Depth Classification of Nine Colorado Lakes
The correlation of inverse Secchi disk transparency measurements
with Landsat MSS data is illustrated in the thematic map in Figure 4-7.
NES acquired Secchi disc measurements for 27 sampling sites on 9 lakes
were numerically ranked and grouped into 4 classes to represent relative
levels of eutrophication. As with the chlorophyll a. content classification,
this thematic map depicts an individual trophic indicator as opposed to
a combination of trophic indicators (i.e.. , multivariate trophic index
such as the PC1). The class accuracies are found in Table 4-32. Class 2
is highly suspect because of its heterogeneity; only 24.8$ of the pixels
used to define it spectrally were identified as belonging to the class.
Class 4 is clearly distinguishable from the others. The correlation
between the ISEC maps and ISEC values in Table 3-5 is very good. It
is very likely that even better correlations would result if additional
effort were expended to further refine the classes.
4-57
-------
MM CHERRY QKIK TO.TB1
fflltW MDUWTAIN
KEUJK
mt. mxm rase
CLASS
CLASS §
CLASS 3
* TO
*•*
CLASS 4
.487 TO*.*
Figure 4-7. Inverse Secchi Depth Classification Maps for Nine
Colorado Lakes
4-58
-------
F. POINTS OF CONCERN
As evidenced from the previous discussion, the use of Landsat MSS
and MMS data in lake classification work is not without its problems, some
relating to the capabilities and limitations of the scanners, others to
the characteristics of the scanner platforms, still others to environ-
mental factors, such as the optical properties of water, the quantity and
quality of the contact-sensed data, the dynamic nature of lacustrine
bodies, and the very nature of the concept of trophic state.
The Landsat MSS was not designed specifically for water quality
studies. It is a low resolution device both spatially and spectrally
speaking. While the MMS has better resolution, both instruments have a
rather limited dynamic range. The predictive capabilities of the scanners
are effectively reduced by the disparity between the range of their data
and those of the trophic indicators. The energy return from water is
relatively small, particularly in the 600 nm to 1100 nm portion of the
spectrum. Thus, the remotely sensed water quality information is
contained in relatively weak signals over a very limited range.
The atmosphere can have a substantial effect on the character of the
signal enroute to and coming from a water body. Atmospheric variability
both over time and space adds variability to the remotely sensed data,
thereby making the elucidation of relationships with the contact-sensed
data much more difficult by increasing the element of uncertainty. The
variability is remotely sensed data, in this case Landsat MSS data, is
illustrated in Table 4-33- William Fork Reservoir and Granby Reservoir
lie in the area of side overlaps; the DN values and pixel counts are those
for two consecutive days. Ground truth is not available to separate the
differences into an atmospheric component and a lacustrine component.
Although these water bodies lie about two kilometers above sea level (and
therefore above a substantial percentage of the atmosphere, this does not
mean that the effects of atmosphere can be ignored. It was anticipated
that the largest differences would be in the CRN band. This is contrary
to the measured average differences of -0.08, -1.67, -2.21, and -2.09 for
the GRN, RED, IR1, and IR2 bands, respectively. The atmosphere has a
greater effect on short wavelengths than on long wavelengths. The above
differences suggest that some other factor or factors (e_.g.. , lacustrine
dynamics, sensor characteristics) are also molding the character of the
data generated from the electromagnetic energy.
Sun glint (i.e.. , specular reflection) is a major problem when col-
lecting multispectral data from lakes using the aircraft-borne MMS. Of the
12 lakes flown over by the aircraft-borne scanner, usable data were only
obtained from five; sun glint effectively marked the volume reflectance of
the remaining lakes. Sun glint has not been demonstrated as being a major
problem with the Landsat MSS.
The Landsat space observatory provides a stable platform from which
to monitor the Earth. In contrast, aircraft platforms are inherently
unstable, which leads to problems of geometric fidelity for which there
are no easy or inexpensive solutions. The spacecraft is tied into a more
or less fixed orbital path. This results in a set pattern of coverage
which cannot take advantage of cloud-free days occurring other than the
4-59
-------
Table 4-33- Changes in DN Values for Two Colorado Lakes Over Two Days
-tr
I
Landsat Scene Date GRN
William Fork Reservoir 5126-16474 8-23-75 42.17
(3.49)
5127-16532 8-24-75 42.36
(3.45)
A-0.19
Granby Reservoir 5126-16474 8-23-75 35.07
(3.80)
5127-16532 8-24-75 35.04
(2.93)
A0.03
RED
24.35
(4.70)
26.52
(4.94)
A-1.75
17-97
(4.48)
19.56
(3.70)
A-1.59
IR1
14.08
(6.29)
16.52
( 6 . 29 )
A-2.44
10.28
(5.71)
12.26
(4.72)
A-1.98
IR2
4.69
(5.94)
7.11
(6.29)
A-2.42
3-37
(5.33)
5.12
(4.78)
A-1.75
Pixels
885
875
A7
4166
4362
A-196
-------
fixed dates of flyover. On the other hand, aircraft can operate on a
more flexible schedule, taking advantage of clear days.
Cloud cover is a major problem in the case of Landsat imagery.
In this study, two of the five scenes of interest were not used because
of excessive cloud cover . The problem becomes particularly acute when
time series studies are attempted. Clouds add an element of chance
to any lake program using satellite and aircraft-acquired remotely
sensed data.
The success of efforts to develop predictive models and classifi-
cation products through supervised approaches (.e.g.., Bayesian maximum
likelihood classifier) is highly dependent upon the quantity and quality
of contact-sensed data (I.e.. , calibration data). The contact-sensed
data used in this study was collected with no thought of their being
used in conjunction with remotely sensed data. The parameter values
are highly skewed, necessitating the use of a transform function.
This problem is frequently encountered in environmental data and must
be addressed if regression is to be employed. Currently, there are
no well recognized criteria to determine the number of sampling sites,
parameters, and sampling techniques necessary to adequately assess
the trophic condition of a lake. Trophic state itself is a multidimen-
sional concept, and the parameters to be included in it are open to
question.
M-61
-------
SECTION V
CONCLUSIONS AND RECOMMENDATIONS
A. CONCLUSIONS
The following conclusions are drawn based on the elucidation
of contact-sensed, Landsat MSS, and MMS data relationships for nine
Colorado lakes.
(1) Both the LANDSAT MSS and the Bendix MMS can give estimates
of lake surface area which are of practical value. Slightly
better results are obtained with the MMS. More accurate
estimates may be obtained using a multiband approach in
place of the single band or channel approach reported here.
(2) Regression models can be developed from both MSS and MSS
data for the prediction of specific trophic indicators.
However, the model estimates do not have the precision and
accuracy of those measurements acquired solely through contact
sensing methods. Generally speaking, the MSS-related models
give better estimates than the Landsat MSS-related models.
(3) The estimation of a numeric trophic state index is possible
through the development of regression models using contact-
sensed data and either MSS or MMS data. The MMS-related
model provides better estimates than the Landsat model.
(4) The production of trophic index photomaps is feasible using
contact-sensed data and either MSS or MMS data. Based on
training site classification results, the MMS product is
more accurate than the LANDSAT product.
It is suggested that the indications of better performance and
results obtained using the MMS are due to its greater spectral and
spatial resolution and less atmospheric interference, a consequence
of the aircraft's relatively low altitude compared to that of Landsat.
B. RECOMMENDATIONS
As is evidenced by this report, the use of multispectral data
in lake classification studies is not without its problems. To overcome
or reduce the magnitude of the problems the following recommendations
are made.
(1) An interactive image processing system (or subsystem) should
be developed specifically for the classification of lakes.
The system should be designed so the resource manager and
lake scientist can employ it in a "hands on" mode; this
should result in a more efficient use of equipment in terms
of time and money and more accurate lake classification
products. Such a system would have to contain a large
5-1
-------
library of digital image preprocessing functions as well
as statistical functions and supervised and unsupervised
classificatory capabilities.
(2) The multivariate trophic indices used in this study are
based on a relatively small sample set. Further considera-
tion of the principal component approach to lake ordination
should be directed toward a model (i.e.. , first principal
component) developed from a large data base. The lakes
included in the data base should not have turbidity problems
attributable to suspended inorganic materials. High concen-
trations of inorganic suspended sediments result in light-
limited conditions, changing the response curve relative to
nutrient levels. It therefore follows that it is inappropriate
to place all inland fresh water lakes on the same trophic
scale. While five to six indicators were incorporated
into the indices used in the investigation, the number
and types may not be the most appropriate. Trophic index
development is one area in need of further consideration.
The principal component technique is intuitively appealing.
However, other indices should also be examined.
(3) A multiband technique should be used to extract the water
pixels from the terrestrial matrix.
(4) It is imperative that an effort be made to remove or reduce the
magnitude of atmospheric effects. This might be accomplished
by a "zeroing out" of the atmospheric component through
use of deep, ultraoligotrophic water bodies, the assumption
being that all of the energy return is from the atmosphere.
More elegant techniques employing a combination of highly
sophisticated equations and additional radiometric data
might also be considered, but these are cumbersome and
may not be cost-effective.
(5) An increase in the MSS's gain would be of value; it would in-
crease the range of MSS data relating to water quality. However,
it will result in sensor saturation for more land features.
(6) The sixth line striping so evident in water bodies is distracting
cosmetically and also contributes "noise" which make MSS-ground
truth relationships more difficult to decipher. Algorithm
development should be undertaken to eliminate the problem or
a MSS developed which is not plagued by it.
(7) This study was limited by circumstances to one point in time.
The need exists to study the capabilities of the LANDSAT MSS
for change detection. This would require the monitoring of
a lake or group of lakes over a span of years.
(8) The spectral signatures identified for Colorado lakes under
study should be extended to other water bodies in the same
scene to determine if extrapolation techniques would be
useful for classification and lake ranking.
5-2
-------
GLOSSARY
AAY
A-space
CDC
CH
DCS
DN
CCT's
CERL
CHLA
cm
COND
EMSL-
Las Vegas
EPA
EHTS
FOV
ha
Hz
CRN
IBM
IFOV
IPL
algal assay yield (dry weight in mg/liter)
attribute space
Control Data Corporation
channel of the modular multispectral scanner. More specifi-
cally, for example, CH4 is channel 4 (540 to 580 nm)
data collection system
digital number level; the Landsat multispectral scanner
data and modular multispectral scanner data were processed
at 8 bits of precision resulting in 256 DN levels (0 to 255)
computer compatible tapes
Corvallis Environmental Research Laboratory, Corvallis,
Oregon
chlorophyll a (^g/liter)
centimeter
conductivity (micromhos/centimeter)
Environmental Monitoring and Support Laboratory, Las Vegas,
Nevada
U.S. Environmental Protection Agency
Earth Resources Technology Satellite, the acronym and name
originally applied to the NASA satellites designed to monitor
the earth's resource; Landsat, an acronym for land satellite,
is currently applied to the series
field of view
hectare (10,000 square meters)
hertz (i.e.. , cycles per second)
Landsat multispectral scanner green band with lower and
upper limits of 500 and 600 nm, respectively
International Business Machines
instantaneous field of view
Image Processing Laboratory, an operational unit at the
Jet Propulsion Laboratory
6-1
-------
IR infrared radiation
IR1 Landsat multispectral scanner near infrared one: band
with lower and upper limits of 700 and 800 nm, respectively
IR2 Landsat multispectral scanner near infrared two: band
with lower and upper limits of 800 and 1,100 nm, respectively
1SEC inverse of Secchi depth (1/m)
JPL Jet Propulsion Laboratory
Landsat land satellite, the acronym currently applied to NASA's
Earth Resources Technology Satellite (ERTS) series
LN natural log-transformation
m meter
rag milligram
MMS modular multispectral scanner
MMSPC1 first principal component resulting from principal component
analysis of modular multispectral scanner data for 13 sites
MSS multispectral scanner found in Landsat
NASA National Aeronautics and Space Administration
NEhC National Environmental Research Center, a no longer functional
unit of the U.S. EPA's Office of Research and Development
NES National Eutrophication Survey
nm nanometer (1 x 10~9m)
NOAA National Oceanic and Atmospheric Administration
PC1 multivariate trophic state index developed through principal
component analysis of several (five or six) trophic indicators
PC1-11 multivariate trophic index developed through principal
component analysis of six trophic indicators for 11 water
bodies
PC1-13 multivariate trophic index developed through principal
component analysis of six trophic indicators for 13 sampling
sites in five Colorado lakes
PC1-2? multivariate trophic index developed through principal
component analysis of six trophic indicators for 27 sampling
sites in nine Colorado lakes
6-2
-------
pixel
PL 92-500
PMT
R
R2
flBV
RED
r-matrix
rps
SEC
SIPS
sr
STORET
TON
TPHOS
W
USGS
VERTSLOG
V/H
mg
mm
picture element, the basic unit of spatial resolution.
The nominal pixel size for the Landsat multispectral scanner
is 57 by 79 m; in this report data preprocessing has resulted
in a Landsat pixel approximately 80 by 80 m. The modular
multispectral scanner pixel is 15 by 15 m.
Congressional act of 1972 entitled Federal Water Pollution
Control Act Amendments
photomultiplier tube
multiple correlation coefficient
multiple correlation coefficient squared
return beam vidicon; each Landsat space observatory contains
three RBVs
Landsat multispectral scanner red band with lower and upper
limits of 600 and 700 nm, respectively
product-moment correlation matrix
revolutions per second
Secchi depth (m)
Statistical Interactive Programming System, an interactive
computer system at Oregon State University
steradian
STOrage and RETrieval. the U.S. EPA's computer-based information
system for water quality data
total organic nitrogen (mg/liter)
Total phosphorus (mg/liter)
watt
U.S. Geological Survey
a VICAR application computer program which converts interleaved
multispectral data into band sequential imagery; the program
also performs radiometric and geometric corrections
velocity/height
microgram
micrometer
6-3
-------
microsecond
1-D one-dimensional space; characterized by a single attribute
/N estimated or predicted parameter value
6-4
-------
REFERENCES
Anderson, T.W. 1958. An Introduction to Multivariate Statistical
Analysis. New York: John Wiley and Sons. 374 pp.
Atkins, W.R.G. and H.H. Poole. 1952. An experimental study of the
scattering of light by natural waters. Proceedings of the Royal
Society (London). Series B. 140:321-338.
Beeton, A.M. 1965. Eutrophication of the St. Lawrence Great Lakes.
Limnology and Oceanography. 10(2):240-254.
Birge, E.A. and C. Juday, 1929. Transmission of solar radiation by
the water of inland lakes. Transactions Wisconsin Academy Science
Arts Letter. 24:509-580.
. 1930. A second report on solar radiation and inland
lakes. Transactions Wisconsin Academy Science Arts Letters.
25:285-335.
1931. A third report on solar radiation and inland
lakes. Transactions Wisconsin Academy Science Arts Letters.
26:383-425.
. 1932. A fourth report. Transactions Wisconsin
Academy Science Arts Letters. 27:523-562.
Blackwell, R.J. and D.H.P. Boland. 1974. Computer analysis of lakes
from ERTS imagery. In: Northwest '74 Pacific Regional Symposium
on Computing the Resources Industry. Canadian Information Processing
Society. 23-24 May 1974. Vancouver, B. C. pp. 147-157.
Blackwell, R.J. and D.H.P. Boland. 1975. The Trophic Classification
of Lakes Using ERTS Multispectral Scanner Data. Proceedings
American Society of Photogrammetry 41st Annual Meeting, March
9-14, 1975.
Boland, D.H.P. and R.J. Blackwell. 1975. The LANDSAT-1 multispectral
scanner as a tool in the classification of inland lakes. Proceedings
NASA Earth Resources Survey Symposium. First Comprehensive Symposium
on the Practical Application of Earth Resources Survey Data.
June 9-12, 1975. Shamrock Hilton. Houston, Texas. NASA TM
X-58168. JSC-09930. I-A: 419-442.
Boland, D.H.P. 1976. Trophic Classification of Lakes Using LANDSAT-1
(ERTS-1) Multispectral Scanner Data. U.S. EPA Ecological Research
Series Report. EPA-600/3-76-037. 263 pp.
Brezonik, P.L. and E.E. Shannon. 1971. Trophic State of Lakes in
North Central Florida. Publication 13. Florida Water Resources
Research Center. Gainesville: University of Florida. 102 pp.
R-1
-------
Collins, J.R. 1925. Change in the infrared absorption spectrum of
water with temperature. Physics Review. 26:771.
Curran, R.J. 1972. Ocean color determination through a scattering
atmosphere. Applied Optics. 11(8):1857-1866 .
Davis, F.J. 1941. Surface loss of solar and sky radiation by inland
lakes. Transactions Wisconsin Academy Science Arts Letters.
33:83-93.
Dawson, L.H. and E.O. Hulburt. 1937. The scattering of light by water.
Journal Optical Society America. 278:199-201.
Donaldson, J.R. 1969. The classification of lakes. Proceedings of
the Eutrophication-Biostimulation Assessment Workshop. 19-21
June 1969. Held at Hotel Claremont, Berkeley, California. Sponsored
by University of California (Berkeley) and Pacific Northwest Water
Laboratory, U.S. Environmental Protection Agency. 171-185 pp.
Duntley, S.Q. 1963.
53:214-233.
Light in the sea. Journal Optical Society America.
Edmondson, W.T. 1974. Review of: Environmental phosphorus handbook.
Edited by E.J. Griffith, A. Beeton, J.M. Spencer, and D.T. Mitchell.
New York: Wiley-Interscience. 1973. 718 pp. Limnology and
Oceanography. 19(2):369-375.
Ewan, T. 1894. On the absorption spectra of dilute solutions. Proceedings
Royal Society (London). 57:117-161.
Freden, S.C. 1973. Introduction: performance of sensors and systems.
In: Earth Resources Technology Satellite-1 Symposium Proceedings.
29 September 1972. NASA (Goddard Space Flight Center). Compiled
by W.A. Finch, Jr. pp. 1-6.
Gerd, W.V. 1957- An attempt at the biolimnological regionalization
of the USSR. Transactions of the Sixth Conference on the Biology
of Inland Waters. Edited by E.N. Pavlovskii. Original in Russian.
Israel Program for Scientific Translations. Jerusalem: IPST
Press. 1969. pp. 138-145.
Hansen, K. 1962. The dystrophic lake type. Hydrobiologia. 19(2)182-191.
Harding, S.T. 1942. Lakes. Hydrology, pp. 220-243. Edited by
O.E. Meinzer. New York: McGraw-Hill. Dover edition by Dover
Publications: New York. 712 pp.
Hasler, A.D. 1947. Eutrophication of lakes by domestic drainage.
Ecology. 28:383-395.
Hasler, A.D. and B. Ingersoll, 1968. Dwindling lakes. Natural History.
77(9):8-19.
R-2
-------
Hooper, F.F. 1969. Eutrophication indices and their relation to other
indices of ecosystem change. In: Eutrophication: Causes, Conse-
quences, Correctives. Proceedings of a Symposium. 11-15 June
1967. University of Wisconsin. Washington, D.C.: National
Academy of Science, pp. 225-235.
Horowitz, H.M., R.F. Nalepka, P.D. Hyde, and J.P. Morgenstern. 1971.
Estimating the proportions of objects within a single resolution
element of a multispectral scanner. Proceedings 7th International
Symposium on Remote Sensing. May 17-21, 1971. Ann Arbor, Michigan,
2:1307-1320.
Hotelling, H. 1933a. Analysis of a complex of statistical variables
into principal components (I. Introduction). The Journal of
Education Psychology. 2*1:417-^1 .
. 1933b. Analysis of a complex of statistical variables
into principal components (II). The Journal of Education Psychology.
24:498-520.
. 1936. Simplified calculation of principal components.
Psychometrika. 1(1):27-35.
Hulburt, E.O. 19^5. Optics of distilled and natural water. Journal
Optical Society America. 35(11) :698-705.
Hulstrom, R.L. 1973- The cloud bright spot. Photogrammetric Engineering.
39(4):370-376.
Hutchinson, G.E. 1957- A Treatise on Limnology. Volume I. Geography,
Physics, and Chemistry. New York: John Wiley. 1015 pp.
. 1967. A Treatise on Limnology. Volume II. Introduction
to Lake Biology and the Liranoplankton. New York: John Wiley. 115 pp.
. 1973- Eutrophication. The scientific background of
a contemporary practical problem. American Scientist. 61:269-279.
James, H.R. and E.A. Birge. 1938. A laboratory study of the absorption
of light by lake waters. Transactions Wisconsin Academy Science
Arts Letters. 31 :1-15M.
Jarnefelt, H. 1958. On the typology of northern lakes. Proceedings
International Association Theoretical Applied Limnology (International
Vereinigung fur Theoretische und Angewandte Limnologie). Volume
XIII. Congress in Finland 1956. Edited by T.T. Macan. Stuttgart.
pp. 228-235.
Jerlov, N.G. 1968. Optical Oceanography. Elsevier Oceanography Series.
Volume 5. Amsterdam: Elsevier. 19^ pp.
Juday, C. and E.A. Birge. 1933. The transparency, the color, and
the specific conductance of the lake water of northwest Wisconsin.
Transaction Wisconsin Academy Science Arts Letters. 28:205-259.
R-3
-------
Larkin, P.A and T.G. Northcote. 1958. Factors in lake typology in
British Columbia, Canada. Proceedings International Association
Theoretical Applied Limnology (Verh. Internat. Ver. Limnol.).
13:252-262.
Lindeman, R.L. 1942. The trophic-dynamic aspect of ecology. Ecology.
23(4):399-4l8.
Lueschow, L.A., J.M. Helm, D.R. Winter, and G.W. Karl. 1970. Trophic
nature of selected Wisconsin lakes. Wisconsin Academy Sciences
Arts Letters. 58:237-264.
Margalef, R. 1958. "Trophic" typology versus biotic typology as
exemplified in the regional limnology of northern Spain. Pro-
ceedings International Association Theoretical Applied Limnology.
13:339-3^9.
McCloy, K.R. 1977. The vector classifier. Proceedings 11th International
Symposium of Remote Sensing of Environment. April 25-29, 1977.
Ann Arbor, Michigan: Environmental Research Institute of Michigan.
1:535-543-
McCluney, W.R. 1976. Remote measurement of water color. Remote Sensing
of Environment. 5:3-33-
Morrison, D.F. 1967. Multivariate Statistical Methods. New York:
McGraw-Hill. 338 pp.
Moyle, J.B. 1945. Classification of lake waters upon the basis of
hardness. Proceedings Minnesota Academy Science. 13:8-12.
. 1946. Some indices of lake productivity.
Transactions of the American Fisheries Society. 76:322-334.
National Aeronautics and Space Administration. 1972. NASA Earth Resources
Technology Satellite Data Users Handbook. Goddard Space Flight
Center. Document Number 71SD4249. Subject to revision.
Naumann, E. 1919. Nagra Synpunkte angaende planktons okologi. Med. sarskild
hynsyn till fytoplanton. Svensk Botanisk Tidskrift. 13:129-158.
. 1931. Limnologische terminologie . Urban and Schwarzenberg,
Berline: Wien. (pp. 153 and 413). 776 pp.
Nordberg, W. 1972. Written communication dated 9 November 1972. NASA
Goddard Space Flight Center. Greenbelt, Maryland. 3 pp.
Pennak, R.W. 1958. Regional lake typology in northern Colorado, U.S.A.
Proceedings International Association Theoretical Applied Limnology.
13:262-283-
Piech, K.R. and J.E. Walker. 1971. Aerial color analysis of water
quality. Journal Survey Mapping Division. Proceedings American
Society Civil Engineers. SU2(8507):185-197.
R-U
-------
Raman, C.V. 1922. On the molecular scattering of light in water and the
color of the sea. Proceedings Royal Society (London). Ser. A. 101:
Rawson, D.S. 1956. Algal indicator of trophic lake types. Limnology
and Oceanography. 1(1):18-25.
. 1960. A liranological comparison of twelve large lakes
in northern Saskatchewan. Limnology and Oceanography. 5(2):195-211.
Rogers, R.H. and K. Peacock. 1973- A technique for correcting ERTS data
for solar and atmospheric effects. Symnposium on Significant
Results Obtained from the Earth Resources Technology Satellite-1.
March 5-9, 1973. NASA (Goddard Space Flight Center). 1-8:1115-1122.
Rogers, R.H., N.J. Shah, J.B. McKeon, and V.E. Smith. 1976. Computer
mapping of water quality in Saginaw Bay with Landsat digital data.
Proceedings of ACSM-ASP convention. February 22-28, 1976. Washington
Hilton. NASA-CR-146057. NTIS E76-10124.
Round, F.E. 1958. Algal aspects of lake typology. Proceedings Interna-
tional Association Theoretical Applied Limnology. 13:306-310.
Ruttner, F. 1963- Fundamentals of Limnology. Third edition. Translated
by D.G. Frey and F.E.J. Fry. Toronto: University of Toronto Press.
295 PP.
Sawyer, W.R. 1931. The spectral absorption of light by pure water and
Bay of Fundy water. Contribution of Canadian Biology and Fisheries
being studied from the Biolgoical Stations of Canada. Series D.
Hydrographic. 7(8):73-89.
Shannon, E.E. 1970. Eutrophication-trophic state relationships in north
and central Florida lakes. Doctoral dissertation. Gainesville:
University of Florida. 257 pp.
Scherz, J.P., D.R. Graff, and W.C. Boyle. 1960. Photographic characteristics
of water pollution. Photogrammetric Engineering. 35:38-43.
Sheldon, A. 1972. A quantitative approach to the classification of inland
waters. In: Natural Environments. Studies in Theoretical and
Applied Analysis. Edited by J.V. Krutilla. Baltimore: John Hopkins
University Press, pp. 205-261.
Stewart, K.M. and G.A. Rohlich. 1967. Eutrophication—A review. Publica-
tion Number 35. California State Water Quality Control Board.
Sacramento. 188 pp.
Taylor, W.D., L.R. Williams, S.C. Hern, and V.W. Lambou. (In preparation).
Phytoplankton Water Quality Relationships in U.S. Lakes, Part VII:
Comparison of Some New and Old Indices and Measurements of Trophic
State. National Eutrophication Survey Working Paper Number 711.
R-5
-------
U.S. Environmental Protection Agency. 1974. National Eutrophication
Survey Methods for Lakes. Sampled in 1972. National Eutrophication
Survey Working Paper Number 1. 40 pp.
U.S. Environmental Protection Agency. 1975. National Eutrophication
Survey Methods 1973-1975. National Eutrophication Survey Working
Paper Number 175. 91 pp.
Veatch, J.O. and C.R. Humphrys. 1966. Water and Water Use Terminology.
Kaukauna, Wisconsin: Thomas Printing and Publishing. 375 pp.
Vollenweider, R.A. 1968. Scientific Fundamentals of the Eutrophication
of Lakes and Flowing Waters with Particular Reference to Nitrogen
and Phosphorus as Factors in Eutrophication. Technical report pre-
pared for the Organization for Economic Cooperation and Development.
Paris, France. 159 pp.
Weber, C.A. 1907- Aufbau and Vegetation der Moore Norddeutschlands.
Beiblatter Botanische Jahrbucker rur Systematik. Pflanzengeschichte
und Pflanzengeographie. 90:19-34. Supplement to Bot. Jahrb. 40.
Welch, P.S. 1952. Limnology. Second edition. New York: McGraw-Hill.
538 pp.
Wezernak, C.T. and F.C. Polcyn. 1972. Eutrophication assessment using
remote sensing techniques. Proceedings of the Eighth International
Symposium on Remote Sensing of Environment. 2-6 October 1972.
Ann Arbor: University of Michigan. 1:541-551.
Wezernak, C.T., F.J. Tannis, and C.A. Bajza. 1975. Trophic Analysis of
Inland Lakes. Ann Arbor: Environmental Research Institute of
Michigan. 27 pp.
Whipple, G.C. 1898. Classifications of lakes according to temperature.
America Naturalist. 32(373):25-33•
Winner, R.W. 1972. An evaluation of certain indices of eutrophy and
maturity in lakes. Hydrobiologia. 40(2)223-245.
Work, E.A. and D.S. Gilmer. 1976. Utilization of satellite data for
inventorying prairie ponds and lakes. Photogrammetric Engineering
and Remote Sensing. 42(5):685-694.
larger, H.L. and J.R. McCauley. 1975. Quantitative water quality
with Landsat and Skylab. Proceedings NASA Earth Resources Survey
Symposium. First Comprehensive Symposium on the Practical Appli-
cation of Earth Resources Survey Data. June 9-12, 1975. Shamrock
Hilton. Houston, Texas. NASA TM X-58168. JSC-09930. I-A:347-370.
Zafar, A.R. 1959. Taxonomy of lakes. Hydrobiologia. 13(3)=287-299
Zumberge, J.H. 1952. The Lakes of Minnesota: Their Origin and
Classification. Minnesota Geological Survey Bulletin. Number 35.
99 pp.
R-6
-------
APPENDIX A
PHYSICAL-CHEMICAL DATA FOR
THE COLORADO LAKES STUDY
A-l
-------
STORET RETRIEVAL DATE 76/01/09
080101
39 57 56.0 105 29 00.0
BARKER RESERVOIR
08013 COLORADO
>
DATE
FROM
TO
75/05/07
75/08/26
75/10/10
TIME
OF
DAY
09 00
09 00
09 00
09 00
09 00
11 15
11 15
11 15
11 15
11 15
U 15
10 15
10 15
10 15
10 15
10 15
DEPTH
FEET
0000
0005
0015
0048
0081
0000
0005
0018
0040
0080
0120
0000
0005
0015
0035
0061
00010
WATER
TEMP
CENT
3.9
3.7
3.8
3.8
3.9
15.6
15.7
15.6
9.7
7.4
7.0
9.8
9.8
9.8
9.7
9.5
00300
DO
MG/L
9.6
9.4
9.4
9.4
9,4
7.6
7.4
7.6
7.0
6.4
5.6
7.8
7.4
7.4
7.6
7.0
00077
TRANSP
SECCHI
INCHES
51
96
120
00094
CNDUCTVY
FIELD
MICROMHO
26
38
38
37
39
29
32
29
29
29
27
27
28
28
27
28
IIEPALES
3
00400 00410
PH T ALK
CAC03
SU _ MG/L
8.80 26
8.80 26
8.70 25
9.00 23
8.70 24
8.35
8.20
8.00
7.40
7.00
6.90
8.25
8.10
8.00
B.SO
8.05
2111202
0085
00610
NH3-N
TOTAL
MG/L
0.020K
0.020K
0.020
0.020
0.020
FEET DEPTH
00625
TOT KJEL
N
MG/L
0.400
0.600
0.400
0.400
0.400
00630
N02&N03
N-TOTAL
MG/L
0.060
0.080
0.080
0.080
0.090
00671
PHOS-OIS
ORTHO
MG/L P
0.012
0.009
0.007
0.006
0.005
STOHET RETRIEVAL DATE 76/01/09
060101
39 57 56.0 105 29 00.0
BARKER RESERVOIR
08013 COLORADO
00665
DATE TIME DEPTH PHOS-TOT
FROM OF
TO DAY FEET MG/L P
75/05/07 09 00 0000 0.037
09 00 0005 0.037
09 00 0015 0.035
09 00 0048 0.037
09 00 0081 0.038
75/08/26 11 15 0000
75/10/10 10 15 0000
32217 00031
CHLRPHYL INCOT LT
A REMNING
UG/L PERCENT
0.6
3.7
6.2
IIEPALES
3
2111202
0085 FEET
DEPTH
-------
STORET RETRIEVAL DATE 76/01/09
080102
39 57 Sl.O 105 29 ".8.0
BARKER RESERVOIR
08001 CQLOHAOO
DATE
FROM
TO
75/05/07
75/08/26
75/10/10
TIME DEPTH
OF
DAY FEET
09 30 0000
09 30 0005
09 30 0017
11 40 0000
11 40 0005
11 40 0015
11 40 00?6
10 40 0000
10 40 0005
10 40 0012
00010
WATER
TEMP
CENT
3.7
3.7
3.7
15.4
15.4
15.1
13.1
9.6
9.6
9.5
00300 00077 00094
DO TRANSP CNDUCTVY
SECCHI FIELD
MG/L INCHES MICROMHO
10.8
9.6
9.2
7.0
7.6
7.8
7.0
7.2
7.4
7.4
48
75
96
39
38
38
32
32
32
28
28
28
11EPALES 2111202
3
00400
PH
su
8.50
8. do
8.40
9.10
8.80
8.40
8.00
8.50
8.15
7.90
0021
00410 00610
T AUK NH3-N
CAC03 TOTAL
MG/L MG/L
24 0.030
23 0.020
24 0.020K
FEET DEPTH
00625
TOT KJEL
N
MG/L
0.400
0.500
0.400
00630
N02&.N03
N-TOTAL
MG/L
0.090
o.oao
O.ObO
00671
PHOS-DIS
ORTHO
MG/L P
0.006
0.005
0.006
-O
STOHET RETRIEVAL DATE 76/01/09
080102
39 57 51.0 105 29 48.0
BARKER RESERVOIR
08001 COLORADO
11EPALES
3
2111202
0021 FEET
DEPTH
DATE
FROM
TO
75/05/07
75/08/26
75/10/10
TIME DEPTH
OF
DAY FEET
09 30 0000
09 30 0005
09 30 0017
11 40 0000
10 40 0000
00665
PHOS-TOT
MG/L P
0.034
0.040
0.035
32217
CHLRPHYL
A
UG/L
9.0
3.7
6.8
00031
INCDT LT
REMNING
PERCENT
-------
STORET RETRIEVAL DATE 76/01/09
080301
39 56 37.0 104 44 53.0
BARR LAKE
08001 COLORADO
DATE
FROM
TO
75/05/05
75/08/36
75/10/10
TIME DEPTH
OF
DAY FEET
14 80 0000
14 20 0005
14 20 0015
14 20 C026
13 35 0000
13 35 0005
13 35 0015
11 30 0000
11 30 0005
11 30 0015
00010
WATER
TEMP
CENT
13.7
12.8
13.9
13.0
2^.4
22.3
21.6
13.7
13.3
00300 00077 00094
DO TRANSP CNDUCTVY
SECCHI FIELD
MG/L INCHES MICROMHO
8.4
8.4
8.8
7.2
16.6
7.6
4.8
8.6
8.0
8.3
130
13
34
753
749
749
731
608
591
591
557
554
551
11EPALES
3
00400
PH
SU
8.15
a. is
8.20
8.30
9.40
9.30
9.00
8.80
8.90
9.00
00410
T AUK
CAC03
MG/L
350
236
333
334
3111303
0030
00610
NH3-N
TOTAL
MG/L
6.460
7.310
6.510
7.160
FEET DEPTH
00635
TOT KJEL
N
MG/L
6.500
7.000
6.900
7.300
00630
N03&N03
N-TOTAL
MG/L
1.510
1.430
1.470
1.470
00671
PHOS-DIS
ORTHO
MG/L P
1.170
1.350
1.680
1.730
STORET RETRIEVAL DATE 76/01/09
080201
39 56 37.0 104 44 53.0
BARR LAKE
08001 COLORADO
00665
DATE TIME DEPTH PHOS-TOT
FROM OF
TO DAY FEET MG/L P
75/05/05 14 20 0000 1.780
14 20 0005 1.750
14 20 0015 1.740
14 20 0026 1.770
75/08/26 13 35 0000
75/10/10 11 30 0000
33317 00031
CHLRPHYL INCDT LT
A REMNING
UG/L PERCENT
4.3
74.4
33.5
HEPALES
3
2111302
0030 FEET DEPTH
-------
STORE! RETRIEVAL DATE 76/01/09
080203
39 56 27.0 104 45 48.0
BARR LAKE
08123 COLORADO
DATE
FROM
TO
75/05/05
75/08/26
75/10/10
TIME DEPTH
OF
DAY FEET
14 55 0000
14 55 0005
14 55 0012
13 40 0000
13 40 0005
13 40 0010
11 15 0000
11 15 0005
11 IS 0016
00010
MATER
TEMP
CENT
11.7
11.8
11.7
22.5
22.0
21.5
1^.0
13.5
13.1
00300
00
MG/L
9.2
9.0
9.2
9.6
7.2
7.4
9.0
8.0
7.2
00077 00094
TRANSP CNOUCTVY
SECCHI FIELD
INCHES MICROMHO
72
24
40
723
724
723
598
591
590
561
556
560
UEPALES
3
00400
PH
Su
e.40
8.35
8.35
9.25
9.15
9.10
9.00
9.00
9.00
09410
T ALK
CAC03
MG/L
234
228
186
2111202
0016
00610
NH3-N
TOTAL
MG/L
6.180
6.590
1.390
FEET DEPTH
00625
TOT KJEL
N
MG/L
8,300
7.900
7.700
00630
N02I.N03
N-TOTAL
MG/L
1.550
1.450
2.140
00671
PHOS-01S
ORTHO
MG/L P
1.660
1.490
1.640
STORET RETRIEVAL DATE 76/01/09
080202
39 56 27.0 104 45 48.0
BARR LAKE
08123 COLORADO
UEPALES
3
2111202
0016 FEET DEPTH
DATE
FROM
TO
75/05/05
75/08/26
75/10/10
TIME DEPTH
OF
DAY FEET
14 55 0000
14 55 0005
14 55 0012
13 40 0000
11 15 0000
00665
PHOS-TOT
MG/L P
1.750
1.810
1.730
32217
CHLRPHYL
A
UG/L
12.8
29.0
19.6
00031
INCDT LT
REMNING
PERCENT
-------
STORE! RETRIEVAL DATE 76/01/09
080401
39 38 58.0 104 51 13.0
CHERRY CHEEK LAKE
08005 COLORADO
>
—i
DATE TIME DEPTH
FROM OF
TO DAY FEET
75/05/07 10 45 0000
10 45 0006
75/08/22 14 25 0000
14 25 0005
14 25 0018
75/10/09 10 55 0000
10 55 0005
10 55 0015
00010
WATER
TEMP
CENT
9.7
9.7
22.9
23.1
22.6
13.9
13.5
13.1
00300 00077 00094
DO TRANSP CNDUCTVY
SECCHI FIELD
MG/L INCHES MICROMHO
8.2
8.0
7.2
6.4
6.0
6.0
8.0
7.6
33
36
24
377
378
574
770
567
422
422
418
HEP ALES
3
00400
PH
SU
8.50
8.50
8.00
8.30
8.35
8.15
8.20
8.25
00410
T ALK
PAC03
MG/L
175
2111202
0010
00610
NH3-N
TOTAL
MG/L
0.030
FEET DEPTH
00625 00630
TOT KJEL N02&N03
N N-TOTAL
MG/L MG/L
0.900 0.030
00671
PHOS-D1S
ORTHO
MG/L P
0.009
STORE! RETRIEVAL DATE 76/01/09
080401
39 38 58.0 104 51 13.0
CHERRY CREEK LAKE
08005 COLORADO
00665
DATE TIME DEPTH PHOS-TOT
FROM OF
TO DAY FEET MG/L P
75/05/07 10 45 0000
75/08/22 14 ?5 0000
75/10/09 10 55 0000
0.057
32217 00031
CHLRPHYL INCDT LT
A REMNING
UG/L PERCENT
12.8
9.8
14.2
11EPALES
3
2111202
0010 FEET
DEPTH
-------
STORET RETRIEVAL DATE 76/oi/o«
080402
39 38 18.0 104 51 57.0
CHERRY CREEK LAKE
06005 COLORADO
00010
DATE TIME DEPTH WATER
FROM OF TEMP
TO DAY FEET CENT
75/05/07 11 05 0000 10.2
11 05 0009 9.9
75/08/22 14 00 0000 24.3
14 00 0005 23.4
75/10/09 11 15 0000 13.9
11 15 0005 13.9
11 15 C010 13.5
00300 00077 00094
DO TRANSP CNDUCTVY
SECCHI FIELD
MG/L INCHES M1CROMHO
8.2
7.6
7.6
5.0
8.0
7.6
7.4
37
26
24
380
378
607
564
425
425
422
HEP ALES
3
00400
PH
SU
8.50
B.50
8.50
8.20
8.25
8.25
00410
T ALK
CAC03
MG/L
175
2111202
0013
00610
NH3-N
TOTAL
MG/L
0.040
FEET DEPTH
00625
TOT KJEL
N
MG/L
0.800
00630
N02&N03
N-TOTAL
MG/L
0.020K
00671
PHOS-OIS
ORTHO
MG/L P
0.009
STORET RETRIEVAL DATE 76/01/09
080402
39 38 18.0 104 51 57.0
CHERRY CREEK LAKE
08005 COLORADO
00665
DATE TIME DEPTH PHOS-TOT
FROM OF
TO DAY FEET MG/L P
75/05/07 11 05 0000
11 05 0009
75/08/22 14 00 0000
75/10/09 11 15 0000
0.056
32317 00031
CHLRPHYL INCDT LT
A REMNING
UG/L PERCENT
13.2
124.6
9.8
11EPALES
3
2111202
0013 FEET
DEPTH
-------
STORET RETRIEVAL DATE 76/01/09
080403
39 33 12.0 104 SO 55.0
CHER«Y CREEK LAKE
08089 COLORADO
00010
DATE TIME DEPTH WATER
FROM OF TEMP
TO DAY FEET CENT
10.3
10.a
10.0
23.4
23.S
22.6
13.7
13.6
13.4
75/05/07
75/08/22
75/10/09
11
11
11
14
14
14
10
10
10
30
30
30
30
30
30
40
40
40
0000
0005
0015
0000
0005
0014
0000
0005
0010
11EPALES
3
00300
00
MG/L
a.o
8.0
8.0
7. a
7.2
5.8
7.6
7.4
7.6
00077
TRANSP
SECCHI
INCHES
36
24
36
00094
CNDUCTVr
FIELD
MICROMHO
382
381
382
579
570
565
426
421
418
00400
PH
SU
8.70
8.50
8.50
8.50
8.50
8.40
8.10
8.25
8.30
60410
T ALK
,CAC03
MG/L
176
176
176
2111202
0019 FEET DEPTH
00610
NH3-N
TOTAL
MG/L
0.030
0.040
0.030
00625
TOT KJEL
N
MG/L
0.800
0.800
0.900
00630
N02tN03
N-TOTAL
MG/L
0.020K
0.020K
0.020K
00671
PHOS-OIS
ORTrtO
MG/L P
0.009
0.010
0.011
STORET RETRIEVAL DATE 76/01/09
080403
39 38 12.0 104 50 55.0
CHERRY CREEK LAKE
08089 COLORADO
DATE
FROM
TO
75/05/07
75/08/22
75/10/09
TIME DEPTH
OF
DAY FEET
11 30 0000
11 30 0005
11 30 0015
14 30 0000
10 40 0000
00665
PHOS-TOT
MG/L P
0.051
0.070
0.065
32217
CHLRPHYL
A
UG/L
2.3
11.6
11.6
00031
INCOT LT
REMNING
PERCENT
11EPALES
3
2111202
0019 FEET
DEPTH
-------
STORET RETRIEVAL DATE 76/01/09
080601
39 36 30.0 106 01 22.0
DILLON RESERVOIR
08117 COLORADO
I
I—'
o
DATE TIME DEPTH
FROM OF
TO DAY FEET
75/08/25 09 20 0000
09 ?0 0005
09 30 0030
09 20 0045
09 20 0071
75/10/09 14 50 0000
14 50 0005
14 50 0023
14 50 0050
14 50 0075
00010
WATER
TEMP
CENT
12.6
12.4
10.1
7.5
6.1
11.2
11.1
11.0
10.4
9.5
00300 00077 00094
DO TRANSP CNDUCTVY
SECCHI FIELD
MG/L INCHES M1CROMHO
7.4
7.2
7.0
6.8
6.6
7.8
8.0
8.2
7.6
6.6
180
252
97
94
82
85
87
1
1
1
1
11EPALES
3
OQ400 00410
PH T ALK
CAC03
SU MG/L
7.60
7.70
7.40
7.20
7.20
7.40
7.60
7.6Q
7.50
7.40
2111202
0075
00610
NH3-N
TOTAL
MG/L
FEET DEI
00625
TOT KJEL
N
MG/L
00630 00671
N084N03 PHOS-OIS
N-TOTAL ORTriO
MG/L MG/L P
STORET RETRIEVAL DATE 76/01/09
080601
39 36 30.0 106 01 22.0
DILLON RESERVOIR
08117 COLORADO
00665
DATE TIME DEPTH PHOS-TOT
FROM OF
TO DAY FEET MG/L P
75/08/25 09 20 0000
75/10/09 14 50 0000
32217 00031
CHLRPHYL INCDT LT
A REMNING
UG/L PERCENT
2.2
3.7
11EPALES
3
2111202
0075 FEET
DEPTH
-------
STORET RETRIEVAL DATE 76/01/09
080602
39 35 00.D 106 03 00.0
DILLON RESERVOIR
08117 COLORADO
DATE TIME DEPTH
FROM OF
TO DAY FEET
75/OB/25 09 45 0000
09 45 0005
09 45 0020
09 45 0038
09 45 0043
75/10/09 U 00 0000
U 00 Q005
14 00 0023
14 00 0044
00010
WATER
TEMP
CENT
13.4
13.2
17.9
6.0
7.7
11.2
11.2
10.6
10.3
00300 OOOtf 00094
bO TRANSP tNDUCTVY
SECCHt flELO
MG/L INCHES MICROMHO
7.6
7.5
7.6
6.0
6.0
8.2
e.o
8.0
8.0
204
206
97
97
95
87
87
97
94
94
94
UEPALES
3
00400 00410
PH T ALK
CAC03
SU Mti/L
T.55
T.T5
7.90
T.55
T.35
7.60
7.75
7.60
7.80
21ii2d2
0047 FEET DEPTH
00610 00625 00630
NH3-N TOT KJEL N02&N03
TOTAL N N-TOTAL
MG/L MG/L MG/L
006tt
PHOS-D1S
ORTHO
MG/L P
STORET RETRIEVAL DATE 76/01/09
080602'
39 35 00.0 106 03 00.0
DILLON RESERVOIR
08117 COLORADO
00665
DATE TIME DEPTH PHOS-TOT
FttOM OF
TO OAY FEET MG/L P
75/08/25 09 45 0000
75/10/09 14 00 0000
322171
CHLRPHYL INCDr LT
A RtMNING
UG/L PERCENT
2.5
4.9
11EPALES
3
2111202
0047 FEET
DEPTH
-------
STORET RETRIEVAL DATE 76/01/09
090603
39 35 00.0 106 04 OT.O
DILLON RESERVOIR
08117 COLORADO
I
I—'
K3
DATE TIME DEPTH
FROM OF
TO DAY FEET
75/08/25 10 15 0000
10 15 0005
10 15 0020
10 15 0040
75/10/09 14 30 0000
14 30 0005
14 30 0023
14 30 0055
00010
MATER
TEMP
CENT
13.0
13.0
13.4
8.7
11.0
10.8
10.8
10.8
00300 00077 00094
DO TRANSP CNOUCTVY
SECCHI FIELD
MG/L INCHES MICROMHO
7.2
7.2
7.0
6.2
7.8
7.8
7.8
8.0
600
252
96
96
94
89
1
1
1
1
11EPALES
3
00400 00410
PH T ALK
CAC03
SU MG/L
8.35
8.15
7.90
7.60
7.80
7.80
7.70
7.70
2111303
0044 FEET DEPTH
00610 00625 00630
NH3-N TOT KJEL N02iN03
TOTAL N N-TOTAL
MG/L MG/L MG/L
00671
PHOS-DIS
ORTHO
MG/L P
STORET RETRIEVAL DATE 76/01/09
030603
39 35 00.0 106 04 07.0
DILLON RESERVOIR
08117 COLORADO
00665
DATE TIME DEPTH PHOS-TOT
FROM OF
TO DAY FEET MG/L P
75/08/25 10 15 0000
75/10/09 14 30 0000
32217 00031
CHLRPHYL INCDT LT
A REMNING
UG/L PERCENT
2.1
2.8
11EPALES
3
2111202
0044 FEET
DEPTH
-------
STORE! RETRIEVAL DATE 76/01/09
080604
39 36 22.0 106 03 45.0
DILLON RESERVOIR
16043 COLORADO
I
H1
OJ
DATE
FROM
TO
75/08/25
75/10/09
TIME DEPTH
OF
DAY FEET
10 45
10 45
10 45
10 45
10 45
13 15
13 15
13 15
13 15
13 15
13 15
13 15
0000
0005
0037
0060
0105
0000
0005
0023
0045
0075
oiao
0165
00010
WATER
TEMP
CENT
13.0
13.1
10.4
6.4
4.2
11.0
11.0
10.8
10.8
9.0
7.1
6.5
00300 00077 00094
DO TRANSP CNOUCTVY
SECCHI FIELD
HG/L INCHES MICROMHO
7.6
7.3
6.6
6.8
6.9
7.8
7.6
8.0
7.8
5.8
6.2
6.6
252
600
96
93
84
82
88
HEP ALES
3
00400 00410
PH T ALK
CAC03
SU HG/L
7.80
7.70
7.40
7.20
7.00
7.80
7.B5
7.85
7.80
7.60
7.50
7.60
2111202
0109 FEET DEPTH
00610 00625 00630 00671
NH3-N TOT KJEL N02iN03 PHOS-OIS
TOTAL N N-TOTAL ORTMO
MG/L MG/L M(j/L MG/L P
STORET RETRIEVAL DATE 76/01/09
080604
39 36 22.0 106 03 45.0
DILLON RESERVOIR
16043 COLORADO
00665
DATE TIME DEPTH PHOS-TOT
FROM OF
TO DAY FEET MG/L P
75/08/25 10 45 0000
75/10/09 13 15 0000
32217 00031
CHLRPHYL INCDT LT
A REMNING
UG/L PERCENT
2.4
4.6
11EPALES
3
2111202
0109 FEET DEPTH
-------
STORET RETRIEVAL DATE 76/01/09
080801
39 52 34.0 106 18 57.0
GREEN MOUNTAIN RESERVOIR
08117 COLORADO
I
h-'
4>
DATE
FROM
TO
75/08/25
75/10/09
TIME
OF
DAY
13 30
13 30
13 30
13 30
13 30
16 50
16 50
16 50
16 50
16 50
16 50
DEPTH
FEET
0000
0005
0022
0040
0066
0000
0005
0018
0045
0090
0129
00010
WATER
TEMP
CENT
13.5
13.8
13.5
12.6
11.0
13.0
12.8
12.5
12.5
12.4
12.5
00300 00077 00094
00 TRANSP CNOUCTVY
SECCHI FIELD
MG/L INCHES MICROMHO
7.2
7.2
7.0
6.8
5.9
7.2
7.2
7.2
7.2
7.4
7.0
96
156
114
110
109
108
103
101
106
1
1
1
1
11EPALES
3
00400 00410
PH T ALK
CAC03
SU MG/L
7.60
8.10
7.90
7.85
7.60
7.75
7.80
7.80
7.60
7.75
7.70
2111203
0070 FEET DEI
00610 00625
NH3-N TOT KJEL
TOTAL N
MG/L MG/L
00630 00-671
N024N03 PMOS-DIS
N-TOTAL ORTHO
MG/L MG/L P
STORET RETRIEVAL DATE 76/01/09
080801
39 52 34.0 106 IB 57.0
GREEN MOUNTAIN RESERVOIR
08117 COLORADO
00665
DATE TIME DEPTH PHOS-TOT
FROM OF
TO DAY FEET MG/L P
75/08/25 13 30 0000
75/10/09 16 50 0000
32217 00031
CMLRPHYL INCDT LT
A REMNING
UG/L PERCENT
7.9
3.1
11EPALES
3
2111202
0070 FEET
DEPTH
-------
STORET RETRIEVAL DATE 76/01/09
080803
39 52 15.0 106 16 57.0
GREEN MOUNTAIN RESERVOIR
08117 COLORADO
DATE
FROM
TO
75/08/25
75/10/09
TIME DEPTH
OF
DAY FEET
14 00 0000
14 00 0005
14 00 0021
14 00 0055
14 00 0091
16 30 0000
16 30 0005
16 30 0016
16 30 00
-------
STOHET RETRIEVAL DATE 76/01/09
080803
39 50 07.0 106 14 30.0
GREEN MOUNTAIN RESERVOIR
08117 COLORADO
I
M
C^
DATE
FROM
TO
75/08/25
75/10/09
TIME DEPTH
OF
DAY FEET
14 30 0000
14 30 0005
14 30 0018
14 30 0034
16 00 0000
16 00 0005
16 00 0015
16 00 0023
00010
WATER
TEMP
CENT
14.6
14.7
14.0
13.2
12.7
12.8
10.0
6.8
00300 00077 00094
DO TRANSP CNDUCTVY
SECCHI FIELD
MG/L INCHES MICROMHO
7.2
7.3
7.2
7.2
7.8
8.0
8.2
9.6
75
108
109
108
108
109
109
1
1
1
11EPALES
3
00400 00410
PH T ALK
CAC03
SU MG/L
7.70
8.05
8.00
8.10
7.55
8.00
8.00
8.00
2111203
0027 FEET DEPTH
00610 00625 00630
NH3-N TOT KJEL N02&N03
TOTAL N N-TOTAL
MG/L MG/L MG/L
00671
PHOS-DIS
ORTHO
MG/L P
STORET RETRIEVAL DATE 76/01/09
080803
39 50 07.0 106 14 30.0
GrtEEN MOUNTAIN RESERVOIR
08117 COLORADO
00665
DATE TIME DEPTH PHOS-TOT
FROM OF
TO DAY FEET MG/L P
75/08/25 14 30 0000
75/10/09 16 00 0000
32217 00031
CHLHPHYL INCDT LT
A REMNING
UG/L PERCENT
9.8
3.1
11EPALES
3
2111202
0027 FEET
DEPTH
-------
STORET RETRIEVAL DATE 76/01/09
081001
38 13 12.0 103 40 13.0
LAKE MEREDITH
08025 COLORADO
00010
DATE TIME DEPTH WATER
FROM OF TEMP
TO DAY FEET CENT
75/05/06 11 00 0000 12.8
75/08/22 11 30 0000 23.8
75/10/07 14 50 0000 17.7
00300 00077 00094
DO TRANSP CNDUCTVY
SECCHI FIELD
MG/L INCHES MICROHHO
7.6
5.4
8.0
11
10
10
3949
7096
7000
11EPALES
3
00400
PH
SU
8.50
9.00
9.10
00410
T ALK
CAC03
MG/L
ice
2111202
0005 FEET DEPTH
00610
NH3-N
TOTAL
MG/L
0.320
00625
TOT KJEL
N
MG/L
3.100
00630
N02tN03
N-TOTAL
MG/L
0.060
00671
PHOS-D1S
ORTHO
MG/L P
0.099
I
M
-^J
STORET RETRIEVAL DATE 76/01/09
081001
38 13 12.0 103 40
LAKE MEREDITH
08025 COLORADO
13.0
00665
DATE TIME DEPTH PHOS-TOT
FROM OF
TO DAY FEET MG/L P
75/05/06 11 00 0000
75/08/22 11 30 0000
75/10/07 14 50 0000
0.326
32217 00031
CHLRPHYL INCDT LT
A REMNING
UG/L PERCENT
46.2
151.3
278.2
11EPALES
3
2111202
0005 FEET
DEPTH
-------
STORET RETRIEVAL DATE 76/01/09
081002
38 12 13.0 103 41
LAKE MEREDITH
08025 COLORADO
14.0
DATE TIME DEPTH
FROM OF
TO DAY FEET
75/05/06 11 40 0000
75/08/22 11 15 0000
75/10/07 15 10 0000
00010
WATER
TEMP
CENT
13.6
23.6
17.9
00300 00077 00094
DO TRANSP CNDUCTVY
SECCHI FIELD
M6/L INCHES MICROMHO
8.0
4.6
7.2
11
11
10
3987
7094
7055
11EPALES
3
00400
PH
SU
8.65
8.90
9.10
00410
T ALK
CAC03
MG/L
106
3111202
0006 FEET DEPTH
00610
NH3-N
TOTAL
MG/L
0.200
00625
TOT KJEL
N
MG/L
3.400
00630
N02&N03
N-TOTAL
MG/L
0.040
00671
PHOS-DIS
ORTHO
MG/L P
0.072
1—1
00
STORET RETRIEVAL DATE 76/01/09
081002
38 12 12.0 103 41 14.0
LAKE MEREDITH
08025 COLORADO
00665
DATE TIME DEPTH PHOS-TOT
FROM OF
TO DAY FEET MG/L P
75/05/06 11 40 0000
75/08/22 11 15 0000
75/10/07 15 10 0000
0.338
32217 00031
CHLRPHYL INCDT LT
A REMNING
UG/L PERCENT
32.8
151.3
271.3
11EPALES
3
2111202
0006 FEET
DEPTH
-------
STORE! RETRIEVAL DATE 76/01/09
081003
38 10 42.0 103 43 31.0
LAKE MEREDITH
35061 COLORADO
00010
DATE TIME DEPTH WATER
FROM OF TEMP
TO DAY FEET CENT
75/05/06 11 55 0000 11.1
75/08/22 11 00 0000 22.0
75/10/07 15 20 0000 19.0
00300 00077 00094
00 TRANSP CNOUCTVY
SECCHI FIELD
MG/L INCHES MICROMHO
7.2
5.3
8.8
11
11
8
3811
7095
7083
11EPALES
3
00400
PH
SU
8.70
8.90
9.40
00410
T ALK
CAC03
MG/L
107
2111202
0004 FEET DEPTH
00610
NH3-N
TOTAL
MG/L
0.320
00625
TOT KJEL
N
MG/L
3.200
00630
N02&N03
N-TOTAL
MG/L
0.050
00671
PHOS-OIS
ORTMO
MG/L P
0.115
I
I—>
^o
STORET RETRIEVAL DATE 76/01/09
081003
38 10 42.0 103 43 31.0
LAKE MEREDITH
35061 COLORADO
00665
DATE TIME DEPTH PHOS-TOT
FROM OF
TO DAY FEET MG/L P
75/05/06 11 55 0000
75/08/22 11 00 0000
75/10/07 15 20 0000
0.333
32217 00031
CHLRPHYL INCDT LT
A REMNING
UG/L PERCENT
63.6
138.0
349.4
11EPALES
3
2111202
0004 FEET
DEPTH
-------
STORE* RETRIEVAL DATE 76/01/09
081001
38 13 12.0 103 40 13.0
LAKE MEREDITH
08035 COLORADO
00010
DATE TIME DEPTH WATER
FROM OF TtMP
TO DAY FEET CENT
75/05/06 11 00 0000 12.8
75/08/22 11 30 0000 23.8
75/10/07 1* 50 0000 17.7
00300 00077 00094
DO TRANSP CNOUCTVY
SECCHI FIELD
MG/L INCHES MICROMHO
7.6
5.4
6.0
11
10
10
3949
7096
7000
11EPALES
3
00400
PH
SU
8.50
9.00
9.10
00410
T ALK
, CAC03
MG/L
108
3111302
0005 FEET DEPTH
00610
NH3-N
TOTAL
MG/L
0.320
00625
TOT KJEL
N
MG/L
3.100
00630
N02t>N03
N-TOTAL
MG/L
0.060
00671
PHOS-DIS
ORTHO
MG/L P
0.099
K>
O
STORET RETRIEVAL DATE 76/01/09
081001
38 13 12.0 103 40 13.0
LAKE MEREDITH
08025 COLORADO
0066S
DATE TIME DEPTH PHOS-TOT
FROM OF
TO DAY FEET MG/L P
75/05/06 11 00 0000
75/08/22 11 30 0000
75/10/07 14 50 0000
0.326
32217 00031
CHLRPHYL INCDT LT
A REMNING
UG/L PERCENT
46.2
151.3
278.2
11EPALES
3
2111202
0005 FEET
DEPTH
-------
STORET RETRIEVAL DATE 76/01/09
081102
40 13 39.0 104 38 25.0
MILTON RESERVOIR
30007 COLORADO
00010
DATE TIME DEPTH WATER
FROM OF TEMP
TO DAY FEET CENT
75/05/06 15 30 0000 13.2
15 30 0005 13.1
15 30 0011 12.6
75/08/36 10 20 0000 22.5
10 20 0004 21.0
75/10/10 09 30 0000 13.0
00300 00077 00094
DO TRANSP CNDUCTVY
SECCHI FIELD
MG/L INCHES MICROMHO
6.2
6.0
5.8
7.2
7.2
6.0
120
40
36
1042
1032
1025
1306
1258
10S7
11EPALES
3
00400
PH
SU
8.35
8.40
9.00
9.05
8.2Q
00410
T ALK
CAC03
MG/L
276
288
276
2111202
001S
00610
NH3-N
TOTAL
MG/L
1.450
1.460
1.770
FEET DEPTH
00625
TOT KJEL
N
MG/L
2.700
2.800
2.700
00630
N02&N03
N-TOTAL
MG/L
0.830
0.820
0.820
00671
PHOS-DIS
ORTHO
MG/L P
1.090
1.000
1.090
>
STORET RETRIEVAL DATE 76/01/09
081102
40 13 39.0 104 38 25.0
MILTON RESERVOIR
30007 COLORADO
11EPALES
3
2111202
0015 FEET
DEPTH
DATE
FROM
TO
75/05/06
75/08/26
75/10/10
TIME DEPTH
OF
DAY FEET
15 30 0000
15 30 0005
15 30 0011
10 20 0000
09 30 0000
00665
PHOS-TOT
MG/L P
1.160
1.170
1.160
32217
CHLRPHYL
A
UG/L
1.4
16.0
4.3
00031
INCDT LT
REMNING
PERCENT
-------
STORET RETRIEVAL DATE 76/01/09
060901
38 03 45.0 103 36 39.0
HOLBHOOK LAKE
08035 COLORADO
00010
DATE TIME DEPTH WATER
FROM OF TEMP
TO DAY FEET CENT
75/05/06 12 30 0000 13.6
75/08/23 10 35 0000 21.6
75/10/07 14 30 0000 18.3
00300 00077 00094
DO TRANSP CNDUCTVY
SECCH1 FIELD
MG/L INCHES MICROMHO
6.0
11
7
11
1650
2368
1812
HEP ALES
3
00400
PH
SU
8.10
8.60
8.90
40410
T ALK
CAC03
MG/L
147
2111202
0005 FEET DEPTH
00610
NH3-N
TOTAL
MG/L
0.050
00625
TOT KJEL
N
MG/L
1.700
00630
N02&N03
N-TOTAL
MG/L
0.020K
00671
PHOS-DIS
ORTHO
MG/L P
0.028
ho
lo
STORET RETRIEVAL DATE 76/01/09
080901
38 03 45.0 103 36 38.0
HOL8ROOK LAKE
08025 COLORADO
00665
DATE TIME DEPTH PHOS-TOT
FROM OF
TO DAY FEET MG/L P
75/05/06 12 30 0000
75/08/22 10 35 0000
75/10/07 14 30 0000
0.127
32217 00031
CHLRPHYL INCOT LT
A REMNING
UG/L PERCENT
28.7
146.9
160.2
11EPALES
3
2111202
0005 FEET
DEPTH
-------
APPENDIX B
PHOTOGRAPHIC FLIGHT LOG OF NASA
AIRCRAFT COVERAGE OF COLORADO
LAKES, AUGUST 25, 1975
-------
td
I
U)
PHOTOGRAPHIC FLIGHT LOG
pA&£ I OF 2.
MISSION
3 / 7 (SITES 18S/X3S3
DATE
CAMERA / POSITION
_EMS / SFRIAL NO.
\ 0.1(0.
FILM /EMULSION MO j - , _ ; |
FILTER/SERIAL NO
SHUTTER SPEED
1/STOP
AEC.
OVERLAP
%
ROLL NUMBER
SITE /
! START
PHOTOGRAPHER B L U N'C K
NOTES
VVX • Clee-r,
\ ROLL &,
M F TF R
u
- -- - I L!NE ! RUN
FLIGHT NO! I STOP
T ' START | START i START \ START i STARTj START , START ; START j_ START [ u
" ~ '
STOP STOP STOP
~ STO
OP , STOP STOP TIME
'- 1
10 \ I
rafm ',3j ,'Rav Dec M )
I 1 - I j - -
r T , '
1 f i 1 1
1
KEVJC'US EDITION ».' A Y OE USLD.
-------
td
I
-o-
PHOTOGRAPHIC FLIGHT LOG (CONTINUATION SHEET)
CAMERA /POSITION G'
SITE/
. _ _ _! LINE i
FLIGHT NO |
Q ~
- -
—
—
\
- -
- -
/4
o
/
s*\
/
_
^
---I
i
1 START
STOP
/T-5^1
1
START START
STOP
' ! ~s~? ~ " 1
/ .56 !
$9\
1 &O\
(o3
j 6>6 i
' 68
1 69i ^
/ 7S
' 78 i
1
[""
STOP
START
STOP
T
j
. _ _ f _ _
^ 1 _
1
1
' !
i-" - I-
r - r f
1 j
I"
p
1
START
MISSION 3 /7/X33"3 j DATE Q/3.S/7J'
1 |
START START
STOP STOP
- 4- -
1
STOP
START
STOP
START 1 START
STOP
i . J
_r
--4--T---
,_ - i . j
')
STOP
i
PAGE 2. °F 2.
TIME J'sVELCR
1949 IS
19 SO 20
19S345
/9S9&0
2.000 i o
&0&50
20091S
30 / 2.2.0
2.01400
2.0/7/0
2.QIB t£
^^
/7
/8
^5-
2.0
20
-6WS™
DRIFT
/85iu'3
.jj- .
18.7
18-6
HEAD._
a-^96.6
7.3
19.2 8-5-
/8.^7-°
/8^
L*
ou.i
178.6
3.30.5
149.3
i
GROUND"
SPEED
336
5/4
i
I
3 D C?
27&
336
COMMENTS: ,; ; f
(Rov BPC > i)
^ EDITION MAY BL USED.
-------
APPENDIX C
REGRESSION MODEL PREDICTED,
RESIDUAL AND ASSOCIATED
OBSERVED VALUES
C-l
-------
Table C-l. Regression Model Predicted, Residual and Associated Observed Values
n
i
OJ
Lake or Site Model 1
STORE! /\
Dumber C11LA CHLA
0801 3.7 4.4
080101
080102
0802 51.7 44.2
080201
080202
0803 4.9 4.8
080301
080302
080303
080304
080305
080306
0804 48.7 47.0
U80401
080402
080403
0806 2.3 4.0
080601
080602
080603
080604
0807 5.5 3.2
080701
080702
0808 8.3 12.4
080801
080802
080803
0811 12.2 11.3
081101
081102
0813
081301 6.2 4.5
081302
081303
Mean 15.9 15.1
Maximum 51.7 47.0
Minimum 2.3 3.2
Range 49.4 43.7
S.D. 19.6 17.6
N 99
Residuals Cl:
-0.7
3.
3.
7.5
74.
29.
0.1
6.
^ .
4.
4 .
5.
5.
1.7
9.
124.
11.
-1.7
2,
2.
2,
2
2.3
5
5
-4.1
7
7
9
0.9
8
16
1.7 8
6
3
0.9 13
7.5 124
-4.1 2
11.6 122
3.2 26
9 27
ILA
7
7
4
0
0
1
6
.2
2
4
.8
.6
.6
.2
.5
.1
.4
.5
.4
.9
.1
.8
.3
.0
.1
.5
.9
.9
.6
.1
.5
.2
Model 2
C1ILA
3.9
4.8
18.5
32.5
6.7
7.6
4.8
5.3
4.5
4.4
20.4
29.8
28.3
4.5
3.0
4.0
2.6
2.6
2.9
13.6
10.1
9.1
9.7
13.5
5.1
3.8
4.2
9.6
32.5
2.6
29.9
8.7
27
Residuals
-0.2
-1. 1
55.9
-3.5
-0.7
-3.5
-0.2
-1.1
0.7
1.0
-10.6
94.8
-16.7
-2.3
-0.5
-1.9
-0.2
2.9
2.5
-5.7
-3.0
0.7
-1.4
2.5
3.0
2.7
-0.3
4.2
94.8
-16.7
111.5
21.6
27
CHLA
74.4
29.0
9.8
124.6
11.6
2.2
2.5
2.1
2.4
5.5
5.4
8.3
16.0
22.6
124.6
2.1
122.5
36.4
13
Model 3
CHLA
£(9.5
72.5
5.1
37.7
20.8
4.0
4.2
3.0
4.0
6.3
2.1
6.8
13.1
17.6
72.5
2.1
70.4
22.1
13
Residuals
24.9
-43.5
4.7
86.9
-9.2
-1.8
-1.7
-0.9
-1.6
-0.8
3.3
1.5
2.9
5.0
86.9
-43.5
130.4
28.7
13
CHLA
74.
29.
9.
124.
11 .
2.
2.
2.
2.
5 (
5.
8
16
22
124
2
122
36
13
4
0
8
6
6
.2
5
.1
.4
.5
, 4
.3
.0
.6
.6
.1
.5
.4
Model 4
CHLA
32.6
49.6
11.2
57.5
15.6
1.9
2.3
3.0
1.3
6.3
10.1
8.1
22.4
17.1
57.5
1.4
56.1
18.6
13
Residuals
41.8
-20.6
-1.4
67.1
-4.0
0.3
0.2
-0.9
1.0
-0.8
-4.7
0.2
-6.4
5.5
67.1
-20.6
87.7
23.0
13
-------
Table C-l. Regression Model Predicted, Residual and Associated Observed Values (Continuation 1)
n
i
Lake or Site
STORET
Number
0801
080101
080102
0802
080201
080202
0803
080301
080302
080303
080304
080305
080306
0804
080401
080402
080403
0806
080601
080602
080603
080604
0807
080701
080702
0808
080801
080802
080803
0811
081101
081102
0813
081301
081302
081303
Mean
Maximum
Minimum
Range
S.V.
N
CHLA
74.4
29.0
9.8
124.6
11.6
2.2
2.5
2.1
2.4
5.5
5.4
8.3
16.0
22.6
124.6
2.1
122.5
36.4
13
Model 5
;*.
CHLA
33.1
29.2
16.4
75.0
17.1
2.4
4.6
2.0
1.0
7.2
6.9
13.1
12.1
16.9
75.0
1.0
74.0
20.1
13
Residuals
41.
-0.
-6.
49.
-5.
-0.
-2.
0.
1.
-1,
-1,
-4,
3
5,
49
-6
56
18
13
3
2
7
6
,5
,2
1
.1
.4
p 7
,5
,8
.9
.7
.6
.6
.3
.0
CHLA
74.4
29.0
9.8
124.6
11.6
2.2
2.5
2.1
2.4
5.5
5.4
8.3
16.0
26.6
124.6
2.1
122.5
36.4
13
Model 6
- —
CHLA
25.9
28.2
15.4
84.7
16.6
2.1
5.0
2.3
1.0
8.6
8.8
11.0
11.2
17.0
84.7
1.0
83.7
22.1
13
Residuals
48.
0.
-5,
39.
-5.
0,
-2.
-0.
1.
-3.
-3.
-2,
4,
5
48,
-5
54
17
13
.5
8
,6
.9
.0
.1
5
^ 2
,4
.1
,4
.7
.8
.6
.5
.6
.1
.4
Model 7
^^\
ISEC ISEC Residuals ISEC
0.461 0.363 0.098
0.
0.
2.187 1.612 0.575
3.
1.
0.490 0.386 0.104
0.
0,
0.
0.
0.
0.
1.373 1.676 -0.303
1 ,
1.
1 .
0.120 0.343 -0.223
0.
0.
0,
0.
0.398 0.298 0.100
0.
0.
0.471 0.708 -0.237
0,
0,
0.
0.787 0.669 0.118
0.
0.
0.562 0.368 0.194
0,
0.
0.
0.761 0.714 0.047 0.
2.187 1.676 0.575 3.
0.120 0.298 -0.303 0,
1.410 1.378 0.878 3.
0.636 0.547 0.272 0,
9 9 9 27
410
525
281
.640
.547
.787
.437
.410
.394
.525
.094
.514
.640
.219
,193
.066
.156
.410
.386
,410
.492
,525
,656
.984
,656
525
.525
.719
.281
,066
,215
,660
Model 8
^^\
ISEC
0.338
0.394
1.147
1.785
0.515
0.569
0.393
0.428
0.376
0.368
1.238
1.669
1.603
0.378
0.271
0.342
0.245
0.243
0.268
0.897
0.712
0.654
0.690
0.894
0.415
0.332
0.353
0.649
1.785
0.243
1.542
0.458
27
Residuals
0.072
0.131
2.134
-0.145
0.032
0.218
0.044
-0.018
0.018
0.157
-0.144
-0.155
0.037
-0.159
-0.078
-0.276
-0.089
0.167
0.118
-0.487
-0.220
-0.129
-0.034
0.090
0.241
0.193
0.172
0.719
3.281
0.066
3.215
0.660
27
-------
Table C-l. Regression Model Predicted, Residual and Associated Observed Values (Continuation 2)
o
Lake or Site
STORET
Number
0801
080101
080102
0802
080201
080202
0803
080301
080302
080303
080304
080305
080306
0804
080401
080402
080403
0806
080601
080602
080603
080604
0807
080701
080702
0808
080801
080802
080803
0811
081101
081102
0813
081301
081302
081303
Mean
Maximum
Minimum
Range
S.D.
N
ISEC
3.281
1.640
1.094
1.514
1.640
0.219
0.193
0.066
0.156
0.410
0.386
0.656
0.984
0.942
3.281
0.066
3.215
0.910
13
Model 9
/~^
ISEC
1.177
2.006
1.290
1.850
1.763
0.308
0.207
0.274
0.183
0.181
0.204
0.637
0.871
0.843
2.006
0.181
1.825
0.699
13
Model 10
Residuals
2.104
-0.366
-0.196
-0.336
-0.123
-0.089
-0.014
-0.208
-0.027
0.229
0.182
0.019
0.113
0.099
2.104
-0.366
2.470
0.629
13
ISEC
3.281
1.640
1.094
1.514
1.640
0.219
0.193
0.066
0.156
0.410
0.386
0.656
0.984
0.942
3.281
0.066
3.215
0.910
13
^^
ISEC
1.177
1.472
1.004
2.783
1.552
0.279
0.245
0.143
0.106
0.280
0.279
0.924
1.153
0.877
2.783
0.106
2.676
0.776
13
Residuals
2.104
0.168
0.090
-1.269
0.088
-0.060
-0.052
-0.077
0.050
0.130
0.1 Q7
-0.268
-0.169
0.065
2.104
-1.269
3.372
0.717
13
ISEC
3.281
1.640
1.094
1.514
1.640
0.219
0.193
0,066
0.156
0.410
0.386
0.656
0.984
0.942
3.281
0.066
3.215
0.910
13
Model 11
^^
ISEC
1.177
1.472
1.004
2.783
1.552
0.279
0.245
0.143
0.106
0.280
0.279
0.924
1.153
0.877
2.783
0.106
2.676
0.776
13
Residuals
2.104
0.168
0.090
-1.269
0.088
-0.060
-0.052
-0.077
0.050
0.130
0.107
-0.268
-0.169
0.065
2.104
-1.269
3.372
0.717
13
ISEC
3.281
1.640
1.094
1.514
1.640
0.219
0.193
0.066
0.156
0.410
0.386
0.656
0.984
0.942
3.281
0.066
3.215
0.910
13
Model 12
^v^
ISEC
1.428
2.171
1.042
1.460
2.496
0.224
0.168
0.099
0.150
0.358
0.331
1.024
0.817
0.905
2.496
0.099
2.396
0.796
13
Residuals
1.853
-0.531
0.052
0.054
-0.856
-0.005
0.025
-0.033
0.006
0.052
0.055
-0.368
0.167
0.036
1.853
-0.856
2.709
0.620
13
-------
Table C-l. Regression Model Predicted, Residual and Associated Observed Values (Continuation 3)
o
I
I akc or Si to
SIORFT
Number
0801
080101
080102
0802
080201
080202
0803
080301
080302
080303
080304
080305
080306
0804
080401
080402
080403
0806
080601
080602
080603
080604
0807
080701
080702
isrc
j
I.
1 ,
1,
1,
0.
0.
0,
0.
0,
0.
.281
.640
.094
,514
.640
.219
193
.066
.156
,410
.386
Model 13
ISEC
1
2
0
2
2
0
0
0
0
0
0
.398
.499
.858
.086
.013
.281
.230
.149
.125
.260
.258
Residuals
1.
-0.
0.
-0.
-0.
-0.
-0.
-0.
0.
0,
0.
.883
.859
.236
572
.373
062
037
083
031
,150
128
SEC
0.
0.
0,
0.
0,
4.
5.
15.
6.
2,
2.
.31
.61
,91
,66
,61
.57
18
,24
.40
.44
59
Mode 1
SFC
0.
0.
1.
0.
0.
3.
4 .
6.
9.
3.
3.
85
68
00
36
65
59
08
99
43
58
59
14
Residuals
-0
-0
-0.
0,
-0.
0.
1.
8,
-3.
-1 ,
-1.
.54
.07
.09
,30
.04
.98
,10
,25
.03
,14
.00
SEC
0
0
0.
0
0
4.
5.
15.
6.
2
2,
.31
.61
.91
.66
.61
.57
.18
.24
.40
,44
.59
Model
SEC
0.70
0.46
0.96
0.69
0.40
4.49
5.96
10.10
6.69
2.79
3.02
15
Res Lduals
-0
0
-0
-0.
0
0
-0.
5
-0.
-0.
-0.
.39
.15
.05
.03
.21
.08
.78
.14
.29
.35
.43
SEC
0
0
0
0.
0.
4.
5.
15.
6 .
2
2 ,
.31
.61
.91
.66
.61
.57
.18
. 24
.40
.44
.59
Model
SEC
0.72
0.40
1.17
0.48
0.50
3.58
4.35
6.72
8.03
3.85
3.87
16
Resid
-0.
0.
-0.
0.
0.
0.
0.
8.
_^
-1.
-1 .
uals
41
21
26
18
11
99
83
52
63
41
28
0.656
0.984
0.712
0.902
-0.056
0.082
1.52
1.02
1.08
0.87
0.44
0.15
1.52
1.02
0.98
1.23
0.54
-0.21
1.52
1.02
1.40
1.11
0.12
-0.09
Mean
Maximum
Minimum
Range
S.D.
N
0.942
3.281
0.066
3.215
0.910
13
0.905
2.499
0.125
2.374
0.832
13
0.036
1.883
-0.859
1.787
0.636
13
3.24
15.24
0.31
14.93
4.11
13
2.83
9.43
0.36
9.07
2.80
13
0.41
8.25
-3.03
11.28
2.59
13
3.24
15.24
0.31
14.93
4.11
13
2.96
10.10
0.40
9.70
3.03
13
0.28
5.14
-0.78
5.92
1.50
13
3.24
15.24
0.31
14.93
4.11
13
2.78
8.03
0.40
7.62
2.52
13
0.45
8.52
-1.63
10.15
2.55
13
-------
Table C-l. Regression Model Predicted, Residual and Associated Observed Values (Continuation 4)
n
i
Lake or Site
STORET
Number
0801
080101
080102
0802
080201
080202
0803
080301
080302
080303
080304
080305
080306
0804
080401
080402
080403
0806
080601
080602
080603
080604
0807
080701
080702
0808
080801
080802
080803
0811
081101
081102
0813
081301
081302
081303
Mean
Max imum
Minimum
Range
S.D.
N
Model 17
TPHOS TPHOS Residuals TPHOS
0.015 0.012 0.003
0
0
0.747 0.490 0.257
0
0
0.022 0.139 -0.117
0,
0
0,
0.
0
0
0.054 0.086 -0.032
0
0
0.
0.009 0.007 0.002
0
0
0
0
0.011 0.010 0.001
0
0.
0.010 0.015 -0.005
0,
0.
0
0.720 0.134 0.586
0
0
0.025 0.024 0.001
0.
0.
0
0.179 0.102 0.077 0
0.747 0.490 0.586 0
0.009 0.007 -0.117 0
0.738 0.483 0.703 0
0.314 0.155 0.215 0
99 27
.016
.014
.761
.733
.059
.020
.019
.022
.025
.020
.041
.089
.043
.008
.009
.011
.006
.012
.010
.009
.010
.013
.714
.728
.049
.021
.018
.129
.761
.006
.755
.258
Model 18
TPHOS
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
27
015
020
106
212
030
035
020
022
018
018
119
191
179
018
Oil
016
009
009
Oil
072
050
044
047
071
021
015
017
052
212
009
203
059
Residuals
0.
-0
0.
0
0.
-0.
-0,
-0.
0.
0
-0.
-0.
-0,
-0.
-0
-0.
-0
0.
-0.
-0,
-0.
-0.
0.
0
0,
0
0
0.
0.
-0.
0.
0,
27
.001
.006
.655
.521
.029
.015
.001
.000
.007
.002
.078
.102
.136
.010
.002
.005
.003
.003
.001
.063
.040
.031
.667
.657
.028
.006
.001
.077
,667
,136
,803
.237
TPHOS
0.
0,
0.
0.
0.
0.
0,
0.
0.
0.
0.
0.
0,
0,
0,
0,
0.
0,
13
,761
,733
,041
,089
,043
,008
.009
,011
.006
.012
.010
.714
,728
,245
.761
.006
.755
.341
Model 19
TPHOS
0,
0
0
0
0,
0
0
0
0
0,
0
0
0
0
0
0
0
0
13
.157
.349
.181
.309
.288
.021
.012
.018
.010
.010
.011
.063
.100
.118
.349
.009
.340
.127
Residuals
0.
0.
-0.
-0.
-0.
-0.
-0.
-0.
-0.
0.
-0.
0.
0,
0.
0,
-0
0.
0
13
604
384
140
,220
245
,013
,003
007
.004
002
.001
.651
,628
,126
.651
.245
.896
.323
TPHOS
0.761
0.733
0.041
0.089
0.043
0.008
0.009
0.011
0.006
0.012
0.010
0.714
0.728
0.245
0.761
0.006
0.755
0.341
13
Model 20
TPHOS
1.491
0.244
0.066
0.163
0.033
0.023
0.007
0.006
0.004
0.015
0.019
0.350
0.487
0.224
1.491
0.004
1.487
0.411
13
Residuals
-0.730
0.489
-0.025
-0.074
0.010
-0.015
0.002
0.005
0.002
-0.003
-0.009
0.364
0.241
0.120
0.489
-0.730
1.220
0.284
13
-------
Table C-l. Regression Model Predicted, Residual and Associated Observed Values (Continuation 5)
n
oo
Lake or Site
STORE!
Number
0801
080101
080102
0802
080201
080202
0803
080301
080302
080303
080304
080305
080306
0804
080401
080402
080403
0806
080601
080602
080603
080604
0807
080701
080702
0808
080801
080802
080803
0811
081101
081102
0813
081301
081302
081303
Mean
Max imum
Minimum
Range
S.D.
N
TPHOS
0.761
0.733
0.041
0.089
0.043
0.008
0.009
0.011
0.006
0.012
0.010
0.714
0.728
0.245
0.761
0.006
0.755
0.341
13
Model 21
TPHOS
0.561
0.153
0.103
0.192
0.074
0.054
0.010
0.006
0.006
0.008
0.008
0.354
0.529
0.158
0.561
0.006
0.556
0.199
13
Residuals
0.
0,
-0.
-0.
-0.
-0.
-0.
0.
0.
0.
0.
0.
0.
0.
0.
-0.
0.
0.
13
,200
,580
062
103
031
046
001
005
000
004
002
360
199
085
580
103
684
198
TPHOS
0.761
0.733
0.041
0.089
0.043
0.008
0.009
0.011
0.006
0.012
0.010
0.714
0.728
0.245
0.761
0.006
0.755
0.341
13
Model 22
TPHOS
0.979
0.629
0.045
0.146
0.038
0.023
0.005
0.008
0.006
0.012
0.021
0.171
0.782
0.220
0.979
0.005
0.974
0.340
13
Residuals
-0.218
0.104
-0.004
-0.057
0.005
-0.015
0.004
0.003
0.000
0.000
-0.011
0.543
-0.054
0.232
0.543
-0.218
0.761
0.171
13
Model 23
TON TON Residuals TON
0.180 0.220 -0.040
0
0
1.623 1.219 0.404
1
1
0.277 0.273 0.004
0
0
0
0
0
0
0.816 1.151 -0.335
0
1
0
0.190 0.191 -0.001
0
0
0
0
0.116 0.141 -0.025
0.
0,
0.237 0.472 -0.235
0.
0.
0.
1.092 0.600 0.492
1.
1.
0.320 0.187 0.133
0.
0.
0.
0.539 0.495 0.044 0.
1.623 1.219 0.492 1.
0.116 0.141 -0.335 0.
1.507 1.077 0.827 1.
0.524 0.419 0.268 0.
99 9 27
.180
.180
.890
.357
.380
.180
.180
.197
.143
.150
.533
.130
.710
.200
.180
.180
.158
,180
.197
180
180
380
050
155
330
347
330
454
890
143
740
458
Model 24
TON
0.201
0.234
0.689
1.076
0.307
0.340
0.234
0.255
0.224
0.219
0.744
1.005
0.966
0.225
0.161
0.204
0.145
0.144
0.159
0.537
0.426
0.391
0.412
0.535
0.247
0.197
0.210
0.388
1.076
0.144
0.932
0.276
27
Residuals
' -0.021
-0.054
1.201
0.281
0.073
-0.160
-0.054
-0.058
-0.081
-0.069
-0.211
0.125
-0.256
-0.025
0.019
-0.024
0.013
0.036
0.038
-0.357
-0.246
-0.011
0.638
0.620
0.083
0.150
0.120
0.066
1.206
-0.357
1.558
0.317
27
-------
Table C-l. Regression Model Predicted, Residual and Associated Observed Values (Continuation 6)
n
i
Lake or Site
STORET
Number
0801
080101
080102
0802
080201
080202
0803
080301
080302
080303
080304
080305
080306
0804
080401
080402
080403
0806
080601
080602
080603
080604
0807
080701
080702
0808
080801
080802
080803
0811
081101
081102
0813
081301
081302
081303
Mean
Max imura
Minimum
Range
S.D.
N
Model 25
TON
1.890
1.357
0.533
1.130
0.710
0.200
0.180
0.180
0.158
0.180
0.197
1.050
1.155
0.686
1.890
0.158
1.732
0.577
13
^^
TON
0.881
1.406
0.955
1.310
1.255
0.272
0.192
0.245
0.172
0.171
0.189
0.515
0.677
0.634
1.406
0.171
1.235
0.476
13
Residuals
1.009
-0.049
-0.422
-0.180
-0.545
-0.072
-0.012
-0.065
-0.014
0.009
0.008
0.535
0.478
0,052
1.009
-0.545
1.554
0.410
13
TON
1.890
1.357
0.533
1.130
0.710
0.200
0.180
0.180
0.158
0.180
0.197
1.050
1.155
0.686
1.890
0.158
1.732
0.577
13
Model 26
TON
1.923
1.149
*-
0.680
1.072
0.798
0.337
0.183
0.147
0.129
0.199
0.192
0.891
0.947
0.666
1.934
0.129
1.805
0.540
13
Residuals
-0.044
0.208
-0.147
0.058
-0.088
-0.137
-0.003
0.033
0.029
-0.019
0.005
0.159
0.208
0.020
0.208
-0.147
0.355
0.116
13
TON
1.890
1.357
0.533
1.130
0.710
0.200
0.180
0.180
0.158
0.180
0.197
1.050
1.155
0.686
1.890
0.158
1.732
0.577
13
Model 27
^^
TON
1.927
1.184
0.629
1.173
0.628
0.317
0.204
0.143
0.114
0.217
0.213
0.921
1.016
0.668
1.927
0.114
1.813
0.550
13
Residuals
-0.037
0.173
-0.096
-0.043
0.082
-0.117
-0.024
0.037
0.044
-0.037
-0.016
0.129
0.139
0.018
0.173
-0.117
0.290
0.091
13
TON
1.890
1.357
0.533
1.130
0.710
0.200
0.180
0.180
0.158
0.180
0.197
1.050
1.155
0.686
1.890
0.158
1.732
0.577
13
Model 28
^^
TON
1.591
1.738
0.521
1.048
0.685
0.273
0.160
0.156
0.131
0.218
0.261
0.726
1.263
0.675
1.738
0.131
1.606
0.567
13
Residuals
0.299
-0.381
0.012
0.082
0.025
-0.073
0.020
0.024
0.027
-0.038
-0.064
0.324
-0.108
0.012
0.324
-0.381
0.705
0.176
13
-------
Table C-l. Regression Model Predicted, Residual and Associated Observed Values (Continuation 7)
n
i
o
Lake or Site
STORE!
Number
0801
080101
080102
0802
080201
080202
0803
080301
080302
080303
080304
080305
080306
0804
080401
080402
080403
0806
080601
080602
080603
080604
0807
080701
080702
0808
080801
080802
080803
0811
081101
081102
0813
081301
081302
081303
Mean
Maximum
Minimum
Range
S.D.
N
Model 29
COND
30
595
152
600
92
7
107
1295
24
322
1295
7
1288
433
9
^^
COND
57
747
156
426
32
11
177
1055
19
298
1055
11
1045
373
9
Residuals
-27
-152
-4
174
60
4
-70
240
5
25
240
-152
392
120
9
COND
29
32
597
593
123
132
160
167
180
180
637
586
571
89
93
91
89
8
5
109
105
109
1304
1282
24
25
24
272
1304
5
1299
359
27
Model 30
^.
COND
44
53
896
589
156
178
109
86
116
89
1149
849
187
96
34
28
35
17
27
394
127
142
668
368
39
35
39
242
1150
17
1133
312
27
Residuals
-15
-21
-299
4
-34
-46
51
81
64
91
-512
-263
384
-7
59
63
54
-9
-22
-285
-22
-33
636
914
-15
-10
-15
29
914
-512
1427
271
27
COND
597
593
637
586
571
89
93
91
89
8
5
1304
1282
457
1304
5
1299
451
13
Model 31
^^
COND
555
1231
637
1091
1015
75
41
63
34
34
40
222
354
415
1232
34
1198
448
13
Residuals
42
-638
0
-506
-444
14
52
28
55
-26
-35
1082
928
43
1082
-639
1721
490
13
COND
597
593
637
586
571
89
93
91
89
8
5
1304
1282
457
1304
5
1299
451
13
Model 32
^^^
COND
173
502
424
680
1860
236
40
34
52
25
27
511
1319
452
1860
25
1835
562
13
Residuals
424
91
213
-94
-1289
-147
53
57
37
-17
-22
793
-38
5
793
-1289
2082
462
13
-------
Table C-l. Regression Model Predicted, Residual and Associated Observed Values (Continuation 8)
o
Lake or Site
STORE!
Number
0801
080101
080102
0802
080201
080202
0803
080301
080302
080303
080304
080305
080306
0804
080401
080402
080403
0806
080601
080602
080603
080604
0807
080701
080702
0808
080801
080802
080803
0811
081101
081102
0813
081301
081302
081303
Mean
Maximum
Minimum
Range
S.D.
N
COND
597
593
637
586
571
89
93
91
89
8
5
1304
1282
457
1304
5
1299
451
13
Model
COND
173
502
424
680
1860
236
40
34
52
25
27
511
1320
452
1860
25
1835
562
13
33
Residuals COND
424
91
213
-94
-1289
-147
53
57
37
-17
-22
793
-38
5
793
-1288
2082
462
13
597
593
637
586
571
89
93
91
89
8
5
1304
1282
457
1304
5
1299
451
13
Model 3
COND
267
463
332
1366
478
69
88
44
28
114
123
290
422
315
1366
28
1339
355
13
14 Model 35
Residuals AAY AAY Residuals MY
330
130
304
-780
93
20
5
47
61
-106
-118
1040
859
143
1014
-780
1794
444
13
0.5 0.4 0.1
0.
0.
186.3 139.4 46.9
186.
186.
0.7 3.6 -2.9
0.
0.
0.
0.
0.
0.
3.2 3.0 0.2
3.
3.
3.
0.3 0.3 0.0
0.
0.
0.
0.
0.2 0.2 0.0
0.
0.
0.3 0.5 -0.2
0.
0.
0,
7.2 2.2 5.0
7.
7,
0.5 0.4 0.1
0
0.
0,
22.1 16.7 5.5 15
186.3 139.4 46.9 186,
0.2 0.2 -2.9 0,
186.1 139.2 49.8 186
61.6 46.0 15.7 49
99 9 27
5
5
3
3
4
4
4
9
9
9
2
2
2
3
3
,3
3
2
,2
,3
.3
,3
, 2
.2
.5
.5
,5
.1
.3
.2
.1
. 4
Model 36
AAY Residuals
0.
0.
9.
29.
0.
1.
0.
2 ,
0.
0.
3,
18,
11,
0.
0.
0.
0.
0,
0,
0,
1,
0,
1,
3
0
0
0
3
29
0
29
6
27
9
6
9
6
3
7
4
4
3
.3
,6
,7
,5
.4
i 2
.3
,2
,3
.2
,7
.4
.9
.6
.7
. 4
. 7
.5
.4
.6
.2
.4
.8
-0.4
-0.1
176.4
156.7
0.1
-1 .3
0.0
-1.5
0.6
0.6
-0.4
-15.5
-8.3
-0.1
0.1
0.0
0.1
-0.1
0.0
-0.4
-1.1
-0.6
5.6
3.5
0.1
-0.2
0.0
11.6
276. j
-15.5
191.9
44.9
27
-------
Table C-l. Regression Model Predicted, Residual and Associated Observed Values (Continuation 9)
n
Lake or Site
STORET
Number
0801
080101
080102
0802
080201
080202
0803
080301
080302
080303
080304
080305
080306
0804
080401
080402
080403
0806
080601
080602
080603
080604
0807
080701
080702
0808
080801
080802
080803
0811
081101
081102
0813
081301
081302
081303
Mean
Maximum
Minimum
Range
S.D.
N
Model 37
AAY
186.3
186.3
3.2
3.2
3.2
0.3
0.3
0.3
0.3
0.2
0.2
7.2
7.2
30.6
186. 3
0.2
186.1
69.1
13
^v^
AAY
192.0
19.4
7.4
14.0
6.3
0.8
0.3
0.7
0.2
0.1
0.2
4.1
3.1
19.1
192.0
0.1
191.9
52.3
13
Residuals
-5.7
166.9
-4.2
-10.8
-3.0
-0.5
0.0
-0.4
0.1
0.1
0.0
3.1
4.1
11.5
166.9
-10.8
177.7
46.9
13
AAY
186.3
186.3
3.2
3.2
3.2
0.3
0.3
0.3
0.3
0.2
0.2
7.2
7.2
30.6
186.3
0.2
186.1
69.1
13
Model 38
^^
AAY
259.3
112.1
1.7
4 . 1
5.0
1 .1
0.3
0.2
0.2
0.3
0.3
4.7
5.5
30.4
259.3
0.2
259.1
75.2
13
Residuals
-73.0
74.2
1.5
-0.9
-1.8
-0.8
0.0
0.1
0.1
-0.1
-0.1
2.5
1.7
0.3
74.3
-73.0
147.2
30.1
13
AAY
186.3
186.3
3.2
3.2
3.2
0.3
0.3
0.3
0.3
0.2
0.2
7.2
7.2
30.6
186.3
0.2
186.1
69.1
13
Model 39
^^
AAY
192.0
19.4
7.4
14.0
6.3
0.8
0.3
0.7
0.2
0.1
0.2
4.1
3.1
19.1
192.0
0.1
191 .9
52.3
13
Residuals
-5.7
166.9
-4.2
-10.8
-3.1
-0.5
0.0
-0.4
0.1
0.1
0.0
3.1
4.1
0.3
74.2
-73.0
147.2
30.1
13
AAY
186.3
186.3
3.2
3.2
3.2
0.3
0.3
0.3
0.3
0.2
0.2
7.2
7.2
30.6
186.3
0.2
186.1
69.1
13
Model 40
^^
AAY
117.0
125.9
1.8
3.2
6.8
1.1
0.1
0.2
0.3
0.2
0.3
4.4
13.8
21.2
125.9
0.1
125.8
44.7
13
Residuals
69.3
60.4
1.4
0.0
-3.6
-0.8
0.2
0.1
0.0
0.0
-0.1
2.8
-6.6
9.5
69.3
-6.6
76.9
24.8
13
-------
Table C-l. Regression Model Predicted, Residual and Associated Observed Values (Continuation 10)
n
Lake or Site
STORE!
Number
0801
080101
080102
0802
080201
080202
0803
080301
080302
080303
080304
080305
080306
0804
080401
080402
080403
0806
080601
080602
080603
080604
0807
080701
080702
0808
080801
080802
080803
0811
081101
081102
0813
081301
081302
081303
Mean
Maximum
Minimum
Range
S.D.
N
Model 41
PC1-11 PC1-11 Residuals PCI
-1.95 -2.16 0.21
-1.
-1.
2.90 2.62 0.28
5.
4.
-1.17 -0.20 -0.97
0.
-0.
-0.
-0.
-0.
-0.
1.11 1.52 -0.41
1.
3.
1.
-2.52 -2.69 0.17
-1.
-1.
-2.
-2.
-2.59 -2.61 0.02
-1,
-2.
-1.49 -1.09 -0.40
-1
-0
-0
1.50 0.60 0.90
2
3
-1.41 -1.59 0.18
-0
-0
-0
-0.62 -0.62 0.00 0
2.90 2.62 0.90 5
-2.59 -2.69 -0.97 -2
5.49 5.31 1.86 7
1.96 1.89 0.67 2
9 9 9 27
-27
41
30
43
49
02
66
88
65
71
59
54
39
96
93
96
.11
,28
.85
.01
.06
.99
.33
.78
.31
.23
.64
.94
.014
.43
.28
.71
.15
Model 42
PC1-27
-1.34
-0.89
2.29
3.60
-0.09
0.21
-0.89
-0.64
-1.02
-1.09
2.51
3.40
3.28
-1.01
-1.99
-1.30
-2.30
-2.32
-2.03
1.56
0.87
0.62
0.78
1.55
-0.73
-1.40
-1.21
0.014
3.60
-2.32
5.92
1.81
27
Residuals
-0.07
-0.41
3.14
0.89
0.11
-0.87
0.01
-0.01
0.31
0.50
-0.97
-0.01
-1.32
_0.92
0.03
-0.81
0.02
0.47
0.02
-2.62
-1.86
-0.95
2.00
1.76
0.50
0.76
0.27
0.00
3.14
-2.62
5.77
1.19
27
PC1-13
5.43
4.49
1.54
3.39
1.96
-1.93
-1.96
-2.51
-2.28
-1.85
-2.01
2.78
3.31
0.80
5.43
-2.51
7.94
2.95
13
Model 43
PC1-13
5.18
4.67
2.04
4.21
1.72
-0.69
-1.73
-2.38
-1.77
-2.05
-2.91
2.03
2.05
0.80
5.18
-2.91
8.09
2.86
13
Residuals
0,
-0,
-0
-0
0
-1
-0
-0
-0
0
0
0
1
0
1
-1
2
0
13
.25
.18
.50
.82
.24
.24
.23
.13
.51
.20
.90
.75
.26
.00
.26
.24
.50
.86
PC1-13
5.43
4.49
1 .54
3.39
1.96
-1.93
-1.96
-2.51
-2.28
-1.85
-2.01
2.78
3.31
0.80
5.43
-2.51
7.94
2.95
13
Model 44
PC1-13
5.41
4.20
1.95
3.52
3.14
-0.88
-2.03
-2.56
-3.15
-1.38
-1.54
1.92
1.76
0.80
5.41
-3.15
8.56
2.85
13
Residuals
0.02
0.29
-0.41
-0.13
-1.18
-1.05
0.07
0.05
0.87
-0.47
-0.47
0.86
1.55
0.00
1.55
-1.18
2.74
0.84
13
-------
Table C-l. Regression Model Predicted, Residual and Associated Observed Values (Continuation 11)
n
Lake or Site
STORE1
Number
0801
080101
080102
0802
080201
080202
0803
080301
080302
080303
080304
080305
080306
0804
080401
080402
080403
0806
080601
080602
080603
080604
0807
080701
080702
r\or\Q
UoUo
080801
080802
080803
0811
081101
081102
0813
081301
081302
081303
Mean
Max imum
Minimum
Range
S.D.
N
PC1-13
5.
4.
1.
3.
1.
-1.
-1.
_2
-2.
-1.
-2.
2.
3.
0.
5.
-2.
7.
2.
13
,43
49
,54
39
,96
93
,96
,51
28
85
01
.78
31
.80
,43
.51
.94
,95
Model 45
PC1-13
5
3
1,
3,
1.
-0,
-1,
-2.
-3,
-1.
-1,
2.
2.
0,
5,
-3,
9.
2.
13
.45
.92
.62
.64
.62
.58
.68
.79
.61
.40
.48
.73
,92
.80
.45
.61
.06
.88
Residuals
-0.02
0.57
-0.08
-0.25
0.34
-1.35
-0.28
0.28
1.33
-0.45
-0.53
0.05
0.39
0.00
1.33
-1.35
2.68
0.86
13
PC1-13
5
4,
1.
3.
1
-1,
-1,
-2,
-2,
-1.
-2,
2 ,
3.
0.
5.
-2.
7.
2.
13
.43
.49
.54
,39
.96
,93
.96
.51
,28
.85
,01
.78
,31
,80
,43
,51
,94
,95
Model 46
PC1-13 Residuals
4
5
1
3
2
-0
-2
-2
-3
-1
-1
2
3
0
5
-3
8
2
13
.70
.05
.13
.15
.22
.78
.51
.56
.00
.63
.14
.06
.67
.80
.05
.00
.05
.88
0
-0
0
0
-0
-1
0
0
.0
-0
-0
0
-0
0.
0.
-1.
1.
0.
13
.73
.56
.41
.24
.26
.15
.55
.05
.72
.22
.87
.72
.36
00
74
15
88
73
-------
Table C-2. Regression Models Developed from Water Truth and MSS and MMS Data
n
i
On
Model Dependent
Number Variable
1 LNCHLA
2 LNCHLA
3 LNCHLA
4 LNCHLA
5 LNCHLA
6 LNCHLA
7 LNISEC
8 LNISEC
9 LNISEC
10 LNISEC
11 LNISEC
12 LNISEC
13 LNISEC
LNSEC
LNSEC
Regression,
Intercept Independent Variables and Associated Residual
Value Coefficients d.f.
-3.036 +0.233 RED 1, 7
-3.367 +0.142 CRN 1, 25
0.728 +0.449 IR1; -1.039 IR2 2, 10
-25.628 +0.358 CHI; -0.220 CH2; +0.207 CH7 3, 9
-9.281 +0.081 CH4; +0.156 CH9 2, 10
-12.901 +0.059 MMSPC1; +0.001 MSSPC2 2, 10
-3.931 +0.150 RED 1, 7
-4.817 +0.112 CRN 1, 25
-5.811 +0.135 CRN 1, H
-8.120 +0.119 CH4 1, 11
-8.120 +0.119 CH4 1, 11
9.474 +0.051 MMSPC1; -0.115 MMSPC3; -1.519 MMSPC7 3, 9
-6.165 +0.051 MMSPC1; -0.115 MMSPC3 2, 10
(NO MODEL DEVELOPED)
(NO MODEL DEVELOPED)
Standard
Error
Calculated of
F-value R xlOO Estimate
64.87 90.26 0.36
44.07 63.80 0.59
15.35 75.43 0.72
20.49 87.23 0.55
30.71 86.00 0.54
24.69 83.16 0.60
13.76 66.28 0.51
36.34 59.25 0.52
27.92 71.73 0.64
52.78 82.75 0.50
52.78 82.75 0.50
30.23 90.97 0.40
29.11 85.34 0.48
Comments
Nine lakes. LAXDSAT MSS.
Twenty-seven sites. LANDSAT
MSS.
Thirteen sites. LANDSAT MSS.
Thirteen sites. MMS. Eight
channel selection
Thirteen sites. MMS. Channel
selection limited to 4,7,8,9.
Thirteen sites. MMS. Eight
principal component-derived
"new" channels.
Nine lakes. LANDSAT MSS.
Twenty-seven sites. LANDSAT MSS.
Thirteen sites. LANDSAT MSS.
Thirteen sites. MMS. Eight
channel selection.
Thirteen sites. MMS. Channel
selection limited to 4,7,8,9.
Thirteen sites. MMS. Eight
principal component-derived
"new" channels.
Thirteen sites. MMS. Selection
limited to first four compo-
nents of principal component
transformed eight channels.
Nine lakes. LANDSAT MSS.
Twenty-seven, sites. LANDSAT
MSS.
-------
Table C-2. Regression Models Developed from Water Truth and MSS and MMS Data (Continuation 1)
Model Dependent Intercept
Number Variable Value
Independent Variables and Associated
Coefficients
Regression,
Residual Calculated
d.f .
F-value R xlOO
Standard
Error
of
Estimate
Comments
n
LNSEC (NO MODEL DEVELOPED)
LNSEC (NO MODEL DEVELOPED)
14 LNSEC 8.119 -0.119 CH4
15 LNSEC -9.520 -0.051 MMSPC1; +0.114 MMSPC3; +1.527 MMSPC7
16 LNSEC .6.196 -0.051 MMSPC1; +0.114 MMSPC3
17 LNTPHOS -3.746 +0.575 IR1; -1.301 IR2
18 LNTPHOS -10.053 +0.176 CRN
19
20
21
LNTPHOS
LNTPHOS
LNTPHOS
-10.785 +0.201 CRN
-16.360
+0.442 CH2; -0.768 CH3: +0.712 CH4
-0.462 CH7
-8.947 +0.437 CH4; -0.463 CH7
22 LNTPHOS -28.507 +0.071 MMSPC1; -0.190 MMSPC3; -0.559 MMSPC4
23 LNTON -6.511 +0.143 CRN
24 LNTON -5.367 +0.113 CRN
25 LNTON -5.363 +0.118 CRN
26 LNTON -4.793 -0.091 CH3; +0.253 CH4; -0.104 CH7
2, 10
1, 7
1, 25
53.24 82.88 0.49
30.78 91.12 0.39
29.29 85.42 0.48
2, 6
1, 25
1, 11
4, 8
7.63
18.26
12.70
16.64
71.75
41.84
53.59
89.27
1.07
1.15
1.41
0.79
20.53
28.38
).42
90.44
25.99 78.78
37.94 60.28
41.83 79.18
58.23 95.10
0.96
0.71
0.45
0.51
0.45
0.24
Thirteen sites. LANDSAT MSS.
Thirteen sites. MMS.
Thirteen sites. MMS. Channel
selection limited to 4,7,8,9.
Thirteen sites. MMS. Eight
principal component-derived
"new" channels.
Thirteen sites. MMS. Selection
limited to first four components
of principal component trans-
formed eight channels.
Nine lakes. LANDSAT MSS.
Twenty-seven sites. LANDSAT
MSS.
Thirteen sites. LANDSAT MSS.
Thirteen sites. MMS. Eight
channel selection.
Thirteen sites. MMS channel
selection limited to 4,7,8,9.
Thirteen sites. MMS. Eight
principal component-derived
"new" channels.
Nine lakes. LANDSAT MSS.
Twenty-seven sites. LANDSAT
MSS.
Thirteen sites. LANDSAT MSS.
Thirteen sites. MMS. Eight
channel selection.
-------
Table C-2. Regression Models Developed from Water Truth and MSS and MMS Data (Continuation 2)
o
Model Dependent Intercept
Number Variable Value
27 LNTON -6.449
28 LNTON -10.732
29 LNCOND -11.690
30 LNCOND -6.277
31 LNCOND -2.618
32 LNCOND 2.251
33 LNCOND 2.251
34 COND -4616.1
35 LNAAY -6.285
36 LNAAY -7.481
37 LNAAY -13.288
38 LNAAY -12.273
39 LNAAY -12.273
40 LNAAY -12.841
41 PC1-9 -5.359
Regression,
Independent Variables and Associated Residual
Coefficients d.f.
+0.167 CH4; -0.190 CH7 ; +0.076 CHS 3, 9
+0.041 MMSPC1; -0.101 MMSPC3; -0.170 MMSPC4 3, 9
+0.882 CRN; -0.807 RED 2, 6
+0.493 CRN; -0.360 RED 2, 24
+0.201 CRN 1, 11
+0.255 CH4; -0.326 CH9 2, 10
+0.255 CH4; -0.326 CH9 2, 10
+15.362 MMSPC1; -48.91 MMSPC2 ; -101.85 3, 9
MMSPC4
+0.823 IR1; -0.954 IR2 2, 6
+0.190 CRN; +0.246 IR1; -0.634 IR2 3, 23
+0.439 CRN; -0.376 RED; +0.392 IR1 3, 9
+0.289 CH4; -0.654 CH7; +0.835 CHS 4, 8
-0.460 CH9
+0.289 CH4; -0.654 CH7; +0.835 CHS 4, 8
-0.460 CH9
+0.081 MMSPC1; -0.0446 MMSPC3; -0.462 MMSPC4 3, 9
+0.218 RED; +0.421 IR1; -1.036 IR2 3, 5
Standard
Error
Calculated „ of
F-value R xlOO Estimate Comments
59.69 95.21 0.24 Thirteen sites. 1>!MS channel
selection limited to 4,7,8,9.
59.80 95.22 0.24 Thirteen sites. MMS. Kight
principal component-derived
"new" channels.
29.53 90.78 0.60 Nine lakes. LANDSAT MSS.
31.38 72.33 0.79 Twenty-seven sites. LANDSAT
MSS.
19.46 63.89 Thirteen sites. LANDSAT MSS.
12.85 72.00 1.05 Thirteen sites. MMS. Eight
channel selection.
12.85 72.00 1.05 Thirteen sites. MMS channel
selection limited to 4,7,8,9.
13.45 81.76 70.33 Thirteen sites. MMS. Eight
principal component-derived
"new" channels.
21.37 87.69 0.79 Nine lakes. LANDSAT MSS.
13.84 64.35 1.16 Twenty-seven sites. LANDSAT
MSS.
15.21 83.53 1.13 Thirteen sites. LANDSAT MSS.
39.22 95.15 0.65 Thirteen sites. MMS. Eight
channel selection.
39.22 95.15 0.65 Thirteen sites. MMS channel
selection limited to 4,7,8,9.
41.32 93.23 0.73 Thirteen sites. MMS. Eight
principal component-derived
"new" channels.
21.06 92.67 0.67 Nine lakes. LANDSAT MSS.
-------
Table C-2. Regression Models Developed from Water Truth and MSS and MMS Data (Continuation 3)
Standard
Regression, Error
Model Dependent Intercept Independent Variables and Associated Residual Calculated ^ of
Number Variable Value Coefficients d.f. F-value R xlOO Estimate
Comme
42 PC1-27 -12.414 +0.331 CRN 1, 25 59.59 70.45 1.19 Twenty-seven
43 PC1-13 -13.153 +0.539 CRN; -0.442 RED; +0.632 IR1 4, 8 33.22 94.32 0.86 - Thirteen
-1.332 IR2
44 PC1-13 -12.617 -0.476 CH3; +0.797 CH4 2, 10 68.14 93.16 0.84 Thirteen
channel
45 PC1-13 -17.912 +0.480 CH4; -0.595 CH7 ; +0.305 CHS 3, 9 60.89 95.30 0.86 Thirteen
site
nts
sites. 1.
s . .
sites .
s e 1 e c
t ion
sites.
selection limited
46 PC1-13 -27.289 +0.126 MMSPC1; -0.348 MMSPC3; -0.497 MMSPC4 3, 9 62.93 95.45 0.73 Thirteen
limited
of princ
formed e
n
sites.
to f i
ipal
ight
rst
ANDSA'l
MSS.
LAND SAT MSS.
MMS.
MMS.
to 4
MMS .
four
component
chan
nels.
Eight
Channe
,7,8,9.
Select
1
ion
component
t rans-
i
oo
------- |