United States
Environmental Protection
Agency
Environmental Monitoring
Systems Laboratory
P.O. Box 15027
Las Vegas, NV 89114
TS-AMD-82073
January 1983
AEPA ACID deposition multispectral
SCANNER SURVEY
ADIRONDACK LAKES, NEW YORK
Phase II
prepared for
U.S. Environmental Protection Agency
Environmental Research Laboratory
Corvallis, Ore.
-------
TS-AMD-8 2073
January 1983
ACID DEPOSITION MULTISPECTRAL SCANNER SURVEY
ADIRONDACK LAKES, NEW YORK
Phase II
by
T. H. Mace
Environmental Programs
Lockheed Engineering and Management Services Company, Inc.
Las Vegas, Nevada 89114
Contract No. 68-03-3049
Project Officer
G. A. Shelton
Advanced Monitoring Systems Division
Environmental Monitoring Systems Laboratory
Las Vegas, Nevada 89114
ENVIRONMENTAL MONITORING SYSTEMS LABORATORY
OFFICE OF RESEARCH AND DEVELOPMENT
U.S. ENVIRONMENTAL PROTECTION AGENCY
LAS VEGAS, NEVADA 89114
-------
NOTICE
This document has not been peer and administratively reviewed within EPA and is
for internal Agency use and distribution only.
ii
-------
ABSTRACT
During the spring of 1982, the U.S. Environmental Protection Agency's Environ-
mental Monitoring Systems Laboratory at Las Vegas, Nevada, conducted an aerial over-
flight and multispectral analysis of a portion of Adirondack State Park, New York.
The purpose of this flight was to correlate airborne multispectral scanner data and
acid deposition-related water quality measurements during the period of spring
freshet. This study was the second in a series, designed to determine the applica-
bility of satellite remote sensing to characterize acid deposition effects in
freshwater lakes.
Water quality measurements of Secchi depth, pH, alkalinity, total suspended
solids, chlorophyll a, conductivity, phaeophyton, and total organic carbon were pro-
vided by Paul Smith's College and the Environmental Protection Agency's Environmental
Research Laboratory at Corvallis, Oregon. A multiple linear regression procedure was
used to relate the 10-channel multispectral scanner data to each of the water
measurements.
Total organic carbon and Secchi depth were found to have the best relationships
with the scanner data. The remaining parameters possessed significant, weak linear
relationships with the scanner data, however, this may be due to relatively limited
ranges in their observed values. The results of this study confirm the observation
of the first study, conducted in the fall of 1981, that the thematic mapper scanner
aboard the Landsat-4 satellite may prove to be a useful tool in monitoring acid
deposition effects in freshwater lakes.
i ii
-------
CONTENTS
Abstract iii
Figures iv
Tables iv
Introduction 1
Conclusions and Recommendations 5
Methodology 7
Results 12
References 26
Appendix
Multispectral Scanner and Analysis System Characteristics 27
FIGURES
Number Page
1 Location of the study area 8
2 Predicted vs observed log^Q Secchi depth 17
3 Predicted vs observed pH 18
4 Predicted vs observed alkalinity (peg/1) 19
5 Predicted vs observed logig total suspended solids (ppm) 20
6 Predicted vs observed conductivity (ymhos) 21
7 Predicted vs observed phaeophyton (mg/m3) 22
8 Predicted vs observed chlorophyll a (mg/m3) 23
9 Predicted vs observed total organic carbon (mg/1) 24
TABLES
Number Page
1 Values of R2 for Several Data Transformations 12
2 Comparison of Standard Errors and Cp/P Ratios for Transformed and
Untransformed Variables 13
3 Dependent Variable Summary Statistics 14
4 Correlation (R) Matrix Between Independent and Dependent Variables . . 14
5 Best Subset Multiple Linear Regression Comparison 15
iv
-------
INTRODUCTION
THE NATURE OF THE PROBLEM
The effects of acid deposition on the environment are among the most pressing
environmental issues facing the industrial countries. Atmospheric pollution, caused
by fossil fuel combustion, is carried across state and national boundaries producing
national as well as international environmental degredation.
Scores of articles have been published outlining the possible effects of acid
deposition on sensitive areas, and no attempt to reproduce them will be made here.
It may, however, be useful to outline some of the effects suggested by the litera-
ture .
Generally, sulfur and nitrogen oxides are produced in the process of fossil fuel
combustion. These oxides combine with other atmospheric constituents to produce
sulfuric and nitric acids. Normally, these acids would be buffered by atmospheric
carbon dioxide to produce a pH of approximately 5.6 in rain. However, the pH of
rainfall throughout most of the eastern United States, for example, has been reported
at about 4.0 (Galloway et_al., 1976).
This acid deposition phenomenon has many effects. Among these effects are
increased erosion of statuary and historic buildings, crop and forest productivity
reduction, increased acidification and leaching of soils, and degredation of biota in
lakes and streams. These effects are particularly severe in those areas of crystal-
line bedrock and naturally acidic soils, where cation exchange capacity (buffering)
is relatively low. Lakes and streams in these areas are very poorly buffered, and
the plant and animal species reduction which results is severe.
As the lakes become more acid, organic matter decomposition decreases. Leaf
1
-------
litter increases on the lake floor and a fungal mat may spread along the bottom.
Additionally, magnesium and aluminum ions are liberated from the bottom sediments,
increasing their concentrations in the lakes, while phosphorous decreases. The
effect on biota is dramatic. The distribution and extent of macrophytes, phyto-
plankton, aquatic invertebrates, and fish decreases markedly (Baalsrud et al., 1976).
These effects have been severe in the Adirondack region of New York both in
extent and rapidity of development. "A recent survey found that 51 percent of
mountain lakes (217 lakes at an elevation above 600 m) have pH values below 5.0; 90
percent of these lakes contain no fish. In contrast, during the period 1929-1937,
only 4 percent of these lakes had a pH under 5.0 or were devoid of fish" (Likens,
1976, p. 43).
This rapid change in environmental conditions creates an environmental
monitoring problem. At one extreme, investigators have suggested that water
chemistry measurements should be made in lakes 5-10 times each year and during
periods of limnological change (Dochinger and Seliga, 1976). However, which lakes
should be sampled, and how can we optimize such a costly program? Remotely sensed
measurements may provide a cost-effective environmental monitoring tool to reduce the
number of water samples required for a regional monitoring program. Such programs
have been successful in the past for lake trophic classification (Witzig and
Whitehurst, 1981), and it is possible that digital analysis techniques may be
applicable to the monitoring of acid rain effects as well.
THE CURRENT STUDY
Perspective
This study represents the second of a series, designed to determine the applica-
bility of satellite remote sensing to detect and characterize acid deposition effects
in freshwater lakes. The first study was conducted on data gathered on an identical
mission flown on October 13, 1981.
2
-------
Follow-on studies are planned to utilize the thematic mapper simulator aboard
the NASA U-2 aircraft and, finally, the Thematic Mapper scanner aboard the Landsat-4
satellite. This multi-stage approach is designed to provide progressive development
of an operational environmental monitoring tool, with the decision to proceed based
on the results of each phase.
The ultimate goal of these studies - taken together - is to develop a quality-
assured data base (to include land cover and quantitative water quality maps) in a
geographic information system (GIS). The GIS would be updated periodically, using
satellite data, to provide baseline inventory as well as indications of environmental
change. It is assumed that these data would be integrated with other graphically-
based information to provide the U.S. Environmental Protection Agency with a compre-
hensive view of acid rain effects.
Scope
There is ample evidence in the open literature to indicate that a variety of
water quality parameters may be estimated directly from MSS imagery (Witzig and
Whitehurst, 1981). It is the object of this study to determine if the effects of
acid loading caused by freshet conditions could be detected by multivariate inter-
pretation of multispectral scanner imagery.
In the first of the series of studies, a portion of Adirondack State Park (see
Figure 1) was imaged with the MSS during the fall of 1981. Statistically significant
linear relationships were found between the MSS data and concurrent water quality
measurements of Secchi depth, pH, alkalinity, total suspended solids, chlorophyll a,
phaeophyton, and dissolved organic carbon. During the second study, the same area
was imaged using identical systems, flight lines, and altitudes in the spring of
1982. Water measurements were made of Secchi depth, pH, alkalinity, total suspended
solids, conductivity, chlorophyll a, phaeophyton, and total organic carbon. Best
subsets multiple linear regression analysis was then used to relate the scanner data
to the water measurements. Comparison of the results from these two studies should
provide valuable information on the optimum timing of scanner overflights, with the
possible limitation being that chlorophyll a, phaeophyton, and total suspended solids
3
-------
may exist in such low concentrations that small errors in laboratory analyses could
produce relatively large errors in the regression analyses. Confirmation of the
results from the first study, however, should provide sufficient evidence to indicate
that the program is worth pursuing to the next platform (the U-2 aircraft and
Thematic Mapper Simulator).
4
-------
CONCLUSIONS AND RECOMMENDATIONS
CONCLUSIONS
Based on review of pertinent literature and the evidence produced by this study,
it is concluded that:
o Linear relationhsips exist between aircraft multispectral scanner data and
water quality measurements.
o Secchi depth and total organic carbon provided the best correlations with
the multispectral scanner data.
o Correlations between the multispectral data and pH and alkalinity show weak,
linear relationships.
o Spring sampling does not provide optimum results in conjunction with
multispectral scanner flights.
o A multivariate approach is necessary to quantitatively predict the measured
parameters.
o While the results of this study are not adequate to propose that these data
be used in an operational monitoring program, the correlations are
sufficiently good to justify continued research.
o Potential gains from continuing this research include the development of an
image-based geographic information system, synoptic maps of the current state
of U.S. freshwater bodies, and a system for continued monitoring of lake
acidification.
5
-------
RECOMMENDATIONS
Based on the findings reported herein, the following recommendations are made:
o The study should be continued using high-altitude Thematic Mapper Simulator
data collected during peak lake productivity.
o An "interior analysis" should be performed on the data already collected, to
determine if there is a better subset of data points to be used in the
regressions.
o Water sampling should be conducted more nearly coincident with the
overflights.
o Judgement of validity of the regression model for predictive purposes should
not be made on R2 alone, but should include Mallow's Cp, standard error,
statistical significance, and signal-to-noise considerations as well.
o Special attention should be placed on the distribution of sampling stations,
to achieve the greatest range and most uniform distribution of water quality
values possible.
6
-------
METHODOLOGY
DATA ACQUISITION
The multispectral data were acquired using a Daedalus DS-1260 multispectral
scanner flown aboard an Aero Commander 680V aircraft from 0903 to 1031 EST on May 1,
1982. The flight altitude was approximately 6,400 meters (21,000 feet) above mean
sea level (MSL). The average terrain level below the aircraft was approximately 600
meters (19,700 feet) MSL, yielding a nadir ground resolution element of approximately
14.5 by 14.5 meters. (More detailed information about the multispectral scanner may
be found in the Appendix.) Four north-south flight lines provided continuous cover-
age of the 97.5 by 25.0 kilometer study area (Figure 1).
Additionally, simultaneous 9 by 9-inch aerial transparencies were collected
using Kodak 2448 color reversal film. The camera used was a Wild RC-8 with a 152-mm
(6-inch) focal length lens and a haze-correcting antivignetting filter. This
produced stereo coverage of the study area at a nominal scale of 1:38,000.
Water sampling of 56 stations at a depth of 1 meter was conducted by Paul
Smith's College, New York over the period of May 3 through May 5, 198 2. Measurements
of Secchi depth, pH, alkalinity, total suspended solids, conductivity, chlorophyll a,
and phaeophyton were made by Paul Smith's College, and measurement of total organic
carbon was made by the USEPA Environmental Research Laboratory at Corvallis, Oregon.
A three-replicate procedure was used to provide precision data for the water quality
parameters.
DATA ANALYSIS
The general data analysis approach used in this study was to relate the
multispectral scanner data to the water quality parameters, one at a time, with a
7
-------
'////"'"
Long.
ADIRONDACK
STUDY AREA
PARK
20 SO
30
MILES
Lake Ontario
Rochestero
OUtica
SyracuseO
[pBuffalo
AlbanyO
Lake
Erie
Finger
Lakes
New York
Figure 1. Location of the study area.
8
-------
multiple linear regression technique. Best subsets multiple linear regression
(Dixon, 1977) was used to quantify the relationships between the 10-channel multi-
spectral data and the water measurements.
Preprocessing and Signature Extraction
Channels 1 through 10 of the multispectral scanner (0.38-1.10 ym) were decom-
mutated from high density digital tape (HDDT) to computer compatible (CCT) format.
Corrections were then made for scan line overscan and tangential distortion. No
other preprocessing steps were undertaken.
Channel 10 (0.92-1.10 pm) data were then displayed on the video screen of a
COMTOL image processing system. Reference maps (1:62,500 scale USGS quadrangles)
provided by Paul Smith's College were used to locate the sampling stations in the
imagery. A five by five picture element (pixel) matrix of data was extracted from
the MSS data set about the location of each water sample station.
The means for each channel were calculated and converted to radiance values
(expressed in yW/cm2-sr-nm). These mean radiance vectors for each sample station
were then input to the Biomedical Computer Programs, P-Series statistics package
(Dixon, 1977), program P9R (All Possible Subsets Multiple Linear Regression) for
computation of the regression equations.
Three of the 56 sampling stations were eliminated from the data extraction
process because of the presence of lake ice. The stations eliminated were:
1) Honnedaga Lake, 2) Raquette Lake (Lonesone Bay), and 3) Raquette Lake (Sucker
Brook).
Data Transformations and Regression Analysis
It has been suggested in the literature that transformations of the multi-
spectral data improve the relationships between MSS data and the water sample data
(Alfoldi and Munday, 1978). Several data transformations were included in the
analysis.
9
-------
Chromaticity ratios were calculated for each channel. The chromaticity ratio
follows the form of equation 1.
Ri = Ch^Chi (1)
i=l,10
Where: = the chromaticity ratio for channel (i)
Ch^ = the radiance for channel (i)
Also, adjacent-channel ratios of the form:
Ri = Chi/Chi+1 (2)
were calculated. The ratio Ch5/Ch10 was also included in this group.
Other transformations included logarithms of the MSS and the water sample data,
inverse transforms of the MSS data, and squares of the MSS data.
The best subsets of MSS data were determined by minimizing Mallow's Cp. This
procedure minimizes the effects of sampling error, producing a less biased estimate
of the optimum subset of variables (Snedecor and Cochran, 1980).
Criteria for Regression Evaluation
In a benchmark article on the use of remote sensing data for water pollution
studies, Whitlock, Kuo, and LeCroy (1982) suggest a series of criteria by which the
multiple linear regression procedure may be evaluated. Most of their criteria are
incorporated here.
First of all, the scanner data must possess a sufficient signal-to-noise ratio
to observe differences caused by changes in the optical response of the water.
Daniel and Woods criteria (Ibid., p. 156), that the variance of the data exceed the
scanner calibration lamp variance by a factor of 10, was adopted. The variance of
the training sets was compared to the variance of the calibration lamp values. All
channels except MSS channel 9 (0.80-0.89 iim) passed this test. Channel 9 failed the
10
-------
test by only a slight amount, and this was due to a relatively low range in the data,
rather than noise in the system. Therefore, although it technically failed Daniel
and Wood's criteria, channel 9 was included in the analysis.
The remaining criteria are as follows:
R2 - the variance about the dependent variable explained by the regression
equation. This statistic varies between zero and one and is called the
coefficieint of determination. Its value should approach one (100 percent
of the variance explained). However, values of 0.7 or even slightly less
may be usable.
F - this statistic is used to compute the significance of the regression
equation (i.e., the regression is linear in form and possesses a non-zero
slope). It has been suggested that the F-value should exceed the critical
F-value (at the 95 percent confidence level) by a factor four (Ibid., p.
166). It is the opinion of the author that the level of significance
should be set at the 99.0 or 99.9 percent level and no multiplication
factor should be used.
SE - this is the standard error of the estimate, and is expressed in units of
the dependent variable. Sixty-eight percent of the errors should fall
within one standard error. Its value will be no better than the precision
of the laboratory data, and the acceptable value depends heavily on how the
remotely sensed measurements are to be used. In an ideal case, it would
approach zero.
11
-------
RESULTS
REGRESSION ANALYSIS
Table 1 contains a list of R2 for each regression. In each case, the R2 was
based on the "best subset" of variables chosen to minimize the Cp statistic.
Therefore, there may have been higher R2 values with other regressions, but these
represent the combinations with the least bias in the independent variables (the MSS
radiance values). Total suspended solids and total organic carbon are represented by
the abbreviations TSS and TOC, respectively, and all logarithms are to base 10.
TABLE 1. VALUES OF R2 FOR SEVERAL DATA TRANSFORMATIONS
MSS
MSS
log
MSS Chromaticity Ratio
log MSS
MSS
MSS
vs
vs
vs
vs
vs
vs
Variable
Variable
Variable
Variable
log Variable Variable
log Variable
Secchi
.622*
.608*
.608*
.702*
.615*
.703*
P«
.462*
.472*
.499*
.489*
.475*
.479*
Alkalinity
.366*
.373*
.362*
-
.365
-
TSS
.225
.225
.140
.272
.224
.286
Conductivity
.452*
.446*
.594*
.297
.423*
.307
Phaeophyton
.329
.326
.088
.340*
.344
.330*
Chlorophyll
a .251*
.268*
.226
.127
.240
.214
TOC
.577*
.614*
.564*
.582*
.576*
.562*
*Signif icant
beyond the 0
.001 level
(99.9 percent) by the F-test
•
Table 2
contains the
standard error (S.E.)
and Cp/P values
for the
untransformed
(MSS versus
variable) and
the "best"
transformed
regressions. (
"Best"
in this case
means highest R2 .)
12
-------
TABLE 2. COMPARISON OF STANDARD ERRORS AND Cp/P RATIOS FOR
UNTRANSFORMED AND TRANSFORMED VARIABLES
Untransformed Transformed
Cp/P S.E. Best Transformation Cp/P S.E.
Secchi
0.
62
1.
223 m
MSS
vs. log Secchi
0.
,55
1.294 m
PH
0.
60
0.
650
MSS
Ratio vs. pH
0.
88
0.634
Alkalinity
0.
68
56
.126 peq/1
MSS
Chromaticity
vs.
Alkalinity
0.
,82
55.825 peq/1
TSS
0.
14
0.
972 ppm
MSS
vs. log TSS
-0.
,04
2.080 ppm
Conductivity
0.
90
15
.280 pmhos
MSS
Ratio vs.
Conductivity
0.
.88
13.948 umhos
Phaeophyton
0.
96
1.
377 mg/m3
log
MSS vs. Phaeophyton
1.
,01
1.362 mg/m3
Chlorophyll a
0.
24
0.
416 mg/m3
MSS
Chromaticity
vs.
Chlorophyll a
-1,
.55
0.411 mg/m3
TOC
0.
62
1.
222 mg/1
MSS
Chromoticity
vs.
TOC
0,
.73
1.180 mq/1
The summary statistics for the dependent variable data are presented in Table 3.
Daniel and Woods criteria for relating the measurement error to the variance of
measured values is also presented. Daniel and Woods ratio values of less than 3.16
indicate noisy measurements relative to the observed values. The coefficient of
variation is also presented. This statistic varies between zero and one, represent-
ing the magnitude of measurement error relative to the mean observation. The error
analysis is missing for Secchi depth because replicate data were not provided. It is
estimated that the measurement error for Secchi depth is between 0.5 and 1.0 m.
Table 4 presents the correlation (R) matrix between independent and dependent
variables. In addition to the multispectral scanner radiances for each channel,
correlations were determined between pH and alkalinity and the remaining water
quality variables.
Table 5 compares the results of the fall study with the current (spring) study
for untransformed data sets. The best subset of MSS channels is also listed.
Conductivity and total organic carbon tests were not made in the fall study, and
dissolved organic carbon (DOC) tests were not performed on the spring data.
13
-------
TABLE 3. DEPENDENT VARIABLE SUMMARY STATISTICS
Mean Daniel Coefficient
Smallest Largest Laboratory Standard and of
Variable Mean Value Value Units Precision* Deviation Wood's Variation
(y) (y) (y min) (y max) (arep.' (ff/°rep.* {arep./y)
Secchi
3.74
1.00
13.00
m
not avail.
1.87
-
-
pH
5.70
4.40
7.20
none
0.02
0.85
42.5
.004
Alkalinity
38.92
o
GO
•
PO
1
245.10
peq/1
3.61
66.94
18.54
.093
TSS
1.45
0.1 0
4.20
ppm
0.34
1.06
3.12
.234
Conductivity
32.78
17.50
159.70
ymhos
0.55
19.81
36.01
.017
Phaeophyton
1.283
0.045
10.147
mg/m^
0.209
1.562
7.27
.163
Chlorophyll a^
0.345
0.000
2.162
mg/m^
0.100
0.470
4.70
.290
TOC
5.8
2.0
10.0
mg/1
0.2
1.8
9.0
.034
•The sum of the standard deviations for the three-replicates divided by the number of sampling
stations.
TABLE 4. CORRELATION (R) MATRIX BETWEEN INDEPENDENT AND DEPENDENT VARIABLES
Independent
Variable
Dependent Variable
MSS Secchi
pH Alkalinity
TSS Conductivity Phaeophyton Chlorophyll a
TOC
1
0.120
-0.1 12
-0.128
-0.053
-0.138
0.002
0.066
-0.135
2
0.106
-0.1 14
-0.1 40
-0.059
-0.132
-0.074
0.259
-0.172
3
0.119
-0.114
-0.110
-0.075
-0.105
-0.101
0.310
-0.192
4
0.1 21
-0.043
-0.051
-0.072
-0.072
-0.104
0.333
-0.204
5
0.065
0.024
0.009
-0.041
-0.003
-0.089
0.356
-0.153
6
-0.018
0.043
0.024
0.007
0.051
-0.076
0.384
-0.060
7
-0.063
0.037
0.041
0.048
0.1 35
-0.052
0.411
0.015
8
-0.078
-0.024
0.027
0.074
0.217
-0.037
0.396
0.059
9
-0.092
-0.020
0.050
0.109
0.300
-0.022
0.363
0.102
10
-0.138
-0.031
0.046
0.143
0.377
-0.029
0.289
0.167
Secchi
-0.081
-0.110
TSS
0.279
0.322
Conductivity
0.517
0.716
Phaeophyton
-0.045
0.078
Chlorophyll a
0.444
0.546
TOC
0.1 77
0.1 24
1 4
-------
TABLE 5. BEST SUBSET MULTIPLE LINEAR REGRESSION COMPARISON
Fall Spring
R2
S.E.
MSS
Best Subset
ChanneIs
R2
S.E.
MSS
Best Subset
Channels
Secchi
0.796*
1.436 m
2,4,5,6,10
0.622*
1.223 m
1,2,4,5,6,9
pH
0.383*
0.593
1,2,3,5,9
0.462*
0.650
2,3,5,6
Alkalinity
0.309*
39.378 jieq/1
1,2,3,5,6,9
0.366*
56.126 weq/1
2,3,5,6,7
TSS
0.494*
0.935 ppm
2,7
0.225
0.972 ppm
1,2,3,5
Chlorophyll a
0.424*
1.649 mg/m3
2,7,10
0.251*
0.416 mg/m3
1,3
Conductivity
-
-
-
0.452*
15.28 ymhos
4,5,6,10
Phaeophyton
0.502*
0.820 mg/m3
1,5,7
0.329
1.377 mg/m3
2,3,5,6,7,8,10
DOC
0.807*
0.824 mg/m3
1,2,4,7,9
-
-
-
TOC
0.577*
1.22 mg/1
3,4,5,7,9
~Significant regression beyond the 0.001
(99.9 percent)
level by the
F-test.
1 5
-------
Scatter plots of predicted versus observed values of the dependent variables for
the best regressions from transformed variables (see Tables 1 and 2) are presented in
Figures 2 through 9. Predicted values refer to those predicted for each point
through the multiple linear regression model. Observed values refer to the
laboratory measurements for the same points.
DISCUSSION
It is evident from the preceding tables and plots that linear relationships do
exist between multispectral scanner data and a variety of acid deposition-related
water quality parameters. It is also clear that these relationships are not as
precise as we would wish them to be for predictive purposes. There is a great deal
of variance present in the data. Techniques to reduce this variance must be found
before an operational monitoring procedure is established. However, sufficiently
significant relationships have been found between the scanner data and many of the
water quality variables to encourage further research in this area.
Improvements in the relationships found may be a question of timing. It is
interesting to note, that the only improvements in R2 from the fall mission to the
spring mission were related to pH and alkalinity, the only two variables having
increased ranges from fall to spring. The effects of narrow ranges in the data may
be seen in Tables 1, 2, and 3. Total suspended solids, phaeophyton, and chlorophyll
a all had high variances with respect to the magnitude of the measurements. As one
might expect, the R2 values (Table 1) were correspondingly low. Spring freshet does
not appear to be the optimum time to measure these variables, due to their low con-
centrations. Substantial improvement may be made by timing the overflight and water
sampling to coincide with maximum productivity. This should allow the lakes to react
to differences in pH and alkalinity caused by acid deposition and increase the range
of related optical effects.
16
-------
a Log 5ECCHI
OT
O
co
a
r-
O
f ) to I
LJ-I o
(_)
H—-*
a
in
a
QC a-
Q_ o
cn
a
ru
o
o -
o
0 —p 11 111111 11 l ll 11 n 111111 l 11 1111111111 i l| 1111111 11| 111111 H l| I M n 1111| 111 111111 11 11 111111[ 11111 i 111| n 11111 11 | 1111 nTTTj
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1 1.2
OBSERVED
29-0EC-82
Figure 2. Predicted versus observed log^g Secchi depth (original values in m) .
17
-------
CO -|
a
UJ
I—
CJ
I—I
~
LU
QC
Q_
OBSERVED
29-0EC-B2
Figure 3. Predicted versus observed pH.
18
-------
RLKRLINITY
|rinin>i|iimnii|iMMiM»|MHinn|riifniinmnim|iiiftM»i|nnn»m'i
-40 -20 0 20 40 60 80 100 120 140 160 180 200 220 240
OBSERVED
29-DEC-82
Figure 4. Predicted versus observed alkalinity (peq/1).
19
-------
Log T.S.S.
i i r i i i i i i [ i i i M i i i > | i i i i i i i i i | ii i i ii i i i | i i i i i ii i i'| i i i i i M n |
11 i i i i i i i i | i i i i i i i i i | i i i i i i i i i
-1.0
-0.8 -0.6
-0.4 -0.2 0.0 0.2
OBSERVED
0.6
1.0
29-DEC-82
Figure 5. Predicted versus observed log^g total suspended solids
(original values in ppm).
20
-------
CONDUCTIVITY
pinnni |inmm| i
60 70 80 90 100
OBSERVED
110 120 130 140
i|iimiiiih
150 160
39-Dcc-ea
Figure 6. Predicted versus observed conductivity (yrahos).
21
-------
PHREOPHTTON
t 111 n 111111 1111 111 [ 1111111111111111111111111111111111111 ri 1111 ri 1111111 1111111111 11 1111 11 m 111 i 11 11111' 111111
01 2 3 4 5 6 7 8 9 10 11
OBSERVED
?9-0EC-82
Figure 7. Predicted versus observed phaeophyton (mg/m3).
22
-------
CHLOROPHTLLa
i t 11 ri i m 11 n 111 n n i n 11 n 11 |i 111H111| 111111111111 il 111M [ 11111 n r 1111111 n 111111 n n n |ll 111 n 11| M H H111| n'' i' i n ["' i' 11H111' 111 n * | n i n i ii f| i n 1111" |" 111" 111
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1. 1 1.2 1.3 1.4 1.5 1.6 1.7 1.8
OBSERVED
29-0EC-82
Figure 8. Predicted versus observed chlorophyll a (mg/ra3).
23
-------
o _
T.O.C.
II I II II 11 | I M I II M I |'M I I I I Ml | I I II I I I M | I I I Ml T M ]» I I II II t I | » I M M I II |» M M I M I | I M
OBSERVED6
29-DEC-B2
Figure 9. Predicted versus observed total organic carbon (mg/1).
24
-------
The relationships may also be improved by a careful analysis of the points to be
included in the regression. This may be illustrated by the scatter plot of conduc-
tivity (Figure 6). The regression is dominated by only one point with a relatively
high value. The decision whether or not to delete that point from the analysis has a
relatively large effect on the results. For the other parameters, careful "interior
analysis" of the data points could considerably improve the model. The current, as
well as future, data sets should be examined, with particular attention to robust
regression estimates (Huber, 1981), to determine the optimum set of points to be
included in the regression computations.
Despite some of the problems revealed by this study, and the one preceding it,
the multispectral, regression-based approach appears to show promise as an eventual
monitoring tool. Table 4 clearly demonstrates that univariate methods will not be
adequate. It is not the absolute response, but the pattern of response which will
reveal the relationship between the optical properties of water and its constituents.
Moreover, the value of the synoptic view must not be minimized. An image-based
geographic information system of acid precipitation effects would be of great value
in acid deposition assessment and monitoring.
Considering the potential for long-term gain, it is recommended that the U.S.
Environmental Protection Agency proceed with the next phase of this project. A mid-
summer flight should be conducted consisting of a high-altitude U-2 platform and the
Thematic Mapper Simulator. Multivariate analysis should be conducted by an inter-
disciplinary team consisting of a remote sensing specialist, a statistician, and a
limnologist to fully exploit the techniques described in this study.
25
-------
REFERENCES
Alfoldi, T. T., and J. C. Munday. 1978. Water Quality Analysis by Digital
Chromaticity Mapping of Landsat Data. Canadian Journal of Remote Sensing.
4:108-126.
Baalsrud, K., T. S. Traaen, M. Laake, and G. Raddum. 1976. Acid Precipitation:
Some Hydrobiological Changes. AMBIO. 5(5-6):224-227.
Dixon, W. J. (ed) 1977. BMDP-77: Biomedical Computer Programs - P Series.
University of California Press, Berkeley, California.
Dochinger, L. S., and T. A. Seliga. (eds). 1976. Acid Precipitation and the Forest
Ecosystem. Unpublished workshop report. USDA Forest Service, Ohio State
University, and the National Science Foundation. 38 pp.
Galloway, J. N., G. E. Likens, and E. S. Edgerton. 1976. Acid Precipitation in the
Northeastern U.S.: pH and Acidity. Science. Vol 194:722-724.
Huber, P. J. 1981. Robust Statistics. John Wiley and Sons, Inc. 308 pp.
Likens, G. E. 1976. Acid Precipitation. C & EN Special Report. :29-44.
Snedecor, G. W. and W. G. Cochran. 1980. Statistical Methods, Seventh Edition.
Iowa State University Press. 507 pp.
Whitlock, C. H. 1977. Fundamental Analysis of the Linear Multiple Regression
Technique for Quantification of Water Quality Parameters From Remote Sensing
Data. PhD Dissertation, Old Dominion University. University Microfilms
International, Ann Arbor, MI. 176 pp.
Whitlock, C. H., C. Y. Kuo, and S. R. LeCroy. 1982. Criteria for the Use of
Regression Analysis for Remote Sensing of Sediment and Pollutants. Remote
Sensing of Environment. (12):151-168.
Witzig, A. S. and C. A. Whitehurst. 1981. Literature Review of the Current Use and
Technology of MSS Digital Data for Lake Trophic Classification. Technical
Papers of the American Society of Photogrammetry: ASP-ACSM Fall Technical
Meeting. ASP, Falls Church, VA. :l-20.
26
-------
APPENDIX
MULTISPECTRAL SCANNER
AND
ANALYSIS SYSTEM CHARACTERISTICS
27
-------
MSS WAVELENGTH BANDS
Wavelength
Channel Band Color/Spectrum
1 0.38-0.42pm New Ultraviolet
2 0.42-0.45pm Blue
3 0.45-0.50pm Blue
4 0.50-0.55pm Green
5 0.55-0.60pm Green
6 0.60-0.65pm Red
7 0.65-0.70pm Red
8 0.70-0.79pm Near Infrared
9 0.80-0.89pm Near Infrared
10 0.92-1.10pm Near Infrared
11 8.00-14.00pm Thermal Infrared
28
-------
MULTISPECTRAL SCANNER
IMAGING CHARACTERISTICS (SIMPLIFIED)
I0I41 FKU) Ot Ytm
a
DIRECTION OF SCAN
GROUND
RESOLUTION
ELEMENT
SCAN LINE
29
-------
MULTISPECTRAL SCANNER DATA PROCESSING
Processing and analysis of multispectral scanner digital data is accomplished on
the EMSL-LV Data Analysis System (DAS). The DAS (see attached figure) consists of a
sensor tape playback unit, a high-speed digital computer, an interactive color TV
display system, and an off-line film recorder. The functional steps in processing
MSS digital data are summarized in the following diagram (MSS Data Processing -
Simplified).
The MSS sensor data must be preprocessed before image analysis can be initiated.
Preprocessing entails converting the data into a digital format compatible with the
DAS processing software. In addition, the data can be calibrated and geometric cor-
rections applied to rectify scan line distortions.
Primary data analysis and processing begin when software programs contained in
the Image Analysis block are implemented. A variety of enhancement and classifi-
cation procedures can be utilized by the image analyst. Single or multi-channel
images, as well as enhanced and classified images can be viewed and analyzed on the
color TV display. In addition, statistical parameters computed from the data can be
extracted for detailed analysis. Hard copy records that can be generated at this
stage in data processing include black and white film images, electrostatic paper
plots from each MSS channel, and statistical printouts.
Following Image Analysis, further processing is required before the final output
product can be produced. Classified data is color-coded and enlargement/reduction
factors are computed by the analyst. Optional programs in this phase include a geo-
graphic rectification routine to match the image to selected UTM (Universal Trans-
verse Mercator) map projection scales, and inputing image annotations.
Final output products from the MSS processing includes hardcopy color and/or
black and white film images (positive or negative), and electrostatic paper plots.
The classified and/or enhanced images can also be viewed in color and black and white
on the TV display. Again, statistics computed from the data can also be viewed on
the display. Class statistics such as pixels per class, acres per class, or square
miles per class are output in the form of tabulated data.
30
-------
MSS DATA PROCESSING (SIMPLIFIED)
TV DISPLAY
• COLOR / BtW IMAGE
• STATISTICS
HARD COPY IMAGE
• COLOR / B & W FILM
• ELECTROSTATIC PLOT
DATA
PRE-PROCESSING
MSS
SENSOR
TAPE
LINE PRINTER
• STATISTICS
TV DISPLAY
• IMAGE
• STATISTICS
HARD COPY
• B 4 W FILM
• ELECTROSTATIC PLOTS
• STATISTICS
POST ANALYSIS
PROCESSING
• COLOR ASSIGNMENT
• ANNOTATION OVERLAYS
• RECTIFY TO MAP BASE
IMAGE ANALYSIS
• image enhancement
• CLASSIFICATION
31
------- |