TVA
EPA
Tennessee
Valley
Authority
Office of Natural
Resources
Norris TN 37828
TVA/ONR/ARP-81/5
United States
Environmental Protection
Agency
Office of Environmental
Engineering and Technology
Washington DC 20460
-X
EPA-600/7-81-113
July 1981
Research and Development
Remote Sensing of
Sulfur Dioxide
Effects on
Vegetation—Final
Report
Volume I.
Summary
-------
RESEARCH REPORTING SERIES
Research reports of the Office of Research and Development, U.S. Environmental
Protection Agency, have been grouped into nine series. These nine broad cate-
gories were established to facilitate further development and application of en-
vironmental technology. Elimination of traditional grouping was consciously
planned to foster technology transfer and a maximum interface in related fields.
The nine series are:
1. Environmental Health Effects Research
2. Environmental Protection Technology
3. Ecological Research
4. Environmental Monitoring
5. Socioeconomic Environmental Studies
6. Scientific and Technical Assessment Reports (STAR)
7. Interagency Energy-Environment Research and Development
8. "Special" Reports
9. Miscellaneous Reports
This report has been assigned to the INTERAGENCY ENERGY-ENVIRONMENT
RESEARCH AND DEVELOPMENT series. Reports in this series result from the
effort funded under the 17-agency Federal Energy/Environment Research and
Development Program. These studies relate to EPA's mission to protect the public
health and welfare from adverse effects of pollutants associated with energy sys-
tems. The goal of the Program is to assure the rapid development of domestic
energy supplies in an environmentally-compatible manner by providing the nec-
essary environmental data and control technology. Investigations include analy-
ses of the transport of energy-related pollutants and their health and ecological
effects; assessments of, and development of, control technologies for energy
systems; and integrated assessments of a wide range of energy-related environ-
mental issues.
This document is available to the public through the National Technical Informa-
tion Service, Springfield, Virginia 22161.
-------
TVA/ONR/ARP-81/5
EPA-600/7-81-113
July 1981
REMOTE SENSING OF SULFUR DIOXIDE EFFECTS ON VEGETATION
FINAL REPORT
VOLUME I - SUMMARY
by
C. Daniel Sapp
Office of Natural Resources
Tennessee Valley Authority
Chattanooga, Tennessee 37401
Interagency Agreement EPA-IAG-D8-E721-DJ
Project No. E-AP 80 BDJ
Program Element No. INE 625C
Project Officer
James Stemmle
U.S. Environmental Protection Agency
401 M Street, SW.
Washington, DC 20460
Prepared for
OFFICE OF ENERGY, MINERALS, AND INDUSTRY
OFFICE OF RESEARCH AND DEVELOPMENT
U.S. ENVIRONMENTAL PROTECTION AGENCY
WASHINGTON, DC 20460
fl,*?, l?nviron--:=ntal Protection Agency
'''"' ''
-------
DISCLAIMER
This report was prepared by the Tennessee Valley Authority and has
been reviewed by the Office of Energy, Minerals, and Industry, U.S.
Environmental Protection Agency, and approved for publication. Approval
does not signify that the contents necessarily reflect the views and
policies of the Tennessee Valley Authority or the U.S. Environmental
Protection Agency, nor does mention of trade names or commercial products
constitute endorsement or recommendation for use.
11
-------
ABSTRACT
Three techniques for detecting and mapping sulfur dioxide (802) effects
on the foliage of sensitive crops and trees near large, coal-fired power
plants were tested and evaluated. These techniques were spectroradiometry,
photometric analysis of aerial photographs, and computer analysis of
airborne multispectral scanner data.
Spectroradiometry is a useful, ground-based technique for measuring
the changes in reflectance that accompany exposure of sensitive crops to
802- Photometric analysis of aerial color-infrared photographs has some
practical advantages for measuring the reflectances of forest species or
for synoptic point-sampling of extensive areas; these tasks cannot be done
effectively by field crews. The relationships among reflectance, foliar
injury, and yield of crops are complex and are affected by many extraneous
variables such as canopy density. The 862 effects are easier to detect on
winter wheat than on soybeans, but in either case they cannot be con-
sistently detected by airborne remote sensors except under near-ideal con-
ditions when the injury is moderate to severe. Airborne multispectral
scanner data covering affected soybean fields were analyzed using three
computer-assisted procedures: unsupervised, supervised, and pseudosuper-
vised; the last method provided the best results. Landsat imagery was
also investigated, but the foliar effects of 802 were too subtle to
detect from orbit.
This report was submitted by the Tennessee Valley Authority, Office
of Natural Resources, in fulfillment of Energy Accomplishment Plan 80 BDJ
under terms of Interagency Agreement EPA-IAG-D8-E721-DJ with the Environ-
mental Protection Agency. Work was completed as of December 1980.
111
-------
CONTENTS
Abstract
Figures v
Tables | ' v
Acknowledgment vi
1. Introduction 1
Background 1
Remote Sensing 1
Spectroradiometry 1
Reflectance Properties and Vegetative Stress. . . 2
Aerial Photography 3
Airborne Multispectral Scanners 3
Purpose and Objectives 4
Scope 4
Hypothesis 6
2. Conclusions and Recommendations 7
Conclusions 7
Recommendations 9
3. Results 11
Laboratory Spectroradiometry 11
General 11
Soybeans 11
Winter Wheat 12
Field Spectroradiometry 13
General 13
Soybeans 14
Winter Wheat 15
Photometric Analysis of Aerial Photographs 15
General 15
Overflights 16
Colbert Site Test 16
Johnsonville Site Test 17
Analysis of Multispectral Scanner Data 18
General 18
Ground Truth 19
Colbert Site Test 19
Shawnee Site Test 19
Optimal Flying Heights 19
Optimal MSS Channels 19
MSS Data Classification 20
Enhancement of Patterns of S02 Effects
Within Fields 22
IV
-------
FIGURES
Number
Western Part of Tennessee Valley Showing Four
Steam Plants
Number
3
4
5
6
TABLES
Simple Correlation Coefficients for Single-Band
Reflectance and Foliar Injury to Soybeans. . .
Simple Correlation Coefficients for Single-Band
Reflectance and Foliar Injury to Winter Wheat
Simple Correlation Coefficients for Single-Band
Reflectance and Foliar Injury to Soybean Plot
Simple Correlation Coefficients for Single-Band
Reflectance and Necrosis in Winter Wheat Plot. .
Optimal MSS Channels for Detecting and Classifying
S02-Affected Soybean Fields Near Colbert in 1977
Optimal MSS Channels for Detecting and Classifying
S02~Affected Soybean Fields Near Shawnee in 1978
Errors Resulting From Procedures for Detecting
and Classifying S02 Effects on Soybeans ....
Page
. 12
. 13
. 14
. 15
. 20
. 21
23
-------
ACKNOWLEDGMENT
This work was conducted as part of the Federal Interagency Energy/
Environment Research and Development Program with funds administered
through the Environmental Protection Agency (EPA Contract No. EPA-IAG-
D8-E721-DJ, TVA Contract No. TV-41967A).
The EPA Project Officer for this research project is James Stemmle,
401 M Street, SW., Washington, DC. His contribution to the direction
of the research and his constructive review of the reported results are
appreciated. The TVA Project Director is Herbert C. Jones, Supervisor,
Air Quality Research Section, Air Resources Program, River Oaks Building,
Muscle Shoals, Alabama.
VI
-------
SECTION 1
INTRODUCTION
BACKGROUND
The effects on vegetation of sulfur dioxide (S02) emissions from
large, coal-fired power plants have been recognized as a potential prob-
lem for several decades. The traditional method for measuring the
effects in the field involves observations of injury to S02~sensitive
indicator species such as ragweed and blackberry. Records from fixed
S02 monitoring stations are also used to determine the spatial charac-
teristics of plume contact with the ground.
Some problems exist with the traditional approach to surveying and
identifying S02 effects. The monitoring network is often inadequate for
mapping the limits of the effects, and botanical surveillance is usually
restricted to readily accessible areas because of the constraints of
time. Highly trained biologists are needed to identify and record the
symptoms of injury to foliage.
REMOTE SENSING
Spectroradiometry
Remote sensing--the detection and measurement of characteristics of
phenomena from a distance, without direct contact—can assist those
engaged in field surveillance of S02 effects on crops and trees. The
technique provides a permanent record on film or magnetic tape. An
instrumented aircraft can continuously cover extensive areas in a matter
of hours.
The state of the art of remote sensing requires that ground
truth—field observations--be gathered to support the analysis of the
-------
-2-
remotely-sensed data. Preliminary but detailed information should be
gathered concerning the differences in spectral reflectance between the
objects of interest (in our study, affected foliage) and the background
(unaffected foliage). Spectral measurements may be made in the laboratory
or in the field, or in both places. Such measurements would allow the
selection of appropriate sensor configurations, films, filters, and air-
borne scanner channels and bandwidths, thus improving the chances of
successful detection of S02 effects.
There are at least two methods for making spectral reflectance
measurements. The traditional method is to make measurements at discrete
points in the field with a portable spectroradiometer. An indirect
method which may be more efficient in some instances is to make point
measurements of the optical density of aerial photographs of S02-affected
areas, and then convert these densities to reflectances. The latter
method, called photometry, entails a complex calibration of the photo-
graphs before the conversion to reflectance can be made. In this study
the investigator used both methods. Moreover, field plots of affected
plants were used to bridge the wide gap between the laboratory and the
uncontrolled environment of crops and trees in the vicinity of the power
plant source.
Reflectance Properties and Vegetative Stress
For detecting the effects of air pollution on vegetation, the
investigator selected an appropriate region of the electromagnetic
spectrum spanning the visible and near-infrared wavelengths. The far-
infrared (thermal) wavelengths were also used for measurements. These
selections were based partly on the inherent properties of the spectrum
and partly on the capabilities and availabilities of remote sensors.
-------
-3-
The actual change in reflectance of a species or variety of plant
under stress from SOg or some other agent is not easily predicted.
Visible reflectance generally increases with stress, but the response of
reflectance in the near-infrared is variable, although it eventually
decreases in advanced senescence. In remote sensing studies, the stress-
causing agent cannot usually be identified without ground truth. Foliar
markings, which indicate the identity of the agent, cannot be resolved
from the distances or altitudes at which the sensor is operated. However,
clusters of stressed plants can often be distinguished from a background
of normal plants.
Aerial Photography
Color and color-infrared photography show promise for detecting
vegetative stress and so were used in airborne cameras to record the
patterns of S02 injury to sensitive crops and trees. Several flying
heights were used; they ranged from 500 m above ground level (AGL) up to
almost 4000 m AGL on various missions. It is generally most efficient to
fly at the highest altitude that enables the interpreter to detect the
phenomena of interest, because more area is photographed per unit of time.
Airborne Multispectral Scanners
The multispectral scanner is at the frontier of remote-sensor tech-
nology. Digital processing of multispectral scanner data is advancing
our capabilities to reduce the output of the scanner to understandable
form. Digital classification and enhancement of detail in the images
help the interpreter detect and measure the patterns of most interest
to him. During the course of this project, an 11-channel multispectral
scanner was employed on several occasions to detect and map SQ% effects,
and two digital image processing systems were used to process the data.
-------
-4-
PURPOSE AND OBJECTIVES
The purpose of this 5-year study was to analyze and evaluate remote-
sensing techniques to detect and map the effects of SO^ emissions from
large, coal-burning power plants on the foliage of sensitive crops and
trees. The objectives were to test, refine, and develop ground-based,
airborne, and satellite-borne remote-sensor instrumentation for this
purpose.
SCOPE
The scope of the project included four coal-fired power plant sites
in the Tennessee Valley region (Figure 1), several experimental plots,
and several species of vegetation. Laboratory-based spectroradiometric
experiments were performed on soybeans, wheat, and cotton. Techniques
included stereoscopic photo interpretation, photometric analysis of
aerial photographs, and digital image analysis. Ground truth was
acquired by experienced surveillance biologists who observed affected
vegetation in the greenhouse, in experimental plots, and in the field
near the power plants. Investigations included the following S02~
sensitive crop and tree species:
Common Name Scientific Name
1. Soybeans Glycine max (L.) Merr.
2. Winter wheat Triticum aestivum
3. Cotton Gossypium hirsatim
4. Virginia pine Pinus virginiana
5. Loblolly pine Pinus taeda
6. White pine Pinus strobus
7. Shortleaf pine Pinus echinata
8. Hickory Garya sp.
9. Northern red oak Quercus rubra (L.)
10. Southern red oak Quercus falcata Michx.
As the project progressed, its scope had to be narrowed to exclude
the hardwoods (hickory and oaks), because S02~affected stands of these
-------
—5—
^aducah
SHAWNEE ^
STEAM PLANT
WIDOWS CREEK
STEAM PLANT
Vs \
.BOUNDARY OF TENNESSEE RIVER WATERSHED
10 0
SCALE
50
100km
Figure 1. Western part of Tennessee Valley showing four steam plants.
-------
-6-
trees were never encountered. Some affected pine stands were found near
the Widows Creek Steam Plant and were studied, but the injury was light
and discontinuous and could not be consistently detected. Affected
wheat and cotton fields were never found, so S0% injury to these species
was induced through the use of experimental plots. SOg-affected soybeans
were studied intensively and extensively.
HYPOTHESIS
The hypothesis of the research performed during this project was
that there is a relationship between the reflectance of the plants and
levels of injury to their foliage from sulfur dioxide. Such a relation-
ship would form a theoretical basis supporting the use of remote sensors
to detect and map the distribution of SC^-affected plants.
-------
-7-
SECTION 2
CONCLUSIONS AND RECOMMENDATIONS
CONCLUSIONS
Spectroradiometry is a useful technique for measuring the changes in
visible and near-infrared reflectance that accompany relatively severe,
S02~induced injury to the foliage of sensitive row crops. The remote-
sensor technique is not practical for measuring reflectances of mature
trees because of the difficulty in scanning such large objects from
specific angles. Spectroradiometry provides valuable information for
planning overflights so that the remote-sensor instruments can be tuned
to detect and discriminate the S02~induced stress.
Laboratory scanning experiments indicate that changes in the total
reflectance spectrum of soybeans accompany necrosis but not chlorosis.
The ratio of near-infrared to red (IR/red) reflectance correlates sig-
nificantly (a=.05) with necrosis of the foliage of these plants. In
scans of winter wheat, the total visible spectrum, as well as the single
bands, green and red, shows close relationships with foliar injury.
(Laboratory-based IR scans of wheat were not made.)
Statistical analysis of scans of experimental plots of soybeans and
wheat fails to verify laboratory findings. No relationship is apparent
when reflectance and observed injury to soybeans are compared, but a
relationship is evident between the two variables for wheat. The total
reflectance spectrum (visible plus near-infrared), as well as the indi-
vidual (green, red, and near-infrared) bands, is associated with S02
injury.
Photometric analysis of aerial color-infrared photographs of 803-
affected soybean fields shows no relationship between single-band
-------
-8-
reflectance and foliar injury. The spectral bands included green, red,
and near-infrared wavelengths.
Analysis of airborne multispectral scanner data indicates that S02-
affected soybean fields can be distinguished from unaffected soybean
fields when conditions are nearly ideal. Such conditions are defined
generally as a continuous foliage canopy, mature stage of growth, and
S02 effects that are moderate to severe. Comparison of three data
classification procedures shows that a pseudosupervised procedure pro-
vides greater accuracy than either supervised or unsupervised. The
pseudosupervised procedure can distinguish moderately to severely affected
soybeans from unaffected soybeans with errors ranging from 11 to 24 percent.
Experience with three aircraft altitudes for acquiring scanner data
indicates that 1800 m AGL flying height is superior to 500 and 3660 m AGL
for detecting moderate to severe chlorosis symptoms on the foliage of row
crops. Light chlorosis may be undetectable by an airborne scanner or
camera regardless of platform altitude. Such effects are certainly
undetectable by orbiting sensors, such as those aboard Landsat. This
study indicates that unless the 862 effects are severe enough to result
in necrosis, they will not be detectable from any altitude greater than
150 m AGL by remote sensors. Even when necrosis exists, detection may be
possible only under nearly ideal conditions.
The hypothesis that there is a relationship between reflectance and
observed levels of S02~induced injury to sensitive plants is neither
accepted nor rejected in an unequivocal sense. The relationship is
apparent when field conditions are nearly ideal for detection and the
injury to foliage is relatively severe. The fact that the association
(1) was generally apparent in data from controlled laboratory experiments;
-------
-9-
(2) was sometimes verifiable in data from semicontrolled experimental
plots; and (3) was seldom found in the uncontrolled data from S02~affected
soybean fields located downwind from power plants suggests that extraneous
variables were affecting the results. These variables included, but were
not limited to, stage of growth, soil moisture, terrain slope, canopy
density, farming practices, level of chlorosis or necrosis, herbicide
effects, weeds, and variety of plant.
RECOMMENDATIONS
Because of the complexity of the relationship between reflectance
and foliar injury from S02, spectroradiometry should be an integral part
of planning for remote-sensor overflights of affected crops. The scan-
ning technique is useful for measuring the spectral differences between
target and background so that success in detecting stressed vegetation
can be predicted. The spectroradiometer is also a valuable laboratory
instrument for quantifying levels of foliar injury.
Photometric analysis of aerial photographs is a potential alter-
native to field-based scanning with a spectroradiometer. It would be
advantageous if reflectances need to be sampled at many points over
a large area. Our negative findings were probably a result of
extraneous variables (e.g., weeds) controlling reflectance.
The color-infrared film type should normally be used instead of
conventional color film because the infrared shows the patterns of
stress better and provides superior penetration of atmospheric haze.
Airborne multispectral scanning should be employed, if appropriate,
to detect and map S02~related injury to row crops whenever the foliar
symptoms are relatively severe, consisting primarily of necrosis, and
the canopy is continuous, dense, and weed-free. These conditions are
-------
-10-
quite restrictive because S02 effects in the field are usually subtle
and consist mainly of chlorosis. These light effects cannot be
detected consistently with currently available airborne or spaceborne
remote sensors.
-------
-11-
SECTION 3
RESULTS
LABORATORY SPECTRORADIOMETRY
General
Uniform groups of soybeans and winter wheat were grown in a green-
house, exposed to controlled doses of S02 in a laboratory exposure
chamber, observed for foliar effects, and scanned with a spectroradiom-
eter. The resulting data were statistically analyzed to determine the
spectral changes plant foliage undergoes when it is affected by S02.
Soybeans
Foliar injury (chlorosis and necrosis) was divided into traditional
classes: unaffected (0 percent); light (1-10 percent); moderate (11-25
percent); severe (26-50 percent); and very severe (>50 percent). Mean
reflectance curves for each class were computed. The areas under the
curves were examined with respect to foliar injury. Shifts in narrow
bands of blue, green, red, and near-infrared (IR) reflectance were also
examined to determine whether statistically significant differences in
mean reflectance existed between and among injury classes.
Some significant results were obtained by comparing the mean reflec-
tances of injury classes through an analysis of variance statistical
procedure (ANOVA). Significant (a=.05) differences in red- and green-
band reflectance were found between unaffected soybeans and affected soy-
beans having greater than 10 percent necrosis. A significant (a=.05)
difference was also found in the ratio of IR to red reflectance (IR/red)
between the unaffected and affected soybeans.
-------
-12-
The strength of the relationships between reflectance and foliar
injury is indicated by the correlation coefficients (r) listed in Table 1
TABLE 1. SIMPLE CORRELATION COEFFICIENTS (r) FOR SINGLE-BAND
REFLECTANCE AND FOLIAR INJURY TO SOYBEANS
Symptom
Chlorosis
Necrosis
Blue
+0.20
+0.89
Green
+0.72
+0.92
Reflectance
Red
+0.36
+0.98
IR
-0.10
0.00
IR/Red
-0.32
-0.94
Underlined coefficients are significant, a=.05.
I
The table warrants a close examination. Except for green reflec-
tance, the r coefficients for chlorosis are below 0.50; for necrosis, the
visible reflectance bands, especially red, correlate much higher. IR
reflectance alone showed no significant relationship to either symptom.
The ratio of IR to red reflectance has been found to be an indirect
indicator of stress in foliage, according to other studies. However,
the correlation of -0.32 between chlorosis and the IR/red ratio does not
indicate a strong relationship. On the other hand, there is a strong
relationship (r = -0.94) between necrosis and the ratio. With increasing
necrosis, the IR component of the ratio decreases and the red component
increases, thus bringing the ratio value down, closer to unity.
Winter Wheat
Foliar injury (chlorosis and necrosis) was divided into traditional
classes: unaffected (0 percent); light (1-10 percent); moderate (11-25
percent); severe (26-50 percent); and very severe (>50 percent). The
range of foliar symptoms was broad, consisting primarily of necrosis.
-------
-13-
As with soybeans, mean reflectance curves were computed. The areas under
the curves were examined with respect to foliar injury, as were changes
in visible reflectance for particular wavelengths (blue, green, and red).
IR reflectances of wheat were not measured.
The area under the reflectance curves increased with increasing
necrosis (r2 = 0.72) and chlorosis (r2 = 0.85). The increase in red
reflectance was greatest at moderate and severe levels of stress. The
increase in green reflectance was greatest at light levels of stress.
Statistical analysis of the visible reflectance curves also included
single bands. Simple correlation coefficients were computed to assess
the relationship between injury and reflectance (Table 2). They range
between +0.73 and +0.90.
TABLE 2. SIMPLE CORRELATION COEFFICIENTS (r) FOR SINGLE-BAND
REFLECTANCE AND FOLIAR INJURY TO WINTER WHEAT
Reflectance
Symptom Blue Green Red
Chlorosis +0.83 +0.90 +0.81
Necrosis +0.73 +0.83 +0.85
Underlined coefficients are significant, (Y=.05.
A one-way analysis of variance showed that significant (F-test, a=.05)
differences in reflectance existed among all injury classes of wheat.
FIELD SPECTRORADIOMETRY
General
Experimental 0.40-hectare (ha) plots of soybeans and winter wheat
were grown and subdivided; then the subplots were exposed to several
-------
-14-
controlled doses of SOz and observed systematically to determine foliar
effects. Then the plots were scanned row by row with a van-mounted spec-
troradiometer. The resulting data were statistically analyzed using pro-
cedures similar to those that were applied to the laboratory-based data
described previously.
Soybeans
The affected subplots had higher green and red reflectance, lower IR
reflectance, and a lower IR/red reflectance ratio. The reflectance mea-
surements were grouped into three classes: those for unaffected (control)
soybeans, those for chlorotic soybeans, and those for necrotic soybeans.
Variations in visible reflectance (green and red) correlated significantly
(a=.05) with necrosis but not with chlorosis (Table 3).
TABLE 3. SIMPLE CORRELATION COEFFICIENTS (r) FOR SINGLE-BAND
REFLECTANCE AND FOLIAR INJURY TO SOYBEAN PLOT
Reflectance
Symptom Green Red IR IR/Red
Chlorosis +0.47 +0.57 -0.62 -0.68
Necrosis +0.83 +0.97 -0.65 -0.84
Underlined coefficients are significant, a=.05.
Analysis of variance was used to compare the differences in reflec-
tance between S02-affected and unaffected soybeans. A significant (a=.05)
difference in IR reflectance was found when chlorotic subplots were com-
pared to unaffected subplots. Similar differences in IR/red reflectance
were discovered. Significant differences in red reflectance, IR reflec-
tance, and the ratio were found when necrotic subplots of soybeans were
compared to unaffected subplots.
-------
-15-
Winter Wheat
Four mean reflectance curves were produced by averaging the individual
curves by necrosis class. The classes represented none or light (<10
percent); moderate (11-25 percent); severe (26-50 percent); and very severe
(>50 percent) necrosis. (No chlorosis was found on the wheat.) The three
reflectance bands, green, red, and IR, as well as the IR/red ratio, were
analyzed. The red band, the IR band, and the ratio seemed to be useful
indicators of necrosis (Table 4).
TABLE 4. SIMPLE CORRELATION COEFFICIENTS (r) FOR SINGLE-BAND
REFLECTANCE AND NECROSIS IN WINTER WHEAT PLOT
Reflectance
Symptom Green Red IR IR/Red
Necrosis -0.06 +0.59 -0.53 -0.71
Underlined coefficients are significant, a=.05.
The trends of the relationships were also noteworthy. Red reflec-
tance increased and IR reflectance decreased as the level of necrosis
rose. The IR/red ratio decreased as necrosis increased.
A one-way analysis of variance was used to compare the differences
in reflectance among the four wheat classes. Significant (a=.05)
differences in red reflectance, IR reflectance, and the IR/red ratio,
but not green reflectance, were found.
PHOTOMETRIC ANALYSIS OF AERIAL PHOTOGRAPHS
General
A method of calibrating the color-infrared (CIR) photographs was
used so that the reflectances of vegetation could be obtained from them.
Uncalibrated photographs contain many systematic errors which affect
-------
-16-
exposure and must be accounted for. The errors result from film process-
ing, atmospheric effects, and variation in illumination. The calibration
process, called photometric analysis, included spot measurements of
image density, conversion of densities to exposure values, and finally,
conversion of these exposures to percent reflectance. Photometric
analysis was especially valuable when photographs of a different flight
line, altitude, or date had to be compared.
Once reflectances were obtained for particular point locations,
they were plotted and compared with ground-truth data on SQ% effects to
ascertain whether any relationships existed.
Overflights
Aerial photographic overflights of areas near 4 of TVA's 12 coal-
fired power plants were performed during the 1977 and 1978 growing seasons
when the foliar effects of S02 on vegetation were still visible to ground
observers. Soybean fields near Colbert Steam Plant in northwestern Alabama,
Johnsonville Steam Plant in western Tennessee, and Shawnee Steam Plant in
western Kentucky were photographed, as were soybeans, winter wheat, and
pine trees growing near Widows Creek Steam Plant in northeastern Alabama.
Several flying heights and film types were used.
Colbert Site Tests
One extensive test and one intensive test of the photometric analysis
technique were conducted using photographs of the Colbert Steam Plant area.
The extensive test focused on five soybean fields that fell within a single
photographic frame. Four of the fields were affected by S02 and one was
unaffected. The effects consisted of light levels of chlorosis, but no
necrosis. The soybean canopies were nearly continuous, but some areas were
infested with weeds and there was evidence of drought-induced stress. A
-------
-17-
microdensitometer was used to measure optical densities of the CIR film at
sample locations within each soybean field. The instrument was filtered
so that measurements were made in the green, red, and near-infrared (IR)
bands. The single-band densities were converted to reflectance and com-
pared with foliar injury levels. No relationship was indicated. However,
the IR/red reflectance ratio decreased as foliar injury increased. Also,
the weed-infested fields showed high standard deviations for IR/red
measurements, and weed-free fields with continuous canopies showed low
standard deviations for IR/red.
The intensive test included a single field of S02~affected, mature,
weed-free soybeans near Colbert Steam Plant. Measurements of optical
density were made systematically at 196 points within the field to deter-
mine possible relationships between reflectance and three other parameters:
chlorosis, plant height, and elevation of the field. After the densities
were corrected to reflectance, regressions of this parameter versus
chlorosis, and chlorosis versus elevation were calculated. None of these
relationships was significant (a=.05), and the r2 coefficients were all
below 0.25. A comparison of three-dimensional plots of the data for the
soybean field showed little similarity between the variations in
reflectance and the other parameters.
Johnsonville Site Test
Several incidents of S02 injury to vegetation occurred near the John-
sonville area during July 1977 and were photographed from the air by TVA
and EPA on different dates. A full range of foliar effects was still
visible to the ground observer in many of the soybean fields at the time
of the overflights. We obtained copies of all of the film for inter-
pretation and photometric analysis.
-------
-18-
Microdensitometer measurements of optical density were made at random
point locations in 15 soybean fields where the S02 plume had contacted the
crop. The number was later reduced to nine because the other six fields
consisted of immature plants and incomplete canopies. The optical densi-
ties were then converted to reflectance.
Statistical analysis of the Johnsonville data included comparison
of the reflectances with ground truth. The IR/red ratio correlated
significantly (a=.05) with injury (r2=0.32), but the direction (positive)
of the relationship was not in accordance with theory. None of the single-
band reflectances showed any relationship to injury.
ANALYSIS OF MULTISPECTRAL SCANNER DATA
General
Three times since 1975 TVA has arranged multispectral scanner (MSS)
overflights of S02-affected soybean fields. The first overflight, which
covered the Shawnee Steam Plant area, was conducted in 1975 by NASA/Earth
Resources Laboratory, from Slidell, Louisiana. The results of analysis
of this MSS data were negative because of the effects of diverse farming
practices and differing stages of crop growth. The second MSS overflight
was conducted in 1977 by the Environmental Protection Agency (EMSL-LV),
who scanned affected soybean fields near Colbert Steam Plant. The third
MSS overflight was done in 1978 by EMSL-LV, this time over affected
fields near Shawnee. Concurrent CIR photography was acquired on all of
these MSS overflights, and it was used along with ground truth to support
the MSS imagery analysis. Analyses of the 1977 and 1978 data are summarized
in this report.
-------
-19-
Ground Truth
Colbert Site Test—Evidence of multiple exposures of vegetation to
SC-2 was observed in a 1,620-ha area north of the Colbert Steam Plant dur-
ing July and August 1977. A scanner overflight was conducted on August 29,
at which time the soybeans still showed the foliar effects.
Shawnee Site Test—Field surveillance by TVA biologists showed that
during early August 1977, vegetation was affected by 862 emissions in
three areas totaling approximately 857 ha located south and east-
southeast of the Shawnee Steam Plant. Effects ranged from very light to
severe.
Optimal Flying Heights
The MSS lines over Colbert and Shawnee were flown at 1800 m and
500 m AGL. The lower altitude provided no improvement in accuracy of the
results of data classification. Since a low altitude line generates more
data per kilometer flown and is therefore more costly to analyze, we then
concentrated on the higher altitude data.
Optimal MSS Channels
Existing computer algorithms developed by NASA/Earth Resources
Laboratory were used to select the best four channels from eight for
detecting and classifying SC^-affected soybean fields using the Colbert
data (Table 5). The selection was required before supervised data classi-
fication could be done by the computer, which required a maximum of four
channels as input. Two procedures were used, the first procedure result-
ing from computation of divergence matrices showing optimal separation of
data classes, and the second resulting from computation of maximum
-------
-20-
TABLE 5. OPTIMAL MSS CHANNELS FOR DETECTING AND CLASSIFYING
S02-AFFECTED SOYBEAN FIELDS NEAR COLBERT STEAM
PLANT IN 1977
Procedure3
1
2
MSS channel,
designation
4
7
8
9
4
6
7
8
Wavelength
(H«0
0.50-0.55
0.65-0.70
0.70-0.79
0.80-0.89
0.50-0.55
0.60-0.65
0.65-0.70
0.70-0.79
Spectral
region
Green
Red
Near-IR
Near-IR
Green
Red
Red
Near-IR
o
^Procedures discussed in text.
MSS channels 3 through 11 (blue through thermal IR) considered.
divergence among individual areas (agricultural fields). The blue and
thermal IR channels were rejected because of their inherently low
contrast with respect to vegetation.
Optimal channels were also selected from the Shawnee data (Table 6).
Basically the same channels were chosen as for Colbert. There was some
open water in the Shawnee north-south flight line, and its influence
probably resulted in selection of the blue channel by the computer.
MSS Data Classification
Classifying digital images involved three procedures: unsupervised,
supervised, and pseudosupervised. The unsupervised procedure is done
without intervention by the analyst, and no preliminary training of the
computer is done. Therefore, the need for a_ priori knowledge of the
scene is not great. The supervised procedure involves programming the
computer with ground truth so it can recognize the phenomena. The
pseudosupervised procedure is an efficient combination of the two others,
-------
-21-
TABLE 6. OPTIMAL MSS CHANNELS FOR DETECTING AND CLASSIFYING
S02-AFFECTED SOYBEAN FIELDS NEAR SHAWNEE STEAM PLANT
IN 1978
MSS channel
designation
3
7
8
9
5
7
8
9
Wavelength
([M)
North-South Flight Lines
0.45-0.49
0.65-0.70
0.70-0.79
0.80-0.89
East-West Flight Lines
0.55-0.60
0.65-0.70
0.70-0.79
0.80-0.89
Spectral
region
Blue
Red
Near-IR
Near-IR
Green
Red
Near-IR
Near-IR
<»
MSS channels 3 through 10 considered. Channel 11 (thermal IR) not
considered. Procedure used was interclass distance separation.
and it uses a minimum of ground truth. The supervised and pseudosuper-
vised procedures use a maximum of four input channels, while the
unsupervised procedure can use eight.
The three procedures were evaluated by comparing their output which
consisted of classified images (maps) depicting patterns of affected
and unaffected soybean fields. The ground truth about the proportion of
each field that was affected by SO^ was compared with the classification
results.
In MSS data classification, there are errors of omission and errors
of commission. The first error results in underclassification and the
second, in overclassification of the phenomena of interest.
The accuracy evaluation showed that the pseudosupervised classifier
could map soybeans without regard to 50% effects with overclassification
-------
-22-
errors of 0.6 to 5.1 percent. This procedure could differentiate moder-
ately to severely affected soybeans and unaffected soybeans with over-
classification errors of 11.3 to 24.4 percent (Table 7). It was not
tested on very light to light S02 effects. The unsupervised classifier
could identify soybean fields without regard to SC>2 effects with over-
classification errors of 7.6 percent. However, it could not separate
S02-affected soybean fields from unaffected soybean fields.
The supervised classification procedure yielded inconclusive results.
Because of time constraints, this classifier could not be tested on
moderately to severely affected soybean fields. Had it been, the classifier
might have yielded better results than the pseudosupervised procedure.
Enhancement of Patterns of S02 Effects Within Fields
The I2S Image Processing System at TVA's Mapping Services Branch in
Chattanooga was used to enhance and display selected scenes of MSS data
covering the Shawnee area. The effects on soybeans ranged from very
light to severe. A density level-slicing procedure was used to display
the background in monochrome and the S(>2 effects in orange. The cor-
respondence of patterns with field observations of injury was fairly
close in some fields,where the soybean canopies were dense and continu-
ous. The scanner system was apparently not successful in detecting very
light and light chlorosis. Moreover, the instrument did not consistently
detect moderate and severe injury to the crop.
Multispectral scanner imagery from the orbiting Landsat vehicle was
obtained for the Shawnee Steam Plant to cover a period when the 862
effects on soybean fields should have been visible to the ground observer.
Preliminary analysis of the four individual MSS bands and the color
composite provided no indication of patterns associated with the effects,
so this task was discontinued.
-------
TABLE 7. ERRORS RESULTING FROM PROCEDURES FOR DETECTING
AND CLASSIFYING S02 EFFECTS ON SOYBEANS
, Site with Light S02 Effects Site with Moderate to Severe S02 Effects
Unsupervised Supervised Unsupervised Pseudosupervised
Separation of soybeans
from other land cover
Separation of S02-
affected from
unaffected soybeans
+7.2%
+7.6%
+142.0%
+101.4%
+5.1% (first
flight line)
+0.6% (second
flight line)
-24.4% (first
flight line)
+11.3% (second
flight Line)
i
N3
OJ
^Inconclusive results, error not determined
+0verclassification
-Underclassification
Zero percent would indicate no error
-------
TECHNICAL REPORT DATA
(Please read Instructions on the reverse before completing)
REPORT NO.
PA-600/7-81-113
2.
3. RECIPIENT'S ACCESSION NO.
TITLE AND SUBTITLE
.emote Sensing of Sulfur Dioxide Effects on Vegetation
inal Report - Volume I - Summary
5. REPORT DATE
July 1981
6. PERFORMING ORGANIZATION CODE
AUTHOR(S)
. Daniel Sapp
8. PERFORMING ORGANIZATION REPORT NO.
TVA/ONR/ARP-81/5
PERFORMING ORGANIZATION NAME AND ADDRESS
)ffice of Natural Resources
ennessee Valley Authority
Morris, TN 37828
10. PROGRAM ELEMENT NO.
INE 625C
11. CONTRACT/GRANT NO.
80 BDJ
2. SPONSORING AGENCY NAME AND ADDRESS
U.S. Environmental Protection Agency
)ffice of Research and Development
Office of Energy, Minerals, and Industry
Washington, D.C. 20460
13. TYPE OF REPORT AND PERIOD COVERED
Final 1976-1980
14. SPONSORING AGENCY CODE
5. SUPPLEMENTARY NOTES
This project is part of the EPA-planned and -coordinated Federal
interagency Energy/Environmental R&D Program.
6-ABSTRACT Three techniques for detecting and mapping sulfur dioxide (S02) effects on
the foliage of sensitive crops and trees near large, coal-fired power plants were testec
and evaluated. These techniques were spectroradiometry, photometric analysis of aerial
photographs, and computer analysis of airborne multispectral scanner data.
pectroradiometry is a useful, ground-based technique for measuring the changes in
reflectance that accompany exposure of sensitive crops to S02. Photometric analysis of
aerial color-infrared photographs has some practical advantages for measuring the
reflectances of forest species or for synoptic point-sampling of extensive areas; these
tasks cannot be done effectively by field crews. The relationships among reflectance,
:oliar injury, and yield of crops are complex and are affected by many extraneous vari-
ables such as canopy density. The S02 effects are easier to detect on winter wheat than
on soybeans, but in either case they cannot be consistently detected by airborne remote
sensors except under near-ideal conditions when the injury is moderate to severe. Air-
>orne multispectral scanner data covering affected soybean fields were analyzed using
three computer-assisted procedures: unsupervised, supervised, and pseudosupervised;
the last method provided the bes,t results. Landsat imagery was also investigated, but
the foliar effects of 862 were too subtle to detect from orbit.
KEY WORDS AND DOCUMENT ANALYSIS
DESCRIPTORS
b.IDENTIFIERS/OPEN ENDED TERMS
COSATl Field/Group
Air pollution *
Electric power plants
Photointerpretation
Remote sensing *
Environmental surveys *
Infrared photography
Photometry
Reflectance
Sulfur dioxide *
Plant pathology
Transport processes
Char., meas. & monit.
Crop & forest species
Digital image analysis
Multispectral scanning
Microdensitometry
Tennessee Valley
18. DISTRIBUTION STATEMENT
Release to Public
19. SECURITY CLASS (This Report I
Unclassified
31
20. SECURITY CLASS (This page)
Unclassified
22. PRICE
EPA Form 2220-1 (Rev. 4-77) PREVIOUS EDITION is OBSOLETE
-------
I It'll II Idll
environmental
Protection
Agency
EPA-335
Official Business
Penalty for Private Use, $300
Special Fourth-Class Rate
Book
Please make all necessary changes on the above label,
detach or copy, and return to the address in the upper
left-hand corner
If you do not wish to receive these reports CHECK HERE D,
detach, or copy this cover, and return to the address in the
upper left-hand corner
EPA-600/7-81-113
------- |