United States
                   Environmental Protection
                   Agency
 Environmental Research
 Laboratory
 Duluth MN 55804
                   Research and Development
 EPA/600/S3-85/034 July 1985
&ERA         Project  Summary
                                                                                   \ A .'.
                   Aerial  Photography and  Ground
                   Verification  at  Power  Plant
                   Sites:  Wisconsin  Power  Plant
                   Impact  Study
                   Sarah L Wynn and Ralph W. Kiefer
                     This study demonstrated and evalu-
                   ated nine methods for monitoring the
                   deterioration of a large wetland on the
                   site of a newly-constructed coal-fired
                   power plant in Columbia County, Wis-
                   consin. Four of the nine methods used
                   data from ground sampling; two were
                   remote  sensing methods  without
                   ground verification;  and three were
                   remote sensing methods which either
                   used ground verification or relied on the
                   analyst's "on-the-ground" knowledge
                   of the area.
                     These methods were evaluated on
                   the basis of  whether  they monitor
                   change at a species or a community
                   level, whether they monitor community
                   change in terms of area or location or
                   both, and whether they provide infor-
                   mation about trends in plant communi-
                   ties. They were also evaluated in terms
                   of time, cost, sensitivity, and reliability.
                   Changes in the wetland over a three-
                   year period  are presented, as deter-
                   mined by each of the methods. Eight
                   appendices  provide  information and
                   raw data for several of the methods,
                   color/texture keys for interpreting air
                   photos, and an annotated bibliography
                   on remote sensing methods.

                     This Project  Summary was  devel-
                   oped by  EPA's  Environmental
                   Research Laboratory, Duluth, MN, to
                   announce key findings of the research
                   project that is  fully documented in a
                   separate report of the same  title (see
                   Project Report  ordering information at
                   back).
 Background
  The full report describes part of a large
 study documenting the environmental
 impact of construction and operation of a
 1050 MW coal-fired power plant in south
 central Wisconsin. A major goal of the
 study was to develop new, less expensive
 and more effective methods for predict-
 ing and measuring environmental
 change.
  This subproject had a dual purpose:
   1. To document the nature and
     extent of changes in vegetation by
     a variety  of ground-based and
     remote sensing methods, and
   2. To compare and  evaluate the
     methods on the basis of their effi-
     ciency, sensitivity,  and reliability.
  Before construction, the site was an
 extensive marsh/sedge meadow with
 areas of floodplain forest and a few low,
 semi-wooded knolls. The marsh included
 small expanses of open water with emer-
 gent vegetation,  and pockets of shrub
 carr and alder thicket. The soil was a peat
 mat overlying  sand.  Construction and
 operation of the power plant in this set-
ting resulted in the elimination of sea-
sonal fluctuations in water level,  an
increased flow of ground water, and
year-round thermal loading of the ground
water. All plant communities responded
to these changed conditions.

 Methods
  Changes were monitored by nine dif-
ferent methods representing both tradi-
tional and new approaches. Four were

-------
ground based; five involved remote sens-
ing with or without ground verification.
Together, they provided information at
many levels. The evaluation of the results
and methods considered which methods
can best:
1. Detect changes at the species
level with time.
2. Detect changes at the community
level in the area and location of
each type of plant community.
3. Document trends in changes in
vegetation.
Ground Sampling Methods
The study site was an area of 33.5 ha.
marked off into transects at 50 m inter-
vals with sampling stations every 50 m
along these transects. Data were col-
lected each summer and fall, from 1974
through 1977.
Diversity Index
The diversity index gives information
about changes in the total number of spe-
cies. The number of species present at
each sampling station was counted, and
these values were summed for each year
of the study. The index was calculated as
the annual percentage gain or loss in
species or as a change in relation to some
base year.

Subjective Classification
This method classifies vegetation by
type of community. Twelve classes of
vegetation were defined empirically,
according to the most prominent species
in each one. (Classes identified in this
way and by the seven other ground-
based and remote sensing methods are
shown in Table 1.) Stem counts were
then made of each species found at a
sampling station, and on this basis the
station was assigned to the appropriate
class. For maximum reliability, the sub-
jective classification method requires
that the analyst adhere strictly to the
criteria established for each class.
Table 1 . Classes of Vegetation Identified by Eight Ground-Based and Remote Sensing
Methods


Vegetation class
Carex stricta
Degraded C. stricta

C. lacustris
Degraded C. lacustris
Transition
Degraded transition
Emergents
Degraded emergents
Spiraea alba
Shrubs
Open water
Open-emergents
Weedy annuals
Transition-emergents
Sedges and grasses
Grasslike
Tall-coarse
Grasslike-tall
Disturbed vegetation
Undisturbed vegetation
Degraded sedges
Shrubs and trees
Typha latifolia
Scirpus fluviatilis
Floating mat
Lemna minor
Trees
Spiraea/sedges
Spiraea/shrubs

* Not all of these classes
.to QJ QS
§. ^ -S
"5 y o
is 2 •?
P. TO +S C
.§ c * §
0) a .0 C fc
.> (0 r; O _
a -SJ .<5 a •£> 2
'•"•s ^ O Q) ^* "C
•Q jo °° Oi <5 P-
3 45 w Q) c- '^
co ^3 <* S § ^
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
XX X
X

X
X
X


X
x



X
X



can be identified on every airphoto.
\rbance mapping
joto grid analysis
1 fc
Q s;
m*
(X)

(X)
(X)
(X)
(X)
(X)
(X)
X
X
X X

(X)
x
X



X
X
X

(X)
(X)
m
(X)
(X)



Parentheses
wto interpreted
tation mapping
juter-assisted
>ing
*-§> ?S
•t Ji ft (0
^ =>
(X)


(X)
(X)
(X)
(X)
(X)
X
X
X

m
x
X





X

(X)
IX)
(X)
(X)
(X)



indicate
0 E
X
X

X
X

X
X

X

X







X




X

X
X

classes
Association Analysis
  Association analysis is a type of cluster
analysis which groups stations according
to their similarities in species composi-
tion. Thirty-four common, visually domi-
nant species were selected as attributes
in identifying clusters. Each station was
rated + or - for the presence or absence of
each attribute. A computer program then
divided the stations into two clusters of
maximum  dissimilarity.  Division  was
based on presence or absence of the sin-
gle species which created the greatest
dissimilarity  in species composition of
                                          merged with others and not included as distinct classes in the final analysis.
the resulting  clusters,  using  sum chi
square  as a  dissimiliarity  coefficient.
Subdivision  into  further clusters  con-
tinued on this basis until the desired
number of clusters had been created. The
result is a hierarchial structure that can
be displayed as a dendrogram. The clus-
ters created by association analysis were
named to correspond as well as possible
with other classifications of vegetation
used in this study (Table  1).

Vegetation Structure Analysis
  Another way to monitor changes in the
study area with time is on the basis of
changes in  gross  vegetation structure.
Five categories of structure were defined
(Table  1). Each station was assigned to
one of these according to the results of
ground sampling, year by year.

Remote Sensing Methods
  For the five remote sensing methods,
airphotos were interpreted or analyzed in
a variety of ways. The photos were taken
with both color and color infrared film, at
scales of from 1:11,500 to  1:120,000.
Color infrared film gave the best,  most

-------
 discriminating results;  the  scales that
 permitted the most satisfactory interpre-
 tations were 1:19,000 and 1:38,000. Air-
 photo data were collected monthly during
 the growing season and several times
 during  the rest of the year.

 Airphoto Monitoring
   Airphotos can be used without ground
 verification to monitor  sites which are
 inaccessible  from  the  ground.  This
 method may provide a high degree of
 detail, but alone it does not map or quan-
 tify information. It gives a simple photo-
 graphic record of change, accompanied
 by any  interpretations the analyst is able
 to make. Six types of features could read-
 ily be  interpreted from airphotos, and
 with some ground verification, two addi-
 tional  classes of vegetation  could  be
 identified (Table 1).

 Airphoto Interpreted Disturbance
 Maps
   Disturbance mapping identifies only
 three types of area: undisturbed  vegeta-
 tion, disturbed  vegetation,  and open
 water.  Disturbed  areas   represent
 degraded vegetation resulting from ero-
 sion of the peat mat. They  are  easy to
 recognize on airphotos as  changes in
 appearance from earlier photos.
   Maps were made by tracing the boun-
 daries of the three types of area on mylar
 overlying enlarged airphotos. No ground
 verification is needed for this method.
 The extent of change is easy to quantify,
 for example, by use of a planimeter.

 Methods Combining Airphotos
 and Ground Sampling Data

 Airphoto  Interpreted  Vegetation
 Maps
   Vegetation  classification  maps were
 drawn  in the same  way as the distur-
 bance maps, with mylar overlays show-
 ing  the locations  of  transects  and
 sampling stations. Patterns seen on the
 photos  were  traced and labeled by refer-
 ring to the ground-based data and classi-
 fication  for  each  station. Classes
 recognizable on  the airphotos did not
 always  correspond exactly  to ground-
 based classifications, and some classes
 identified on the  ground or in individual
 airphotos had to be merged so that com-
 parisons could be made among the air-
 photo interpreted vegetation maps (Table
 1).
   A color and texture key was assembled
 for each airphoto to facilitate consistent
 mapping of the various communities. The
, total area of each community was deter-
mined as for the disturbance maps, and
expressed as a percentage of the study
area. One problem with this kind of map-
ping  is  that  communities sometimes
grade into one another, and the boundar-
ies between them do not appear distinct.
This detracts from the precision of the
method.

Airphoto Grid Analysis
  In. airphoto grid  analysis,  airphotos
were overlaid with grids to scale ( a cell
representing 50 m2 on the ground), with
the locations of ground stations and tran-
sectS also shown. Classes of vegetation
were identified (Table 1) and a color/tex-
ture key  was assembled  as in airphoto
interpreted vegetation mapping. The dif-
ferences between the two methods  is
that,  in grid analysis, assigning vegeta-
tion classes and determining the percen-
tage  of cover of each class  are done
simultaneously on a cell by cell basis.
Results are  presented  as percentage
cover or change in percentage cover for
each  class of vegetation, when  the
method is used to monitor change over
time.

Computer  Assisted   Vegetation
Mapping
  Color and color infrared photos were
scanned with a densitometer using three
different filters to obtain analyses of color
bands in the red, green, and blue spectral
regions. The size of the  scanning unit
(pixel) corresponded to a spot size of 1.9
m (3.6 m2) on the ground. Continuous
output from the photomultiplier tube of
the  densitometer  was  converted  to
integer values, pixel by pixel. Through a
series of computer programs, these data
underwent various transformations and
corrections. Sets  of  data were  then
selected for "training" the computer to
distinguish among different classes of
vegetation.  This  part  of  the  analysis
requires  a high degree  of interaction
between the analyst and the computer.
  After the limits of each class had been
defined and the probability distribution of
data values within each class had been
determined, the spectral response signa-
ture of each pixel was compared to the
spectral response signatures of the final
training sets and classified for the best fit.
Three  models  for  classification were
used in this study. They involved different
amounts  of time and expense and gave
somewhat different results.
  This method offers consistent classifi-
cation of vegetation and quantification of
the area of each class, once the criteria
for classification have been established.
 Classes  are named  by  reference  to
 ground-based data (Table 1). Results can
 be displayed  with  computer-printed
 maps or color photo maps.

 Results
   The combined results from all methods
 give information at several levels. The
 method of choice for any study depends
 on what kind of information is desired.

 Changes at  the Species Level
   Of the methods employed in this study,
 only the diversity  index provides direct
 information  on  changes occurring in the
 number of species  in the area. However,
 the diversity index gives no information
 at the community  level and cannot be
 used  to locate or quantify  areas  of
 change. Furthermore, the method is use-
 ful only if there  is a definite trend toward
 net gain or loss of  species. If some spe-
 cies disappear  from parts of the study
 area while others invade or increase, the
 diversity index may show no change. This
 particular  study  revealed  a  sharp
 decrease in  diversity, as summarized  in
 Table  2.

 Changes   at  the   Community
 Level
   All  methods  other than the diversity
 index give some information on changes
 in the area or the location of plant com-
 munities  or both.  Ground  sampling
 methods show  point locations of com-
 munities and quantify changes in the rel-
 ative percentage of stations in each class.
 Airphoto methods,  whether quantitative
 or purely descriptive, are better able  to
 show the area and location of plant com-
 munities than ground sampling methods.

 Ground Sampling Methods
   Both  subjective  classification  and
 association  analysis demonstrated
 changes with time  in the number of sta-
 tions assigned to each  vegetation class
 (Table 3). These changes could be corre-
 lated with changes in water temperature,
 volume of flow, and erosion of the peat
 mat. The results show a  successional
 trend toward deeper water species. Both
 methods indicate that  Carex  lacustris,
transition, and  emergent communities
 are more fragile than  C.  stricta  and
 shrubby communities.
  Vegetation structure analysis revealed
 that the grasslike and  grasslike  tall
 classes were most sensitive to impact. Of
the 29 stations  classified as grasslike in
 1974, 21 were reclassified in 1977, and
 only one of twelve stations that were
grasslike tall in 1974 retained this classi-

-------
Table 2. Diversity Index, 1974-1977
Number of species (sum
Year of numbers at 62 stations)
1974
1975
1976
1977
379
357
296
266
Change (number of
species/ year)
-22
-61
-30
Diversity index*
(% change/year)
-5.8
-17.1
-10.1
*A positive index number represents an overall gain in number of species; a negative number
 indicates a loss.
fication three  years  later.  The  shift
toward a tall coarse and open water vege-
tation structure occurred throughout the
study area.

Remote Sensing Methods
  The four remote sensing methods cor-
roborate results of the ground sampling
methods, that is, a trend from healthy,
predominantly  marsh/sedge  meadow
communities  to eroded and  disturbed
wetland vegetation. Airphoto monitoring
provides a photographic record of the
change. The  increasing and  enlarging
areas of open water and disturbed vege-
tation  between  September 1975 and
October 1977 are shown in the airphoto
interpreted disturbance maps (Figure 1).
Table 4 shows results for the same
period, for airphoto grid analysis and air-
photo interpreted mapping, in terms of
changes in percentage of cover of each
vegetation class.
  The discrepancies between airphoto
interpretated  vegetation mapping and
airphoto grid analysis result from the dif-
ference in scale at which the analyses
are done. Grid analysis, in which the per-
centage of each vegetation class is deter-
mined on  a cell by cell  basis, gives the
more accurate information on changes in
total area of each type of vegetation, but it
cannot reveal changes in the location of
the various  communities.  Vegetation
mapping shows the predominant type of
vegetation  on  a  broader  scale, and
changes in the location of a plant com-
munity can be seen at a glance on maps
similar to the disturbance maps of Figure
1.
  Computer assisted  mapping  offers
some  advantages  of each  of the other
Table 3.   Changes in Classification of 62 Stations from 1974 TO 1977, by Methods of
          Subjective Classification and Association Analysis

                                       Classification in 1977
Classification
in 1974,
community
Carex lacustris






C. stricta




Transition



Emergents


Shrubs

Spiraea

Subjective classification
Community
C. lacustris
Weedy annuals
Degraded C. stricta
Degraded C. lacustris
C. stricta
Open water
Emergents
C. stricta
Degraded C. stricta
Transition


Transition
Degraded transition
Emergents
Open water
Emergents
Degraded emergents
Open water
Shrubs
C. stricta
Spiraea

No. of
stations
1
2
1
3
3
4
1
5
9
1


1
4
5
2
2
6
3
5
1
3

Association analysis
Community

Weedy annuals
Degraded C. stricta
Transition
Degraded transition
Emergents
Emergents-open

Degraded C. stricta
Spiraea
Weedy annuals
Shrubs
Transition
Degraded transition
Emergents
Emergents-open
Emergents
Emergents-open

Shrubs
Emergents-open

C. stricta
No. of
stations
0
2
8
2
4
1
3
4
10
1
1
1
1
1
4
1
2
3

2
2
1
2
remote sensing methods.  Because the
analyst can choose the size of the unit
area to be classified (the pixel), the ana-
lyses can be made on as large or small a
scale as  desired.  Results can  be pres-
ented as the overall area of each vegeta-
tion class (Table 5) or  as a map on which
each pixel is  shown  in the color desig-
nated for the corresponding  class of
vegetation. Comparison of Tables 4and 5
shows general agreement among  the
three methods. Of the three, computer
assisted mapping  offers the most con-
sistent classification  of vegetation and
quantification  of results and the most
readable  visual product  in the form of
maps.

Evaluation of Methods
  The nine methods applied in this study
were evaluated according to the kind of
information provided, the expertise
required, efficiency (requirements in
time, capital equipment, materials), and
sensitivity and reliability.
  The diversity index  is the only method
which provides information on  changes
in numbers of species, but it provides no
information at the community level.
Ground sampling methods show point
locations of communities and  quantify
changes in the numbers of stations clas-
sified as to type of community, but they
cannot show changes in the area or loca-
tion  of communities  as well as remote
sensing methods can.
  All methods require expertise in botany
and  ecology. Computer facilities and
expertise are necessary  for association
analysis and computer assisted mapping.
The five remote sensing methods
demand skills in the visual interpretation
of airphotos, and,  in addition, computer
assisted mapping involves interpretation
of computer generated images. Drafting
skills are necessary for disturbance map-
ping and  vegetation mapping.
  Detailed analyses compared the effi-
ciency, sensitivity, and reliability of the
nine methods. Table  6 summarizes the
results. There  is a fourfold  difference in
cost between the least expensive and the
most expensive methods, and a twelve-
fold difference in the requirement for
time between  the fastest and the most
time-consuming methods.
  The greatest amount of information is
obtained by combining ground based
sampling methods with airphoto map-
ping methods. Ground sampling data
give information on  the nature of the
occurring change, whereas  mapping
methods show where the change is
occurring  and how extensive it is. For

-------
     September 25.  1975

         D Undisturbed
         H Disturbed
         • Open Water
September 24, 1976
small data sets, subjective classification
is effective in demonstrating trends  in
plant communities and can be combined
effectively with airphoto-interpreted
vegetation mapping. Both of these
methods are costly and time consuming,
however, and would not be the methods
of choice for large data sets. For  large
amounts of data, association analysis
can be used effectively in conjunction
with grid analysis or computer assisted
mapping. The choice of remote sensing
method would depend on the nature  of
the results desired - whether highly
quantitative in terms of community area,
or both quantifiable and visual, such as a
map.
  Grid analysis is also compatible with
disturbance mapping, for use in large
areas. Grid analysis measures the per-
cent change in cover of each vegetation
class, and disturbance mapping shows
the extent and location of gross change.
Similarly, association analysis and dis-
turbance mapping provide complemen-
tary information for large data sets. If
facilities are available, the ideal combina-
tion of methods for extensive monitoring
would be association analysis, to identify
community change, and computer
assisted mapping, for visual displays and
quantitative presentation of results.
Figure 1.    Changes in vegetation at site of Columbia Generating Station, 1975-1977, as
           shown by airphoto disturbance mapping.

-------
  Table  4.
 Percentage Cover of Each Vegetation Class,  as Determined by Airphoto Grid
Analysis and Airphoto Interpreted Mapping

                                      % of each class
                                     Grid analysis
                                               Airphoto interpreted mapping
Vegetation class
Shrub
Spiraea
Open water
Transition-Emergents
Sedges
Degraded sedges
Other disturbance
Unclassified
Sept. 1975
7.4
9.5
5.2
25
50
-
2.2
-
Oct. 1977
7.8
9.2
10
25
11
12
24
-
Sept 7975
;/
11
1.5
22
45
-
7.5
1.9
Oct. 1977
5.7
7.5
10
15
22
20
17
1.3
        Table 5.    Summary of Results of Computer Assisted Vegetation Mapping
Class of vegetation
Shrubs
Spiraea/Sedges
Open water
Transition -Emergents
Sedges
Degraded Sedges
Other Disturbance
Other, and unclassified

September
9.7
18
0.5
22
39
-
1.8
9
% of total area
1975 June 1977
6.7
9.6
8.1
17
13
21
18
5.8
Table 6.    Comparison of Methods Based on Efficiency. Sensitivity, and Reliability
Sensitivity2
Method
Diversity Index
Subjective Classification
Association Analysis
Vegetation Structure
Analysis
Airphoto Monitoring
Disturbance Mapping
Airphoto Grid Analysis
Airphoto Vegetation
Mapping
Computer Assisted
Mapping
Efficiency'

Time Cost
2
3
2

2
1
1
3

3

2
1
2
1

1
1
1
3

3

3
No. classes
defined
-
1
2

3
2
3
1

1

2
Type
of data
2
1
2

2
3
3
3

3

3
Reliability3
1
3
1

3
3
3
2

2

2
'A rating of 1  represents high efficiency, i.e.. low time or cost requirements. A rating of 3
 represents relatively low efficiency.
2A rating of 1 represents a high degree of sensitivity to subtle changes. It requires that a large
 number of vegetation classes be defined, or that data be based on stem counts. A rating of 3
 represents few classes distinguished or airphoto data. Presence-absence data receive a rating of
 2.
3The reliability rating is based on the repeatability of data collection, the level of subjectivity (the
 degree to which the analyst must interpret the data),  and whether results are quantitative or
 qualitative. A rating of 1 represents a high level of repeatability, a high degree of objectivity, or
 computer quantitative results. In this table, the methods were rated most reliable (11, moderately
 reliable (2), and least reliable (31, according to their combined scores on  the three criteria for
 reliability.
                                                                                U. S. GOVERNMENT PRINTING OFFICE: 1985/559-111/20618

-------
     Sarah L  Wynn and Ralph  W. Kiefer are with the University of Wisconsin,
       Madison, Wl 53706.
     Gary E. Glass is the EPA Project Officer (see below).
     The complete report, entitled "Aerial Photography and Ground Verification at
       Power Plant Sites: Wisconsin Power Plant Impact Study," (Order No. PB 85-197
       358/AS; Cost: $23.50, subject to change) will be available only from:
            National Technical Information Service
            5285 Port Royal Road
            Springfield, VA 22161
             Telephone: 703-487-4650
     The EPA Project Officer can be contacted at:
            Environmental Research Laboratory
            U.S. Environmental Protection Agency
            6201 Congdon Blvd.
            Duluth, MN 55804
United States
Environmental Protection
Agency
Center for Environmental Research
Information
Cincinnati OH 45268
     BULK RATE
POSTAGE & FEES PAIt
        EPA
   PERMIT No G-35
Official Business
Penalty for Private Use $300
EPA/600/S3-85/034

-------