• TABLE OF CONTENTS
HIDE
 Front Cover
 Title Page
 Acknowledgement
 Table of Contents
 Executive summary
 Introduction
 Plot study
 Methods
 Results
 Discussion
 Reference
 Appendix A: Satellite image...
 Appendix B: Image rectification...
 Appendix C: Erdas programs...
 Appendix D: Individual plot data...






Group Title: Florida Cooperative Fish and Wildlife Research Unit Technical report 52
Title: Methods for determining change in Florida wetland habitats using SPOT satellite data
CITATION PAGE IMAGE ZOOMABLE
Full Citation
STANDARD VIEW MARC VIEW
Permanent Link: http://ufdc.ufl.edu/UF00073895/00001
 Material Information
Title: Methods for determining change in Florida wetland habitats using SPOT satellite data
Series Title: Technical report
Physical Description: ii, 67 p. : ill., map ; 28 cm.
Language: English
Creator: Silveira, Jennifer E
Publisher: Florida Cooperative Fish and Wildlife Research Unit
Place of Publication: Gainesville FL
Publication Date: <1996.>
 Subjects
Subject: Wetlands -- Florida   ( lcsh )
Artificial satellites in surveying -- Florida   ( lcsh )
Genre: government publication (state, provincial, terriorial, dependent)   ( marcgt )
bibliography   ( marcgt )
non-fiction   ( marcgt )
 Notes
Bibliography: Bibliography: p. 36.
Statement of Responsibility: Jennifer E. Silveira.
General Note: "January, 1996."
Funding: This collection includes items related to Florida’s environments, ecosystems, and species. It includes the subcollections of Florida Cooperative Fish and Wildlife Research Unit project documents, the Sea Grant technical series, the Florida Geological Survey series, the Coastal Engineering Department series, the Howard T. Odum Center for Wetland technical reports, and other entities devoted to the study and preservation of Florida's natural resources.
 Record Information
Bibliographic ID: UF00073895
Volume ID: VID00001
Source Institution: University of Florida
Holding Location: University of Florida
Rights Management: All rights reserved, Board of Trustees of the University of Florida
Resource Identifier: aleph - 002277277
oclc - 34518252
notis - ALN0356

Table of Contents
    Front Cover
        Front cover
    Title Page
        Title page
    Acknowledgement
        i
    Table of Contents
        ii
    Executive summary
        Page 1
        Page 2
    Introduction
        Page 3
    Plot study
        Page 4
        Page 5
    Methods
        Page 6
        Satellite image acquisition
            Page 6
            Page 7
            Page 8
        Satellite image processing
            Page 9
        Plot overlay digitization
            Page 10
        Display of images with overlays
            Page 10
        Evaluation of interpretation
            Page 11
    Results
        Page 12
        Page 13
        Page 14
        Page 15
        Page 16
        Page 17
        Page 18
        Page 19
        Page 20
        Page 21
        Page 22
        Page 23
        Page 24
    Discussion
        Page 25
        Page 26
        Page 27
        Page 28
        Page 29
        Page 30
        Page 31
        Page 32
        Page 33
        Page 34
        Page 35
    Reference
        Page 36
    Appendix A: Satellite image translation
        Page 37
        Page 38
        Page 39
        Page 40
    Appendix B: Image rectification and overlay registration
        Page 41
        Page 42
        Page 43
    Appendix C: Erdas programs used
        Page 44
    Appendix D: Individual plot data notes
        Page 45
        Page 46
        Page 47
        Page 48
        Page 49
        Page 50
        Page 51
        Page 52
        Page 53
        Page 54
        Page 55
        Page 56
        Page 57
        Page 58
        Page 59
        Page 60
        Page 61
        Page 62
        Page 63
        Page 64
        Page 65
        Page 66
        Page 67
Full Text







Methods for Determining Change in Florida Wetland Habitats
Using SPOT Satellite Data




Jennifer E. Silveira


Technical Report #52


Florida Cooperative Fish and Wildlife Research Unit
Gainesville, Florida 32611-0450







I 1-




,!'~~~~~ ~~~ '.r.---.-. ..' \''
";: """:" !











Methods for Determining Change
in Florida Wetland Habitats
Using SPOT Satellite Data

January, 1996

Technical Report #52







Completed for the U.S. Fish and Wildlife Service,
National Wetlands Inventory


Jennifer E. Silveira














Citation: Silveira, J. E. 1996. Methods for determining change
in Florida wetland habitats using SPOT satellite data. Fla. Coop.
Fish and Wildl. Res. Unit, U.S. Nat. Biol. Serv. Tech. Rep. 52.
Gainesville, FL.












ACKNOWLEGMENTS


This work was funded by the U.S. Fish and Wildlife Service National Wetlands
Inventory. Funding was administered by the Florida Cooperative Fish and Wildlife
Research Unit, National Biological Service, which also provided laboratory space and
computer support.
I am very grateful to Tom Dahl, of the National Wetlands Inventory, for his
essential guidance in this work. Special thanks to Wiley Kitchens, unit leader of the
Florida Cooperative Fish and Wildlife Research Unit, and Don Woodard, of the National
Wetlands Inventory, for making this work possible. Thanks also to Rich Young, of the
National Wetlands Inventory Status and Trends, and Leonard Pearlstine and Barbara
Fesler of the Florida Cooperative Fish and Wildlife Research Unit.









TABLE OF CONTENTS

EXECUTIVE SUMMARY 1

INTRODUCTION 3

PILOT STUDY 3

METHODS 5

Satellite Image Acquisition 6

Satellite Image Processing 9

Plot Overlay Digitization 10

Display of Images with Overlays 10

Satellite Image Interpretation 10

Evaluation of Image Interpretation 11

RESULTS 12

DISCUSSION 25

REFERENCES 36

LIST OF TABLES AND FIGURES 36

APPENDIX A SATELLITE IMAGE TRANSFORMATION 37

APPENDIX B IMAGE RECTIFICATION AND OVERLAY REGISTRATION 41

APPENDIX C ERDAS PROGRAMS USED 44

APPENDIX D INDIVIDUAL PLOT DATA AND NOTES 45












EXECUTIVE SUMMARY


Procedures used by the National Wetlands Inventory Status and Trends Program
for wetlands change detection were used with satellite imagery rather than aerial
photography to test the utility of satellite imagery for this purpose.


The satellite images used were SPOT panchromatic and multispectral satellite data,
merged to produce 10-meter resolution data covering green, red and near-infrared
reflectance. Forty-four Status and Trends monitoring plots, found within four satellite
images in Central and South Florida, were examined for change.


To be as consistent as possible with current Status and Trends change detection
techniques, the satellite images were interpreted visually, rather than being classified by
computer. Overlay maps from the latest Status and Trends evaluation of wetland change,
(up to the 1980's), were converted to digital format and compared with the satellite
images on a computer monitor. Land covers were identified on the satellite images, and
changes were successfully detected and measured with the combined aid of aerial
photographs, land cover interpretations and topographic quadrangle maps from past
Status and Trends updates.

The visual interpretation of the satellite images differed from past Status and Trends
photo-interpretations in three ways:

1) The details of individual tree crowns and buildings could not be used as
interpretation clues because they were not visible at the images' 10-meter
resolution.

2) The lack of stereoscopic vision did not allow forest and shrub identification
based upon height.

3) The ability to examine green, red and infrared satellite data separately, and the
use of a red/infrared ratio (a normalized difference vegetation index or NDVI)
highlighted the appearance of wetland vegetation.









The minimum sizes of wetland and upland polygons reliably detected from the
satellite images are show below:


Cover Types Minimum Reliably Detected Size:
(pixels) (m2) (acres)
Non-Forested
Uplands 4-8 400-800 0.1-0.2
Estuarine 8-12 800-1200 0.2-0.3
Riverine, Lacustrine 8 800 0.2
Palustrine 8 800 0.2
Non-vegetated
.Palustrine Vegetated 12-40 1200-4000 0.3-1
Forested :
Upland and Palustrine 12 1200 0.3
Forest Surrounded by
Non-forest Covers
Upland Forest Within 40-200 400-2000 1-5
Palustrine Forest, and
Vice Versa


The spatial resolution provided by the merged panchromatic and multispectral
SPOT data was critical for land cover interpretation. Additional spectral resolution from
other remotely sensed data sources would be useful, provided it was available at 10 meter
or finer spatial resolution.


The use of satellite data alone for wetlands change detection is not recommended.
Either extensive ground truth must be collected, or additional data resources such as
interpreted aerial photography and topographic maps must be used If satellite imagery is
used for Status and Trends wetlands change updates, alternating updates using satellite
imagery with updates using photography would take advantage of the strengths of both
data types. Satellite imagery is not recommended for use on new plots which have not
previously been interpreted using aerial photography.









INTRODUCTION
The National Wetlands Inventory (NWI) monitors, maps, evaluates and
disseminates information on the wetlands of the United States. An important facet of this
work is carried out by the Wetlands Status and Trends program, which gathers
information on wetland acreage status, and analyzes trends in wetland changes. The NWI
currently uses aerial photography in both its national wetlands mapping and the Status and
Trends program, but has been investigating the potential of satellite imagery for use in
these applications.
In May 1992, funding was approved to investigate the potential use of satellite
imagery in the Status and Trends program. This document is a report on the results of this
research. A research work order was carried out at the Florida Cooperative Fish and
Wildlife Research Unit, based at the University of Florida. The Florida Cooperative Fish
and Wildlife Research Unit specializes in wetland ecology, and has experience using
satellite imagery for wetland vegetation mapping.
The research work order contained two tasks aimed at answering the following
questions:
Can satellite imagery be analyzed to detect acreage changes consistent
with past Status and Trends work?
What are the size limitations (acres) inherent in this process?
Can cover type changes among wetland and upland types be consistently
identified?
Can gains in wetland acreage be identified and measured using SPOT?
What are the recommended procedures for using satellite data alone, and
in combination with aerial photography?


Task 1 was an adaptation of the current Status and Trends change detection
process, using SPOT satellite imagery along with aerial photography. Task 1 included the
preparation of satellite data, the preparation of Status and Trends monitoring plot data,
the detection of wetland change within the plots, and the evaluation of image
interpretation. Task 2 was a report on the results of Task 1.









PILOT STUDY


In the months before the work order started, a preliminary study of techniques for
incorporating satellite imagery into Status and Trends' wetlands change detection
procedures was done. A set of Status and Trends wetland monitoring plots was selected
for the initial study. The set included 25 four-square-mile plots in Florida, for which
1950's, 1970's and 1980's aerial photography and plot overlays and 1990's airborne video
data were available. The plot overlays were polygon maps of the wetlands and uplands in
each plot, interpreted from the aerial photography, and used by Status and Trends to
estimate wetland change.
Several SPOT panchromatic and multispectral satellite images were borrowed
from the University of Florida Map and Image Library and evaluated to see whether they
contained any of the Status and Trends plots. Two images were selected for study: one
covering the Ocala area, dated 1987, and one covering the Crystal River area, dated 1989.
In addition, a Landsat TM image covering the Ocala area and a vegetation classification of
that image done by the Florida Department of Game and Fish were borrowed from the
Florida Cooperative Fish and Wildlife Research Unit's image collection.
SPOT panchromatic image data has 10 meter resolution, the best spatial resolution
in satellite data available to the public at the time of the study. SPOT multispectral data
has 20 meter resolution, while Landsat TM has 30 meter resolution. However, the
Landsat TM sensor records data from 7 separate portions (data bands) of the visible and
infrared parts of the electromagnetic spectrum. The SPOT multispectral sensor records
data from 3 bands of the visible and near-infrared parts of the spectrum, while the SPOT
panchromatic sensor records only a single broad band covering the visible green and red
parts of the spectrum (see Appendix A). Though it has less spectral resolution than
Landsat TM, SPOT satellite data was chosen as the focus of both the preliminary and
main studies because of its finer spatial resolution. Both the spectral and spatial
resolutions of SPOT data can be maximized by merging the panchromatic and
multispectral data. Merged SPOT data have been successfully used by the Florida
Cooperative Fish and Wildlife Research Unit to map wetland species at the Loxahatchee
National Wildlife Refuge and Lake Okeechobee (Richardson et al. 1990).
For the pilot study, SPOT images were merged by computer using the process
described in Appendix A. A normalized difference vegetation index (NDVI) was
calculated for each pixel from the red and infrared image bands (see Appendix A). The
NDVI is a difference ratio of red to infrared reflectance that highlights vegetation and was
used in addition to the three image bands. The images were then rectified using the









process described in Appendix B. All computer processing for both the preliminary study
and the research work order was done using ERDAS 7.5 software. (A list of ERDAS
programs used is given in Appendix C.)
The locations of five Status and Trends plots were identified on the satellite
images. These were plots numbered 1924 (Irvine), 2025 (Silver Springs), 2079
(Lochloosa Lake), 327 (Sand Slough), and 2057 (Rainbow Springs). The SPOT satellite
images of these plots were compared with aerial photographs from the 1970's and early
1980's, and with the plot overlays. The airborne video data were not used, because they
were too variable spectrally and difficult to register to the plots. Although the time
between the most recent aerial photographs and the satellite images was only 2 to 4 years,
wetland change was clearly visible in plot 1924, due to the clearing of palustrine forest.
The SPOT images were also compared with the Landsat TM image and the TM-derived
vegetation classification, to see whether additional information was found in these data.
No additional information was found, and fewer landscape details were visible. The
results of this preliminary work suggested that wetland change could be detected from
comparisons of SPOT satellite imagery and aerial photography.









METHODS


Based upon procedures used by the Florida Cooperative Fish and Wildlife
Research Unit for satellite image-based wetland mapping, the study was planned in six
phases shown in Figure 1.

Figure 1. Study flow chart


Satellite Image Acquisition
The first phase of the research began with a request to the SPOT Image
Corporation to search their archives for simultaneous pairs of multispectral and
panchromatic satellite images that covered the same area on the earth. The image pairs
were also required to: 1) be recent images of Florida, 2) cover a variety of wetland and
upland cover types, and 3) contain as many Status and Trends plots as possible. Digital
tapes of three image pairs that met these criteria were purchased from SPOT for the study.
In addition, the 1989 Crystal River scene used in the preliminary study was included. The
location of the scenes are shown in Figure 2. (See Appendix A for SPOT Scene
Identification Numbers).









METHODS


Based upon procedures used by the Florida Cooperative Fish and Wildlife
Research Unit for satellite image-based wetland mapping, the study was planned in six
phases shown in Figure 1.

Figure 1. Study flow chart


Satellite Image Acquisition
The first phase of the research began with a request to the SPOT Image
Corporation to search their archives for simultaneous pairs of multispectral and
panchromatic satellite images that covered the same area on the earth. The image pairs
were also required to: 1) be recent images of Florida, 2) cover a variety of wetland and
upland cover types, and 3) contain as many Status and Trends plots as possible. Digital
tapes of three image pairs that met these criteria were purchased from SPOT for the study.
In addition, the 1989 Crystal River scene used in the preliminary study was included. The
location of the scenes are shown in Figure 2. (See Appendix A for SPOT Scene
Identification Numbers).









Figure 2. Location of satellite scenes


Each satellite scene contained 11 plots, and covered the areas described below:

SCENE 1
This image extends from Hawthorne southward to Ocala, and from Orange Lake
eastward to the St. John's River. The date is May 1, 1990. Due to the satellite path
alignment, the coverage is not a square aligned north-south, but rather a parallelogram
aligned northeast to southwest. The area covered by both the panchromatic and
multispectral data has the following corner points:


N 29 46' 29"
W 82 12' 38"


N 29 41' 01"
W8131' 24"


N 29 14' 52"
W 82 20' 02"


N 29 09' 15"
W 8139' 01"









SCENE 2
This image extends from Ocala southward to Wildwood, and from the
Withlacoochee State Forest eastward to the Ocala National Forest. It is immediately
south of the Scene 1 image, on the same flight path, taken on the same date: May 1,
1990. The area covered by both the panchromatic and multispectral data has the
following corner points:


N 29 16'45"
W 82 19' 34"


N 29 11' 09"
W 81 38' 32"


N 28 45' 08"
W 82 26' 57"


N 28 39'32"
W 8146' 07"


SCENE 3
This image extends from the southeastern edge of Lake Okeechobee southward to
Water Conservation Area 2, and from the Everglades Agricultural Area eastward to the
Atlantic coast. The date is April 5, 1990. The area covered by both the panchromatic and
multispectral data has the following corner points:


N 26 47' 48"
W 80 43' 57"


N 26 42' 32"
W 80 09' 16"


N 26 16' 04"
W 80 51' 47"


N 26 10' 47"
W 80 17' 15"








SCENE 4
This image extends from Lebanon to Homosassa Springs, and from offshore in the
Gulf of Mexico eastward to the Withlacoochee State Forest. It is adjacent to Scene 2,
lying on the flight path just to the west, but is taken on an earlier date. The date is April
26, 1989. It was obtained from the University of Florida library. The area covered by
both the panchromatic and multispectral data has the following corner points:

N 29 17' 36" N 29 12' 03"
W 82 51'48" W 82 18'39"








N 28 45' 55" N 28 40' 24"
W 83 00' 13" W 82 27' 14"



Satellite Image Processing
The satellite data were downloaded from the tapes using a 1/2 inch magnetic tape
drive linked to a Gateway 486 33 MHz microcomputer, with a 1.2 gigabyte hard drive and
a super VGA monitor, running VGA ERDAS version 7.5 image processing software. The
satellite data were computer processed to enhance their spatial resolution, including
registering the pairs of multispectral and panchromatic images for each scene together,
and merging the pairs using a resampling and color transformation process. A normalized
difference vegetation index (NDVI) was calculated for each merged image. (See
Appendix A for additional information on the SPOT satellite and a detailed description of
the image processing).
Once the image pairs were merged, the locations of all Status and Trends plots
were identified by displaying the images on the computer monitor and referring to
photographs and USGS 1:24,000 topographic quadrangle maps marked with the plot
outlines. Eleven plots were found in each of the four images. The locations were
recorded by image file row and column numbers, and subsets containing each plot were
extracted from the images and placed in separate files for convenience in handling. Each
plot file was rectified to Universal Transverse Mercator (UTM) map coordinates by
matching control points on USGS 1: 100,000 scale topographic maps (see Appendix B).











Plot Overlay Digitization
The preparation of the Status and Trends monitoring plot data was carried out at
both the National Wetlands Inventory and the Florida Cooperative Fish and Wildlife
Research Unit. Plot overlays interpreted from 1980's photography were digitized by the
National Wetlands Inventory using WAMS software for the plots in the study. The
digitizing was georeferenced using USGS 1:24000 topographic quadrangle maps with the
plot outlines drawn on them. The digital files were converted to "digital line graph"
(DLG) format, and sent to the Florida Cooperative Fish and Wildlife Research Unit. The
digital files were converted from DLG format to ERDAS "digfile" format, to allow them
to be displayed on the computer monitor over the images of the plots. Although the
format conversion of the digital files was generally successful, some corrections were
necessary, as described below.


Display of Images with Overlays
As an initial check of each Status and Trends plot, the satellite image was first
displayed on the computer monitor, then the polygons of the digitized plot overlays were
displayed over the image, with the image showing through. A color palate file was made
for the digitized files so that the wetland and upland classes in all the plot overlays were
displayed with a consistent and easily interpreted color scheme.
The initial check showed that many of the digital files were spatially offset from the
images. The reasons for this registration problem are addressed Appendix B. The mis-
registered files were shifted in the x and y directions until they matched the image. In
addition, it was found that when the overlays were digitized, the numerical values assigned
to wetland classes were not consistent; for example, palustrine forest was assigned class
numbers 6 and 14 in one digital file, and 2 and 7 in another digital file. Once this problem
was detected, the digitizing procedure was modified to prevent it. The affected digital
files (approximately half of the plots in the study) were edited using a text editor to make
all class values consistent.


Satellite Image Interpretation
The interpretation of land covers and the detection of wetland changes within the
plots required a familiarity with each plot, and with the appearance of wetland classes
throughout all the images. For each plot, the satellite image was displayed on the monitor











Plot Overlay Digitization
The preparation of the Status and Trends monitoring plot data was carried out at
both the National Wetlands Inventory and the Florida Cooperative Fish and Wildlife
Research Unit. Plot overlays interpreted from 1980's photography were digitized by the
National Wetlands Inventory using WAMS software for the plots in the study. The
digitizing was georeferenced using USGS 1:24000 topographic quadrangle maps with the
plot outlines drawn on them. The digital files were converted to "digital line graph"
(DLG) format, and sent to the Florida Cooperative Fish and Wildlife Research Unit. The
digital files were converted from DLG format to ERDAS "digfile" format, to allow them
to be displayed on the computer monitor over the images of the plots. Although the
format conversion of the digital files was generally successful, some corrections were
necessary, as described below.


Display of Images with Overlays
As an initial check of each Status and Trends plot, the satellite image was first
displayed on the computer monitor, then the polygons of the digitized plot overlays were
displayed over the image, with the image showing through. A color palate file was made
for the digitized files so that the wetland and upland classes in all the plot overlays were
displayed with a consistent and easily interpreted color scheme.
The initial check showed that many of the digital files were spatially offset from the
images. The reasons for this registration problem are addressed Appendix B. The mis-
registered files were shifted in the x and y directions until they matched the image. In
addition, it was found that when the overlays were digitized, the numerical values assigned
to wetland classes were not consistent; for example, palustrine forest was assigned class
numbers 6 and 14 in one digital file, and 2 and 7 in another digital file. Once this problem
was detected, the digitizing procedure was modified to prevent it. The affected digital
files (approximately half of the plots in the study) were edited using a text editor to make
all class values consistent.


Satellite Image Interpretation
The interpretation of land covers and the detection of wetland changes within the
plots required a familiarity with each plot, and with the appearance of wetland classes
throughout all the images. For each plot, the satellite image was displayed on the monitor









with the line-work of the overlay displayed over it. The three satellite data bands and the
normalized difference vegetation index (NDVI) were displayed individually and in several
combinations to highlight and separate different land cover classes. The image was visually
interpreted, and examined for changes. The plot's aerial photography and topographic
quadrangle map were used as references. This referencing of previous sources of data is
consistent with the wetland change detection procedures currently used by Status and
Trends. A description of each plot and any wetland or upland changes were noted. The
types and number of wetland classes, the number of polygons, the offset of the digital file,
and the date of the topographic quadrangle map used in digitizing were also noted.
Once all plots had been described, similar plots were again examined in close
succession, to gain an understanding of the variation in the appearance of wetland classes
within a region, and between regions. Then the image of each plot was again displayed on
the monitor with the line-work of the overlay map. Using an ERDAS program that
measures areas digitized directly off the monitor screen, changes were measured and
noted. (See Appendix C for a list of ERDAS programs used in this project.)


Evaluation of Image Interpretation
The final phase of the research was an evaluation of image interpretation,
particularly the ability to interpret small patches of a land cover type. A visual guideline
for 1-acre polygons on the overlay map was established by measuring small polygons until
several polygons close to 1 acre were found. These were used as references to identify all
polygons smaller than 1 acre. If a polygon was unusually shaped or appeared close to 1
acre in size, it was measured to verify its size. The small polygons were counted, and then
each one was evaluated to see how well it could be interpreted on the image.
If the land cover class of a small polygon could be identified with confidence
without referring to the plot's aerial photography or topographic quadrangle map, it was
categorized as "identified using the image alone". If the class could be confidently
identified after referring to the photography or the topographic map, it was categorized as
"identified using the image and ancillary data". If there was no feature on the image linked
by size and location to the polygon, and the surrounding area did not indicate that a
change had taken place, then the polygon was categorized as "not identified". If the
surrounding area indicated that a land cover change had taken place, then the polygon was
excluded from the count, because presumably it no longer existed. In addition,
interpretation problems with specific classes were noted, both where change did and did
not occur.











RESULTS


A list of the plots analyzed is given in Table 1. Information on each plot is
included in the table: the digital overlay file(s) for each plot, the topographic quadrangle
map used in digitizing the overlay and its date, and the number of meters in the x and y
direction that the digital plot was shifted to match the image. Plots required two digital
overlay files if they straddled the border between two topographic quadrangle maps. An
"a" or "b" after the plot number indicates the digital file is one of two needed for that plot.
An "x" indicates the digital file was adjusted for registration offset.










Table 1. Status and Trends plots analyzed, their quadrangle map sheets and plot offsets
Scene Plot # Digfile 1 Quad name-date Digfile 2 Quad name-date Shift x,y
(in meters)
1 1735 1735x Fairfield-68 -30, -20
1 1924 1924x Flemington-69 2144xx Mclntosh-68 -30, 0
1 1949 1949a Eureka Dam-81 1949b Lk. Delancy-70
1 1978 1978ax Ft. McCoy-70 1978bx Lk. Kerr-70 -40, 30
1 1982 1982x Welaka-80 -40, 80
1 2079 2079ax Hawthorne-66 2079bx Rochelle-66 -60, 40
1 2120 2120x Baywood-70 -80, 50
1 2144 2144x Mclntosh-68 -10, 20
1 2182 1978ax Ft. McCoy-70 -40, 30
1 2246 2246x Putnam Hall-70 -80, 70
1 2259 1982x Welaka-80 -40, 80
2 359 359x Emeralda Is.-66 0, -25
2 366 366a Juniper Spring-72 366b Farles Lk.-72
2 1709 1709 Dunnellon-54
2 1758 1758x Ocala West-68 -30, 10
2 1930 1930x Belleview-67 -20, -30
2 1948 1948 Lecanto-54
2 1975 1975x Stokes Ferry-54 1709 Dunnellon-54 -20, -10
2 2025 2030b Ocala East-67
2 2030 2030a Lynne-70 2030b Ocala East-67
2 2096 2096 Shady-68
2 2236 2030ax Lynne-70 2030bx Ocala East-67 -20, -30
3 350 350a Loxahatch. NW-70 350b Sixmile Bend-70
3 370 370x Ft. Laud. 2SE-69 20, 20
3 376 376x Ft. Laud. 2SW-73 20, 0
3 1907 1907 Belle Glade-70
3 1926 1926 Greenacres-67
3 2026 2026 Evergl. INW-74
3 2077 2077x Evergl. 1NE-74 -30, 30
3 2092 2092x Ft. Laud. 2NW-74 376xx Ft. Laud. 2SW-73 110, 0
3 2132 2132 Loxahatch. SE-70
3 2247 2247 Okeelanta-70
3 2264 2264x Evergl. 2NE-74 0, -60
4 327 327 Tidewater-55
4 416 416 Red Level
4 432 432x Ozello
4 461 461ax Withlacooch. Bay 461bx Yankeetown 20, 40
4 482 482 Ozello 416 Red Level
4 1741 1741 Romeo-54
4 1803 1803 Ozello 0, -20
4 2057 2057 Dunnellon-54
4 2147 2147 Crystal River-54
4 2176 2208 Lecanto-54
4 2208 2208x Lecanto-54 -40, 30








Table 2 lists all the wetland and upland classes found in the Status and Trends
plots studied, and the abbreviations for them used on the overlay maps.


Table 2. Cover type classes


Abbreviation Cover type class
Elab estuarine subtidal aquatic bed
Elub estuarine subtidal unconsolidated bottom
E2ab estuarine intertidal aquatic bed
E2em estuarine intertidal emergent
E2fo estuarine intertidal forested
E2ss estuarine intertidal scrub/shrub
E2us estuarine intertidal unconsolidated shore
Lac lacustrine
Pab palustrine aquatic bed
Pem palustrine emergent
Pfo palustrine forested
Pss palustrine scrub/shrub
Pub palustrine unconsolidated bottom
Pus palustrine unconsolidated shore
Riv riverine
Ua upland agriculture
Ub urban
Uo upland other


Table 3 lists the number of polygons contained in each plot, and the number and
types of classes in each plot. The total number of polygons in the study was 2,508. In
general, Scene 1 plots contained a mixture of forests and wetlands, and Scene 2 plots
contained forests, urban, agriculture and fewer wetlands. Scene 3 plots contained either
sugarcane agriculture, Everglades wetland or dense urban, and Scene 4 plots were either
similar to Scene 2 plots or contained Gulf Coast wetlands.










Table 3. Number of polygons and wetland classes in each plot


Scene Plot # # polygons # classes Classes
1 1735 65 8 Pab, Uo, Ua, Ub, Pfo, Pem, Pss, Pub
1 1924 76 8 Uo, Ua, Ub, Pfo, Pss, Pab, Pem, Pub
1 1949 1 lu Uo
1 1978 77 7 Uo, Lac, Ub, Pem, Pfo, Pub, Pss
1 1982 25 5 Uo, Riv, Pfo, Pem, Pss
1 2079 119 6 Uo, Pfo, Pem, Pss, Pub, Lac
1 2120 28 4 Uo, Pfo, Pss, Pem
1 2144 81 8 Pem, Ua, Uo, Lac, Ub, Pss, Pfo, Pub
1 2182 44 8 Uo, Lac, Ub, Pss, Pfo, Pem, Pub, Pab
1 2246 20 6 Ub, Uo, Lac, Pem, Pfo, Pub
1 2259 97 9 Lac, Uo, Ub, Riv, Pfo, Pem, Pss, Pab, Ua
2 359 30 7 Ua, Uo, Lac, Riv, Pfo, Pss, Pem
2 366 1 lu Uo
2 1709 4 3u Uo, Ua, Ub
2 1758 10 4 Uo, Ua, Ub, Pfo
2 1930 56 7 Ub, Uo, Ua, Pab, Pub, Pem, Pfo
2 1948 1 lu Uo
2 1975 20 5 Uo, Ua, Ub, Per, Pub
2 2025 57 7 Uo, Pfo, Ua, Ub, Riv, Pss, Pem
2 2030 40 8 Pss, Riv, Pfo, Pub, Pab, Pem, Uo, Ub
2 2096 18 7 Ua, Uo, Ub, Pem, Pss, Pub
2 2236 61 6 Uo, Pfo, Ua, Riv, Pss, Pem
3 350 5 2u Riv (canal), Ua
3 370 48 7 Ub, Riv, Pub, Ua, Uo, Pss, Pem
3 376 38 3w Pfo, Pss, Pem
3 1907 30 9 Ua, Ub, Uo, Riv (canal), Lac, Pub, Pem, Pfo, Pss
3 1926 87 7 Ub, Ua, Uo, Pfo, Pss, Pem, Pub
3 2026 1 lu Ua
3 2077 4 2 Ua, Pss
3 2092 4 2w Pss, Pem
3 2132 193 3w Pss, Pfo, Pem
3 2247 1 lu Ua
3 2264 9 3 Ua, Pem, Pss
4 327 105 7 Pfo, Pss, Pem, Lac, Ua, Uo, Pub
4 416 135 8 Elab, Elub, E2em, E2fo, Ub, E2us, Uo, E2ss
4 432x 57 6w Elab, Elub, E2em, E2ss, E2ab, E2us
4 461 315 7 Elub, E2em, E2f0, E2us, E2ss, Uo, Pem
4 482 85 4w Elab, E2us, E2ss, Elub
4 1741 9 4 Uo, Ua, Ub, Pem
4 1803 315 4w Elub, E2em, E2fo, E2us
4 2057 8 5 Ua, Uo, Ub, Pem, Riv
4 2147 122 8 Riv, Pub, Pab, Ub, Uo, Pfo, Pem, Pss
4 2176 3 2u Uo,Ub
4 2208 3 2u Uo, Ua
"w" indicates the plot contains wetland classes only
"u" indicates the plot contains upland classes only










A description of the landscape in each plot, the image bands used for
interpretation, how well the overlay aligned with the image after adjustment, data on small
polygons and areas of change, and comments on the difficulty of interpretation are given
in Appendix D. The total number of acres in all 44 plots was 111,630. Two plots (2092
and 2264) could not be evaluated for change because of extreme overlay mis-registration,
(see Appendix D), so the number of acres evaluated for change was 106,510. The total
number of acres changed in the plots was 1,450. A breakdown of this number by class is
shown in Table 4. Net changes from one class to another appear in the summary box at
the bottom of Table 4. Classes not appearing on this list did not change.
The changes found in the study plots differed somewhat from trends in Florida as a
whole. Agriculture decreased in the area of study due to urbanization, even though it had
been increasing overall in the state. Fewer forested and other undeveloped uplands were
converted to urban use compared to the state as a whole, because large tracts of federal
and state forest lands were in the study area. There were also larger gains in palustrine
scrub/shrub than might be expected. This was a result of palustrine forest logging, where
palustrine vegetation was allowed to regenerate rather than being converted to an upland
use. While there was a net loss of palustrine forest in the study as a whole, palustrine
forest increased in the Everglades plots, where water management had changed
vegetation.









Table 4. Acres changed, by class.
Rows = Class in 1980's. Columns = Class in 1990.
Row sums = Loss by class. Column sums = Gains by class.


1990
Pfo Pem Pub Pss Lac Ua Uo Ub Elab Elub
Pfo 435.46 26.2 1.7
Pem 27.9 2.42 241.41 6.9 2.5 20.63 1
Pub 9.2
Pss 50.63 5 4.2
Lac
Ua 56.5 47.2 263.4
Uo 38.34 140.9
Ub 7.5
Elab 60.67
Elub___

Pfo Pem Pub Pss Lac Ua Uo Ub Elab Elub
Loss 463 303 9 60 0 367 179 7 61 0
Gain 79 0 105 677 7 2 108 411 0 61
Net -384 -303 96 617 7 -365 -71 404 -61 61


Data on the distribution of polygons in the plots are given in Table 5. In the upper
half of the table, the mean number of polygons of all sizes per plot, the mean number of
small polygons per plot (polygons less than 1 acre), and the percentage of small polygons
are shown. This information is summarized graphically in Figure 3. Data on the
interpretability of small polygons are listed in the lower half of Table 5. This information
is summarized graphically in Figure 4.
The most common classes of small polygons in the study were palustrine
emergent, palustrine unconsolidated bottom, and palustrine scrub/shrub. Additional small
polygons were palustrine forested, estuarine intertidal forested, estuarine intertidal
scrub/shrub, estuarine intertidal unconsolidated shore, estuarine intertidal emergent,
upland other, and estuarine subtidal unconsolidated bottom.









Table 5. Distribution of polygons and their interpretability


For all plots For all plots For Gulf Coast
except Gulf plots only
Coast plots
Mean number of all 57.00 41.05 181.40
polygons per plot

Mean number of small 24.98 15.15 101.60
polygons per plot

Percentage of polygons that 44 37 56
were small
Interpretability of
polygons < 1 acre:
Percentage of small 34 23 47
polygons identified by
image alone

Percentage of small 45 49 41
polygons identified by
image and ancillary data

Percentage of small 21 29 12
polygons not identified









Figure 3. Average number of polygons per plot, showing polygons larger and smaller than
1 acre


not gulf
gulf only


Figure 4. Interpretability of polygons smaller than 1 acre from all plots


O percent polygons
identified by
image alone
* percent polygons
identified with
aid
* percent polygons
not identified


O > 1 acre
S< acre


140

120

100

80


all
plots


60
50
40
30
20
10


0 ---


FIR









The five plots along the coastline of the Gulf of Mexico were more complex than
the other plots in the study, which increased up the mean number of polygons per plot.
Therefore, the small polygon data from the Gulf Coast plots are also presented separately
in Table 5, to show results with and without their effect.
The difference in spatial complexity between the Gulf Coast plots and the plots in
the rest of the study is illustrated in Figure 3. Not only were there more polygons per Gulf
Coast plot, but a larger proportion of them were less than 1 acre.
Differences in the spatial complexity of the landscape among the four satellite
images are also apparent when the same data are presented by scene, as shown in Table 6.
The data are shown graphically in Figure 4. The Scene 1 and Scene 2 landscapes were a
mixture of forests, wetlands, urban, and agriculture. Scene 1 plots generally contained
more polygons than Scene 2 plots, particularly small polygons, because Scene 1 contained
more wetlands. However, the percentage of small polygons that could not be identified
was similar in both scenes (28% and 29%). Most problems identifying small polygons
arose from small patches of palustrine forest within upland forest, and small wetlands
surrounded by forest. In Scene 3, intensive agriculture and dense urban development
simplified the landscape pattern, reducing the number of polygons in the scene. This
simplification of the landscape, plus the distinctive appearance of Everglades wetland
types kept the percentage of unidentified small polygons low (15%). Scene 4 contained
urban areas, agriculture and forest, with similar small polygon identification problems to
Scenes 1 and 2, but also contained coastal wetland types where small polygons were easily
identified. This combination resulted in an intermediate percentage of unidentified small
polygons (19%).









Table 6. Distribution of polygons and their interpretability, by satellite scene


Scene 1 Scene 2 Scene 3 Scene 4
Mean number of all 57.55 27.09 38.18 105.18
polygons per plot

Mean number of small 23.36 8.18 12.55 55.82
polygons per plot

Percentage of polygons that 41 30 33 57
were small
Interpretability of
polygons < 1 acre:
Percentage of small 7 7 75 40
polygons identified by
image alone

Percentage of small 65 64 9 42
polygons identified by
image and ancillary data

Percentage of small 28 29 15 19
polygons not identified



Figure 5. Average number of polygons per plot by satellite scene, and the interpretability
of small polygons

120

100
0 polygons > 1 acre
80
Spolo goni < 1 acre,
60 identified by image
alone
40 polygons < 1 acre,
identified with aid
20
] polygons < 1 acre, not
0 4 identified
Scenel Scene 2 Scene 3 Scene4









Concern over the difficult interpretation ofpalustrine forest and small wetland
polygons within forest cover prompted a closer examination of palustrine forest data.
Table 7 shows data on the distribution of palustrine forest by scene and for the entire
study area. Data on small polygons are included. Table 8 shows data on the distribution
ofpalustrine unconsolidated bottom. This cover type is typical of wetlands found in small
polygons surrounded by forest (although it is also found in urban and agricultural
settings).
Palustrine forest was widely distributed and a major wetland class, making up
7.35% of the study area. Twenty percent ofpalustrine forest polygons were less than 1
acre. However these small polygons made up only 0.4% of the acreage in palustrine
forest, because some palustrine forest polygons were very large (over 1000 acres). In
contrast, palustrine unconsolidated bottom is a small class: 0.27% of the study area. Fifty
one percent of palustrine unconsolidated bottom polygons were less than 1 acre, making
up 11% of the palustrine unconsolidated bottom acreage. Palustrine unconsolidated
bottom is a small wetland class consisting of small polygons, however small wetlands
should not overlooked. Their wildlife habitat value is great, particularly when surrounded
by large tracts of uplands.









Table 7. Distribution of palustrine forest (Pfo)


Scene 1 Scene 2 Scene 3 Scene 4 Entire
Study
Number of plots 10 6 4 2 22
containing Pfo

Total area of Pfo (in 5,412.93 1,723.42 166.99 902.59 8,205.93
acres)

Percent area in Pfo 19.2 6.1 0.6 3.2 7.35

Total number of 139 53 20 56 268
Pfo polygons

Number of Pfo 15 3 1 2 21
polygons <.5 acres

Number of Pfo 32 9 7 6 54
polygons < 1 acre

Total area of Pfo 17.51 5.77 5 4.19 32.47
polygons < 1 acre
(in acres)

Percent of Pfo in 0.33 0.33 2.99 0.46 0.40
polygons < 1 acre

Percent of scene in Pfo 0.062 0.020 0.018 0.015 0.029
polygons < 1 acre









Table 8. Distribution ofpalustrine unconsolidated bottom (Pub)

Scene 1 Scene 2 Scene 3 Scene 4 Entire
Study
Number of plots 7 4 3 2 16
containing Pub

Total area of Pub (in 95.89 26.42 155.85 28.29 306.45
acres)

Percent area in Pub 0.341 0.0938 0.574 0.101 0.2745

Total number of Pub 37 23 70 20 150
polygons

Number of Pub 14 10 15 10 49
polygons <.5 acres

Number of Pub 22 14 28 13 77
polygons < 1 acre

Total area of Pub 10.24 5.65 14.56 4.55 35
polygons < 1 acre
(in acres)

Percent of Pub in 10.7 21.4 9.3 16.1 11.42
polygons < 1 acre

Percent of scene in 0.36 0.02 0.054 0.016 0.031
Pub polygons < 1 acre









DISCUSSION
Satellite imagery was used in this study to identify wetland acreage changes,
following current Status and Trends change detection procedures. The use of satellite
imagery, as would any alteration in the technique used for change detection, affected
wetland acreage figures by introducing new errors into the trend data and eliminating
others. To keep the effects of this new technique to a minimum, and to allow the
comparison of past wetland acreage changes with those found using satellite imagery, the
current Status and Trends procedures were emulated as closely as possible. As part of
this effort, the satellite data were interpreted visually rather than classified by computer to
identify land cover types.
Satellite image computer classification is based upon digital data of spectral
reflectance (light reflected from the surface of the earth). Aerial photographs are also
records of spectral reflectance-- analog records. In this respect, satellite image computer
classifications and photo-interpretations are similar. However, the interpretations
performed by Status and Trends photo-interpreters use more information than what is
visible in a set of aerial photographs from a single point in time. The information used
includes photographs from earlier dates, interpretations of these earlier photographs, and
topographic maps. In addition, interpreters use logic and experience to synthesize this
ancillary information, allowing them to make interpretations even when visual clues are
lacking. This human factor, combined with the fine spatial resolution of photography and
the ability to see in stereo, gives Status and Trends plot interpretations greater detail and
accuracy than would be possible in a wetland classification done using satellite imagery
alone.
The advantages of this human factor were illustrated in many of the photo-
interpreted plot overlays used in this study. For example, a barren area in one plot, where
palustrine forest had been cleared, appeared identical in the photograph to both an empty
lot in an urban area and a cleared agricultural field. The photo-interpreter used logic,
other visual clues, past interpretations, and a topographic map to classify the area as
palustrine scrub/shrub. The classification decision was probably based on the assumption
that the hydrology of the site was still palustrine. The topographic quadrangle map
indicated a low spot where water collected, and there were no visual clues on the
photograph to indicate drainage or other alterations in hydrology. If the hydrology of the
site was still palustrine, then vegetation succession would produce the palustrine
scrub/shrub cover type.









In examples found in other plots, no clear transitions between palustrine and
upland forest were visible on the photographs. Photo-interpreters drew boundaries
between the two land covers along elevation contours found on topographic maps. In still
other plots, the color quality of the plot photographs was so poor that cover types could
not be identified. Here photo-interpreters based classification on interpretations of
photographs 10 years older, modifying boundary changes only.
These same techniques used on the satellite imagery in this study also allowed
interpretation when the image did not clearly show the land cover type. In the study, a
land cover polygon was labeled "not identified" if no difference could be seen between the
polygon and the surrounding area on the image. In practice, however, a change detection
judgment could still be made on about one third of these polygons. Imagine, for example,
a 1 acre patch of palustrine forest lying within a 100 acre upland forest. If no new roads,
clearings, or indicators of hydrologic change were visible in the forest, the polygon could
reasonably be designated "unchanged", even though it was "not identified" on the image.
The human judgment used in photo-interpretation may lead to errors if poorly
applied, (see Appendix D, Scene 3, plot 376), but it can also allow efficient use of the
data resources at hand. Computer classification of satellite images is usually favored over
human visual interpretation of images, because it seems more objective and requires less
labor for large scenes. However, human judgment is also used in computer classifications
during the selection and identification of training sites. Classification problems such as
those mentioned above, may go detected. If detected, these problems must be addressed
after computer classification by manually editing the classified image. While a computer
classification algorithm could be developed to use ancillary data along with spectral
reflectance to assign land covers for the examples mentioned above, all the photography,
overlays, and topographic quadrangle maps would have to be digitized and referenced in a
geographic information system. The time and costs required to achieve this would be
extremely large.


The questions posed at the outset of the study will be addressed below:


Can satellite imagery be analyzed to detect acreage changes consistent with past
Status and Trends work?
With the aid of detailed past Status and Trends photo-interpretations, satellite
imagery was successfully analyzed to detect acreage changes on existing plots consistent
with past Status and Trends work. However, the SPOT satellite imagery used in this
study does not have adequate spatial resolution to perform a land cover classification









consistent with Status and Trends standards on an area never before interpreted. By using
all available Status and Trends plot information, the satellite images were successfully
interpreted visually. With some corrections, the overlay maps from the latest Status and
Trends update were successfully converted to digital format. They were compared with
the satellite images in the same manner that an acetate overlay is placed over a
photograph. Photographs from earlier dates, interpretations of these earlier photographs,
and topographic maps in hardcopy form were consulted. Changes from one wetland type
to another, conversions of wetlands to uplands, and gains in wetlands were successfully
identified and measured from the computer screen.
The visual interpretation of the satellite images differed from past Status and
Trends photo-interpretations in three ways. First, the details of individual tree crowns and
buildings could not be used as interpretation clues because they were not visible at the
images' 10-meter resolution. This loss of detail inhibited the detection of vegetated
polygons which were less than 1 acre and surrounded by forest cover. Second, the lack of
stereoscopic vision did not allow forest and shrub identification based upon height. This
was not an important factor in the study, because shrubs and forest had different spectral
signatures. However, it could be a significant problem if shrub and forest land covers
shared the same species, if their spectral signatures were very similar, and/or large areas of
shrub were undergoing succession to forest cover. Third, the ability to individually display
the images' green, red and infrared bands and the normalized difference vegetation index
(NDVI) allowed the interpreter to choose bands which best highlighted the appearance of
wetland vegetation. This improved the identification of wetlands, particularly in wetland
complexes such as the Everglades and Gulf coast, and revealed errors in some past photo-
interpretations.


What are the size limitations (acres) inherent in this process?
Theoretically, the smallest feature reliably detected in a 10 meter resolution
satellite image is a square object with an area of 400 square meters : the size of four pixels
(see Figure 6A). This is approximately equivalent to 0.1 acre. A feature must be is as
large as four pixels before it is certain that at least 1 pixel will be a record of that cover
type alone, rather than the feature being entirely composed of "mixed pixels". However, a
circular feature of 400 square meters, could still be entirely composed of mixed pixels (see
Figure 6B). Due to its shape, a circular feature must be larger than a square one to be
reliably detected: 628 square meters or 0.15 acres (see Figure 6C). Long, narrow
features, and features with convoluted shapes can have a greater area than this and still be
entirely composed of mixed pixels (see Figure 6D).









Figure 6. Pixel configuration examples for different shapes and sizes
(Dashed lines show pixel boundaries)


2
400 m square
Contains at least 1 unmixed pixel


J




L I I

A


628 m2 circle
Contains at least 1 unmixed pixel


400 m2 circle
May contain no unmixed pixels
I I I





K- i




B


628 m2 (approx.) complex shape
May contain no unmixed pixels


I I I I I
0 10 20 30
meters


In this study, successful identification of small polygons was not only dependent
upon the size and shape of the polygon, but also the spectral differences between the
polygon and surrounding cover types and the growth form of the vegetation. Certain
features with extreme spectral signatures, such as bright buildings and open water, were
identified even though they were smaller than a single pixel (.024 acres). This was
because their spectral signatures were so extreme that they dominated the pixel they
appeared in, even though other cover types were present in it. In the Gulf Coast plots,









where many islands of vegetation as small as 0.1 acre were surrounded by spectrally
dissimilar water or sand, the percentage of unidentified small polygons was 12%. In
Scene 3, where many small wetlands were surrounded by spectrally dissimilar urban
development, or small islands of woody vegetation were surrounded by marsh, 15% of
small polygons went unidentified. However, in the remaining study area, the percentage
of unidentified small polygons was twice as great. These unidentified small polygons were
largely found in upland forest and were vegetated palustrine land covers: forest,
scrub/shrub, emergent, and aquatic bed.
Forest cover is spectrally variable when it contains a mix of species at a variety of
densities. The texture of the forest canopy creates shadows and bright spots, and the
understory may show through the canopy where the tree density is low. In an aerial
photograph, this texture aids in the identification of tree and understory species. Available
satellite sensors, however, reduce this variation to a single value within each pixel, losing
the advantage of texture. Variation between pixels is still present, but this variation may
hinder rather than help in the interpretation of forest cover.
Forest cover in this study was spectrally distinct and easily separated from all other
land covers.. Small polygons of palustrine forest and even palustrine scrub/shrub were
very distinct when surrounded by palustrine emergent (as is common in the Everglades).
Small polygons of palustrine forest were very distinct from surrounding palustrine
scrub/shrub when viewing the normalized difference vegetation index (NDVI). Small
polygons of palustrine forest or palustrine scrub/shrub surrounded by non-forested uplands
were also easily identified. Difficulties in identifying forest were confined to separating
between the two forest types: upland forest and palustrine forest. (This problem will be
discussed later).
In contrast, small polygons of palustrine shrub, palustrine emergent and palustrine
aquatic bed were difficult to identify when found within forests. After examining the aerial
photographs, it was found that these polygons were often partially obscured by
surrounding trees. The boundaries of these polygons were delineated by photo-
interpreters by observing small bits of the wetland cover type showing through the trees,
and with the aid of topographic quadrangle maps. When these small wetlands were
translated by the satellite into pixels, the reflectance of the trees and the reflectance of the
small wetland were combined. The spectral signatures of these mixed pixels were not
distinct enough from the variable surrounding forest to be identified. A canopy of large
trees at a small wetland's edge could extend up to 10 meters over the wetland. If the tree
canopy extended only 5 meters out over a theoretical circular pond, the pond would have
to be 1151 square meters (0.28 acres) to be reliably detected, (rather than 628 square









meters or 0.15 acres as in Figure 6C). If the tree canopy extended 10 meters out over a
theoretical circular pond, it would have to be 1831 square meters (0.45 acres) to be
reliably detected. In the study, vegetated wetlands within forest that were larger than 1
acre had enough pure pixels present that the polygons were identified.
Small polygons containing non-vegetated land covers within forest (such as urban,
palustrine unconsolidated bottom, lacustrine and riverine) were more spectrally distinct
from forest than were vegetated land covers. Therefore even when combined with the
reflectance of overhanging trees, their mixed pixels could be distinguished from forest
cover, allowing polygons was as small as 0.1 acres to be identified.
The most intractable land cover identification problem in the study was not size-
dependent: the separation between palustrine and upland forest. This problem was not a
simple lack of spectral resolution either. In four plots found in Scenes 1 and 2, palustrine
and upland forest could not be separated, no matter how large the polygon. More
commonly, particularly in the Ocala National Forest and Silver-Oklawaha River
floodplains, the boundaries between these two forest types shown on the overlays only
partially agreed with what appeared on the image. The image showed more complex
boundaries between upland and palustrine forest, and showed some gaps in both forest
types not shown on the overlays.
The reason for this disagreement lies in the definition of wetland versus upland
forest, and the difficulty in separating the two, even on the ground. "For the purposes of
this classification wetlands must have one or more of the following three attributes: (1) at
least periodically, the land supports predominantly hydrophytes; (2) the substrate is
predominantly undrained hydric soil, and (3) the substrate is non-soil and is saturated with
water or covered by shallow water at some time during the growing season of each year"
(Cowardin et al. 1979). For the purpose of delineating palustrine forest, the third
criterion does not apply. The second criterion cannot be determined by remote sensing,
although it can be estimated from topographic or soils maps (not generally well surveyed
in swamps). Palustrine forest is usually delineated by the first criterion: the dominance of
wetland plants. The difficulty is in determining the presence of hydrophytes using remote
sensing.
In photo-interpreted plot overlays, palustrine forest boundaries were frequently
drawn along the edges of cypress stands, or following the outlines of low elevation areas
appearing on the topographic map. The plot photographs were taken in February, when
bald cypress trees had a distinct blue-green hue in color infrared. Broad-leaved
hydrophytic trees were leafless at that time of year, and hard to distinguish from the pines
found mainly in the uplands. The photo interpretations relied on elevation rather than the









presence of broad-leaved hydrophytic trees to delineate palustrine forest that was not
cypress. On the other hand, the satellite images were taken in April, after the cypress and
broad-leaved hydrophytic trees had both produced leaves. The cypress was no longer
distinct from all other vegetation, but resembled the broad-leaved hydrophytic trees.
However, the broadleaf-cypress combination was more easily distinguished from pines on
the image, particularly with the normalized difference vegetation index (NDVI).
Complicating identification further is the fact that in this area pines occasionally occur in
palustrine forest, palms occur in both upland and palustrine forest, and broad-leaved
upland trees may be spectrally similar to broad-leaved hydrophytes.
Many palustrine forest polygons as large as 5 acres could not be seen on either the
image or the plot photographs. They had been identified as palustrine on the basis of
criterion 2 above (from the topographic quadrangle map). These polygons fell in the
"unidentified" category in this study, because they did not appear on the image, even
though they actually were identified.
Similarly, in plots offshore along the Gulf Coast, estuarine intertidal aquatic bed
polygons could not be distinguished from surrounding estuarine subtidal aquatic bed on
the image nor in the plot photographs. They also fell in the "unidentified" category.
However these two classes are separated by their relation to the mean high water mark,
which must be determined by topographic quadrangle maps rather than by remote sensing.
In summary, the size limitations of interpreting Status and Trends land cover
classes from satellite imagery are shown below in Table 9.









Table 9. Size limitations
imagery


of interpreting Status and Trends land cover classes from satellite


Can cover type changes among wetland and upland types be consistently identified?
Changes between upland and wetland cover types were very clear on the satellite
imagery, and could be consistently identified. Twenty-one different types of change were
identified (see Table 4). Changes in wetlands were largely due to human activity such as
road-building, construction, clearing for agriculture, logging and drainage. As mentioned
above, these activities were easily identified on the image, even if their extent was less
than 0.1 acres. Where water levels dropped, areas of non-vegetated substrate exposed by
the receding water stood out clearly on the imagery (see Appendix D, plot 2144).
Because human activity was such a clear indication of change, the boundary confusion
between palustrine and upland forest did not pose a problem for change detection. As
mentioned above, a change detection judgment could still be made on forest polygons
even though they were "not identified" on the image, based upon the presence or absence
of new roads, clearings, and other signs of hydrologic change in the forest. If palustrine


Cover Types Minimum Reliably Detectable Size:
(pixels) (m2) (acres)
Non-Forested
Uplands 4-8 400-800 0.1-0.2
Estuarine 8-12 800-1200 0.2-0.3
Riverine, Lacustrine 8 800 0.2
Palustrine Non- 8 800 0.2
vegetated (Pub, Pus)
.Palustrine Vegetated 12-40 1200-4000 0.3-1
(Pem, Pss, Pab)
Forested :
Upland and Palustrine 12 1200 0.3
Forest Surrounded by
Non-forest Covers
Upland Forest Within 40-200 400-2000 1-5
Palustrine Forest, and
Vice Versa









forest changed to upland forest and vice versa without human interference, the confusion
would be a significant problem.
Detecting changes among wetland cover types that occurred without obvious
human activity required more background knowledge and careful observation, but was still
accomplished successfully. This type of change included changes in wetland vegetation in
the Everglades due to water management, and shoreline changes on lakes and the Gulf
coastline. Change due to succession from palustrine scrub/shrub to palustrine forest was
found in the Everglades, but not elsewhere, even though it was searched for in the study.
This type of change may not have existed outside the Everglades, or it may not have been
detected without the use of stereoscopic vision to estimate vegetation height.
Changes from one upland cover type to another were numerous and easily seen on
the imagery. However, defining how much residential or low-density commercial
development could be introduced into an agricultural or other upland area before it was
reclassified as urban was difficult. Fortunately, this distinction was not essential to the
purpose of Status and Trends.


Can gains in wetland acreage be identified and measured using SPOT?
Gains in wetlands were successfully identified and measured from the computer
screen (see Table 4). In the study plots, wetland acreage gains were generally the result of
habitat degradation. The greatest gains were in palustrine scrub/shrub acreage, but these
came at the expense of palustrine forest (due to logging), and palustrine emergent (due to
drainage). Gains in wetland acreage from upland classes (Ua, Ub and Uo), were made by
human-made ponds (Pub) rather than vegetated wetlands.


What are the recommended procedures for using satellite data alone? In
combination with aerial photography?
The use of satellite data alone is not recommended. Either extensive ground truth
must be collected (not a part of current Status and Trends procedure), or the plot data
resources already possessed by Status and Trends (interpreted photography and
topographic maps) must be used. Topographic quadrangle maps should be the most
recent editions available. The photography does not need to be the same date as the
satellite imagery, but having some photographs that are less than 20 years older than the
imagery is recommended. A time sequence of color infrared photographs is optimal. If
satellite imagery is used for Status and Trends updates, alternating updates using satellite
imagery with updates using photography would take advantage of the strengths of both
data types.









Because spatial rather than spectral resolution limited the identification of small
wetlands, it is recommended that satellite data with the highest spatial resolution available
be used, provided it has at least green, red and near-infrared bands. Additional spectral
resolution would be useful, provided it was available at 10 meter or finer spatial
resolution. While additional infrared spectral information such as that found in Band 5 of
Landsat TM data might have been useful in distinguishing small wetlands from
surrounding forest, the minimum detectable feature at the 30 meter resolution of TM data
would be nine times larger than it was in this study. No advantage would be gained for
the detection of small wetlands.
Merged SPOT data were very satisfactory, and are recommended unless a superior
product is introduced in the future. The calculation and use of a vegetation index band is
highly recommended. The normalized difference vegetation index used in this study was
extremely useful. It is recommended that the satellite data be purchased pre-rectified.
This roughly doubles the cost of the data, but rectification requires experience and is time-
consuming. Contractors are also available for the merging of SPOT panchromatic and
multispectral data. Although merging SPOT data is less time-consuming than
rectification, it also requires a trained analyst.
It is strongly recommended that ArcInfo or some other software package with
similar capabilities be used to display the overlay files over the images and digitize changes
from a computer monitor. This would greatly increase options for the analysis of change
data. ERDAS 7.5 has practically no ability to edit and analyze vector files, and its
program for measuring areas directly from the monitor (used for measuring changes) is
imprecise.
While ERDAS 7.5 was a satisfactory image processing software package, it has
now been replaced by an improved ERDAS product. It is recommended that whatever
image processing package is used, it should be designed for use with satellite imagery, be
completely compatible with the vector software used to display the overlays (or have
sophisticated vector capability itself) and have the ability to quickly and easily display
individual bands of data and band combinations defined by the user. Recommended band
combinations for identifying wetlands are listed below in Table 10.










Table 10. Band combinations used to identify land covers


Bands Displayed Visual Effect
Band 2 (red) or Band 1(green) only Highlights shorelines, sandbars, suspended
sediment and some submerged vegetation
Band 3 (infrared) only Shows open water as black, shows
saturated soil as dark gray, highlights non-
vegetated dry areas
NDVI only Highlights lush biomass, separates broad-
leaved and coniferous trees (aids in the
identification ofpalustrine, estuarine and
upland forest)
Band 3 in red video channel, Band 2 in Approximates color infrared photography
green channel, Band 1 in blue channel
NDVI in red video channel, Band 2 in Separates palustrine vegetated classes
green channel, Band 1 in blue channel from each other
NDVI in red video channel, Band 3 in Separates wetlands from uplands,
green channel, Band 2 in blue channel especially useful for high-contrast mixes of
wetlands with all three upland types


It is possible to merge satellite imagery with aerial photography that has been
scanned into a digital form. Scanning photographs, joining adjacent photographs to make
up a scene, and registering and merging them with satellite imagery is a time consuming
process. This might be desirable if either a panchromatic or a multispectral satellite image
were unavailable, or if very fine spatial resolution was needed. However, this technique
was not tested in this study. As long as satellite imagery is being interpreted visually,
hard-copy photographs are recommended as references rather than scanned phonographs.
Several hard-copy photographs can easily be examined at one time and compared with an
image displayed on a monitor, while scanned photographs require a monitor for display.









REFERENCES


-- 1991. ERDAS Field Guide, Second Edition, Version 7.5. ERDAS, Inc., Atlanta,
Georgia.
Paine, David P. 1981. Aerial Photography and Image Interpretation for Resource
Management. John Wiley & Sons, New York.


Cowardin, Lewis M., Virginia Carter, Francis C. Golet, and Edward T. LaRoe. 1979.
Classification of Wetlands and Deepwater Habitats of the United States.
FWS/OBS-79/31 U.S. Fish and Wildlife Service, Office of Biological Services,
Washington DC.


Richardson, John R., Wade L. Bryant, Wiley M. Kitchens, Jennifer E. Mattson, and Kevin
R. Pope. 1990. An Evaluation of Refuge Habitats and Relationships to Water
Quality, Quantity and Hydroperiod. A Synthesis Report. November 1990.
Florida Cooperative Fish and Wildlife Research Unit, University of Florida.


LIST OF TABLES AND FIGURES

Figure 1. Study flow chart 6
Figure 2. Location of satellite scenes 6
Table 1. Status and Trends plots analyzed, their quadrangle map sheets and plot offsets 13
Table 2. Cover type classes 14
Table 3. Number of polygons and wetland classes in each plot 15
Table 4. Acres changed, by class. 17
Table 5. Distribution of polygons and their interpretability 18
Figure 3. Average number of polygons per plot, showing polygons larger and smaller than
1 acre 19
Figure 4. Interpretability of polygons smaller than 1 acre from all plots 19
Table 6. Distribution of polygons and their interpretability, by satellite scene 21
Figure 5. Average number of polygons per plot by satellite scene, and the interpretability
of small polygons 21
Table 7. Distribution of palustrine forest (Pfo) 23
Table 8. Distribution of palustrine unconsolidated bottom (Pub) 24
Figure 6. Pixel configuration examples for different shapes and sizes 28
Table 9. Size limitations of interpreting Status and Trends land cover classes from satellite
imagery 31
Table 10. Band combinations used to identify land covers 35









APPENDIX A-- SATELLITE IMAGE TRANSFORMATION:
TECHNIQUES AND BACKGROUND


SPOT Imagery
The satellite data used in this study came from the French SPOT Image
Corporation. The images used are listed by their SPOT Scene Identification Numbers
below.


Scene in Study Image Type SPOT Scene ID Number
scene 1 multispectral 12X618291900206161844
scene 1 panchromatic 11P618291900206161842
scene 2 multispectral 12X618292900206161853
scene 2 panchromatic 11P618292900206161851
scene 3 multispectral 1 1X623297900405160335
scene 3 panchromatic 12P623297900405160333
scene 4 multispectral 12X617292890426161806
scene 4 panchromatic 11P617292890426161804


The two SPOT satellites currently deployed move in sun-synchronous orbit 823
km above the earth. The swath width covered by the satellite sensors is 60 km, so each
satellite image covers a parallelogram that is 60 km on a side. Each satellite has two
sensors: one panchromatic and one multispectral.
The panchromatic ("black and white") sensor measures reflectance of visible light
in the green and red portion of the spectrum (.51-.73 micrometers). It senses an area on
the ground that is 10m by 10m, and then records this panchromatic reflectance value as a
data point or pixel. In other words, it has a 10 meter pixel resolution.
The multispectral sensor collects three data values for each pixel, and has a 20
meter pixel resolution. The three data values are recorded in separate files or bands.
Band 1 is the reflectance of green visible light (.50-.59 micrometers) Band 2 is the
reflectance of red visible light (.61-.68 micrometers). Band 3 is the reflectance of near
infrared radiation (.79-.89 micrometers).
SPOT multispectral sensor data can be merged with the panchromatic
sensor data. This creates an enhanced dataset that looks like a color image with 10m
resolution when it is displayed. The first stage of the merging process is to resample the
multispectral data to 10m resolution (cut the 20m pixels up into quarters), and register it









to the panchromatic data (make the pixels of the multispectral image line up with the
panchromatic).


Resampling
Every pixel has a data value for each band, and a row number (y coordinate) and
column number (x coordinate) to show its location in space. Resampling involves
reassigning these spatial coordinates, as shown below, which increases the size of the data
file by a factor of 4.

20 m pixel Resampled to 10 m pixels

x=1 x=2
y=l y=1
column (x) = 1 d=5 d=5

row (y) =1 x = x=2
y=2 y=2
data (d)= 5
d=5 d=5


Resampling of the multispectral data can be done simultaneously with registering
(matching the alignment of) the multispectral to the panchromatic. Registering involves
finding points on the two images which correspond (control points), and pairing the x and
y coordinates on each image for each point. These spatial coordinate pairs are run
through an algorithm which identifies the mean geometric offset of the points, and creates
an algebraic matrix to transform the coordinates of the multispectral pixels to the
coordinates of panchromatic pixels. However, it is important to note that while the
multispectral pixels are spatially redefined, the spectral data values are not altered. The
spectral data values of the original pixels are just reassigned to the nearest spatially
redefined pixel ("nearest neighbor" resampling).


Image Transformation
Once the two images are aligned with 10m resolution, the second stage of the
merging process is accomplished by incorporating the panchromatic data into the
multispectral data through a transformation procedure. The procedure used here involves
a color transformation, and is described below.









The data values in each of the three multispectral bands are usually assigned color
values for red, green and blue for display purposes. Color values range from 0 to 255.
Band 1 is given blue values, band 2 green, and band 3 red. When the color values of each
pixel's three bands are plotted in 3-dimensional color space, one of 16,777,216 possible
colors is defined. This pixel color can also be defined by an alternate colorspace, a polar
coordinate system known as intensity, hue and saturation (IHS).
To merge the panchromatic and multispectral data, the multispectral pixel colors in
red, green and blue colorspace are first transformed to intensity, hue and saturation values.
Then only the intensity value for each pixel is deleted and replaced by the panchromatic
reflectance datum for that location. Finally, the intensity, hue and saturation are
converted back to red, green and blue colorspace, but now the pixel colors contain
intensity values derived from the panchromatic information.
The advantage of this transformed data set is that it adds data with higher spatial
resolution to the multispectral data. (It is not possible to increase the resolution of the
multispectral data itself). When viewing the image, this greatly improves the ability of the
viewer to identify features such as small ponds, narrow roads and channels, buildings and
parking lots. The spectral resolution of the data is not increased because the panchromatic
sensor records data from the green and red portions of the spectrum already sensed by the
multispectral sensor.
During the color transformation process some data are also being lost (the
multispectral intensity values). In most cases, the lost data are better described by the
panchromatic data which replace them. Generally, when satellite data are being classified
based on statistics, unnecessary alterations of the original data should be avoided.
However, for images of very heterogeneous landscapes, or for applications which require
the finest resolution available, this particular alteration of the data is justified.


Vegetation Index
A normalized difference vegetation index, or NDVI, is a commonly used ratio of
the red to infrared bands which highlights vegetation with high biomass or photosynthetic
productivity. An NDVI data band, (band 4), was added to the images by calculating the
following formula for each pixel:

(band 3 band 2)/ (band 3 + band 2 + 0.5) 100


The 0.5 is added to avoid values of 0, and the multiplier of 100 is a scaling factor to put
values in the same range as the other bands.









The NDVI band is used in addition to the 3 other spectral bands, and its calulation
does not affect the data in the other bands.









APPENDIX B-- SATELLITE IMAGE RECTIFICATION AND
OVERLAY REGISTRATION: TECHNIQUES AND ERRORS


Rectification of a satellite image is the process of linking a map coordinate system,
(in this case UTM coordinates), to the pixels of the image. For this study, images were
rectified using 25 to 30 control points for each satellite image. These points were human-
made features, such as canals, road intersections and bridges, which could be clearly
identified on both the image and 1:100,000 scale USGS topographic maps. An attempt
was made to distribute the control points evenly throughout the image. A portion of the
image was displayed on the computer monitor, and a corresponding map was "locked
down" to the digitizing table by entering several UTM tic marks along the edge of the map
into the computer. The accuracy of the lock-down was tested, and if the map had greater
than 10 meters of distortion, it was not used. Control points were then digitized on the
image and the map to create a set of corresponding image file coordinates and map
coordinates.
From these points a transformation matrix was calculated for each image. The
program that generates the transformation matrix also calculates the root mean square of
the distance between the actual coordinates of the control points and the coordinates
predicted for each point by the transformation matrix. This measure identifies control
points contributing the greatest error. Only control points with a root mean square
distance less than 1 pixel (10m) were included in the calculation of the transformation
matrix.
A linear rectification of each image was done. Using the transformation matrix,
image file coordinates were converted to map coordinates. Spectral values were not
altered, but were assigned to the nearest rectified pixel.
It was unnecessary to consider elevation in this rectification, due to the height of
the satellite and the lack of terrain in Florida. The SPOT satellite's push broom sensor
does not have the distortion of a camera lens, and the data are corrected for the curvature
of the earth. A rectification using surveyed ground coordinates or GPS points is more
accurate than one using control points taken from a map. However, the accuracy of this
type of rectification is appropriate to the 10-meter resolution of the data when the control
points are numerous, well distributed and checked for error.
The Status and Trends land cover overlays were registered to map coordinates by
first locking down 1:24000 topographic quads to the digitizing table. The topographic
quads already had the boundaries of the plots penciled on them from the original photo-
interpretation. Tic mark coordinates along the edge of the topographic quad were entered









into the computer, and the tic marks were digitized. Then the acetate land cover overlay
of the plot was affixed to the topographic quad, making sure it matched the penciled plot
boundary, and the polygons on the overlay were digitized.
Achieving good registration between the digitized land cover overlays and the
images was essential for change detection, so that real change was not confused with class
discrepancies along polygon borders due to mis-registration. The registration problems
encountered in this study required additional labor to correct, but most were correctable.
Of the 44 plots in the study, 24 were not adequately registered to the image and
were corrected by shifting them in the x and y directions until they matched the image.
There are two possible sources of this mis-registration: error in the rectification of the
images, and error in the digitization of the overlays. The majority of registration problems
resulted from error in the digitization of the overlays.
Image rectification errors arise from the fact that ground control points on a map
are identified as points, but they must be matched to a 100 square meter pixel on the
image. This is an unavoidable consequence of the resolution of the satellite data. In
addition, there is error introduced when paper maps, which can be stretched and warped,
are used to identify ground control points. The analyst may also err in the identification
and digitization of control points.
Because a linear rectification to the UTM grid was done in this study, and SPOT
images themselves are not distorted, errors in an image's rectification would be expected
to be consistent throughout all the plots within that scene. For example, if the image
rectification were the only source of error, when a plot in the northwest part of the scene
was offset 1 pixel in the x direction and -2 pixels in the y direction, a plot in the center of
the scene would be offset in the same way. The plots would align well after being shifted
10 meters to the east and 20 meters to the north. This was somewhat the case with Scene
1, in which all the plots needed shifting, and nearly all were shifted in the same directions.
(A careful reexamination of the error associated with each control point used in
rectification, the rejection of bad points, the collection of new points, and a repeat of the
rectification might improve the alignment of the plots in this scene.) Even after shifting,
however, four plots in Scene 1 still had poor or fair alignment. These plots, and all the
plots requiring shifting in the three other images, have errors generated from the overlay
digitization.
Many errors in the overlay digitizations arose from use of the topographic quads to
lock down the overlays to map coordinates. The topographic maps with the plot
boundaries drawn on them were old, had been folded many times, and were certainly
warped. This distorted the spatial relationship between the tic marks on the edges of the









map and the plot boundaries. Human error during lock down may also be a factor.
Several plots overlapped onto two quads, so each half had to be locked down and
digitized separately. In one instance, (plot 1975), the two digitized halves of a plot did not
match up; this was a result of lock-down error on one of the quads.
Another error associated with digitizing lock-down occurred when the penciled
outline of a plot on the topographic quad did not correspond to the actual area interpreted
on the photographs. This error originated when the plots were first located and
interpreted. If the penciled outlines were in the wrong place (see Appendix D, Scene 3,
plot 2092), there was no match between the overlay and image at all. If the area
interpreted on the photograph was larger or smaller than the penciled outline, then the
center of a plot matched the image while the edges of the plot were displaced. The use of
fresh topographic maps with carefully penciled and checked plot outlines would reduce
these lock-down errors.
Less troublesome but irreparable errors arose during the photo-interpretation and
digitizing of the polygons. Unavoidable error is introduced because the width of the lines
on the overlay represents a much greater width on the ground. The digitizer's hand may
also stray, or the acetate overlay may slip during digitizing. These errors result in a mis-
registration that varies from polygon to polygon.









APPENDIX C-- ERDAS PROGRAMS USED
(VGA ERDAS Version 7.5)

PROGRAM Function

LDDATA- downloads an image from tape
BSTATS- generates image statistics
LISTIT- lists image statistics
READ- displays up to 3 bands of an image on the monitor at once
HISTOEQ- equalizes color contrast of image displayed on monitor
DISPOL- displays a vector (overlay) file
DIGUTIL- lists vector file statistics, offsets a vector file
COLORMOD- edits colors (palate file) of a vector file
ALGEBRA- performs algebraic functions on an image (image transformation and
vegetation index calculation)
GCP- identifies ground control points
COORDN- uses ground control points to generate a transformation matrix
LRECTIFY- uses a transformation matrix to rectify an image
SUBSET- extracts a subset of an image
SMEASURE- calculates linear and area measurements from a displayed image or vector
file
CURSES- lists locational and spectral image data at the cursor point









APPENDIX D-- INDIVIDUAL PLOT DATA AND NOTES

Note: The small polygons discussed are less than 1 acre.
"Complex" plots have >50 polygons in their overlay files.
"Very complex" plots have >100 polygons in their overlay files.


SCENE 1 PLOTS

1735
Landscape: agriculture with some upland forest, urban residential development (Emathia),
a racetrack and wetlands.

Image processing used: Bands 3,2,1.
Band 1 only.
NDVI and Bands 2,1.
NDVI and Bands 3, 2.
Unsupervised classification.

Number of Polygons: 65

Overlay Alignment: Poor.

This is a complex plot with 19 small palustrine polygons. Of these it is possible to
identify 9 using the image and ancillary data. It is not possible to identify 10, which
include Pfo, Pss, Pem, Pab and Pub polygons.
This plot was difficult to interpret because of great contrast between bright cleared
agricultural fields and dark forest. The NDVI and bands 3 and 2 displayed together, and
band 3 alone were the most useful for interpretation. Pfo could not be distinguished from
upland forest in this plot, even though many band combinations were tried and a 27-class
unsupervised classification was done. Pfo was a difficult class to separate from upland
forest on many plots (see comments on plots 2025 and 2030), but unless the polygon was
small, some spectral difference coinciding generally with the polygon was present. In this
plot the upland forest appeared wet, and no difference showed between Uo and Pfo.
The alignment of the overlay on the image is nonlinear. Approximately half of the
polygons line up with features on the image, but polygons near the eastern plot border are
displaced. A road and a 2.54 acre pond have been made in upland forest. No other
changes are apparent.









1924
Landscape: agriculture and upland forest, with a racetrack, ponds, urban residential
development (Irvine) and interstate highway 75.

Image processing used: Bands 3,2,1.
Band 1 only.
NDVI and Bands 2,1.

Number of Polygons: 76

Overlay Alignment: fair.

This plot is complex, with 23 small palustrine polygons. Of these it is possible to
identify 1 using the image alone, and to identify 16 using the image and ancillary data. It
is not possible to identify 6, which include Pfo, Pem and Pss polygons.
Pfo cannot be distinguished from upland forest using any of the band
combinations. The alignment of the polygons is not good enough to tell whether slight
boundary discrepancies are real changes or not. A large area of palustrine forest (57.4
acres) has been logged and would now be classified as Pss.

1949
Landscape: Ocala National Forest, upland forest only

Image processing used: Bands 3,2,1.

Number of Polygons: 1

Overlay Alignment: n.a.

This plot contains only upland forest in a single polygon. With only one polygon,
alignment cannot be judged. No change is apparent.

1978
Landscape: upland forest in timber management, urban residential development (Cedar
Creek), roads, palustrine forest along Eaton Creek, Lake Eaton, and many circular
wetlands and ponds.

Image processing used: Bands 3,2,1.
Band 1 only.
NDVI and Bands 2,1.

Number of Polygons: 77

Overlay Alignment: fair.









This complex plot contains 31 small polygons, mostly Pem. Of these it is possible
to identify 29 using the image and ancillary data. It is not possible to identify 2 Pem
polygons.
In this plot large Pfo polygons can be clearly distinguished from upland forest, but
most of the small ones cannot. Major changes in the upland forest from logging can be
clearly identified. One Pfo polygon (2 acres) has been logged, and would now be
classified as Pss.

1982
Landscape: Little Lake George, surrounded by palustrine forest and shrubs, and cleared
uplands.

Image processing used: Bands 3,2,1.
Band 1 only.
NDVI and Bands 2,1.

Number of Polygons: 25

Overlay Alignment: excellent.

This plot contains 6 small polygons, both Pem and Pss. Of these it is possible to
identify 5 using the image and ancillary data. It is not possible to identify 1. The small
polygons in this plot are very small (as little as 0.2 acres).
Since the 1980's, the upland forest in this plot has been cleared. While Pfo is easily
distinguished from cleared uplands, regenerating Pfo (Pss) is hard to distinguish from
regenerating uplands. In the image there are zones of shrubs found within cleared Uo
polygons bordering Pfo. These areas look like Pss, but it is not possible to determine
whether they are actually Pss or regenerating upland areas that are spectrally similar.
Since it is unlikely for Uo to change to Pss, either these zones were originally mis-
classified as upland when they really were palustrine, or the image does not spectrally
separate the two classes in this area. The latter is assumed to be the case.
There is a line of bad data in this plot.

2079
Landscape: the north shore of Lochloosa Lake, forested upland in various stages of forest
management, palustrine forest, many scattered emergent wetlands, and roads.

Image processing used: Bands 3,2,1.
Band 1 only.
NDVI and Bands 2,1.
NDVI and Bands 3,2.

Number of Polygons: 119

Overlay Alignment: good.










This plot is complex, containing 71 small polygons. Of these, it is possible to
identify 58 using the image and ancillary data. It is not possible to identify 13, which
include Pem, Pub, Pfo, and Pss polygons.
The majority of small polygons not identified are Pss and Pfo polygons >0.4 acres.
The large polygons of Pfo in this plot can be clearly distinguished from upland forest using
the NDVI and bands 3 and 2. There are extensive areas in the uplands that have been
logged. Portions of 3 Pfo polygons have been cleared by logging, and would now be
classified as Pss. These changes cover areas approximately 12.6, 8.5, and 5.5 acres.
There are also complexes of wetlands surrounded by upland forest that have been logged
and changed by drainage and pond construction. Four polygons changed from Pem to Uo
are 11.88, 3.8, 1.37 and .16 acres in size. Pfo changed to Uo is 15.4 acres. Pond
construction changed 1.52 acres of Pem, and 1.99 acres of Uo to Pub.
In this plot there are areas which appear to be Pfo spectrally, but were classified
as Uo on the overlay. Around these areas the upland forest was recently logged, except
all Pfo polygons within the forest were left intact. These areas were also left intact. They
closely resemble the other Pfo polygons spectrally and by shape. Therefore, the overlay
interpretation must be incorrect. Logging can reveal previously undetected patches of
Pfo.

2120
Landscape: Rice Creek Swamp (palustrine forest), with some upland forest, palustrine
shrubs and roads.

Image processing used: Bands 3,2,1.
Band 1 only.
NDVI and Bands 2,1.

Number of Polygons: 28

Overlay Alignment: poor.

This plot contains 10 small polygons. Of these, it is possible to identify 5 using the
image and ancillary data. It is not possible to identify 5.
It is difficult to distinguish Pfo from upland forest in some portions of this plot, but
in other areas the distinction is quite clear. Large areas of Pfo have been cleared,
presumably logged. These areas are approximately 150.3 acres, 60 acres, 53.6 acres and
41.6 acres. They would now be classed as Pss.









2144
Landscape: the southern shore of Orange Lake, with surrounding marshes, Hawthorne
Prairie (lake and marsh), agriculture, palustrine forest, other uplands, roads, and a
small urban development.

Image processing used: Bands 3,2,1.
Band 1 only.
NDVI and Bands 2,1.

Number of Polygons: 81

Overlay Alignment: good.

This complex plot contains 28 small polygons. Of these, it is possible to identify
18 using the image and ancillary data. It is not possible to identify 10.
The water levels in all the ponds, marshes and lakes in this plot are very low,
because a normally blocked sink in Orange Lake opened up naturally. This is why so
many small polygons cannot be identified. Large areas of Lac now appear to be Pus.
When this phenomenon occurred in the past, eventually the sink refilled with debris and
water levels rose again. Evidence of vegetation succession does not appear in this image
yet, (Pem does not appear to have upland species in it) so it is too soon to conclude a
permanent change has occurred.

2182
Landscape: upland forest under timber management, palustrine forest, the town of Cedar
Creek, and several small lakes.

Image processing used: Bands 3,2,1.
Band 1 only.
NDVI and Bands 2,1.

Number of Polygons: 44

Overlay Alignment: excellent.

This plot contains 18 small polygons. Of these, it is possible to identify 9 using the
image and ancillary data. It is not possible to identify 8.
It is possible to distinguish Pfo from upland forest in most of this plot. Some Pfo
(1.7 acres) has been cleared for urban development. Other uplands (67.5 acres) have also
been converted to urban.









2246
Landscape: urban residential development near Interlachen, sparsely forested uplands, and
several small lakes and wetlands.

Image processing used: Bands 3,2,1.
Band 1 only.
NDVI only.
NDVI and Bands 2,1.

Number of Polygons: 20

Overlay Alignment: excellent.

This plot contains 6 small polygons. Of these, it is possible to identify 3 using the
image alone; they are Pub and Pem polygons. It is possible to identify 3 using the image
and ancillary data.
Pfo can be distinguished from upland forest in this plot, although the polygon
boundaries do not exactly agree with the apparent extent. The NDVI band alone is useful
for this. There is no apparent change.

2259
Landscape: the town of Fruitland on the St. Johns River, with urban development,
agriculture, other uplands, palustrine forest along the river and in patches, and
many small marshes.

Image processing used: Bands 3,2,1.
Band 1 only.
NDVI and Bands 2,1.

Number of Polygons: 97

Overlay Alignment: excellent.

This complex plot contains 43 small polygons. Of these, it is possible to identify
18 using the image alone; they are Pub and Pem polygons. It is possible to identify 14
using the image and ancillary data. It is not possible to identify 21, which are Pem and Pfo
polygons < 0.1 acre, the limit of the image's resolution.
Spectrally, Pem stands out very clearly in this area. There is no apparent change.









SCENE 2 PLOTS


359
Landscape: a portion of Lake Griffin and Haines Creek, agriculture, marshes, and
palustrine shrubs.

Image processing used: Bands 3,2,1.
NDVI and Bands 3,2.
Band 1 only.
Band 3 only.

Number of Polygons: 30

Overlay Alignment: excellent.

This plot contains 3 small polygons. Of these, it is possible to identify 1 using the
image and ancillary data. It is not possible to identify 2.
Four polygons of Pem where Haines Creek enters Lake Griffin have changed to
Lac; they are 4.7, 1.5, 0.4, and 0.3 acres in size. From the photographs this appears to be
fast-changing vegetation. In each year of photography, ('53, '84 and '87), the distribution
of vegetation in this area is different. Some of this vegetation is probably floating aquatic
beds, rather than Pem, which are being sprayed for aquatic weed control. Aquatic
vegetation could also change because lake levels fluctuate in central Florida with drought
cycles. Otherwise, there is no apparent change.

366
Landscape: upland forest with timber management, and a distinctive pattern of roads on a
U.S. Naval Reservation.

Image processing used: Bands 3,2,1.

Number of Polygons: 1

Overlay Alignment: n.a.

This plot contains no wetlands. With only a single polygon, alignment cannot be
judged, and no change is apparent.









1709
Landscape: uplands with sparse forest, Rolling Hills urban residential development with a
regular pattern of roads, a golf course, and agriculture.

Image processing used: Bands 3,2,1.

Number of Polygons: 4

Overlay Alignment: good.

This plot contains no wetlands.
There is no apparent change, but the landscape probably will experience more
development soon, since most of the residential lots are still empty.

1758
Landscape: uplands with sparse forest, Ocala Ridge urban residential development with a
regular pattern of roads, and agriculture.

Image processing used: Bands 3,2,1.

Number of Polygons: 10

Overlay Alignment: good.

This plot contains no small polygons or wetlands, except for the edge of a stand of
Pfo that is mostly off the plot.
There is no apparent change, but the landscape probably will experience more
development soon, since most of the residential lots are empty.


1930
Landscape: sparsely vegetated uplands, agriculture, a racetrack, the outskirts of the town
of Belleview, and scattered small wetlands including small lakes and ponds.

Image processing used: Bands 3,2,1.
Band 1 only.
NDVI and Bands, 3,2.
NDVI and Bands 2,1.

Number of Polygons:56

Overlay Alignment: excellent

This complex plot contains 18 small polygons. Of these, it is possible to identify 9
using the image and ancillary data. It is not possible to identify 9.









The large contrast between bright urban and agricultural features and dark
vegetation and wetlands, and the complexity of their mixture make this a difficult plot to
interpret. No change is apparent.

1948
Landscape: the Withlacoochee State Forest, with sparse upland forest and strips of
clearing for forest management.

Image processing used: Bands 3,2,1.

Number of Polygons: 1

Overlay Alignment: n.a.

This plot contains no wetlands.
With only a single polygon, alignment can't be judged, and no change is apparent.

1975
Landscape: sparsely forested uplands with in the Withlacoochee State Forest, agriculture,
and low density urban residential development near Stokes Ferry with a regular
pattern of roads and building construction.

Image processing used: Bands 3,2,1.
Band 1 only.

Number of Polygons:20

Overlay Alignment: fair

This plot contains 11 small Pem and Pub polygons. Of these, it is possible to
identify 9 using the image and ancillary data. It is not possible to identify 2. Some of
these polygons are < 0.1 acre.
There are two overlay files for this plot, which straddled the border of a quad
sheet. The overlay files had to be shifted in opposite directions to align with the image,
indicating that the registration of the digitized overlays, not the satellite image, is in error.
No change was detected.









2025
Landscape: Silver Springs and the Silver river, with dense urban development
surrounding the spring, palustrine forest lining the river, non-forested and forested
uplands in timber management, and small wetlands.

Image processing used: Bands 3,2,1.
Band 1 only.
NDVI and Bands 2,1.
NDVI and Bands 3,2.

Number of Polygons: 57

Overlay Alignment: fair.

This plot is complex. It contains 17 small polygons. Of these, it is possible to
identify 13 using the image and ancillary data. It is not possible to identify 4.
It is difficult to distinguish Pfo from upland forest in this plot, even though the
NDVI band was helpful in bringing out spectral detail in the vegetation. In some places,
Pfo polygons concur with the bright red color and band values usually associated with
swamps, but in other places pixels with the band values typical of Pss or cleared Uo are
found within the Pfo polygons. These anomalies occur in linear patches scattered
throughout large polygons, and along borders where change might occur. There are also
bright red pixels in the Uo, near the boundaries of Pfo polygons. It is difficult to tell
whether these discrepancies in upland and Pfo delineation are due to: generalization in the
original classification method; spectral reflectance of species shared between the two
forest types; or an actual vegetation change. In spite of this confusion, however, it is still
possible to tell whether any major changes have taken place (see also comments on plot
2030).
One 8 acre polygon of Pfo has been logged and would now be classed as Pss.
While overall alignment is good, some polygons are poorly aligned.

2030
Landscape: the Silver River, lined with a wide band of palustrine forest, forested uplands
in timber management, patches of palustrine shrubs, and roads.

Image processing used: Bands 3,2,1.
Band 1 only.
NDVI and Bands 2,1.
NDVI and Bands 3,2.


Number of Polygons: 40

Overlay Alignment: good.









This plot contains 14 small polygons. Of these, it is possible to identify 3 using the
image alone, all of which are Pub (springs). It is possible to identify 7 using the image and
ancillary data. It is not possible to identify 4.
The NDVI band helps to distinguish Pfo from the upland forest, but there are
questionable areas. The Pfo is generally very bright red, compared to the upland forest,
which is darker red. Regenerating areas or shrubs are greenish gray. There are patches of
greenish gray found within Pfo that have the identical band values as pixels in Pss
polygons. There are also strands of darker red found within Pfo that have the identical
band values as pixels within Uo polygons. These correspond to color differences in the
photography, but it is not possible to tell whether this is spectral variation due to a mix of
Pfo species, or patches of Pss and upland forest trees that were overlooked in the photo-
interpretation. It is clear that the spectral variation is not due to human-made change. No
other change is apparent.

2096
Landscape: mainly agriculture, with some other uplands, roads, 2 racetracks, Bahia Oaks
urban residential development, and ponds.

Image processing used: Bands 3,2,1.
Band 1 only.


Number of Polygons: 18

Overlay Alignment: excellent.

This plot contains 10 small Pub and Pem polygons. Of these, it is possible to
identify 3 using the image alone, all of which are Pub. It is possible to identify 7 using the
image and ancillary data.
No change is apparent.

2236
Landscape: the Oklawaha river, with surrounding palustrine forest, upland forest in timber
management, roads, small patches of agriculture, and other uplands.

Image processing used: Bands 3,2,1.
NDVI and Bands 2,1.
NDVI and Bands 3,2.

Number of Polygons: 61

Overlay Alignment: fair.

This plot is complex. It contains 17 small polygons. Of these, it is possible to
identify 12 using the image and ancillary data. It is not possible to identify 5.









The Pfo is not distinct from some well-forested upland areas; both are bright red.
Logging activity is distinct in both Pfo and Uo. Two areas, approximately 26 and 13
acres, have been cleared in Pfo, and would now be classified as Pss.

SCENE 3 PLOTS

350
Landscape: sugarcane agriculture and a canal.

Image processing used: Bands 3,2,1.

Number of Polygons: 5

Overlay Alignment: good.

Apart from the canal, there are no wetlands in this plot, no small polygons, and no
change.

370
Landscape: Coral Springs urban residential development, a golf course, agriculture,
canals, and scattered ponds throughout the development.

Image processing used: Bands 3,2,1.
NDVI and Bands 2,1.
NDVI and Bands 3,2.

Number of Polygons: 48

Overlay Alignment: good.

This plot contains 30 small Pem, Pub and Pss polygons. Of these, it is possible to
identify 24 using the image alone. It is possible to identify 3 using the image and ancillary
data. It is not possible to identify 3.
This region of Florida (where the southern Atlantic coastal cities encroach on
inland agriculture) is experiencing rapid development and extreme wetland change. The
region's wetlands were originally drained to permit agriculture, but some remnants
remained on unused property or were used as impoundments. To convert agricultural
land to urban development requires additional drainage, filling, and the creation of deep
retention ponds. This results in drastic change, greatly disturbing or obliterating all
remaining wetlands.
There are three changes in this plot are from Ua to Ub: 48.8, 21.6, 19 and 15
acres. A new Pub pond in the urban area is 3.9 acres and new Pub ponds in the
agricultural area are 33.7 and 19 acres. One of these ponds also destroyed 0.9 acres of
Pem. There are many canals that do not appear on the overlay, but it is not clear why
these were excluded from the interpretation, since they appear on the 1980's photographs.










376
Landscape: Water Conservation Area 2a, with Everglades marsh and shrub islands.

Image processing used: Bands 3,2,1.
NDVI and Bands 2,1.

Number of Polygons: 38

Overlay Alignment: poor.

This plot contains 21 small polygons, of which it is possible to identify 15 using the
image alone. It is not possible to identify 6.
It is difficult to assess change on this plot, due to the questionable classification of
the overlay. This is also a problem for several other Everglades plots in this scene (2092,
2132 and 2264). For these plots the 1980's photography is very poor, with widespread
glare and very little infrared definition. The 1970's photography is not much better; the
color is very pink and washed out, and the scale is small. The 1950's photography is black
and white. The limitations of these photographs inhibited the photo-interpreter's ability to
identify vegetation in the 1950's and 1970's. The change detection update of the 1980's
maintained any classification errors from the earlier interpretations; it relied heavily on the
1970's overlay. Only changes in the boundaries of the 1970's polygons were noted and
revised; an evaluation of the accuracy of the classification of the polygons was not made.
As a result, classification errors in the 1980's overlay become obvious when viewing the
image. Even though the photographic quality is poor, the correct classes can be identified
on the 1980's photography with careful examination and reference to the image. These
areas changed sometime between 1970 and 1990, so the changes are noted now. The
Everglades plots are undergoing a slow conversion of wetland types due to changes in
hydrology, water quality and fire frequency. No drastic change from clearing or dredging
has occurred.
There are 11 areas on the overlay that are classed as Pem, but appear to be Pss on
the image. These are 205, 15.9, 7.09, 3.26, 2.85, 2.08, 1.22, 1.19, 1.27, 1.11, and 0.44
acres. Four polygons classed as Pss on the overlay appear to be Pfo on the image; these
were 1.3, 1.3, 1.2 and 0.6 acres. Six polygons classed as Pem appear to be Pfo on the
image; these were 16, 8.8, 1.6, 1.6, 0.6, 0.5, 0.2, and 0.2 acres.
The different signatures of these wetland classes are clear on the image,
particularly in band 3 and the NDVI band. Extensive ground-truthing in this area has been
done for other research projects, so confidence on the interpretation of the image is high.
In addition, 1990's Status and Trends photography, which has not yet been photo-
interpreted, confirmed the class identifications from the image.









1907
Landscape: the shore of Lake Okeechobee, with agriculture and urban development, other
uplands (levees), marsh, canals, and ponds.

Image processing used: Bands 3,2,1.
NDVI and Bands 2,1.

Number of Polygons: 30

Overlay Alignment: excellent.


This plot contains 9 small polygons. Of these, it is possible to identify 1 (Pub)
using the image alone. It is possible to identify 5 using the image and ancillary data, and it
is not possible to identify 3.
Two ponds in the plot have very little water in them, but still are probably Pub.
No change is apparent.

1926
Landscape: agriculture, urban residential development, other uplands undergoing
conversion from agriculture to urban, some remnant wetlands and many ponds.


Image processing used: Bands 3,2,1.
NDVI and Bands 2,1.

Number of Polygons: 87

Overlay Alignment: good.

This is a complex plot, containing 25 small polygons. Of these, it is possible to
identify 11 using the image alone, even though some of these have changed. It is possible
to identify 5 using the image and ancillary data. It is not possible to identify 9.
Approximately 1/3 of the image of this plot is missing; it lies off the edge of the
scene. The total number of acres of the plot covered by the image is 1550.
There are many changes in this plot, (see comments for plot 370, above). This
type of change is easy to identify from the image because it is so extreme. Each of the 21









polygons changed are listed by the type of change and their size in acres in the following
table.

Ua>Ub Ua>Uo Pfo>Uo Pem>U Pub>Uo
o
159 26.7 10.8 3.5 3.9
12.3 2.6
8.2 1.7
.8
.2
Pss>Uo Uo>Pub Ua>Pub Ub>Pub
1.5 14.1 3.4 3.6
1.3 8.5
1.3 8.5
.9 1.7


2026
Landscape: sugarcane agriculture.

Image processing used: Bands 3,2,1.

Number of Polygons: 1

Overlay Alignment: n.a.

This plot contains no wetlands.
With only a single polygon, alignment can't be judged, and no change is apparent.

2077
Landscape: sugarcane agriculture, and a single remnant block of Everglades palustrine
shrub.

Image processing used: Bands 3,2,1.

Number of Polygons: 4

Overlay Alignment: excellent.

This plot contains no small polygons. No change is apparent.









2092
Landscape: Water Conservation Area 2a, with Everglades marsh and shrub islands.

Image processing used: Bands 3,2,1.
NDVI and Bands 2,1.

Number of Polygons: 4

Overlay Alignment: poor.

This plot contains 4 small polygons, all of which can be identified as Pfo by the
image alone, although the overlay shows them as Pss.
The digital file of this overlay was greatly misplaced. The map coordinates of the
corner points of the plot did not correspond to the area interpreted on the photographs.
Displayed over the image as digitized, the polygons were completely inappropriate. This
error could have been made if the plot corner points were entered incorrectly during
digitizing, or if the plot outline was penciled on the original quad sheet in a different
location from the actual corner point coordinates. An error like this could only be made in
a landscape like the Everglades, with practically no topography or man-made features.
The only man-made features on the plot are airboat trails, and these often change over the
years.
By carefully examining the 1980's and 1990's photographs of the plot and the
orthophoto quad sheet, the actual location of the plot was located. The digital overlay
was shifted 110 meters to the east to this location.
This plot overlay also has the same classification problems as plot 376 (see
comments above). The overlay shows the plot is mostly Pem with several small Pss
polygons. The image, however, shows many more areas of Pss. There is also a scar from
a recent burn on the plot. Change acreage data from this plot were not recorded,
however, due to the classification uncertainty and the misplacement of the overlay.

2132
Landscape: the Loxahatchee National Wildlife Refuge, with Everglades marsh and islands
of palustrine shrubs and trees.

Image processing used: Bands 3,2,1.
NDVI only.
NDVI and Bands 2,1.

Number of Polygons: 193

Overlay Alignment: excellent.

This is a very complex plot, containing 49 small Pss polygons, all of which can be
identified by the image alone, even though 2 of them are now Pfo.









Several of the larger Pss polygons on the plot overlay appear on the image to be
Pfo instead. The sizes of these polygons are 33, 3.22, 2.83, 2.5, 1.69, 1.51, and 1.48
acres. The polygons delineate what are known locally as "tree islands". They probably
represent classification errors rather than change (see comments for plot 376). Most of
the medium and small Pss polygons have small boundary discrepancies in which the area
of shrubs on the image is slightly smaller than the polygon. Rather than real change, these
discrepancies are probably due to a higher water level in the image than in the 1980's
photographs, which would inundate small shrubs and make them appear spectrally similar
to Pem.

2247
Landscape: sugarcane agriculture.

Image processing used: Bands 3,2,1.

Number of Polygons: 1

Overlay Alignment: n.a.

This plot contains no wetlands.
With only a single polygon, alignment cannot be judged, and no change is
apparent.

2264
Landscape: sugarcane agriculture, with a remnant block of Everglades marsh and
palustrine shrubs.

Image processing used: Bands 3,2,1.
NDVI and Bands 2,1.

Number of Polygons: 9

Overlay Alignment: unknown.

There are no small polygons in this plot.
Either this plot is misplaced or great change has occurred within the remnant block
of Everglades which occupies about one fifth of the plot. The overlay shows several Pss
and Pem polygons, but no features following the outlines of the polygons can be seen on
the image. What appears on the image is a heterogenous mix of cattail (Pem), recently
burned marsh (Pem), and Pss. This type of radical change is possible when an Everglades
wetland is located just across a canal from an agricultural field which disrupts fire, nutrient
loading and hydrologic patterns. However, because of the lack of landmarks in this plot,
misplacement of the overlay or classification errors are also likely (see comments for plots
376 and 2092). It is impossible to tell which is the case without a detailed review of the
plot digitization.










SCENE 4 PLOTS
327
Landscape: Sand Slough, with many shallow lakes, marshes and swamps, agriculture, and
upland forest.

Image processing used: Bands 3,2,1.
NDVI only.
Band 3 only.
NDVI and Bands 2,1.
NDVI and Bands 3,2.

Number of Polygons: 105

Overlay Alignment: fair

This very complex plot has 50 small polygons. Of these it is possible to identify 1
by the image alone, which is Pub. It is possible to identify 28 using the image and ancillary
data. It is not possible to identify 21.
The small polygons which could not be identified were mostly Pss polygons which
were less than 0.5 acres and surrounded by Pfo.
It is possible to distinguish the upland forest from the Pfo in most of this plot from
the image alone, using the NDVI and bands 3 qand 2. Unlike most plots, the NDVI
values for Pfo are lower than for forested Uo in this plot, because the Pfo is flooded and
reflects little infrared.
No change is apparent in the plot.

416
Landscape: Crystal River nuclear power plant, gulf hammock islands, salt marsh, sloughs,
mud flats, and sand bars and seagrass beds in the Gulf of Mexico.

Image processing used: Bands 3,2,1.
Band 1 only.
Band 3 only.
NDVI and Bands 2,1.


Number of Polygons: 135

Overlay Alignment: excellent.

This very complex plot has 82 small polygons. Of these it is possible to identify 72
using the image alone, most of which are forested islands of Uo and E2fo in the salt
marsh. It is not possible identify 10, most of which are < 0.1 acre, less than the image's
resolution. Photographs were not available for this plot at the time of interpretation.








It is difficult to distinguish E2fo from Uo, and Elub from E2em when all three
bands of the image data are displayed. However, when only band 3 is displayed the
differences become very clear. When only band 1 is displayed, sand bars and sandy shores
are distinct.
In one subtidal area near the mouth of a slough, 2.5 acres of Elab (seagrass bed)
has changed to Elub unconsolidatedd bottom). In addition, there appear to be very small
boundary shifts in a few E2us polygons along the interface between the salt marsh and the
gulf This would be expected over time due to shifting unconsolidated sand and mud, and
could also be an artifact of changes in tide level between the photography and the image.
Because these changes are small compared to possible polygon alignment errors, they
should not be considered real change.


432
Landscape: the mangrove-covered St. Martins Keys, seagrass beds, and the shallow water
of the Gulf of Mexico.

Image processing used: Bands 3,2,1.
Band 1 only.
Band 3 only.
NDVI and Bands 2,1.

Number of Polygons: 57

Overlay Alignment: good.

This complex plot contains 22 small polygons. Of these it is possible to identify 8
using the image alone, all of which are E2ss. It is possible to identify 5 using the image
and ancillary data. It is not possible to identify 9, which include Elub, E2us and E2ss.
Photographs were not available for this plot at the time of interpretation.
In this plot, most of the area is Elab. There are two large areas of E2ab that lie
next to mangrove keys. It is not possible to see any difference between the Elab
(subtidal) and E2ab (intertidal) polygons on the image. No special band combinations
improve this problem. Since only the difference between the two classes is depth of
inundation, it is likely that photo-interpreters used ancillary data on elevation to classify
these. There are several large polygons of E2us adjoining the keys, and in some places
these are also indistinguishable from Elab. Rather than representing real change, it is
more likely that remote sensing cannot detect the differences in water depth and
submerged vegetation that separate these classes. The topographic quad would be useful
in analyzing this problem, but is unavailable.
The keys themselves are mangroves, E2ss, and are clearly seen, especially using
band 3. Several small polygons of Elub in the interiors of two keys, however, are not
visible on the image.









461
Landscape: Withlacoochee Bay, gulf hammock islands, and salt marsh.

Image processing used: Bands 3,2,1.
Band 1 only.
Band 2 only.
Band 3 only.

Number of Polygons: 315

Overlay Alignment: fair

This very complex plot has 193 small polygons. Of these it is possible to identify
74, which are mainly E2fo and Uo, with some E2us polygons. It is possible to identify
109 using the image and ancillary data. It is not possible to identify 10. Photographs were
not available for this plot at the time of interpretation.

Considering the complexity of this plot, interpretation of the image is quite clear.
The only problems are separating islands with Uo from those with E2fo, which both occur
in the salt marsh and share the same species. The photo interpretation may have relied on
the topographic quad to separate the two. In addition, Elub and E2us cannot be
distinguished where they coexist along sloughs in the salt marsh. Again, the topographic
quad may have been used to separate the two. There is no apparent change in this plot.

482
Landscape: the mangrove-covered Shark Point, Sandy Hook and Bird Keys, seagrass
beds, and the shallow water of the Gulf of Mexico.

Image processing used: Bands 3,2,1.
Band 1 only.
Band 3 only.
NDVI and Bands 2,1.

Number of Polygons: 85

Overlay Alignment: fair.

This complex plot contains 46 small polygons. Of these it is possible to identify
36, which are mainly E2ss polygons. It is not possible to identify 10. Photographs were
not available for this plot at the time of interpretation.
The mangroves, (E2ss), are very clear on the image, along with small polygons of
Elub in the interiors of the islands. Sandbars have increased in size or appeared in some
areas offshore: Elab has changed to Elub. The polygons are 56.16, .38, .35, .84, and .44
acres. Polygons of E2us adjacent to the islands cannot be distinguished from Elab using









the image alone. Therefore it is not possible to determine whether there is change in this
area.

1741
Landscape: agriculture, Rainbow Lakes Estates low density urban residential
development, and some sparsely vegetated upland.

Image processing used: Bands 3,2,1.
Band 1 only.
NDVI and Bands 2,1.

Number of Polygons: 9

Overlay Alignment: good.

This plot is not complex, and contains only two small polygons (< 0.2 acres), both
Pem. It is possible to identify these using the image and ancillary data.
There are not other wetlands in the plot. No change is apparent in this plot.

1803
Landscape: the Homosassa River and Shivers Bay, with gulf hammock forest, salt marsh
and many sloughs.

Image processing used: Bands 3,2,1.
Band 1 only.
Band 3 only.
NDVI and Bands 2,1.

Number of Polygons: 315

Overlay Alignment: excellent.

This plot is very complex. Although it contains polygons of only 4 classes, it has
the greatest number of polygons of all the plots in this study. Many of these polygons are
smaller than 0.5 acre. The plot has 223 small polygons. Of these it is possible to identify
221 using the image alone; they are mainly E2em and E2fo polygons. It is not possible to
identify 24. Of the total number of small polygons, 119 are E2em, 72 are E2fo, 29 are
Elub, and 3 are E2us. Photographs were not available for this plot at the time of
interpretation.

In spite of its complexity, this plot is easy to interpret on the computer monitor
because the overlay is very well aligned, and the number of classes is low, so the polygon
display constraint is not a problem. No change is apparent in the plot.








2057
Landscape: sparse upland forest, Rainbow Lakes Estates low density urban residential
development, agriculture, urban development with a golf course, and Rainbow
Springs.

Image processing used: Bands 3,2,1.
Band 1 only.
NDVI and Bands 2,1.

Number of Polygons: 8

Overlay Alignment: good.

This plot contains two small polygons, and both Pem, both of which cannot be
identified.
There is no apparent change of classes in the plot, but development is intensifying
within the urban area.

2147
Landscape: the town of Crystal River, highways and roads, cleared uplands, upland and
palustrine forest, ponds, canals, marshes and palustrine shrubs.

Image processing used: Bands 3,2,1.
Band 1 only.
NDVI and Bands 2,1.

Number of Polygons: 122

Overlay Alignment: fair.

This is a very complex plot with 51 small polygons. Of these, it is possible to
identify 4 using the image alone, all Pub. It is possible to identify 17 using the image and
ancillary data. It is not possible to identify 30.
The large contrast between bright urban features and dark vegetation, and the
spatial complexity of this plot make it difficult to interpret. The large number of polygons,
of 8 different classes, is difficult to reference due to the polygon display constraint.
Several areas have been converted to urban: 4.2 acres ofPss, 1 acre of Pem, and
8.8, 23.6 and 41 acres of uplands.









2176
Landscape: upland forest with a large clearing in the Withlacoochee State Forest, and a
small urban area.

Image processing used: Bands 3,2,1.
NDVI and Bands 2,1.

Number of Polygons: 3

Overlay Alignment: good.

This plot contains only one small wetland, Pub, which can be identified using the
image alone.
There is a stand of forest in this plot that looks similar to Pfo, but cannot be Pfo in
this location. Two Pub ponds have appeared in the Uo: 0.9 acres and 0.2 acres. No other
change is apparent on the plot.

2208
Landscape: upland forest and some agriculture at the edge of the Withlacoochee State
Forest.

Image processing used: Bands 3,2,1.
NDVI and Bands 2,1.

Number of Polygons: 3

Overlay Alignment: good.

This plot contains no wetlands or small polygons.
As with plot 2176, there is some upland forest that looks similar to Pfo, but cannot
be Pfo. No change is apparent.




University of Florida Home Page
© 2004 - 2010 University of Florida George A. Smathers Libraries.
All rights reserved.

Acceptable Use, Copyright, and Disclaimer Statement
Last updated October 10, 2010 - - mvs