This item is only available as the following downloads:
LANDSCAPE DYNAMICS IN THE EVERGLADES:
VEGETATION PATTERN AND DISTURBANCE
IN WATER CONSERVATION AREA 1
JENNIFER ENOS SILVEIRA
A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL
OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY
UNIVERSITY OF FLORIDA
UNIVERSITY OF FLORIDA LIBRARIES
Jennifer Enos Silveira
To my son, Whitney Mattson:
Your sacrifices, love and encouragement
are bound into this work.
To my parents, Janice and Richard Enos:
My favorite teachers for over 38 years.
I would like to thank the University of Florida School of
Forest Resources and Conservation, and the U.S. Fish and
Wildlife Service National Wetlands Inventory, for financial
support during the course of my degree. In addition, the
Florida Cooperative Fish and Wildlife Research Unit and the
Arthur R. Marshall Loxahatchee National Wildlife Refuge provided
equipment and facilities that made this work possible.
I offer many thanks to the members of my academic
committee. Dr. Wiley Kitchens' holistic thinking nourished many
of my ideas, but he was my advisor in a greater sense. His
patient understanding of the difficulties of combining
parenthood with a Ph.D. prevented me from giving up the task.
Without his respectful and enthusiastic counsel I could not have
Dr. John Richardson's research provided the starting point
for this work. I was inspired by his innovation and persistence
with computers and data analysis. By teaching me some things,
and forcing me to figure others out on my own, he passed a bit
of this on to me.
Dr. C.S. Holling's teaching on cross-scale interactions
was essential to this work, and through his Everglades Adaptive
Management Workshops I met many experienced Everglades
Drs. Loukas Arvanitis and George Tanner were thoughtful
and thought-provoking advisors, helping me to see my work in a
Many of the image processing and GIS techniques used in
this work were not found in any textbook; they were learned by
peering over John Richardson's or Leonard Pearlstine's shoulders
as they worked. I thank Leonard for his emergency technical
advice, always cheerfully provided, and for his attendance at my
Thanks go to my fellow graduate students Wade Bryant,
Howard Jelks, Frank Jordan, Cyndy Loftin, Laura Brandt, and Dr.
Lance Gunderson for sharing the airboating, flying and/or
proposal brainstorming with me.
Special thanks go to Don Flickinger, Karen Richardson,
Barbara Fesler and Kevin Mattson for their ever-present help and
TABLE OF CONTENTS
LIST OF TABLES .....................................
LIST OF FIGURES ...............................................xi
The Study of Landscape Pattern Dynamics .
Landscape Ecology ....................
Models of Landscape Pattern ..........
Appropriate scaling in models ......
Spatial models .....................
The Landscape of Water Conservation Area
The History of Water Conservation Area
The Substrate of Water Conservation Ar
Peat......... ..... ..................
Water in Water Conservation Area 1 ...
The natural hydrology ..............
Flood control hydrology ............
The Tree Islands of Water Conservation
Tree island formation ..............
Fire in Water Conservation Area 1 ....
Fire and the Everglades ecosystem...
Seasonal probability of fires ......
Fire behavior factors ..............
Fire severity ......................
Patterns of fire events ............
Fire in Water Conservation Area 1..
Vegetation community types .........
Environmental factors in vegetation
Hydrologic factors .................
Nutrient concentrations ............
Vegetation Succession ................
LANDSCAPE DATA COLLECTION ..................
Introduction ...... ........................
Ground Observations .....................
Pre-fire Field Observations ..........
Post-fire Field Observations .........
1 ................. 13
1 ................ 14
ea 1 ...............18
Area 1 ...........25
............ .... 38
Aerial Photographs ......................
Aerial Photographs, 1952 .............
Aerial Photographs, 1991-1992 ........
Satellite Images ........................
SPOT Satellite Image Information .....
Image Merging ........................
Image transformation ...............
Vegetation index ...................
Image Classification .................
Unsupervised Classification ........
Class Identifications ..............
ANALYSIS OF TREE ISLAND DISTRIBUTION .......
Introduction .................... .......
GIS Analysis of Tree Island Distribution
Analysis of Size Distribution ........
Island Population Model Analysis ........
LANDSCAPE MODEL ..........................................
Introduction ..................... ....................
Conceptual Model of Tree Island Growth and Landscape
Processes Contributing to Island Growth ............
Vegetation succession ............................
Peat accumulation ................................
Tree islands .....................................
Conceptual Model ...................................
Phase 1: vegetation succession ...................
Phase 2: fire ....................................
Phase 3: post-fire vegetation recovery ...........
Fire and landscape pattern .......................
Spatial Model of Vegetation Succession and Disturbance
by Fire ............................................
Model Design .......................................
GIS software .....................................
Model structure ..........
Model base map ...........
Model rules and rates ....
Model Verification .........
Improved Model Simulations .
Tree islands in the spatial model ........
Comparison of Model with Independent Data
Model validation .......................
Comparison of model and satellite data.
SUMMARY AND CONCLUSIONS .... ............................ .... 159
LIST OF REFERENCES ........................................... 168
BIOGRAPHICAL SKETCH... .........................................178
LIST OF TABLES
Table 1-1 Large fires recorded in Water Conservation Area
1 .......................................................... 35
Table 1-2 Common plant species of Water Conservation Area
1, by vegetation community ................................. 39
Table 2-1 Results of tree island sample, 1952 photography... .57
Table 2-2 Percent area in sawgrass and wet prairie of two
photographic sample plots in Water Conservation Area 1 ..... 62
Table 2-3 SPOT satellite images of Water Conservation
Area 1 ..................................................... 67
Table 2-4 Classes in the 1987 vegetation map of Water
Conservation Area 1 found in less than 1% of the
successional class map area. ............................... 83
Table 2-5 Description of classes in the 1987 vegetation
map of Water Conservation Area 1 found in the
successional class map area ................................ 84
Table 2-6 Rules for successional class identifications
based on the 1987 vegetation map of Water Conservation
Area 1 ..................................................... 85
Table 2-7 Classes in the successional map of Water
Conservation Area 1 ........................................ 86
Table 4-1 Subdivisions of seral stages in model's
starting base map ......................................... 130
Table 4-2 Initial landscape model rules ..................... 131
Table 4-3 Burn iterations for different fire interval
model simulations ......................................... 135
Table 4-4 Improved landscape model rules .................... 140
Table 4-5 Landscape composition after 20-year fire
interval simulations, (percentage of area by seral
stage) .................................................... 151
Table 4-6 Number of patches after 20-year fire interval
simulations ............................................... 152
LIST OF FIGURES
Figure 1-1 Satellite image of Water Conservation Area 1 ....... .2
Figure 1-2 Diagram of approach to landscape analysis ..........5
Figure 1-3 Map of South Florida showing the Water
Conservation Areas ......................................... 16
Figure 1-4 Map of the historic Everglades system ............. 17
Figure 1-5 Bedrock topography of the Northern Everglades ..... 19
Figure 1-6 Rainfall and water regulation schedule of
Water Conservation Area 1 ......... ......................... 23
Figure 1-7 Satellite image of 1989 fire in Water
Conservation Area 1 ........................................ 37
Figure 1-8 Successional sequence in Water Conservation
Area 1 ..................................................... 44
Figure 2-1 Diagram of landscape data sources ................. 48
Figure 2-2 Map of the 1989 fire, with the location of the
successional vegetation map used in the model in Chapter
4 ........................... ...............................51
Figure 2-3 Map of three islands observed on the ground in
February, 1991 ............................................. 55
Figure 2-4 Areas of vegetation change in Water
Conservation Area 1 between 1952 and 1990, and locations
of photoplots from two vegetation studies .................. 59
Figure 2-5 Pixel coordinate values before and after
resampling ................................................. 70
Figure 2-6 Diagram of merging of images using color
transformation process ..................................... 73
Figure 2-7 Sample pixel values before and after the
transformation process ..................................... 75
Figure 2-8 Band ratios before and after the image
transformation process .....................................76
Figure 3-1 Diagram of GIS landscape analysis .................88
Figure 3-2 Tree island distribution by size .................. 92
Figure 3-3 Number of islands in each size class (detail) ..... 92
Figure 3-4 Island size class distribution .................... 93
Figure 3-5 Island size classes (detail) ....................... 93
Figure 3-6 Total area of islands within each size class ...... 94
Figure 3-7 Diagram of energy hierarchy ....................... 98
Figure 3-8 Basic model code to generate island
distributions: incremental and area-dependent growth ...... 101
Figure 3-9 Basic model code to generate island
distributions: edge-dependent growth ...................... 102
Figure 3-10 Distribution from incremental growth: island
size classes .............................................. 103
Figure 3-11 Distribution from incremental growth with
high constant island recruitment .......................... 103
Figure 3-12 Distribution from incremental growth with
increasing island recruitment: island size classes and
numbers ................................................... 105
Figure 3-13 Distribution from incremental growth with
increasing island recruitment: total area in each size
class ..................................................... 105
Figure 3-14 Distribution from incremental growth with
decreasing island recruitment: island size classes and
numbers ................................................... 106
Figure 3-15 Distribution from incremental growth with
decreasing island recruitment: total area in each size
class ..................................................... 106
Figure 3-16 Distribution from area-dependent growth:
island numbers ............................................ 108
Figure 3-17 Distribution from area-dependent growth:
island size classes ....................................... 108
Figure 3-18 Distribution from area-dependent growth:
iland size classes (detail) ............................... 109
Figure 3-19 Distribution from area-dependent growth:
total area in each size class ............................. 109
Figure 3-20 Distribution from area-dependent growth:
total area in each size class (detail) .................... 110
Figure 3-21 Distribution from edge-dependent growth:
island size classes ....................................... 111
Figure 4-1 Information used to develop the conceptual
model of vegetation change and island dynamics in Water
Conservation Area 1 ....................................... 114
Figure 4-2 Diagram of processes contributing to tree
island growth ............................................. 116
Figure 4-3 Conceptual model of vegetation and island
dynamics within the interior of Water Conservation Area
1 .......................................................... 121
Figure 4-4 Flow chart of spatial modeling and analysis ...... 125
Figure 4-5 Diagram of model structure ....................... 127
Figure 4-6 Composition of the base map during the initial
run of the spatial model (percentage of map in each
seral stage) .............................................. 133
Figure 4-7 Initial landscape model using different fire
intervals: percentage of base map in each seral stage
after 190 years ........................................... 136
Figure 4-8 Initial model results using different fire
intervals: percentage of base map in each seral stage at
selected iterations ....................................... 138
Figure 4-9 Initial model results using different fire
intervals: patch density and the largest patch at
selected iterations ....................................... 138
Figure 4-10 Composition of the base map during a run of
the improved landscape model (percentage of map in each
seral stage) .............................................. 141
Figure 4-11 Improved model results using moderate fire
severity and different fire intervals: percentage of
base map in each seral stage at selected iterations ....... 143
Figure 4-12 Improved model results using high fire
severity and different fire intervals: percentage of
base map in each seral stage at selected iterations ....... 144
Figure 4-13 Improved model results using moderate and
severe fires and different fire intervals: landscape
heterogeneity at selected iterations ...................... 145
Figure 4-14 Tree islands in landscapes simulated by the
spatial model ............................................. 147
Figure 4-15 Improved model results using moderate and
severe fires and different fire intervals: number of
tree islands at selected iterations ....................... 151
Abstract of Dissertation Presented to the Graduate School
of the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Doctor of Philosophy
LANDSCAPE DYNAMICS IN THE EVERGLADES:
VEGETATION PATTERN AND DISTURBANCE
IN WATER CONSERVATION AREA 1
Jennifer Enos Silveira
Chairperson: Dr. Wiley M. Kitchens
Major Department: Wildlife Ecology and Conservation
Water Conservation Area 1 is a 57,234 ha remnant of the
Northern Everglades, presently used as both a flood control
retention area and a wildlife refuge. It is a unique wetland
ecosystem in which the dynamics of water, vegetation, fire, and
peat are tightly linked. These interacting factors produce a
landscape pattern that provides excellent habitat for wading
birds and other Everglades wildlife.
This doctoral research focused on the landscape of Water
Conservation Area 1 and the processes determining its pattern.
Few ecological studies had been conducted in Water Conservation
Area 1, so previous studies of the Everglades and the Okefenokee
Swamp were reviewed for relevant data. The landscape was
observed and described in the field. Aerial photography,
satellite images, and a computer geographic information system
(GIS) were used to quantify the landscape.
The GIS was also used to perform a spatial analysis of
tree islands in Water Conservation Area 1. The analysis showed
more small islands than large ones, and a steep increase in the
sizes of islands above a certain size. Simple models built to
generate island size and frequency distributions demonstrated
mathematical functions that could create a size distribution
similar to that found in Water Conservation Area 1. A
conceptual model proposing an explanation for the tree islands'
size distribution was developed. The theory proposed that tree
islands grow in size over time, but fire reverses island growth.
The ability of fire to limit island growth is reduced once
islands reach some size threshold. Therefore, the growth rate
of large islands is greater than that of small islands.
A spatial model of vegetation succession and fire was made
to illustrate the conceptual model in two dimensions, using a
map of the actual landscape. The GIS was used to construct and
run the spatial model using a classified satellite image as
input. The model produced maps of landscape pattern under
different fire frequency and severity regimes. The landscape
composition and heterogeneity of the maps were compared with the
current landscape. The results illustrated the importance of
fire in determining landscape pattern.
Twenty-five years ago, as the Apollo 8 space mission
prepared to orbit the moon, many people envisioned that the
future of humanity lay in outer space. For three decades the
public imagination had entertained visions of a future society
in space. It was expected that the achievement of a lunar
landing would lead to new efforts in space exploration and
colonization. However, while the television images of Apollo 11
astronauts on the moon inspired awe and pride, another image was
just as powerful--the planet Earth, suspended in the void of
space. Along with the astronauts, viewers saw the beauty and
isolation of the earth, and felt a new appreciation of our
dependence upon it. Since that time, public interest in space
colonization has waned, while the use of space technology to
monitor our home planet has increased. Photos from manned space
craft and images from satellites have shown us the colors and
patterns of the atmosphere, the oceans, and the land as never
It was such an image of Earth that inspired the work that
follows: a satellite image of the Florida Everglades showing a
remarkable pattern of tree-covered islands emerging from a
background of marsh. The islands resemble a school of fish
swimming upstream. A particularly striking example of the
r4* 9. ;< .'^ "^' rv 1^
S. ', *~^ *': i o,.., o /^ ..:- /-" '
, ,,, ^: ,, :l L ,' ^
.. ) .. .' s'- *, "- ,'
"" ; ; ,.* ,' "" ;. : ..A. i .' .
.* "> *'* 4 ".
V *' *1"**, ...-.
^ ;'9 ^'.9 '-.*9 9 *!!,^ 1^ ^
.4v -,. '4, y. -. ;
i ,. ,, ~ ~ ~.: "jo ../.-! ,'
Satellite image of Water Conservation Area 1
pattern, contrasting with the pattern of nearby urban
development, appears in a northern remnant of the Everglades
known as Water Conservation Area 1.
The islands' pattern inspired many questions. How were
these islands formed? Why were the islands oriented in the same
direction? Why were a few islands large and oblong, while most
others were smaller and round? Were the islands changing in
size over time and were they subject to disturbances? Were the
processes responsible for the islands' pattern still operating,
and if so, what effects would recent human alteration of the
region's ecology have on them?
A review of the ecological literature on Everglades
islands showed that these questions have not been definitively
answered, and are very difficult if not impossible to answer
using standard ecological methods. Island formation is too slow
or infrequent to document in the time frame of a scientific
study. Some of the processes that formed and disturbed this
landscape have ceased functioning. Examination and dating of
the peat record could provide clues to the islands' histories,
but collecting and analyzing an adequate sample size is
These constraints are common in landscape ecology. Yet
the "big-picture" viewpoint that remote sensing has given us,
and a new emphasis on conserving whole ecosystems rather than
species, increasingly raise questions to be answered at the
landscape scale. New methods must be developed to address these
questions. In addition, as we seek to understand larger
evolving ecosystems, our expectation of definitive answers may
have to be revised. We must develop skills to extract useful
knowledge from ecosystems without statistical significance and
in spite of uncertainty and unpredictability.
In an Indian folktale, three blind men attempt to identify
an elephant. Because each man examines only one part of the
animal, they in turn describe it as a snake, a brush, and a
tree. By feeling all three parts, a fourth blind man correctly
identifies the elephant, even though he cannot feel the entire
elephant at once. This doctoral work took a similar approach,
attempting to synthesize an understanding of the dynamics of the
landscape of Water Conservation Area 1 by piecing together data
from parts of the system.
Rather than trim the subject of island pattern in this
piece of the northern Everglades into a smaller question on a
smaller scale, this work remained focused on the landscape. The
tools of remote sensing and a geographic information system
(GIS) were used to observe and quantitatively analyze the
landscape (see Figure 1-2). Data from existing ecological
studies, many of which were carried out at the community level,
were synthesized with ground observations to form a conceptual
model of the interaction between ecosystem processes and
structures. A portion of this conceptual model was translated
into a landscape model on the computer using remotely sensed
data as input. A geographic information system ran the model,
simulating fire and vegetation succession.
Data From Remote Sensing,
Field Studies and the
:Map (J.R. Richardson)
GIS Analysis of
... ... ...... Hypotheses and
i Vegetation M Conceptual Model
SSuccession Map i
GIS Model of Disturbance
and Vegetation Succession D
Lj = Data
= Classified Image
Comparison of 1990
Model Output and Satellite Mdlig n
1990 Image Image = Modelling and
Figure 1-2. Diagram of approach to landscape analysis
This dissertation describes the progression of this
undertaking. Chapter 1 gives an overview of previous studies of
landscape pattern, and describes the ecosystem of Water
Conservation Area 1. Chapter 2 describes the collection of data
from Water Conservation Area 1 and the processing of satellite
images used to quantify landscape pattern.
Chapter 3 describes a spatial analysis of the tree islands
in Water Conservation Area 1. The analysis focused on the
distribution of tree islands of different sizes, as recorded in
a 1987 classified satellite image. The size and frequency
distributions of tree islands raised three questions. Could the
ongoing formation and growth of islands create a size
distribution like that found in the landscape of Water
Conservation Area 1? Why were more small islands than large
islands found in the satellite image of the landscape? What
might be causing a discontinuity found in the size distribution
The first two questions were addressed in chapter 3 with
simple models built to generate island size and frequency
distributions for different growth functions. The models
demonstrated that the creation and growth of islands over time
could create a distribution like that found in the landscape of
Water Conservation Area 1. The third question was not easily
answered. Theories of the relationship between ecosystem
processes and ecosystem architecture suggested that a change of
some process affecting the system could cause a discontinuity in
the island size distribution.
Chapter 4 outlines a conceptual model of landscape pattern
dynamics that proposes a mechanism for the discontinuity in the
island size distribution. The conceptual model was also
converted into a spatial model of vegetation disturbance and
landscape pattern. Chapter 4 describes the methods and results
of the spatial model, used to simulate landscapes under
different fire frequency and severity regimes.
Chapter 5 summarizes and discusses the entire work. The
work produced both an example of a holistic, landscape approach
to an ecosystem, and a tool to spatially visualize a theory of
The Study of Landscape Pattern Dynamics
The study of the spatial distribution of vegetation and
its relationship to environmental processes can be carried out
at many levels of resolution: from the study of plants in a
vegetation community, to communities within a landscape, to
landscapes on the globe. A landscape is defined as area of land
containing multiple ecological communities, "a heterogeneous
land area composed of a cluster of interacting ecosystems that
is repeated in similar form throughout" (Forman and Godron
1986:594). Landscape ecology examines and analyzes the spatial
patterns and dynamics of landscapes over time.
One area of study in landscape ecology is the relationship
between vegetation structure (i.e., the size of community
patches, patch connectivity, edge length and diversity), and
environmental disturbance, (i.e., fire, wind and human activity)
(Weins et al. 1985, Turner 1987). The majority of research on
this topic has focused on the effect of disturbance on landscape
pattern (Potter and Kessell 1985). For example, Suffling et al.
(1988) investigated the control of landscape diversity by fire
disturbance in boreal forests. They found that forests with an
intermediate fire frequency had a higher landscape diversity
than forests with high or low fire frequencies.
Fewer researchers have addressed the reciprocal effect of
landscape pattern on disturbance (Franklin and Forman 1987).
Turner et al. (1989) used spatial percolation theory models on
simple abstract landscapes to study how landscape pattern
affects the spread of disturbance. They found that the
frequency of disturbance (the probability of a disturbance event
occurring in a given time period) and the intensity of
disturbance (the probability of a disturbance spreading)
influence the area of landscape disturbed. However, this effect
differs according to the amount and connectivity of patches of
habitat susceptible to disturbance. Disturbance and landscape
structure, therefore, have a complex two-way interaction.
Simulation modeling is required to gain an understanding of
these interactions, due to the complexity of landscape dynamics.
Models of Landscape Pattern
The study of landscape dynamics is difficult because of
the large scale in space and time over which landscapes change.
Scientific methods for the study of single species or
communities within ecosystems are well developed, while there
are few methods designed for landscapes. The amount of
information contained in a large natural system is too great to
be analyzed using standard ecological sampling and analysis
techniques. Therefore landscapes are frequently studied by
simplifying them into models, such as ecosystem/energy models
(Odum 1983), vegetation/forest stand models (Pearlstine et al.
1985, Menaut et al. 1990), patch dynamics (gap) models (Shugart
and Noble 1981, Shugart and West 1977) and gradient models
The term "landscape model" is sometimes used to refer to
models which simulate processes that determine where entire
landscapes of different types are found. These models operate
at a higher hierarchical level than the models referred to in
this work. "Landscape model" is used here to indicate models
which simulate dynamics within a landscape.
Appropriate scaling in models
Ecological computer models can be packed with information
and simulate a great number of processes, but this complexity
does not necessarily produce realistic behavior or
understandable results. A model which limits information and
processes to the temporal and spatial scales pertinent to the
model's subject is more comprehensible and efficient than one
which attempts to explicitly model as many processes as
possible. In general, models are best restricted to variables
ranging over no more than three orders of magnitude in space or
time (Gunderson 1992). This limits a model to variables within
one or two scale domains.
The scale of a variable or process can be identified by
its frequency, by its area of influence, or by its energy.
Areas best measured in centimeters to tens of meters, and time
frequencies of minutes to decades, fall into the domain termed
the "microscale." Areas of tens of meters to hundreds of meters
and time frequencies of decades to centuries fall into the
"mesoscale" domain. Areas of hundreds of meters to thousands of
kilometers and time frequencies of centuries to millennia fall
into the "macroscale" domain (Holling 1992a).
In the Everglades, for example, the species composition of
a wet prairie community changes seasonally during an annual
cycle (Kushlan 1990) at the microscale. However, this
fluctuation is not indicative of long-term community change at
the landscape level, because the species involved are all
normally members of the wet prairie community. To add this
fluctuation to a landscape pattern model would unnecessarily
increase model complexity. Instead, community types rather than
species can be used in the model, understanding that species
composition varies within a community.
Landscape patterning processes in the Everglades operate
at the mesoscale (Gunderson 1992). These processes include
vegetation succession and disturbances like fire, droughts and
storms (DeAngelis and White 1994). A landscape pattern model
should focus on changes at the mesoscale. It should track a
measure of pattern or other landscape attribute which reflects
At the macroscale, slow climatic and geologic processes,
such as sea level rise and geologic uplift, determine where
entire landscape types are found within the Everglades (Davis et
al. 1994). Rather than explicitly include macroscale processes
in a landscape pattern model, these slow global changes can be
assumed static for the time period of the model.
Most ecological models simulate population dynamics or
ecosystem functions: the flows of energy, materials, species, or
individuals in the system. The majority of these models are
spatially aggregated. This means that while they calculate the
quantities and flows of variables within a system, they ignore
the location of those variables (Sklar and Costanza 1991, Turner
and Dale 1991). Spatial models differ from spatially aggregated
models in that they explicitly incorporate the influence of
spatial structure into model functions, usually by means of a
grid. While including spatial information can mean the
sacrifice of functional details, this is justifiable when
spatial factors such as edge effects are critical. A landscape
model must be spatial if landscape pattern is to be analyzed.
Individual cell models. One approach to spatial modeling
is to create a function model for each individual cell within a
spatially-referenced grid (Richardson 1988). Each individual
cell model uses information from neighboring cells to simulate
flows between it and other cells and to change the value of what
is stored within the cell. Rates of change are therefore a
function of spatial relationships.
The cell model method can perform complex computations for
each cell and handles continuous variables well. It is often
used in models in which an important component is the simulation
of water flow (Costanza et al. 1990, Walters et al. 1992,
Fennema et al. 1994). However, the computing requirements to
run a model of this type are very high if numerous cells are
used. For example, a cell model of the Atchafalaya delta
(Costanza et al. 1990) covered 2479 km2, and used a 1 km2
gridcell. A typical 22-year run of the model required
approximately 24 hours on a VAX 11/780 computer, or 15 minutes
on a CRAY X/MP supercomputer. If the model had instead used a
10 m2 gridcell capable of showing landscape pattern but was
otherwise unchanged, the same run would take approximately 27
years on the VAX and 104 days on the CRAY (Costanza et al.
The number of gridcells in a model can be reduced by
modeling small areas (less than 20 km2) or using a coarse
resolution (greater than 1 km2). However, if a natural
landscape is being modeled, a small area may not adequately
illustrate landscape behavior. A course resolution may not
capture spatial variation in the environment necessary to model
processes important in determining landscape pattern. A cell
model of the hydrology of 1fe Everglades using a 4 km2 cell
(Walters et al. 1992) simulated hydrologic behavior well, but a
vegetation component added to the model did not predict
vegetation change satisfactorily (Gunderson 1992).
Cartographic models. Another variety of spatial model
uses a geographic information system (GIS). The GIS approach
emerged from the landscape architecture and geography
disciplines. Some GIS models use vector (line) maps as input.
However, the majority of ecological GIS models use sets of
georeferenced gridcells known as rasters or "data layers." This
approach is also known as cartographic modeling (Tomlin 1990).
Rather than using a single grid, with a model and multiple
storage for each cell, cartographic models store different
types of values in separate layers and use a single model to
superimpose and manipulate them. These models can perform
spatial functions on values associated with a gridcell and its
neighboring cells within a layer, such as calculating proximity
and diversity measures, to generate new layers. Multiple layers
are combined by the logic statements and mathematical equations
of a model, and the resulting information forms another layer.
This process can be repeated to simulate changes over time.
GIS models have been most frequently used to simulate
single events, with simpler calculations and/or fewer iterations
than applications of the individual cell model method described
above. However, GIS software programs have been rapidly
evolving and improving, and now many have the capability to
iteratively model a landscape over time. The spatial model of
Water Conservation Area 1 developed as part of this work
(described in Chapter 4) is a GIS model.
The Landscape of Water Conservation Area 1
Water Conservation Area 1 lies in Palm Beach County,
Florida, and consists of 57,234 hectares of peat wetlands (see
Figure 1-3). The majority of these wetlands are distributed in
a complex mosaic of sloughs, wet prairie and sawgrass (Cladium
jamaicense) marshes, shrub patches, and forested islands.
Disturbances, particularly floods and fires, have shaped this
mosaic of vegetation communities (Gunderson 1994, DeAngelis
1994). The juxtaposition of wetlands has made Water
Conservation Area 1 a high quality wildlife habitat,
particularly for small fish, amphibians, alligators (Alligator
mississippiensis), and wading birds such as herons, egrets, ibis
and wood storks (Mycteria americana) (Thompson 1972). Hoffman
et al. (1994) identified this wetland mosaic as the "Loxahatchee
tree island marsh", and found that it supported the highest
density of wading birds of all the water conservation areas of
The pattern of the landscape mosaic in Water Conservation
Area 1 resembles other peatlands in the Everglades and
elsewhere, but is unique in its abundance of small forested
islands known as bayheads or tree islands. Tree islands
contrast with the marsh vegetation communities around them, so
they highlight the pattern of the landscape mosaic well.
Because they are easily delineated on aerial photography and
satellite imagery, tree islands are useful indicators with which
to quantify the landscape pattern.
The following sections review the environmental factors
that created and continue to influence the landscape of Water
Conservation Area 1 and its mosaic pattern. These factors are
both natural and human in origin.
The History of Water Conservation Area 1
A century ago, Water Conservation Area 1 was known as the
"Hillsborough Lake", although it was not a lake. It was part of
an uninterrupted expanse of wetlands extending from Lake
Okeechobee to the southern tip of Florida (see Figure 1-4).
These wetlands--the Everglades--were geologically young, and had
been in a state of flux since their formation. They originated
approximately 5000 years ago, after glaciers retreated and sea
level rose at the beginning of the Holocene epoch (Gleason and
Stone 1994). The southern Florida peninsula had probably been
an oak savanna at the end of the Pleistocene epoch, but as sea
level rose, fresh water backed up into low lying areas.
Hillsborough Lake was one such area. Wetland vegetation grew in
these new wetlands, and peat formed, displacing water and
increasing the flooded area. Sea level continued to rise, and
eventually the wetlands spread to their current extent (Wanless
et al. 1994, Gleason and Stone 1994).
The Hillsborough Lake area was first affected by human
hydrologic manipulations in the 1880's, when the Caloosahatchee
River canal linked Lake Okeechobee to the Gulf coast, reducing
the volume of water flowing south to the Everglades (see Figure
1-3). Flooding duration (hydroperiod) in the Hillsborough Lake
area was reduced (Thompson 1970). In 1915 the West Palm Beach
and Hillsboro canals were dug to the north and south of
Hillsborough Lake (Figure 1-3). These canals further altered
the volume and flow pattern of water in the area.
Drastic hydrologic changes came to Hillsborough Lake in
the 1950's, when flood control work began to enclose the area
with levees, canals and pump stations. Impoundment was
completed in 1960, creating Water Conservation Area 1. The
South Florida Water Management District began to use the area to
store runoff from adjacent agricultural lands. Since 1951, the
U.S. Fish and Wildlife Service also has administered Water
Conservation Area 1 as a migratory bird refuge (Light and Dineen
1994, VanArman et al. 1984).
Figure 1-3. Map of South Florida showing the Water Conservation
Location of Water
Conservation Area 1
Figure 1-4. Map of the historic Everglades system
Today flood control structures restrict the hydrologic
connectivity of Water Conservation Area 1 to the rest of the
Everglades ecosystem. Flood control and water management have
allowed rapid growth of the agriculture and urban development
now surrounding the area. It remains a northern refuge for
Everglades species threatened by habitat destruction.
The Substrate of Water Conservation Area 1
Water Conservation Area 1 lies over limestone bedrock
known as the Fort Thompson Formation. Originally formed on the
flat floor of a shallow sea, the Fort Thompson Formation has
since been eroding and dissolving. At a fine scale the bedrock
surface is uneven, etched with a topography of depressions and
ridges. At a larger scale, a wide trough in the bedrock
surface, known as the Loxahatchee Channel, runs roughly north-
south from Lake Okeechobee to the southern end of Water
Conservation Area 1 (Jones 1948). Water overflowing from Lake
Okeechobee followed the trough, which is still evident in the
grain of the landscape pattern in Water Conservation Area 1.
Before water management, the trough was relatively wetter than
surrounding areas, inspiring the name "Hillsborough Lake." The
earliest Everglades peat was formed in the Loxahatchee trough
(Gleason and Stone 1994).
DATUM IS MEAN SEA LEVEL
Figure 1-5. Bedrock topography of the Northern Everglades
Source: Jones 1948
On top of the bedrock structure lies a layer of peat,
varying between 124.5 and 457.2 cm deep (Richardson and
Kitchens, unpublished data). The peat layer fills in and masks
the microtopography of the bedrock below it. Its surface slopes
gradually from north to south, dropping less than four feet with
respect to sea level over approximately 30 miles (Richardson et
Peat is built up by the deposition of dead vegetation at a
faster rate than decomposition occurs. Estimates of peat
accumulation in the Everglades ranged from 7.62 to 10.77 cm per
century (Bond et al. 1986). The oldest peat samples dated from
Water Conservation Area 1 are approximately 4800 years old
(Gleason and Stone 1994). The peat found at Water Conservation
Area 1 is derived mainly from water lily, Nymphaea odorata,
(classified as Loxahatchee peat), sawgrass, (classified as
Everglades peat), and trees, Ilex cassine and Persea borbonia,
(classified as Gandy peat) (Gleason et al. 1980).
While the peat surface of Water Conservation Area 1
appears flat when observed from the air, its micro-topography is
quite uneven. An observer on the ground might stand high and
dry on a tree island, but wade knee-deep in water in an adjacent
slough. While this elevational difference may not be
significant in an upland ecosystem, its ecological effect in a
wetland is amplified by flooding. The distribution of
vegetation communities at Water Conservation Area 1 is
correlated with this peat surface topography (Pope 1989).
Several forces shape the micro-topography of peat in Water
Conservation Area 1. These include: the buildup of peat by
vegetation; the disturbance and transportation of peat by
flowing water and alligators; and the removal of peat by fires.
These processes are discussed below.
Water in Water Conservation Area 1
The natural hydrology
Before the era of flood control, Everglades hydrology was
driven by rainfall (Parker 1984). Periods of high rainfall in
the Kissimmee watershed of Lake Okeechobee caused flooding in
the Kissimmee River Basin. As a result, Lake Okeechobee
expanded westward into wetlands along Fisheating Creek. Water
from Okeechobee also rose over a natural berm along the lake's
south shore, and flowed south into what is now the Everglades
During the annual wet season (June November), water from
the Kissimmee/Lake Okeechobee watershed connected with the
Everglades water body to create a continuous inclined hydrologic
gradient extending from Lake Okeechobee to Florida Bay (Parker
1984, Fennema et al. 1994). The entire Everglades, with the
exception of tree islands, were inundated. Water followed this
gradient in a sheet flow across the vegetated landscape,
inspiring the names "Pa-hay-okee" (grassy waters) and "river of
grass" (Douglas 1988). Unlike a river, however, water flow was
so slow and evapotranspiration so high that it is unlikely water
from Lake Okeechobee ever reached Florida Bay. In Water
Conservation Area 1, rainwater fell, flowed slowly south along
the slight elevational gradient, was evaporated or transpired,
and fell again.
In dry years, during the dry season (December May), the
hydrologic gradient in the Everglades was broken. In Water
Conservation Area 1, water evaporated and transpired without
replenishment, until the only water remaining above the peat
surface was found in sloughs and alligator-excavated ponds.
Below the peat surface, water was held like a sponge.
Although a relatively small amount of water actually
reached Water Conservation Area 1 via overland flow from the
north, the role played by "upstream" water was important.
Because peat is loosely consolidated and can be temporarily
suspended in water, the sheet flow of water probably transported
bits of disturbed peat and vegetation. Evidence of this
phenomenon lies in the shape of Water Conservation Area l's
larger tree islands (Gleason and Stone 1994). All large tree
islands have an oblong shape, aligned along the north/south
direction of flow. Sheet flow may have eroded the heads and
flanks of these islands, and/or deposited peat at the downstream
ends. Smaller tree islands in Water Conservation Area 1 do not
show this effect (see Figure 1-1).
Hurricanes, tropical storms and prolonged high rainfall
probably caused deeper flooding in the Everglades, which
completely inundated tree islands and flowed at faster rates
The effects of high-energy
infrequent flooding events like hurricanes on the landscape of
Water Conservation Area 1 have not been documented.
Flood control hydrology
The advent of flood control technology transformed the
hydrology of Water Conservation Area 1 (Richardson et al. 1990).
Today the hydrologic regime is determined by a regulation
schedule set by the South Florida Water Management District (see
Figure 1-6). The schedule's goal is to allow the storage of
Average Monthly Rainfall (1970-1985)
NF M// I7Y ///1 A S O
JAN FEB MAR APR MAY JUN JLY AUG SEP OCT NOV DEC
Figure 1-6. Rainfall and water regulation schedule of Water
Conservation Area 1
Source: Richardson et al. 1990
than the usual sheet flow.
drainage water from agricultural areas during the wet season,
particularly in the event of a hurricane. Stored water is held
during the dry season, and then released at the end of the dry
season to allow room for next year's storage. This pattern of
flooding and dry-down opposes the region's natural hydrology.
The construction of canals and dikes halted the southward
sheet flow of water across the Everglades. Since the
impoundment of Water Conservation Area 1, water has moved from
the canals at the perimeter of the area inward to the interior
and been held there. Rather than lying evenly like a blanket
over the landscape, as it did when sheet flow existed, water now
pools behind the dike at the low southern end of the area.
Water at the northern end of the area drains south, but with no
water flowing from the north to replace it, the northern end
dries. Therefore hydroperiods in the south have lengthened
while hydroperiods in the north have shortened relative to
natural regimes. The high energy floods caused by hurricanes no
longer occur, due to flood control in and around Lake
The alteration of the hydrologic pattern of Water
Conservation Area 1, together with lower water quality, is
changing the landscape mosaic. Studies of vegetation (Thompson
1972, Alexander and Crook 1975, Richardson et al. 1990) have
documented the loss of sawgrass and tree islands at the southern
edge of the area, the invasion of dense cattails (Typha sp.)
along the western edge, and consolidation of shrub patches at
the area's northern tip.
The Tree Islands of Water Conservation Area 1
Tree islands at Water Conservation Area 1 are not found on
top of limestone bedrock pinnacles as are the tree islands of
the southern Everglades. Rather they are "bayhead" islands:
topographic high points made only of the peat itself. These
islands are formed by a more complex process than the pinnacle
islands farther south, an effect of the deep peat substrate
(Gleason and Stone 1994). Because peat is a product of
vegetation, and the vegetation living on bayhead tree islands
contributes to island formation, Gleason et al. (1980) termed
the islands "unique geobotanical landforms."
Tree island formation
The shapes of the tree islands in Water Conservation Area
1 bear a strong resemblance to tree islands found in boreal
patterned peatlands in the Glacial Lake Agassiz region of
central Canada and Minnesota (Glaser 1987). There, peat-based
tree islands are found in sedge-dominated fens. Like Water
Conservation Area l's islands, they are oblong and oriented
parallel to the direction of water flow, with rounded heads and
tapering tails like islands in the southern Everglades.
However, these islands were formed differently from those in
Water Conservation Area 1. The boreal peat islands are forest
fragments left after minerotrophic seeps emerged on the slopes
of raised forested bogs. The seeps generated extensive "water
tracks", wide treeless fen channels dominated by sedges and
Sphagnum species (Glaser 1987). The water tracks formed a
network similar to a braided stream, then expanded and merged,
leaving islands of trees in the interstices. Some water tracks
are also patterned with ridges ("strings") and sloughs
("flarks") oriented perpendicular to the direction of water
Tree islands in Water Conservation Area 1 are formed by an
additive process rather than the subtractive process of the
boreal peatlands mentioned above. Gleason and others (1980)
found radiocarbon evidence in two peat cores of islands in Water
Conservation Area 1 that suggested the islands originated as
peat "batteries" (floating rafts of peat and live vegetation).
In peat prairies of the Okefenokee Swamp, an ecosystem similar
to Water Conservation Area 1 but without the influence of sheet
flow, a connection between peat batteries and tree islands has
also been documented. Research at Okefenokee (Cypert 1972; Rich
1979; Duever and Riopelle 1983, Glasser 1986) indicated that
islands there are formed under wet conditions when large
batteries of peat rise to the surface of the water. Plant
succession occurs on the floating batteries, further
consolidating the peat structure until woody vegetation is
While the mechanisms of peat battery formation at Water
Conservation Area 1 have not been documented, newly formed
batteries have been observed during high water following a very
dry period (W. M. Kitchens, personal communication). Trapped
gasses (Rich 1979) or buoyant plant rhizomes (Gleason and Stone
1994) may be involved in floating the batteries.
Three types of peat batteries have been described in the
Okefenokee Swamp: mats, bulges and aggregates (Rich 1979). Mats
are sheets of peat as large as 20 feet in diameter, which
dislodge from the substrate and rise to the water surface. They
may or may not completely detach from the substrate. If a mat
battery completely detaches, water can carry it to a new
location. When the water level falls, the mat settles back down
to the peat surface, creating a topographic high which becomes
The displacement of mat batteries is not common in the
Okefenokee Swamp, where sheet flow is rare. Thus, Okefenokee
tree islands are round or irregularly shaped, and are not
oriented parallel to water flow. They are similar to small
islands in Water Conservation Area 1. Transportation of mat
batteries has been repeatedly observed in Water Conservation
Area 1, where the shapes of large tree islands show the
influence of water flow (Baker 1952; M. Maffei, personal
communication, Gleason and Stone 1994).
Bulge batteries are sheets of peat which rise to the
surface but remain entirely attached to the substrate. These
are common at Okefenokee, but not in Water Conservation Area 1.
Aggregate batteries are accumulations of small floating
bits of peat, which are gathered together by wind or water flow.
Unlike mats and bulges, aggregates appear gradually rather than
suddenly. They occur in a wide range of sizes and are most
often found at the edges of alligator-excavated ponds
("alligator holes"). They may play an important role in the
expansion of existing islands in addition to the creation of new
Glasser (1986) traced all batteries and islands visible in
Chase Prairie, Okefenokee Swamp, using aerial photographs
covering a 40-year period. She documented the formation of 82
islands during this period, in an area of approximately 30
square kilometers. Of these 82 islands, 64 derived from
aggregate batteries at the margins of alligator holes. During
the period of study, all aggregate battery islands persisted and
developed woody vegetation in 10-30 years, even if the original
alligator hole had disappeared. After 40 years the islands
supported trees. The remaining 18 islands Glasser observed
derived from bulges and mats, but only 10% of these persisted
long enough to develop woody vegetation. Most bulges and mats
subsided below the water surface.
The relationship between alligator holes, aggregate mats
and tree islands has not been investigated at Water Conservation
Area 1. Alligators are very numerous there, and alligator holes
are frequently found adjacent to tree islands. The role of
alligators in shaping peat topography is clearly important and
deserves study, although it is not explicitly addressed in this
Fire in Water Conservation Area 1
Fire and the Everglades ecosystem
A major factor shaping the Water Conservation Area 1
landscape is fire. Fire is an omnipresent process in the
Everglades ecosystem, since
the climate and the vegetation of south Florida,
coupled with the area's elaborate system of drainage
ditches and canals, have created one of the highest
fire potentials in the United States. Abundant
summer rainfall (even during extremely dry years) and
warm temperatures from May through October promote a
rank buildup of fuel that becomes explosive during
the 6- to 8-month dry season. There are more
days with lightning recorded in south Florida than
elsewhere in the Nation. Over 6,000 lightning
strikes were recorded in inland south Florida during
one 6-hour period in the summer of 1976. (Wade et al.
Seasonal probability of fires
The probability a fire will be ignited in the Everglades
varies seasonally. In Everglades National Park most fires are
ignited by humans, in winter, spring and fall (Taylor 1981).
Humans have been present and lighting fires in the Everglades
since its genesis, although human-ignited fires were probably
less frequent before Europeans arrived (Robertson 1953). Fires
not caused by humans are ignited by lightning, which is most
common during the summer storm season. While the two types of
fires in Everglades National Park generally occur at different
times of the year, both types burn the greatest area in the
month of May (Taylor 1981) because of factors determining fire
propagation and intensity.
Fire behavior factors
Propagation and intensity of fires in the Everglades, as
in all wildland fires, are influenced by fuel density and
moisture content, atmospheric humidity, wind speed and
direction, and soil moisture/water level (Anderson 1982,
Rothermel 1983, Burgan and Rothermel 1984, Andrews 1986, Andrews
and Chases 1989). Topographic factors such as slope do not
influence Everglades fires as they do in more variable terrain,
but canals and open water may stop or direct fire movement.
The most important factor in fire behavior is the amount
and moisture content of fine fuel found on a site (Rothermel
1983). Rothermel defined fine fuel as dead or living plant
material less than 1/4 inch in diameter that can change moisture
content by a specified amount within 1 hour. Needles, leaves,
grass and dead herbaceous stems are fine fuel. Because the
moisture content of fine fuel changes rapidly in response to
meteorological and soil moisture conditions, these conditions
are also important in determining the probability and behavior
of a fire.
Fire potential variables in Water Conservation Area 1 vary
daily, seasonally and between wet and dry years. During the dry
season (December May), evapotranspiration lowers the water
level, exposing peat in some areas. Moisture decreases in the
exposed peat, in live vegetation, and in dead fuel of all sizes.
Some plant species senesce during the winter, increasing the
amount of dead fuel. Rare winter frosts also kill vegetation.
The optimal conditions for fires occur at the end of the dry
season, when water levels and soil moisture are at their lowest,
dead fuel is abundant and dry, and the first summer storms
provide lightning as a source of ignition. Fires at this time
of year propagate well and may be moderate to severe.
During the wet season (June November) lightning is
common and ignites fires, but in most years Water Conservation
Area 1 is inundated with water at this time. Fuel is moist, so
fires burn only a limited area around the strike location at a
During drought years both the wet and dry seasons are dry,
creating conditions for severe fires. Minimal rain falls during
summer storms and winter weather fronts. Water evaporates and
transpires without replenishment until it stands only in sloughs
and alligator holes. Vegetation desiccates, and the peat
surface itself may become dry enough to burn. The absence of
standing water permits annual plants adapted to shorter
hydroperiods to grow in the wet prairies. The resulting dense
vegetative cover increases fuel density. Fires ignited under
these dense, dry fuel conditions are intense, and spread quickly
over large areas (Wade et al. 1980, Duever 1984).
The terms "fire intensity" and "fire severity" are
sometimes used interchangeably. More specifically, intensity
refers to the energy expended by a fire, while severity refers
to the effect of a fire on the landscape. The severity of fires
in the Everglades landscape varies, depending upon the intensity
of the fire. The mortality of foliage caused by fire is a
function of plant moisture content, the temperature of the fire
and the time the plant is exposed to the fire (Wright and Bailey
1982). The mortality of plant individuals after a fire is a
function of many factors, including their species, their age,
the season of the fire, plant competition and the post-fire
Fires are low-intensity during the Everglades wet season,
since plant moisture is high and damage to living plant tissues
is reduced. Live foliage above the water surface may burn along
with dead dry fuel, but the submerged crowns of plants are left
intact. Everglades vegetation is adapted to frequent, low-
intensity fires. Unless a low-intensity fire is followed by a
long period of deep flooding, most species resprout quickly
(Wade et al. 1980). Low-intensity fires hasten the release of
nutrients from dead biomass and slow the rate of peat
accumulation by consuming fine material which would otherwise
become peat. They maintain, rather than change, vegetation
patterns at the landscape level.
Moderately intense fires during the dry season also
maintain the existing landscape pattern by slowing vegetation
succession and reducing peat buildup. Moderate dry-season fires
burn a greater proportion of aboveground live vegetation than
wet-season fires. The destruction of living foliage requires
plants to expend energy reserves to resprout. While sawgrass
resprouts quickly after a moderate fire (Forthman 1973), small
woody seedlings with little energy stored in their root systems
may not survive. This serves to set back the vegetation
succession of woody species. By burning dead fuel which would
otherwise be incorporated into the peat, moderate fires maintain
the lower elevations and longer hydroperiod conditions preferred
by wet prairie. In the absence of fire, peat will accrete and
create conditions more favorable to sawgrass and brush (Loveless
1959, Wade et al. 1980).
Extremely intense fires, occurring during drought years
or where water levels have been artificially lowered, cause
long-term changes in the vegetation pattern of the landscape
(Alexander and Crook 1975). Intense fires burn all desiccated
aboveground vegetation susceptible to ignition and mortality.
The high heat generated by the consumption of the ample dry fuel
penetrates into the soil, killing the roots of some plants.
Intense fires can kill even sawgrass and mature trees. They can
burn into the peat, lowering elevation. The lower elevation
increases hydroperiod once the drought ends, leading to a long-
term shift in vegetation community (Craighead 1971, Loope and
Patterns of fire events
The frequency of fires is important, because fires can
cause plant mortality if they occur too frequently, even if
their intensity is low. Gunderson and Snyder (1994) analyzed
fire records from Everglades National Park from 1948 to 1990,
and from Water Conservation Areas 2 and 3 from 1980 to 1990 to
search for patterns of fire occurrence. They performed a
Fourier analysis on the log of fire sizes over time and found
that the dominant temporal cycle was an annual one. A 10 to 14
year cycle accounted for the next largest amount of variation.
This longer cycle "occurs at similar time ranges as other
environmental variables in the system; water levels, water
flows, and pan evaporation all fluctuate on cycles of similar
time frames (Gunderson 1992). These variables indicate longer
term fluctuations in wet or dry conditions and, hence,
susceptibility to burning." (Gunderson and Snyder 1994:302)
The cycles found by the Fourier analysis do not indicate
how often a point on the ground might be expected to burn, and
neither do fire frequency (the mean number of fires per time
period), and fire return interval (the mean time period between
fires). The frequency at which vegetation at any given location
is subjected to a fire is more pertinent to the formation of
landscape pattern than the number of fire events in a region.
While fire frequency is often tabulated within a known area,
such as a park or management area, this statistic does not
incorporate the area burned by each fire. Fire rotation time is
calculated by dividing the total area from which fire data are
collected by the mean area burned annually. Rotation time
estimates how often any point on the ground is expected to burn,
or the period required for the entire area to be burned if no
area burns twice. The rotation time calculated from fire
records for Everglades National Park from 1948 to 1979 (Taylor
1981) is 33.6 years. This includes wildfires and controlled
Fire in Water Conservation Area 1
Detailed fire records for Water Conservation Area 1 do not
exist. Until recently, only very large fires were noted by
wildlife refuge managers, and only rough acreage estimates were
possible. Some large recorded fires are listed in Table 1-1
The most severe fires burned in 1962 and 1989. These fires
ignited in drought years in which large fires also burned in
Everglades National Park (Gunderson and Snyder 1994).
Table 1-1. Large fires recorded in Water Conservation Area 1
Year Hectares Burned
1962 > 40,470
The most recent severe fire in Water Conservation Area 1
burned in April of 1989, at the end of a dry season following
eight months of drought. A winter freeze had heightened the
fire potential by increasing dead vegetation. The fire ignited
when a controlled burn just outside the area jumped across the
canal and started a wildfire. Over 18,000 hectares (40,000
acres) burned in the northern and western portions of the Water
Conservation Area 1 (M. Maffei, personal communication).
Aerial photographs taken soon after the 1989 fire showed
that large tree island edges, small and medium sized tree
islands, brush, sawgrass and wet prairie were severely burned
(Loxahatchee National Wildlife Refuge, unpublished data). No
green vegetation was visible in the burned area, with one
exception--the photographs showed that tree crowns in the
centers of large tree islands were green and undamaged.
It appears that the 1989 fire in Water Conservation Area 1
swept quickly through the prairie and sawgrass. When it met
with an island, the fire swept around the island and continued
on, but ignited the island's perimeter in the process. Fire
then burned inwards on all sides of the island. This behavior
is evident in a panchromatic SPOT satellite image of Water
Conservation Area 1 taken while the fire was burning (see Figure
1-7). If a tree island was small, its vegetation was consumed
by the fire. If it was large, the interior of the island did
not burn, perhaps because the wetter, sparser and/or calmer
interior environment extinguished the fire. This resulted in
the fire's energy being released differently on large and small
islands. If this phenomenon occurs in all severe fires, then
fire is an important factor determining the sizes of tree
islands, and thus the landscape pattern in Water Conservation
Area 1. (This hypothesis is explored in detail in Chapter 4).
The interiors of large islands may have escaped the fire
because their understory vegetation and microclimate are quite
different from other vegetation communities in Water
Conservation Area 1. The leaf area index of a tree island
overstory is significantly higher than sawgrass or wet prairie
vegetation (Gunderson 1992). Therefore tree island canopies
provide dense shade, lowering air temperature and conserving
litter moisture. Insolation, air temperature and dead fuel
moisture are factors in fire potential (Andrews and Chases 1989,
Rothermel 1983, Andrews 1986). Tree canopies also act as a
windbreaks to reduce wind speed (Fons 1940, Marston 1956),
another important factor in fire potential (Schroeder and Buck
..... : '.* .' ... :..... '
,.' ., '7, '
S. .. .. ... '. ,. ^ '...o r'
.,. ... jf .*
* ', /c
4 -*- 'S *
;. *. : **
Satellite image of 1989 fire in Water Conservation
Transpiration on tree islands is also higher than in
sawgrass or wet prairies (Gunderson 1992). Tree roots reach
lower into the water table than the roots of grasses, sedges,
and herbs, providing the trees with moisture during drought
after shallow-rooted marsh vegetation has begun to desiccate.
Higher transpiration on the islands elevates relative humidity
under the shaded canopy relative to the open marsh, further
conserving the moisture of the understory and litter on tree
islands. While the understory at the perimeters of tree islands
consists of dense sawgrass, vines and brush, providing ample
fire fuel, the understory in the interior of large islands is
sparse, consisting mainly of ferns (Pope 1989, Wade et al.
Vegetation community types
Plant species in the Everglades can be organized into
associations, or communities, which share similar environmental
requirements. An early classification of Everglades species
into community types was based upon groups of species observed
growing together, but did not use hydrologic regime as an
organizing criterion (Davis 1943). Later, Loveless (1959)
characterized a set of dominant communities in the Everglades,
tree islands, sawgrass marshes, wet prairies and aquatic
sloughs, and discussed the influence of hydrologic regime in
More detailed classifications of Everglades plant
communities were made later (Craighead 1971, Wade et al. 1980,
Olmsted and Loope 1984). Gunderson (1994:327) divided
Everglades freshwater wetlands into forested communities,
graminoid associations and areas with "little or no emergent
vegetation." Forested communities include bayheads, willow
heads and cypress forests. Graminoid associations include
sawgrass marshes, peat wet prairies and marl wet prairies.
Little or no emergent vegetation includes ponds, creeks and
sloughs. Using this community classification, the common plant
species of Water Conservation Area 1 are listed in Table 1-2.
Table 1-2. Common plant species of Water Conservation Area 1,
by vegetation community
Vegetation Species Common Name
Bayhead Persea borbonia Red Bay
(Tree Island) Ilex cassine Dahoon Holly
Blechnum serrulatum Swamp Fern
Acrostichum danaeifolium Leather Fern
Osmunda regalis Royal Fern
Lygodium japonicum* Japanese Climbing
Smilax laurifolia Bamboo Vine
Myrica cerifera Wax Myrtle
Chrysobalanus icaco Coco Plum
Willow Head Salix caroliniana Willow
Ludwigia octovalis Water Primrose
Mikania scandens Climbing Hemp-vine
Phragmites communis Common (Giant) Reed
Sawgrass Marsh Cladium jamaicense Sawgrass
* indicates exotic species
Sources: Richardson et al. 1990, Silveira 1988 (unpublished
data), Thompson 1970, Alexander and Crook 1975, and Browder et
Vegetation Species Common Name
Peat Wet Rhynchospora traceyi Tracy's Beakrush
Prairie Eleocharis montevidensis Spikerush
Eleocharis quadrangulata Spikerush
Panicum hemitomon Maidencane
Eriocaulon compressum Hat-pins
Fuirena scirpoides Rush Fuirena
Xyris spp. Yellow-eyed Grass
Hypericum spp. St. John's Wort
Micro-algae: Periphyton Mats
Pontederia cordata Pickerelweed
Sagittaria latifolia Arrowhead
Slough Utricularia spp. Bladderworts
Potamogeton illinoensis? Pondweed
Najas guadalupensis Southern Naiad
Nuphar advena(luteum?) Spatterdock
Nymphoides aquaticum? Floating Heart
Nymphaea odorata White Water Lily
Eichhornia crassipes* Water Hyacinth*
Hydrocotyle umbellata Water Pennywort
_Salvinia spp. Water Fern
Invasive Typha domingensis Cattail
Species Melaleuca quinquenervia* Punk Tree, Cajeput*
* i ndl ic-a=t-es ovnt- i e- c i eso." ioc
Sources: Richardson et al. 1990, Silveira
data), Thompson 1970, Alexander and Crook
1975, and Browder et
Environmental factors in vegetation community distribution
The distribution of plant communities in the Everglades
is a function of hydrologic and nutrient factors, along with
fire history, animal disturbance, hurricanes, and previous
vegetation type (Davis 1943, Loveless 1959, Craighead 1971,
McPherson 1973, Davis 1994).
Plant species in Water Conservation Area 1 have varying
tolerances for depth, duration and variability of flooding (Pope
1989, Richardson et al. 1990). The spatial distribution of
Everglades vegetative communities is largely a function of
hydroperiod (McPherson 1973). In semi-permanently flooded
areas, hydroperiod is largely a function of elevation. In the
Everglades "very minor changes in elevation can cause major
shifts in the location of plant communities" (Alexander and
Submerged and floating aquatic plants found in the slough
community tolerate continuous and deep flooding. The species of
the wet prairie require shorter hydroperiods and shallower water
than slough, but tolerate longer hydroperiods than sawgrass.
Tree islands are higher in elevation than the surrounding
communities, and have the shortest hydroperiod.
Hydroperiod at a site is determined by rainfall,
evapotranspiration, the elevation of the peat surface,
surrounding topography, and water flow patterns. While these
environmental parameters influence the vegetation community, at
the same time vegetation also influences its environment.
Vegetation increases elevation by adding dead biomass to the
peat and intercepting floating peat and litter. It reduces
standing water via transpiration, impedes water flow, and
provides fire fuel. Interactions between plants and the
environment drive vegetation succession (Loveless 1959, Vogl
Before drainage and impoundment, Water Conservation Area
1's waters were oligotrophic: nutrients such as nitrogen and
phosphorus entered the system via rainfall, and therefore were
limited. Low nutrient levels gave oligotrophy-adapted species
such as sawgrass a competitive advantage (Steward and Ornes
1975). Historically, nutrient levels had little influence on
the distribution of plant communities, except perhaps in the
vicinity of faunal activity sites such as alligator holes and
bird rookeries (Richardson et al. 1990). Cattail (Typha
domingensis) and willow (Salix caroliniana) grow at these sites
as a result of the nutrient concentrations or physical
disturbance caused by the animals. Since the construction of
the canals on the perimeter of Water Conservation Area 1 and the
resulting influx of nutrient-rich agricultural runoff, cattail
and willow head species have also become dominant near the
canals (Richardson et al. 1990).
The first theory of vegetation succession was put forward
by Frederick Clements, based on his knowledge of forests in the
Great Lakes region (Clements 1898). After observing the species
composition and structure of old growth forest and the regrowth
of cleared areas, Clements proposed that land in its natural
state passes through a succession of vegetation community types
over time (seral stages), until it reaches a climax stage. Each
seral stage, as it matures, alters the conditions of its
environment until the environment is more favorable to other
species. Colonization and conversion to the next seral stage
then follow. Early seral stages are composed of pioneer species
proficient at rapid reproduction and resource exploitation. As
time goes on, succeeding stages have more slower-growing,
resource-conserving species. Climax species are adapted for
equilibrium conditions. Each geographic region has its own
Clements' theory was widely accepted, but some later
ecologists, notably Henry A. Gleason, objected to the rigid
concept of a fixed seral progression and climax. They proposed
that species composition at a site was determined only by the
summation of a variety of stochastic factors, such as seed
dispersal, weather and fire history (Gleason 1926). Gleason
denied the existence of a linear set of seral stages leading to
a regional climax. Disturbance was a peripheral part of
Clements' theory, but as the field of ecology grew, ecosystems
were described in which disturbance played a critical role. In
these systems a regional climax was never achieved. Terms like
"fire climax" were coined to keep these systems under the
umbrella of succession theory.
The concept of succession is useful for understanding the
spatial pattern of vegetation in Water Conservation Area 1.
Vegetation community change in accordance with Clements' theory
can be observed in peat wetland systems (Vogl 1969). In Water
Conservation Area 1, water lilies in sloughs and alligator holes
build up peat, raising elevation and shortening hydroperiod
until the grasses and rushes of the wet prairie can invade.
Sawgrass builds up peat until the hydroperiod is short enough to
favor shrubs; shrubs alter the environment to allow the invasion
of trees (Gleason and Stone 1994, Loveless 1959). We do observe
a loosely predictable process of vegetation community change
after disturbance, and it is convenient to organize these
changes into types or seral stages (i.e. prairie, shrub,
forest). Based on the peat record and plant-environment
relationships, a successional sequence from slough to wet
prairie, to sawgrass, to shrubs and then trees has been
documented in the Everglades (see Figure 1-8) (Loveless 1959,
Pesnell and Brown 1977, Olmsted and Loope 1984, Duever 1984,
S : TREES
Figure 1-8. Successional sequence in Water Conservation Area 1
However, Clements' emphasis on a linear progression
through the seral sequence, ending in equilibrium at a stable
climax is not applicable to the Everglades ecosystem.
Disturbances and other factors perceived as stochastic
(droughts, floods, fires, alligator activity, seed dispersal and
nutrient concentration by birds) all influence plant community
distribution in the Everglades. For a young, disturbance-
adapted ecosystem like the Everglades, the concept of a stable
climax at equilibrium is not appropriate. Tropical hardwood
hammock forest has been identified as the climax community of
the Everglades (Olmsted and Loope 1984), but it is not found
anywhere in Water Conservation Area 1.
A more useful non-linear concept of succession and
disturbance is found in C.S. Holling's cyclical theory of
ecosystem dynamics (Holling 1992a). In this theory, disturbance
is part of a continuum with succession. An ecosystem progresses
from an exploitation phase (corresponding to early seral stages)
to a conservation phase (late seral stages), as in the
succession concept. However, while the system becomes more
stable and organized as it approaches climax, it also becomes
less resilient. Although more able to absorb small
perturbations, it is more vulnerable to a major disturbance.
Inevitably disturbance releases the stored energy and material
in the system and disorganizes it. The system goes through a
release phase and then a reorganization phase, from which it may
follow the same exploitation and conservation series it followed
before. If conditions have changed, however, it may follow a
different path. This last concept is especially applicable to
the Everglades ecosystem. As humans change the hydrology and
introduce exotic species to the Everglades, vegetation
communities have a greater potential to change in ways not
predicted by the our current understanding of succession in the
LANDSCAPE DATA COLLECTION
The foundation of an understanding of landscape pattern is
a thorough familiarity with the landscape. Many sources of
information were used in this work to gain knowledge of the
landscape of Water Conservation Area 1. This chapter will
describe the acquisition and processing of this information (see
Figure 2-1). Data sources included field observations made by
airboat during 1988-1991 and aerial photographs taken in 1992.
Aerial photographs from 1952 were compared with satellite images
from 1987 and 1990 to assess recent landscape changes.
Satellite images were studied in both their raw form and as
vegetation maps. Historical accounts of the vegetation of the
area were used as clues to the landscape in the past (Davis
1943, Thompson 1972). Literature on the natural history of the
Everglades, other Florida wetlands and the Okefenokee swamp also
was used to provide additional information on the species found
in the area (Dressler et al. 1987, Craighead 1971, Glasser 1986,
Kushlan 1990, Olmsted and Loope 1984, Pesnell and Brown 1977,
Wade et al. 1980).
Using the knowledge gained from the above information, a
map of Water Conservation Area 1 vegetation was made from a 1987
satellite image. The image processing required to create this
map is described in the latter part of this chapter. Vegetation
was computer-classified from the satellite data using field
observations, aerial photographs and a vegetation classification
of the same image by J.R. Richardson (Richardson et al. 1990).
Vegetation classes were based on a successional sequence and
will be referred to as "successional classes" to distinguish
them from the classes in the Richardson et al. classification.
The creation of this successional vegetation classification
required a synthesis of all the landscape and community ecology
information gathered in this work.
1952 Aerial 1990 Multispectral 1989 Raw
Photographs & 1990 Panchromatic Panchromatic
Satellite Images Image
1987 Multispectral Field Data and 1992 Aerial
& 1987 Panchromatic Ecological Literature Photographs
18-Class Vegetation i 24-Class Vegetation
:Map (J.R. Richardson) Succession Map
S= Data Source
= Classified Image
Figure 2-1. Diagram of landscape data sources
Because travel on the deep peat of Water Conservation Area
1 is difficult, few vegetation studies of the area exist, and
most are based on inadequately sized samples (Richardson et al.
1990 and Pope 1989 are exceptions). To gain a familiarity with
the landscape as a whole and with the vegetation at a finer
scale than remote sensing allows, observations of the vegetation
communities of Water Conservation Area 1 were made in the field.
Georeferenced observations also were made to be used in
classifying satellite imagery.
Drought conditions and a major fire interrupted and
delayed these observations, but these events also provided the
opportunity to observe fire effects and vegetation recovery.
Pre-fire Field Observations
Between 1988 and 1991, Water Conservation Area 1 was
explored on the ground by airboat. A Loran-C navigational unit
and paper prints of an April 1987 satellite image were used to
track the path of the airboat. The Loran-C unit used signals
from microwave stations to calculate latitude and longitude
coordinates within an error range of 100 m. The Loran-C
coordinates were recorded at different sites, along with the
vegetation types and estimates of the coverage of those types
within a 50 feet radius of the airboat.
Ground exploration began in the spring of 1988. At that
time, the water level in Water Conservation Area 1 was high.
Vegetation communities and corresponding Loran coordinates were
documented at numerous locations, and located on the 1987
satellite image. During 1988 however, drought conditions
lowered the water level in the area. Once the water level
dropped below the peat surface, airboat travel was limited to
the perimeter canals and areas near them. A high concentration
of alligators seeking water in the canals prompted the wildlife
refuge managers to close Water Conservation Area 1 to visitors.
Ground data collection became impossible.
Drought persisted through the winter of 1988-89, and a
freeze further heightened the fire potential by increasing the
dead fuel load. As described in the introductory chapter, on
April 11, 1989, at the end of the dry season, firebrands from a
controlled burn just outside the northwestern boundary of Water
Conservation Area 1 blew across the canal and started a large
wildfire. Over 18,000 ha (40,000 acres) burned in the northern
and western portions of the area (see Figure 2-2). The fire was
controlled by firefighters at the very northern tip of Water
Conservation Area 1 to prevent its spread to private property,
but was allowed to burn freely elsewhere (M. Maffei, personal
communication). Because access to the area was still
impossible, ground observation of the immediate effects of the
fire could not be made.
Figure 2-2. Map of the 1989 fire, with the location of the
successional vegetation map used in the model in Chapter 4
Post-fire Field Observations
Although the initial effects of the fire could not be
observed, vegetation recovery could be observed. The drought
did not end until 1991. However, in November 1989 the water
level temporarily rose enough in Water Conservation Area 1 to
allow limited airboat travel. The status of vegetation recovery
after the fire was observed from an airboat along a wide east-
west corridor running through the center of Water Conservation
Area 1 (varying between 260 30' and 260 32' latitude). Even
eight months after the fire, the boundaries and effects of the
fire were obvious. Recovery was limited, because there had not
been a drought-free growing season since the fire.
In February 1991, after vegetation recovery was well
underway, a larger percentage of the burned area was visited by
airboat. Three burned and three unburned tree islands were
explored on foot, measured and described in detail. A plan to
describe 18 burned and 18 unburned islands was not feasible and
had to be abandoned. Brushy resprouting trees, fallen dead
trees, and dense growth of the exotic Japanese climbing fern
(Lygodium japonicum) that had invaded burned islands made
traversing burned islands by foot extremely difficult and time-
The six islands described were selected at random using
pairs of latitude/longitude coordinates chosen at random within
the burned area, and randomly chosen cardinal directions.
Coordinates and cardinal directions were assigned before
sampling began. A sampling team traveled to a pair of ground
coordinates by airboat, then from there traveled in the assigned
cardinal direction until an island of the desired type was
reached. Two transects were set up on the island, one running
south to north, the other west to east. The samplers followed
these transects on foot, recording whether or not the vegetation
showed signs of burning. The distance along the transect was
recorded where the vegetation changed from burned to unburned
and vice versa. The numbers of tree individuals touching the
transect were recorded and tallied by: burned and living, burned
and dead, unburned and living, unburned and dead. It was assumed
that any burned trees that had not resprouted (22 months post-
fire) were dead.
While the information gathered on the ground after the
fire was limited, it was nonetheless essential as ground
training for the interpretation of aerial photographs taken
after the fire. Depictions of selected surveyed islands are
shown in Figure 2-3.
On the small islands visited in the unburned area of Water
Conservation Area 1, a central core of trees with an understory
of ferns was surrounded by a border of shrubs, sawgrass and
vines 3 to 4 m wide. On the small islands visited in the burned
area, all vegetation had been burned. However, all trees along
the survey transects were sprouting from their bases. A ring of
sawgrass less than 1 m wide surrounded the islands.
On the large burned island surveyed, the core of the
island, made up of trees with a fern understory, was not touched
by fire. The margins of this tree core were burned. The width
of this burned zone varied from 7 m to 77 m. The wider width
was found where small peninsulas of trees extended out from the
general ovoid shape of the island (see Figure 2-3). Seventeen
percent of the burned trees along the transects were dead, and
the rest were living. It was not possible to tell whether the
dead trees had been killed by the fire. Although wax myrtle
(Myrica cerifera) is not typically killed by fire, the wax
myrtle shrubs making up the southern "tail" of the island were
burned, with complete mortality. An extensive band of sawgrass
around the island had been burned, but had recovered with no
sign of mortality.
Aerial photographs were used as a link between the fine
scale of ground observations and the course resolution of
satellite images. When laid out in order with their edges
matching, the photographs provided a landscape view, while still
showing details lost to satellite images such as the texture of
vegetation and alligator trails. Photographs were used to
provide visual information on the landscape, to aid in the
identification of vegetation at different sites for satellite
classification, and to detect vegetation change. A set of 1952
photographs provided an approximation of vegetation before
impoundment and was compared with recent data to detect change.
Approx. scale: 1:1700 ..*
I Approx. direction of
N fire movement 240 m
* Trees, not burned Large burned island
Trees, burned Approx. scale: 1:8600
EH Shrubs, burned
Sawgrass: burned on large island
unburned, combined with shrubs, on small island
Figure 2-3. Map of three islands observed on the ground in
Photographs taken in 1991 and 1992 were taken to observe
disturbance and recovery from fire. The disadvantage of the
aerial photographs used was that they could not be
georeferenced. Due to the lack of man-made features in Water
Conservation Area 1 and the repetitive nature of the landscape
pattern, the matching of the photographs to map coordinates or
other data could not be verified. While an approximate scale of
the 1952 photography was known, the actual scale of each
photograph could not be calculated.
Aerial Photographs, 1952
A set of black and white aerial photographs of Water
Conservation Area 1 was taken in 1952. The aerial photograph
set contained 29 photographs at a scale of approximately
1:16,500, covering approximately 90% of Water Conservation Area
1. The L-7 canal on the northwest border of Water Conservation
Area 1 was still under construction at the time the photographs
were taken, but the remaining perimeter canals were complete.
The 1952 photographs were examined and visually compared
with a satellite image dated April 5, 1990 to detect vegetation
change. Obvious change between 1952 and 1990 was identified.
All islands larger than 1000 m2 in the photographs were searched
for on the satellite image to detect island degradation and
Twenty-five of the 1952 aerial photographs were of
sufficient quality to sample for tree islands. A quadrat 1 km2
was randomly placed on each photo, using a random number table
to assign x and y distances from the upper left corner of the
photo. Within the quadrat, tree islands were delineated and
assigned a category: small (less than 160 m2), medium to large
(between 160 and 2500 m2) and very large (greater than 2500 m2).
The results of the sample of tree islands in the 1952
photographs are listed in Table 2-1.
Table 2-1. Results of tree island sample, 1952 photography
Small islands 93.81
(less than 160 m )
Medium to large islands2 5.62
(between 160 and 2500 m )
Very large islands 0.48
(greater than 2500 m )
All islands 99.91
A careful comparison between the 1952 photographs and the
1990 satellite image showed that all tree islands greater than
1000 m2 in the 1952 photographs, could be identified by shape in
the 1990 image. In some cases the island's trees had been
replaced by other vegetation, or woody vegetation surrounded the
island, but the outline of the island was still visible.
Other vegetation communities were difficult to identify in
the photographs, due to the lack of color and the variation in
exposure between photographs. In particular, sawgrass (Cladium
jamaicense) could not be distinguished from cattail (Typha
domingensis), so the extent of cattail in 1952 could not be
estimated. (Cattail was found to increase in photoplots and
vegetation transects near canals in Water Conservation Area 1
between the 1960's and 1987 (Richardson et al. 1990, Davis et
Two types of vegetation change were apparent however. The
first of these was flooding-induced change. In the 1952
photographs, water was impounded north of the Hillsborough
canal, along the southern border of the future water
conservation area. This water created a linear pond devoid of
emergent vegetation. A network of wide sloughs extended north
from the pond approximately 1 km. The sloughs cut through
vegetation community patches, and were clearly a recent
phenomenon altering the landscape pattern. Tree island canopies
in this aquatic zone were difficult to delineate. This
phenomenon was documented later, in 1970 (Thompson 1970). The
satellite image from 1990 showed a similar aquatic zone, not
only in the south, but along all the canals surrounding Water
Conservation Area 1 (shown in Figure 2-4 as zones of change).
The second type of change was related to reduced flooding.
The 1990 satellite image shows dense concentrations of shrubs at
the northern tip and in the south central portion of Water
Conservation Area 1 (also part of the zones of change shown in
Figure 2-4). The dominant shrubs in these areas are wax myrtle
and buttonbush (Cephalanthus occidentalis). The same shrub
areas appear in a 1987 satellite image. However, neither of
these shrub communities are visible in the 1952 photographs.
They occurred on topographic high points (Richardson et al.
1990), and may be the products of fire frequency change or
shorter hydroperiods since the loss of sheet flow.
t". Z A..'U''
'iDavis et al. plots
Zones of change
SRichardson et al.
',i",- *, *'. .
!? .i; i
'l ,:' i
Figure 2-4. Areas of vegetation change in Water Conservation
Area 1 between 1952 and 1990, and locations of photoplots from
two vegetation studies
Sources: Richardson et al. 1990 and Davis et al. 1992
The 1952 photographs were analyzed visually for change, in
a general, non-quantitative fashion. A conscious decision was
made to avoid standard photogrammetric change detection methods,
based on several factors. First, the precise matching or
registration of photographs of the same site from different
years is critical for change detection. The interior of Water
Conservation Area 1 is devoid of man-made feature or permanent
geographic landmarks such as mountain peaks. The vegetation,
including tree islands, is in flux. Therefore, accurate
registration of photography from different years is not
possible. The problem is compounded by the different scales of
available photography and imagery.
Second, the consistent delineation of vegetation polygons
from remotely sensed data is difficult in Water Conservation
Area 1. The apparent boundaries of vegetation communities are
strongly influenced by water level. Patches of emergent
vegetation appear to shrink when water is high. Water level in
Water Conservation Area 1 fluctuates constantly, introducing
interpretation differences between seasons or years. In
addition, the available types of remotely-sensed data from
different years may vary from black and white photographs, to
color infrared photographs, to satellite imagery. These
different data types may yield different interpretations due to
their different spectral qualities.
Finally, the photo-interpretation of the entire Water
Conservation Area would be too labor-intensive to undertake in
the scope of this work. A sampling of the landscape by
photographic plots could have been done to reduce the labor of
interpretation. Change detection photo-sampling compares
detailed interpretations of vegetation on sample plots using
photographs or satellite images over time (Silveira 1996).
Interpretations of vegetation communities are usually recorded
as overlay maps. Areas changed between dates on each plot are
calculated by comparing overlay maps. The results are often
used to draw conclusions regarding vegetation change in the
ecosystem as a whole. However, important landscape changes may
be missed when an area is sampled rather than examined in its
entirety, particularly if the sample size is small.
The oversight of landscape change was realized in previous
photoplot studies of Water Conservation Area 1 (Thompson 1972,
Richardson et al. 1990, Alexander and Crook 1975, Davis et al.
1994). The studies failed to detect the large area of sawgrass
that changed to brush in the south-central part of Water
Conservation Area 1, because no plots were located in this
region. In addition, these studies found conflicting trends in
the wet prairie and sawgrass communities in central Water
Conservation Area 1.
Thompson (1972) estimated change in vegetation communities
in Water Conservation Area 1 between 1948 and 1968 using a time
series of 7 40.5-ha photoplots. Richardson et al. (1990)
compared these same plots with a 1987 classified satellite image
(see Figure 2-4). Alexander and Crook (1975) estimated
vegetation change between 1940-53 and 1965-71 on 100 259.2-ha
photoplots throughout the Everglades, including 5 plots in Water
Conservation Area 1. Davis et al. (1994) compared 25 of these
plots with 1987-1987 photographs, including all the plots in
Water Conservation Area 1 (see Figure 2-4).
Thompson (1972) and Richardson et al. (1990) found small
vegetation community changes in their 7 photoplots, except in 2
central plots. In plot 4, 25% of the plot changed from sawgrass
to wet prairie between the years 1962 and 1968. By 1987
however, sawgrass and wet prairie had returned to their 1962
distribution. In plot 3, a similar switch from sawgrass to wet
prairie dominance occurred between 1968 and 1987 (see Table 2-
2). Richardson et al. hypothesized that the changes found in
plot 4 between 1962 and 1968 were due to fire. A large fire
burned in Water Conservation Area 1 in 1962 (see Table 1-1).
The later changes in plot 4 may have been post-fire vegetation
Davis et al. (1994) analyzed a plot in the center of Water
Conservation Area 1 (plot 75) that overlapped the above-
mentioned plot 3. The changes in plot 75 are shown in Table 2-
2, and are the opposite of trends found by Richardson et al. in
Table 2-2. Percent area in sawgrass and wet prairie of two
photographic sample plots in Water Conservation Area 1
Plot 75 Plot 75 Plot 3 Plot 3
Year Sawgrass Wet Prairie Sawgrass Wet Prairie
1968 0.0% 89.6% 58.0% 21.0%
1984/87 37.7% 47.3% 5.7% 69.8%
Source: Richardson et al. 1990:134 and Davis et al. 1994:441
Two factors may account for the discrepancy between the
results of the Richardson et al. and Davis et al. studies. One
factor is the interpretation of the 1968 photography by Thompson
(1972) and Alexander and Crook (1975). Alexander and Crook
found no sawgrass in plot 75 (see Table 2-2), while Thompson
interpreted 58% of his plot as sawgrass. Since plot 75
contained plot 3, it should have held some sawgrass if
Thompson's interpretation of plot 3 was correct. The water
level in Water Conservation Area 1 in 1968 was high, and the
area was flooded nearly all year (Thompson 1972). When the
lower portions of sawgrass plants are submerged, sawgrass and
wet prairie appear similar in aerial photographs. The
difference in plot composition must have been due to the
judgment of the photo-interpreters.
The other factor contributing to the conflicting trends is
the interpretation of the plots at the later dates. Some
differences between these interpretations are expected, because
Richardson et al. used satellite imagery instead of photography,
the interpretations are separated by 3 years, and rainfall
varied between these years. If a fire burned in the area of the
plots during this interval, then a real change in vegetation
composition could have occurred.
Based upon the data from their plots in Water Conservation
Area 1 and 9 other plots in the Everglades, Davis et al.
concluded that a significant increase in sawgrass and a
significant decrease in wet prairie had occurred throughout the
Everglades' "'wet prairie/slough-sawgrass-tree island mosaic,"
(Davis et al. 1994:432). However, the different results of
photoplot studies in Water Conservation Area 1 illustrate the
limitations of using a small sample of photographic
interpretations to characterize vegetation community change
within an Everglades landscape.
Aerial Photographs, 1991-1992
A satellite image was used to assess the effects of the
1989 fire on the landscape of Water Conservation Area 1. The
first post-fire cloud-free image available for this purpose was
dated April 1990. To assess the progress of vegetation recovery
after 1990, a low-cost landscape overview was needed.
Photography was preferred to distinguish resprouting trees from
shrubs. It was not necessary for the photographs to be
vertical; in fact, oblique angles were useful for vegetation
identification. Because professional aerial photography was too
costly, a photographic survey was made using a 35 mm. camera
held out the window of a small plane.
On December 1991, 20 months after the 1989 fire, an aerial
photographic survey of Water Conservation Area 1 was conducted.
Photographs were taken from an Cessna 172 airplane, using a
hand-held 35 mm camera. Both near-vertical and oblique color
photographs were taken. The survey flight was part of a wading
bird survey (H. Jelks, unpublished data), but was also well
suited for a landscape inventory. East-west transects were
flown on two consecutive mornings, at a constant speed and 305 m
altitude. The transects followed known coordinates, spaced 0.5
minutes latitude apart. Photographs were taken from the
aircraft at regular intervals, Loran coordinates were noted for
distinctive landmark features, and additional notes were taken
between transects. Weather conditions were clear.
On March 24, 1992, a return flight was made to document
the continued recovery of tree islands after the fire. The time
of day, aircraft, and method of photographing were the same as
the first survey. Rather than following a systematic flight
path however, individual islands were revisited and photographed
at altitudes ranging from 300 to 750 m.
Three hundred and eighty-eight photographs were taken
during the 1991 survey. The quality of the photographs was
good, and the lighting and weather conditions were excellent for
recording the texture of the vegetation. In addition to showing
the condition of the vegetation, the photographs were useful in
identifying vegetation types for the classification of a 1987
satellite image, as described in the next section.
During the second survey 60 photographs were taken. The
weather was hazy, so the color quality of the photographs was
poor. The purpose of the survey was to view the extent of
resprouting by the tree stumps and snags left by the fire. The
lack of color definition, combined with the dense growth of the
exotic Japanese climbing fern (Lygodium japonicum) over tree
snags prevented this.
Five satellite images of Water Conservation Area 1 were
used for a variety of purposes in this work. The images were
used in their raw form for visual analysis in the same way
aerial photographs might be used: as navigation aids, to
identify vegetation, and to identify the extent and behavior of
the 1989 fire. Images also were classified into two vegetation
maps. The maps were used in a geographic information system
(GIS) to analyze tree island distributions, (see Chapter 3), and
to model landscape dynamics, (see Chapter 4). The satellite
data were processed on the computer to make the most of their
spatial and spectral resolution and to create the vegetation
maps. This image processing is described in detail below.
SPOT Satellite Image Information
The satellite data were acquired from the French SPOT
Image Corporation. SPOT Image Corporation currently has two
satellites deployed in sun-synchronous orbits 823 km above the
earth. The swath width covered by the satellite sensors is 60
km, so each satellite image covers a parallelogram 60 km on a
The SPOT satellites have two sensors: one panchromatic and
one multispectral. The panchromatic sensor measures reflectance
of visible light in the green to red portion of the spectrum
(.51-.73 micrometers). It has a 10 m pixel resolution, so it
senses and records one panchromatic reflectance value for each
10-m by 10-m square of earth it scans. The multispectral
sensor has a 20 m pixel resolution, and collects three data
values for each pixel. 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).
The images used in this work are listed in Table 2-3.
Table 2-3. SPOT satellite images of Water Conservation Area 1
Scene Date Image Type SPOT Scene ID Number
April 4, 1987 multispectral 12X623297870404155912
April 4, 1987 panchromatic 11P623297870404155910
April 11, 1989 panchromatic 11P623297890411155903
April 5, 1990 multispectral 11X623297900405160335
April 5, 1990 panchromatic 12P623297900405160333
A SPOT image dating from April 11, 1989 was taken during
the severe fire that burned over 18,000 ha of Water Conservation
Area 1. Smoke obscured part of Water Conservation Area 1, and
created spectral variation in the image data, preventing
vegetation interpretation or classification of the multispectral
image. The panchromatic image was nonetheless useful for visual
analysis. It was used as a documentation of fire behavior
during the 1989 fire, particularly as it affected tree islands.
The multispectral SPOT image dating from April 4, 1987 was
merged with the panchromatic image from the same date to improve
the utility of the imagery. Likewise, the April 5, 1990
multispectral image was merged with the panchromatic image from
April 5, 1990. The result of merging the panchromatic and
multispectral images was an enhanced data set with the 3-band
spectral resolution and 10-m spatial resolution.
The 1987 merged SPOT image was used in several ways. The
image was rectified to State Plane Coordinates, and classified
into an 18-class vegetation map by J. R. Richardson, as part of
an evaluation of vegetation-water relationships in Water
Conservation Area 1 (Richardson et al. 1990). Paper printouts
of this vegetation map were used to navigate Water Conservation
Area 1 by airboat.
Both the unclassified 1987 merged image and the vegetation
map were used to gain familiarity with the appearance of the
landscape in the imagery medium and to identify the spectral
signatures of different vegetation communities. The sites
visited by airboat in Water Conservation Area 1 were located on
the vegetation map and the unclassified merged image using the
Loran-C coordinates of the sites. The spectral characteristics
of the site were compared with field notes taken at the site.
The vegetation map was also used to analyze the size
distribution of tree islands in Water Conservation Area 1, using
a Geographic Information System (see Chapter 3).
In a separate procedure, the multispectral SPOT image from
April 4, 1987 was merged with the panchromatic image from the
same date, and a subset of the merged image was classified into
24 classes based on vegetation type and successional stage (see
Table 2-7). The Richardson et al. vegetation map, the 1992
aerial photographs, field notes and ecological information were
used to define the 24 classes (see Figure 2-1). One copy of
this successional class map was rectified to State Plane
coordinates, and a second copy was rectified to Universal
Transverse Mercator (UTM) coordinates. This allowed the
successional class map to be referenced to data in both State
Plane and UTM coordinates. The successional class map was used
for modeling landscape change in response to fire (see Chapter
4), and was compared with the 1990 SPOT images.
The 1990 SPOT images were merged and rectified to UTM
coordinates. The 1990 merged image was compared with 1952
aerial photographs as described above in this chapter. It was
also used to assess the performance of the landscape model by
providing an example of post-fire vegetation recovery (see
The image processing techniques used in this work included
image merging, vegetation index calculation, rectification and
classification. All image processing tasks and GIS analyses in
this work were performed using ERDAS Inc. computer software,
versions 7.4 and 7.5 (ERDAS 1991). The image merging and
vegetation index calculations were performed similarly for the
1987 panchromatic-multispectral image pair and the 1990
panchromatic-multispectral image pair, and so are described only
Image merging was a two-step process. The first step was
the resampling of the multispectral data at 10-m resolution by
dividing the 20-m pixels into quarters. The resampled
multispectral data were then registered to the panchromatic
data. Once the two images were matched together with 10-m
resolution, the second stage of the merging process was an
incorporation of the panchromatic data into the multispectral
data through a color transformation procedure. A normalized
difference vegetation index, (NDVI), was also calculated from
the multispectral image, and added to the merged image as a
fourth data band.
Each pixel in a SPOT image 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 the multispectral
data reassigned these spatial coordinates, as shown in Figure 2-
5. This increased the size of the data file by a factor of 4.
The same 20-meter pixel
A 20-meter pixel resampled to 10-meter pixels
Figure 2-5. Pixel coordinate values before and after resampling
The multispectral data were resampled at the same time the
multispectral data were registered to the panchromatic data.
Registration is the process of matching the x and y coordinates
column (x) = 1
row (y) =1
data (d)= 5
of the multispectral and panchromatic images. Twenty-five
corresponding points were identified on the two images (control
points). These points were man-made features that could be
clearly identified on both images. For each control point, x
and y coordinates on each of the images were recorded and paired
together. The pairs of coordinates for all the control points
were run through a computer algorithm that identified the mean
geometric offset of the points, and created an algebraic matrix
to transform the coordinates of the multispectral pixels to the
coordinates of the panchromatic.
The computer program that generated the transformation
matrix also calculated 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. The transformation matrix for both the 1987 and 1990
image pairs had a root mean square distance less than one half
of a pixel (5 m).
While the multispectral pixels were spatially redefined
during the resampling and registering process, the multispectral
data values were not altered. "Nearest neighbor" resampling was
used, so the spectral data values of the original pixels were
simply reassigned to the nearest spatially redefined pixel.
Once the two images were matched together with 10-m
resolution, the panchromatic data were incorporated into the
multispectral data via a color transformation procedure
(Richardson et al. 1990).
By convention, the data values in each of the three SPOT
multispectral image bands are designated as color values of red,
green and blue to emulate a color infrared photograph for
display purposes. Multispectral values range from 0 to 255.
Band 1 data are assigned as blue values, band 2 as green, and
band 3 as red. When the red, green and blue color values of
each pixel's three data bands are displayed on a monitor, one of
16,777,216 possible colors appears. This color can be plotted
in 3-dimensional color space, with axes for red, green and blue
and values ranging from 0 to 255. In addition to red-green-blue
(RGB) color space, this pixel color can 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 values in RGB colorspace were transformed to
IHS values. Next the intensity value for each multispectral
pixel was deleted and replaced by the value found at that pixel
location on the panchromatic image. Finally, the IHS
coordinates of the pixels were converted back to RGB colorspace.
The band 3 (red), band 2 (green), and band 1 (blue) values had
been altered by the inclusion of the panchromatic information.
This process, using the spectral values of a sample pixel, is
shown in Figure 2-6.
The advantage of a merged image is the addition of data
with higher spatial resolution to the multispectral data. It is
not possible to increase the spatial resolution of multispectral
data itself. The addition of panchromatic data to the
multispectral image greatly enhances the definition of small
vegetation patches, ponds, sloughs, roads, and buildings.
i- r- i Pixel
Figure 2-6. Diagram of merging of images using color
The spectral resolution of the multispectral data was not
increased by the merging. The panchromatic sensor records data
in a single band that lumps together reflectance from the green
and red portions of the spectrum already sensed by bands 1 and 2
of the multispectral sensor. Some spectral information was
actually lost; the intensity values from the infrared portion of
the spectrum (band 3) were replaced by panchromatic data that
did not include infrared reflectance. However, the infrared hue
P = panchromatic brightness
R = red (band 3)
G = green (band 2)
B = blue (band 1)
I = intensity
H = hue
S = saturation
and saturation data which contain the majority of information
describing infrared reflectance were retained.
While all band values decreased as a result of the
transformation process (see Figure 2-7), they decreased
consistently. The relationships between the band values changed
little, as shown by the band ratios in Figure 2-8.
A normalized difference vegetation index, or NDVI, is a
ratio of the red and infrared bands which highlights vegetation
with high biomass and/or photosynthetic productivity (Jensen
1986). 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
To avoid values of 0, 0.5 was added to the denominator, and a
multiplier of 100 was used as a scaling factor to put the NDVI
in the same range as the other bands.
The NDVI band was used in addition to the 3 other spectral
bands, and its calculation did not affect the data in the other
Panchromatic values (band 1)
of pixels in mesic forest
66 66 64 64
17 17 18 18
30 30 30 30
66 66 64 64
17 17 18 18
30 30 30 30
67 67 67 67
17 17 17 17
31 31 30 30
67 67 67 67
17 17 17 17
31 31 30 30
Multispectral red, green and blue values (bands 3, 2, and 1) of
pixels registered to same the locations
green anci blue values
(bands 3, 2, and 1)
Figure 2-7. Sample pixel values before and after the
24 25 24 24
24 26 26 23
26 27 27 25
26 27 27 26
38 39 37 37
9 10 10 10
17 18 17 17
38 41 40 35
9 10 11 10
17 18 19 16
41 43 43 39
10 10 10 10
19 19 19 17
41 43 43 41
10 10 10 10
19 19 19 18
Multispectral band ratios before and
ege at on n ex (NDVI) c e
.Lf. VILL tIIlU-L.
:ispectral data before and after merging
Figure 2-8. Band ratios before and after the image transformation process
a) Multispectral band ratios for pixels in Figure 2-7 before and after merging
b) Vegetation index (NDVI) calculated from multispectral data for pixels in Figure 2-7
before and after merging
Before/after Before/after Before/after Before/after
Band 3/band 2 0.24 0.26 0.26 0.26 0.28 0.27 0.28 0.27
Band 2/band 1 0.56 0.52 0.56 0.55 0.60 0.59 0.60 0.59
Band 1/band 3 0.45 0.44 0.45 0.46 0.46 0.44 0.46 0.44
Band 3/band 2 0.26 0.24 0.26 0.24 0.28 0.27 0.28 0.29
Band 2/band 1 0.56 0.52 0.56 0.55 0.60 0.58 0.60 0.62
Band 1/band 3 0.45 0.44 0.45 0.44 0.46 0.47 0.46 0.46
Band 3/band 2 0.25 0.24 0.25 0.23 0.25 0.23 0.25 0.26
Band 2/band 1 0.55 0.53 0.55 0.53 0.57 0.53 0.57 0.59
band 1/band 3 0.46 0.46 0.46 0.44 0.45 0.44 0.45 0.44
Band 3/band 2 0.25 0.24 0.25 0.23 0.25 0.23 0.25 0.24
Band 2/band 1 0.55 0.53 0.55 0.53 0.57 0.53 0.57 0.55
Band 1/band 3 0.46 0.46 0.46 0.44 0.45 0.44 0.45 0.44
Before merging 59 59 56 56
After merging 61 59 57 57
Before merging 59 59 56 56
After merging 61 60 56 68
Before merging 59 59 59 59
After merging 60 62 62 59
Before merging 59 59 59 59
After merging 60 62 62 60
V17 4 7 i1 1 t4-
Rectification of a satellite image is the process of
linking a map coordinate system to the pixels of the image. The
1990 merged image was rectified to Universal Transverse Mercator
(UTM) coordinates without altering the pixel size. The
vegetation map made from the 1987 merged image of Water
Conservation Area 1 (Richardson et al. 1990) was rectified to
State Plane coordinates. During the rectification, the pixel
size was redefined from 10 m to 30 ft. The unclassified 1987
merged image was rectified to UTM coordinates by registering it
to the rectified 1990 merged image, using the procedure
described in the Registration section above. A copy of the 1987
merged image was also rectified to State Plane coordinates using
the same algebraic transformation matrix used by Richardson et
al. to rectify the vegetation map. This allowed the
unclassified 1987 merged image to be compared on a pixel-by-
pixel basis with both the 1990 merged image and Richardson et
al. 1987 vegetation map.
To rectify the 1990 merged image, 25 to 30 control points
were used. These points were man-made features, such as canals,
road intersections and bridges, which could be clearly
identified on both the image and 1:24,000 scale USGS topographic
maps. The control points were distributed as evenly as possible
throughout the image. There were no suitable points in the
interior of Water Conservation Area 1, but many points were
located along the perimeter canals.
To identify the coordinates of control points, the
locations of the image control points were digitized on the
computer monitor. The locations of the map control points were
digitized from a digitizing table. To achieve this, a portion
of the image was displayed on the monitor while a corresponding
map was "locked down" to the digitizing table by digitizing
several UTM tic marks along the edge of the map. The accuracy
of the lock-down was tested by digitizing points with known
coordinates and comparing the digitizer's estimate of the
points' coordinates with their actual coordinates. If the
digitized coordinates varied more than 10 m from the actual
coordinates, the map lock-down procedure was repeated. If
repeated lock-downs did not reduce the discrepancy to less than
10 m, the map was assumed to be distorted and was not used.
Thirty control points were digitized on the image and the
map to create a set of corresponding image file coordinates and
map coordinates, in the same way that the panchromatic and
multispectral images were registered together. The pairs of
coordinates for all the control points were run through a
computer algorithm which identified the mean geometric offset of
the points, and created an algebraic matrix to rectify the image
to UTM coordinates. The program that generated the
transformation matrix also calculated 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. The error contributed by each control
point was also identified. If the root mean square distance was
greater than 1 pixel (10 m), the control point contributing the
greatest error was discarded, and the transformation matrix was
recalculated. This process was repeated until the root mean
square distance was less than 1 pixel.
The 1990 merged image was rectified using a linear
rectification, i.e. the algebraic transformation matrix was a
linear function. Using the transformation matrix, the image
file coordinates were converted to map coordinates. Spectral
values were not altered, but were assigned to the nearest
It was unnecessary to adjust for elevation in this
rectification, due to the height of the satellite and the flat
terrain. The SPOT satellite has a fixed array of sensors
aligned in the x direction, and the data are corrected for the
curvature of the earth. Therefore, SPOT images do not have the
displacement and distortion found in aerial photographs.
Distortion in the image can occur in the y direction, if the
speed of the satellite varies while the image is being sensed.
A rectification using control point coordinates identified
from a ground survey or Global Positioning System is preferable
to one using control points taken from a map. However, the
accuracy of a map-based rectification is appropriate to the 10-m
resolution of the data when the control points are numerous,
well distributed and checked for error.
During satellite image classifications, pixels are
assigned to classes based upon their spectral values.
Classifications may be unsupervised, supervised, or a
combination of both. In an unsupervised classification, the
classes to which pixels are assigned are derived from the
spectral attributes of the data and are defined by a clustering
algorithm. Only the number of classes in the classification are
specified beforehand. The pixels are assigned to these classes,
and the analyst studies the locations of the different classes
on the image and identifies their vegetation or land cover type.
In a supervised classification, the analyst first defines
known areas of vegetation on the image. Statistics on the
spectral data of the pixels in each area are used to define
spectral signatures for each class. An algorithm such as
maximum likelihood is used to assign the image pixels to the
classes (ERDAS 1991, Jensen 1986).
The Richardson et al. 18-class 1987 vegetation map was
classified using a supervised procedure in which approximately
25 classes were initially defined, and then narrowed down to 18
classes after testing and evaluation (Richardson et al. 1990).
The classification process gave special attention to the
identification of cattail (Typha domingensis) and other
vegetation in the areas of greatest change.
To create a map suitable for modeling the least-changed
vegetation of Water Conservation Area 1, a successional class
map was classified from a subset of the 1987 merged image using
an unsupervised procedure. The unsupervised classification
focused on identifying vegetation in the interior of Water
Conservation Area 1, and on defining classes which elucidated
the transitions between vegetation communities.
A subset was extracted from the 1987 merged image which
contained a 5000 ha portion of the interior of Water
Conservation Area 1 (see Figure 2-2). This subset was run
through an unsupervised clustering routine to produce a 28-class
classification (ERDAS 1991). The clustering was based on the
values in the four spectral data bands of each pixel: green
reflectance (band 1), red reflectance (band 2), near infrared
reflectance (band 3) and the NDVI, (a vegetation index of the
relationship between bands 2 and 3). The main factors
influencing spectral reflectance in Water Conservation Area 1
are the vegetation species, the density of vegetation, and the
amount of water present. Conveniently, these factors also
define vegetation communities and the gradations between them.
It was expected, for example, that the community of wet prairie
species would be divided between several classes separated by
their proportions of open water, wet prairie vegetation, and
sawgrass. These classes represent stages along a successional
progression from slough to wet prairie to sawgrass.
The 28 unsupervised classes had to be identified and
ranked into a seral progression. This was accomplished by
comparing the classes with gecreferenced ground observations,
with the aerial photographs of Water Conservation Area 1 taken
in December 1991, and with the locations of classes in the 1987
The 18-class 1987 vegetation map was made as part of an
effort to investigate relationships between water quality and
quantity and vegetation change in Water Conservation Area
l(Richardson et al. 1990). The study required detailed mapping
of vegetation near the canals, particularly sawgrass and
cattail. However, to model the landscape of the interior of the
Water Conservation Area 1, a map of the canal zone was
unnecessary. A map showing different densities of wet prairie,
sawgrass, brush and trees was needed instead. The 1987
vegetation map was unsuitable for this purpose, and thus for use
in the landscape model. However it was a valuable tool used in
identifying the classes of the unsupervised classification.
Seven of the 18 classes in the 1987 vegetation map did not
appear in the 5000-ha area of Water Conservation Area 1 used in
the unsupervised classification (see Table 2-4). These classes
were not used in this task. The 11 classes found in the 5000 ha
subset are described in Table 2-5.
The unsupervised classification was cross-referenced with
the 1987 vegetation map using ERDAS software's SUMMARY program
(ERDAS 1991). For each unsupervised class, the program
summarized what the pixels in that class were classified as on
the 1987 vegetation map. By analyzing the locations of the
unsupervised classes, field notes, the aerial photography, and a
nearest neighbor analysis of the vegetation map (Richardson et
al. 1990:46), a set of rules for identifying the unsupervised
classes based on the summary results was defined. The rules
specified the successional class of the unsupervised classes,
based on the percentage of each class in the vegetation map
cross-referenced to it (see Table 2-6).
The 28 classes of the unsupervised classification were
identified according to the rules, renamed, and ranked along a
successional continuum (see Table 2-7). This created a
successional class map suitable for use as a base map for the
spatial model of landscape pattern (see Chapter 4).
Table 2-4. Classes in the 1987 vegetation map of Water
Conservation Area 1 found in less than 1% of the successional
class map area.
1 High density sawgrass with some fern tussocks
2 Sawgrass with invasion of cattail near canal
10 Cattail community close to canal
11 Open deep water mostly along Hillsborough canal
12 Slough deeper or less vegetated than class 9
15 Sawgrass with invasion of wax myrtle close to canal
18 Cattail community
Richardson et al. 1990:40-41
Table 2-5. Description of classes in the 1987 vegetation map of
Water Conservation Area 1 found in the successional class map
3 Sawgrass- ...primary sawgrass class occurring on
all parts of the refuge including the vast
sawgrass areas on the west side of the refuge
4 Brush/sawgrass- ...primarily sawgrass with large
amounts of wax myrtle. Some tree islands which may
have been burned out previously are made up
entirely of the class, particularly in the
southern part of the refuge
5 Tree island- ...lower stature tree island
community made up of a mix of wax myrtle, dahoon
holly and red bay. Occurs along edges of large
tree islands with some small islands composed
entirely of this class
6 Wet prairie- ...largest area of the refuge. This
class is the denser wet prairies occurring over
all of the refuge but the primary community type
of the central portion of the refuge. This class
may contain small tree islands smaller than 30
feet across and areas with small sawgrass strands
7 Tree island- ...core of larger tree islands,
larger stature trees made up primarily of dahoon
holly and red bay with lesser amounts of way
myrtle than class 5
8 Brush- ...smaller brush clumps primarily in wet
9 Wet prairie- ...sparser wet prairie community
often with sparse sawgrass
13 Willow/brush- ...predominantly willow but with
some mixed classification whit wax myrtle brush
14 Brush/tree island- ...class of many of the smaller
tree islands with mostly wax myrtle, some
sawgrass, occasionally dahoon holly and red bay
16 Sawgrass- ...slightly higher elevation sawgrass
than class 3. Core of sawgrass ridges
17 Willow-...willow along canal edge, some
misclassified floating aquatics along Hillsborough
Source: Richardson et al. 1990:40-41
xml version 1.0 encoding UTF-8
REPORT xmlns http:www.fcla.edudlsmddaitss xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.fcla.edudlsmddaitssdaitssReport.xsd
INGEST IEID ECA6528N1_6AWO5D INGEST_TIME 2013-03-21T13:50:17Z PACKAGE AA00013550_00001
AGREEMENT_INFO ACCOUNT UF PROJECT UFDC