Landscape dynamics in the Everglades


Material Information

Landscape dynamics in the Everglades vegetation pattern and disturbance in Water Conservation Area 1
Physical Description:
xvi, 177 leaves : ill. ; 29 cm.
Silveira, Jennifer Enos, 1958-
Publication Date:


Subjects / Keywords:
Wetland ecology -- Florida -- Everglades   ( lcsh )
Conservation of natural resources -- Florida -- Everglades   ( lcsh )
Wildlife Ecology and Conservation thesis, Ph. D
Dissertations, Academic -- Wildlife Ecology and Conservation -- UF
bibliography   ( marcgt )
non-fiction   ( marcgt )


Thesis (Ph. D.)--University of Florida, 1996.
Includes bibliographical references (leaves 168-176).
Statement of Responsibility:
by Jennifer Enos Silveira.
General Note:
General Note:

Record Information

Source Institution:
University of Florida
Rights Management:
All applicable rights reserved by the source institution and holding location.
Resource Identifier:
aleph - 023835822
oclc - 35815494
System ID:

This item is only available as the following downloads:

Full Text








Copyright 1996


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

larger context.

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

doctoral defense.

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

moral support.


ACKNOWLEDGMENTS ...............................................iv

LIST OF TABLES .....................................

........... ix

LIST OF FIGURES ...............................................xi

INTRODUCTION ...............................
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 ...........................
Vegetation community types .........
Environmental factors in vegetation
distribution ....................
Hydrologic factors .................
Nutrient concentrations ............
Vegetation Succession ................

Introduction ...... ........................
Ground Observations .....................
Pre-fire Field Observations ..........
Methods ............................
Post-fire Field Observations .........
Methods ............................

.................... 1
.................... 8
.................... 9
................... 11
1 ................. 13
1 ................ 14
ea 1 ...............18
................... 18
Area 1 ...........25
................... 31
............ .... 38
................... 38
................... 40
S. ................42

S. ...............52



Results ............................
Aerial Photographs ......................
Aerial Photographs, 1952 .............
Methods ............................
Results ............................
Discussion .........................
Aerial Photographs, 1991-1992 ........
Methods ............................
Results ............................
Satellite Images ........................
SPOT Satellite Image Information .....
Image Merging ........................
Resampling .........................
Image transformation ...............
Vegetation index ...................
Rectification ........................
Image Classification .................
Unsupervised Classification ........
Class Identifications ..............

Introduction .................... .......
GIS Analysis of Tree Island Distribution
Analysis of Size Distribution ........
Methods ............................
Results ............................
Discussion .........................
Island Population Model Analysis ........
Methods ............................
Results ............................
Discussion .........................

LANDSCAPE MODEL ..........................................
Introduction ..................... ....................
Conceptual Model of Tree Island Growth and Landscape
Pattern ............................................
Processes Contributing to Island Growth ............
Vegetation succession ............................
Peat accumulation ................................
Fire .............................................
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 .

..... 113
..... 113

.... 113
.... 115
..... 115
.... 117
.... 117
.... 119
.... .120
.... .121
.... 122
.... 123
.... 123




..... 53
..... 54
..... 56
..... 56
..... 57
..... 59
..... 64
..... 64
..... 65
..... 66
..... 66
..... 69
..... 70
..... 71
..... 74
..... 77
..... 80
..... 81
..... 81

Methods ................................
Results ................................
Tree islands in the spatial model ........
Comparison of Model with Independent Data
Model validation .......................
Comparison of model and satellite data.
Discussion ...............................


SUMMARY AND CONCLUSIONS .... ............................ .... 159

LIST OF REFERENCES ........................................... 168

BIOGRAPHICAL SKETCH... .........................................178




Table page

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


Figure page

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



Jennifer Enos Silveira

August, 1996

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^

I'9. .'A
S. ', *~^ *': i o,.., o /^ ..:- /-" '
, ,,, ^: ,, :l L ,' ^

..y~.. lS.^:

k. I
.. ) .. .' s'- *, "- ,'
"" ; ; ,.* ,' "" ;. : ..A. i .' .
.* "> *'* 4 ".
V *' *1"**, ...-.

^ ;'9 ^'.9 '-.*9 9 *!!,^ 1^ ^
.4v -,. '4, y. -. ;
i ,. ,, ~ ~ ~.: "jo ../.-! ,'

Figure 1-1.

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

prohibitively expensive.

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
Ecological Literature

:Map (J.R. Richardson)

GIS Analysis of
Spatial Patterns

Formulation 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

of islands?

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

landscape dynamics.

The Study of Landscape Pattern Dynamics

Landscape Ecology

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

(Kessel 1979).

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

these processes.

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.

Spatial models

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 Everglades.

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

Kissimmee River
and Watershed

Fisheating Creek

Location of Water
Conservation Area 1

Big Cypress

The Everglades

Mangroves and
Coastal Marshes

Figure 1-4. Map of the historic Everglades system

20 mi

32 km

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).




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

al. 1990).

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

Agricultural Area.

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)

- Stage

---.iii -t17

NF M// I7Y ///1 A S O

*/ /*11


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

an island.

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

low intensity.

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).

Fire severity

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

Urban, 1980).

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

1955 4,047

1962 > 40,470

1981 2,631

1989 18,211

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 *
;. *. : **

Figure 1-7.
Area 1

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

structuring communities.

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
Cephalanthus Buttonbush
Willow Head Salix caroliniana Willow
Ludwigia octovalis Water Primrose
Mikania scandens Climbing Hemp-vine
Polygonum Smartweed
hydropiperoi des
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
al. 1994

Table 1-2--continued
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
Bacillophyceae and
Chlorophyceae groups
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
al. 1994

1988 (unpublished
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).

Hydrologic factors

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

Crook 1974:63).

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


Nutrient concentrations

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).

Vegetation Succession

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

climax community.

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,

Kushlan 1990).

I iI



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

Everglades landscape.



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
Satellite Images


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

Ground Observations

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

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.


36 m

21 m
Small unburne

25 m

Small burned

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
February, 1991

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
Island density
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

al. 1994)).

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.

2 1


',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

plot 3.

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.

Satellite Images

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

Chapter 4).

Image Merging

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

once below.

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

x=1 x=2
y=l y=l
d=5 d=5

x=l x=2
y=2 y=2

d=5 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.

Image transformation

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
transformation process

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.

Vegetation index

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
after merging

(bands 3, 2, and 1)

Figure 2-7. Sample pixel values before and after the
transformation process

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

after merging

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

rectified pixel.

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.

Image Classification

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.

Unsupervised Classification

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.

Class Identifications

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

vegetation map.

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.
Class Description
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
Class Description
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

Full Text
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