Citation
Landscape dynamics in the Everglades

Material Information

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

Subjects

Subjects / Keywords:
Everglades ( jstor )
Fires ( jstor )
Landscapes ( jstor )
Modeling ( jstor )
Peat ( jstor )
Pixels ( jstor )
Spatial models ( jstor )
Vegetation ( jstor )
Water conservation ( jstor )
Wet prairies ( jstor )
Conservation of natural resources -- Florida -- Everglades ( lcsh )
Dissertations, Academic -- Wildlife Ecology and Conservation -- UF
Wetland ecology -- Florida -- Everglades ( lcsh )
Wildlife Ecology and Conservation thesis, Ph. D
The Everglades ( local )
Genre:
bibliography ( marcgt )
non-fiction ( marcgt )

Notes

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

Record Information

Source Institution:
University of Florida
Holding Location:
University of Florida
Rights Management:
Copyright [name of dissertation author]. Permission granted to the University of Florida to digitize, archive and distribute this item for non-profit research and educational purposes. Any reuse of this item in excess of fair use or other copyright exemptions requires permission of the copyright holder.
Resource Identifier:
023835822 ( ALEPH )
35815494 ( OCLC )

Downloads

This item has the following downloads:


Full Text










LANDSCAPE DYNAMICS IN THE EVERGLADES:
VEGETATION PATTERN AND DISTURBANCE
IN WATER CONSERVATION AREA 1















By


JENNIFER ENOS SILVEIRA


A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL
OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY



UNIVERSITY OF FLORIDA

1996


UNIVERSITY OF FLORIDA LIBRARIES




























Copyright 1996

by

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.




















ACKNOWLEDGMENTS


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

succeeded.

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

researchers.

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.





















TABLE OF CONTENTS



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


Bedrock


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 ................


LANDSCAPE DATA COLLECTION ..................
Introduction ...... ........................
Ground Observations .....................
Pre-fire Field Observations ..........
Methods ............................
Post-fire Field Observations .........
Methods ............................


.................... 1
....................7
....................7
.................... 8
.................... 9
................... 11
1 ................. 13
1 ................ 14
ea 1 ...............18
................... 18
...................20
...................21
...................21
...................23
Area 1 ...........25
...................25
...................29
...................29
...................29
...................30
................... 31
...................33
...................35
............ .... 38
................... 38
community
................... 40
...................41
S. ................42
....................42


...................47
....................47
...................49
...................49
...................49
S. ...............52
...................52


E


.......................o.....






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 ..............

ANALYSIS OF TREE ISLAND DISTRIBUTION .......
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


.124
.125
.126
.127
.129
.130
.132
.139


vii


.87
.87
.89
.89
.89
.91
.94
.98
.99
100
111


..... 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 ...............................


.139
.142
.150
.152
.152
.154
.156


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

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

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


Viii


.............
.............
.............
.............
.............
.............
.............













LIST OF TABLES


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














LIST OF FIGURES

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


xii






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


xiii







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


xiv












Abstract of Dissertation Presented to the Graduate School
of the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Doctor of Philosophy


LANDSCAPE DYNAMICS IN THE EVERGLADES:
VEGETATION PATTERN AND DISTURBANCE
IN WATER CONSERVATION AREA 1

By

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.


xvi












CHAPTER 1
INTRODUCTION



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

before.

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^
.F.

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






iVegetation
: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
Analysis


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.

1990).

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
Areas
















Kissimmee River
and Watershed


Fisheating Creek


Location of Water
Conservation Area 1


Big Cypress
Swamp


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


Bedrock

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).




















WEST
PALM
BEACH














FORT
LAUDERDALE


DATUM IS MEAN SEA LEVEL


Figure 1-5. Bedrock topography of the Northern Everglades
Source: Jones 1948









Peat

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)
Inches


- Stage


---.iii -t17









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




*/ /*11



JAN FEB MAR APR MAY JUN JLY AUG SEP OCT NOV DEC


Figure 1-6. Rainfall and water regulation schedule of Water
Conservation Area 1
Source: Richardson et al. 1990


Level
Feet


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

Okeechobee.

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

flow.

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

established.

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

islands.

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

work.







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.
1980:13)

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

hydrology.

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

burns.








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

1970).





..... : '.* .' ... :..... '


,.' ., '7, '
S. .. .. ... '. ,. ^ '...o r'
.,. ... jf .*


* ', /c
{;p





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.

1980).





Vegetation


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
Community
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
Fern*
Smilax laurifolia Bamboo Vine
Myrica cerifera Wax Myrtle
Chrysobalanus icaco Coco Plum
Cephalanthus Buttonbush
occidentalis
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
Community
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
Myxophyceae,
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

1969).






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

S : TREES
SHRUBS
WET SAWGRASS
SLOUGH
SLOUGH PRAIRIE

SEAL STAGES


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.













CHAPTER 2
LANDSCAPE DATA COLLECTION


Introduction


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

1990
Merged
Image




1987 Multispectral Field Data and 1992 Aerial
& 1987 Panchromatic Ecological Literature Photographs
Satellite Images

1987
Merged
Image






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


Methods

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.

































4.jr


W


A^


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


Methods

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-

consuming.

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.

Results

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.










540


36 m




21 m
Small unburne
island


25 m



Small burned
island
d


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.

Methods

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

disappearance.

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).







Results

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
(islands/km2)
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.














4

t". Z A..'U''


75


3
6


'iDavis et al. plots
Zones of change
SRichardson et al.
plots


















2 1


2
.,,,,,


8
',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




Discussion

The 1952 photographs were analyzed visually for change, in

a general, non-quantitative fashion. A conscious decision was


4

6








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

recovery.

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.

Methods

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.

Results

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

side.

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.

Resampling

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.


Original
Pixels
Transformed
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

bands.




















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


Multispectral


red,


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


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
Number
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


Source:








Table 2-5. Description of classes in the 1987 vegetation map of
Water Conservation Area 1 found in the successional class map
area
Class Description
Number
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
prairie
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
areas
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
canal
Source: Richardson et al. 1990:40-41




Full Text

PAGE 1

LANDSCAPE DYNAMICS IN THE EVERGLADES: VEGETATION PATTERN AND DISTURBANCE IN WATER CONSERVATION AREA 1 By JENNIFER ENOS SILVEIRA A DISSERTATION PRESENTED T O THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSIT Y OF FLORIDA 199 6 UNIVE S ITY O -L.v .ID Lluru~ E.

PAGE 2

Copyright 1996 by Jennifer Enos Silveira

PAGE 3

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.

PAGE 4

ACKNOWLEDGMENTS I would like to than k 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 c o urse of my degree. In addition, the Florida Cooperative Fish and W ildlife Research Unit and the Arthur R. Marshall Lo x ahatchee National Wildlife Refuge provided equipment and facilities that made this work possible. I offer man y than ks to th e members of m y 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 gi v ing up the task. Without his respectful and enth usiastic counsel I could not have succeeded. Dr. John Richards on s rese arch provided the starting point for this work. I was inspired by his innovati o n and persistence with computers and data analysis. By teaching me some things, and forcing me t o figur e other s out on m y own, he passed a bit of this on to me. iv

PAGE 5

Dr. c.s Holling's teaching o n cross-scale interactions was essential to this work, an d thr o ugh his Everglades Adaptive Management Workshops I met many e xperience d Everglades research ers Drs. Loukas Arvanitis and Ge orge T a nner were thoughtful and thought-provoking advis ors helping me t o see my work in a larger context. Man y of the image processing and GIS techniques used in this wor k were not foun d in any te xtbook ; they were learned by peering ov er John Richa rdson s or Le o nard Pearlstine's shoulders as they worked. I thank L eo nard f o r his emergency technical advice, always cheerful ly provi ded, and fo r his attendance at my doctoral defense. Thanks go to my fel low graduate students Wade Bryant, Howard Jelks Frank Jordan, Cy ndy L oftin Laura Brandt, and Dr. Lance Gunderson f or sharing the airb oatin g, flying and/or proposal brainstorming with me Special thanks go to Don Flic kinger Karen Richardson, Barbara Fesler and Kevin Mattso n f or their ever-present help and moral support V

PAGE 6

TABLE OF CONTENTS ACKNOWLEDGMENTS ............................................... iv LIST OF TABLES ................................................ ix LIST OF FIGURES ............................................... xi INTRODUCTION ................................................... 1 The Study of Landscape Pattern Dynamics ..................... 7 Lands cape Ecology ........................................ 7 Models of Landscape Pattern .............................. 8 Appropriate scaling in models .......................... 9 Spatial models ........................................ 11 The Landscape of Water Conse rvati on Area 1 ................. 13 Th e History of Water Conse rvati o n Area 1 ................ 14 The Substrate of Water Co nser vation Area 1 .............. 18 Bedrock ............................................... 18 Peat .................................................. 20 Water in Water Conservation Area 1 ...................... 21 The natural hydrology ................................. 21 Flood control hydr o l ogy ............................... 23 The Tree Islands of Water Conservation Area 1 ........... 25 Tree island forma tion ................................. 25 Fire in Water Conser va ti o n Area 1 ....................... 29 Fire and the Everglades ecosystem ... ................. 29 Seasonal probability of fires ......................... 29 Fire behavior fact ors ................................. 30 Fire severity ......................................... 31 Patterns of fire events ............................... 33 Fire in Water C onservat ion Area 1 ..................... 35 Vegetation .............................................. 38 Vegetation communit y types ............................ 38 Environmental fact ors in vegetation community distribution ....................................... 40 Hydrologic factors .................................... 41 Nutrient concentrati ons ............................... 42 Vegetation Successi on ................................... 42 LANDSCAPE DATA COLLECTION ..................................... 4 7 Introduction ............................................... 4 7 Ground Observati o ns .... ...... .............................. 4 9 Pre-fire Field Obser v ations ............................. 4 9 Methods ............................................... 4 9 Post-fire Field Obser vatio ns ............................ 52 Methods ............................................... 52 vi

PAGE 7

Results ............................................... 53 Aerial Ph otographs ............. ........................... 5 4 Aeri al Photographs, 1952 ... ............................. 56 Methods ............................................... 5 6 Results ............................................... 57 Discussi on ............................................ 5 9 Aerial Phot ograp h s 1991-199 2 ........................... 64 Methods ............................................... 6 4 Results ............................................... 65 Satellite Images ........................................... 66 SPOT Satellite Image Information ........................ 66 Image Merging ........................................... 69 Resampling ............................................ 7 0 Image transformati on .................................. 71 Vegetation inde x ...................................... 7 4 Rectification ........................................... 77 Image Classification .................................... 80 Unsuper vised Classificati on ........................... 81 Class Id entifications ................................. 81 ANALYSIS OF TREE ISLAND DISTRIBUTION .......................... 87 Introduction ............................................... 87 GIS An al ysis of Tree Island Distribution ................... 89 Ana lysis of Size Distributi on ........................... 89 Methods .............. ........ ....................... 8 9 Results ............................................... 91 Discussi on ............................................ 94 Island Populati on Model Analysis ........................... 98 Methods ............................................... 99 Results .............................................. 100 Discussi on ........................................... 111 LANDSCAPE MODEL .............................................. 113 Introduction .............................................. 113 Conceptual Model of Tree Island Growth and Landscape Pattern ................................................ 113 Processes Contributing to Island Growth ......... ...... 115 Vegetati on successi on ................................ 115 Peat accumulation .................................... 11 7 Fire ................... ... ........................... 11 7 Tree is lands ......................................... 119 Conceptual Model ....................................... 12 O Phase 1: vegetation succession ....... ............... 121 Phase 2 : fire .............. ......................... 122 Phase 3 : post-fire vegetation recovery .. ............ 123 Fire and landscape pattern . ............... ......... 123 Spatial Model of Vegetation Succession and Disturbance by Fire ................................................ 124 Model Design ........................................... 125 GIS software ......................................... 126 M odel structure ...................................... 127 Model base map ......... .. ............ ............... 129 Model rules and rates ................................ 130 Model Verification ........................... .. .. .... 132 Improved Model Simulations ............. .. .. ........... 139 vii

PAGE 8

Methods .............................................. 13 9 Results .............................................. 142 Tree islands in the spatial model ...................... 150 Comparison of Model with Independent Data .. .. .......... 152 Model validation .. ... . ... .......................... 152 Compar ison of model and satellite data ............... 154 Discussion ............................................. 156 SUMMARY AND CONCLUSIONS ...................................... 159 LIST OF REFERENCES ........................................... 168 BIOGRAPH ICAL SKETCH .......................................... 178 Vlll

PAGE 9

LIST OF TABLES Table Table 1-1 Large fires recorded in Water Conservation Area 1 .......................................................... 35 Table 1-2 Comm on plant species of Water Conservation Area 1, by vegetati on community ................................. 3 9 Table 2 -1 Results o f 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 th e 1987 vegetation map of Water C o nservation Area 1 found in less than 1% of the successional class map area ................................ 83 Table 2-5 Descripti on 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 ..................................................... 8 5 Table 2 -7 Classes in the successional map of Water Conservation Area 1 ...................... ................. 8 6 Table 4-1 Subdi visions 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 simulati ons ........................ ................ 135 Table 4-4 Impr oved landscape model rules .................... 140 Tabl e 4-5 Lands cape composition after 20-year fire interval simu lations (percentage of area by seral stage) .................................................... 151 i x

PAGE 10

Table 4-6 Number of patches after 20 year fire interval simulations ............................................... 15 2 X

PAGE 11

LIST OF FIGURES 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 ..................................................... 4 4 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 February, Map of three islands observed on the ground in 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 re sampling ................................................. 7 0 Figure 2-6 Diagram of m e rging of images us i ng c o l or transformation process ..................................... 73 Figure 2-7 Sample pi x el values b e for e and aft er th e transformation process ..................................... 75 xi

PAGE 12

Figure 2-8 Band ratios before and after the image transformation process ..................................... 7 6 Figure 3-1 Diagram of GIS landscape analysis ................. 88 Figure 3-2 Tree island distribution b y 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 o f islands within each size class ...... 94 Figure 3-7 Diagram of energy hierarch y ....................... 98 Figure 3-8 Basic model c o de to generate island distributions: incremental and area-dependent growth ...... 101 Figure 3-9 Basic model c o de to generate island distributions: edge-dependent growth ...................... 102 Figure 3-10 Distribution from incremental growth: island si z e 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 numb e rs . . . . . . . . . . . . . . . . . . . . . . . . . 1 0 5 Figure 3-13 Distribution from incremental growth with increasing island recruitment: total area in each size class ..................................................... 105 Figure 3-14 Distributi o n fr o m incremental growth with decreasing island recruitment: island size classes and numb e r s . . . . . . . . . . . . . . . . . . . . . . . . . 1 0 6 Figure 3-15 Distributi o n fr o m incremental growth with decreasing island recruitment: total area in each size class ..................................................... 106 Figure 3-16 Distributi o n from area-dependent growth: is 1 and numbers . . . . . . . . . . . . . . . . . . . . . . 1 0 8 Figure 3-17 Distribution from area-dependent growth: island size classes ....................................... 108 Figure 3-18 Distributi o n from area-dependent growth: iland size classes ( detail ) ............................... 109 X ll

PAGE 13

Figure 3-19 Dist ribution 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 (d etai l ) .................... 110 Figure 3-21 Dist ribution from edge-dependent growth: island size classes ....................................... 111 Figure 4-1 Informati on used to de velop 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 gr owt h ............................................. 116 Figure 4-3 Conceptual model of vegetation and island dynamics within the interior of Water Conservation Area 1 ......................................................... 121 Figure 4-4 Fl ow chart of spati al modeling and analysis ...... 125 Figure 4-5 Diagram of model st ructu re ....................... 127 Figure 4-6 Comp osi ti on of the base map during the initial run of the spatial model (pe rcen tage of map in each seral stage) .......... ................................... 133 Figure 4-7 Initial landscape model using diffe rent fire intervals: percentage of base map in each seral stage after 190 years ..... ..................................... 13 6 Figure 4-8 Initial model results using different fire intervals: percentage of base map in each seral stage at selected iterati ons ............................ .......... 138 Figure 4-9 Initial model resul ts using different fire intervals: patch density and th e largest patch at selected it erations ............ .... ...... .............. . 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 x iii

PAGE 14

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 ............................................. 14 7 Figure 4-15 Improved model results using moderate and severe fires and different fire intervals: number of tree islands at selected iterations ....................... 151 x iv

PAGE 15

Abstract of Dissertati on Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy LANDSCAPE DYNAMICS IN THE EVERGLADES: VEGETATION PATTERN AND DISTURBANCE IN WATER CONSERVATION AREA 1 By 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, presentl y used as both a flood control retention area and a wildlife refuge. It is a unique wetland ecosystem in which the dynamics of water, ve getation, fire, and peat are tightly linked. These interacting factors produce a landscape pattern that provides excellent habitat for wading birds and other Everglade s wildlife. This doctoral research focused on the landscape of Water Conservation Area 1 and the processes determining its pattern. Few ecologi cal 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) wer e used to quantify the landscape xv

PAGE 16

The GIS was also used t o perf o rm a spatial analysis of tree islands in Water C o nser v ati o n Area 1. The analysis showed more small islands than large o nes, 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 c o uld create a size distribution similar to that found in Water C o nservation Area 1. A conceptual model proposing an explanation for the tree islands' size distribution was devel o ped. The theory proposed that tree islands grow in size over time, but fire reverses island growth. The abilit y o f fire to limit islan d gr o wth is reduced once islands reach some size thresh o ld. 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 m o del 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 o f landscape pattern under different fire frequenc y and severity regimes. The landscape composition and heterogeneit y of the maps were compared with the current landscape. The results illustrated the importance of fire in determining landscape patt e rn. x vi

PAGE 17

CH A PTER 1 INTR O DUCTION Twenty-five years ago, a s the Apollo 8 space mission prepared to orbit the moon, m any people envisioned that the future of humanity lay in out e r space. For three decades the public imagination had enter t a i ned visions of a future society in space. It was expected th at the achievement of a lunar landing would lead to new ef for ts in space e x ploration and colonization. However, while the television images of Apollo 11 astronauts on the moon inspi red a w e and pride, another image was just as powerful--the planet Ear th, suspended in the void of space. Along with the astr onauts viewers saw the beauty and isolation of the earth, and f e lt a new appreciation of our dependence upon it. Since t hat time, public interest in space colonization has waned, whil e t he use o f space technology to monitor our home planet has incre a se d. Ph o t o s fr o m manned space craft and images from satell ites ha v e sh o wn us the colors and patterns of the atmosphere, the oce ans, and the land as never before. It was such an image of Earth that inspired the work that follows: a satellite image of the F l ori da Ever glad e s sh o wing a remarkable pattern of tre e covered isla n ds e m er g i n g f r o m a background of marsh. The islands resemble a school of fis h swimming upstream. A part icularly striking example of the 1

PAGE 18

2 Figure 1-1. Satellite image of Water Conservation Area 1

PAGE 19

3 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

PAGE 20

4 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 piec~ng 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.

PAGE 21

,,-------------i Vegetation lMap (J R. Richardson)) GIS Analysis of Spatial Patterns -------------\ iVegetation ; Succession Map GIS Model of Disturba nce and Veget ation Succession Comparison of Model Output and i ..-----, 1990 Image 5 Data From Remote Sensing Field Studies and the Ecological Literature Formulation of Hypotheses and Conceptual Model 1990 Satellite Image D r:~~:J D Data Classified Modelling Analysis Figure 1 2 Diagram of approach t o landscape analysis This dissertation describes the progression of this Image and undertaking. Chapter 1 gives an overv iew 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 Con servation Area 1 The analysis focused on the distribution of tree islands of different sizes as recorded in

PAGE 22

6 a 1987 classified satellite image. The size and frequency Could the distributions of tree islands raised three questions. ongoing formation and growth of islands create a size distribution like that found in the landscape of Water Conservation Areal? Why were more small islands than large islands found in the satellite image of the landscape? What might be causing a disc o ntinuity f o und in the size distribution of islands? The first two questions we re addressed in chapter 3 with simple models built t o generate island size and frequency distributions f o r different grow th functions. The models demonstrated that the creation and growth o f islands over time could create a distributi on like that found in the landscape of Water Conservation Area 1. Th e 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 distributi on Chapter 4 outlines a conceptual model o f 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 frequ ency and severity regimes. Chapter 5 summari zes and discusses the entire work. The work produced both an example o f a holistic, landscape approach

PAGE 23

7 t o an ecosystem and a tool to spatially vis uali ze a theory of landscape dynamics. The Stud y of Landscape Pattern D y namics Landscape Ecology The study of the spatial distribution o f vegetation and its relationship to environmental processes can be carried out at many levels of resolution: from the study of plants in a vegetation communit y to communities wit hin a landscape, to landscapes on the globe. A landscape is defined as area of land containing multiple ecological c omm unities, "a heterogeneous land area composed o f a cluster of in teracting ecosystems that is repeated in similar form throughout (Forman and Godron 1986:594). Landscap e ecology examines and analyzes the spatial patterns and dynamics of landscapes ove r time. One area of stud y in landscape eco l ogy is the relationship between vegetati o n structure ( i.e., the size of community patches, patch c o nnecti vity 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 landscap e pattern (P otter and Kessell 1985). For example Suffling et al (1988) investigated the control of landscape diversity by fire disturbanc e in boreal forests They found that forests with an intermediat e fire frequency had a higher landscape diversity than forests with high or low fire frequencies

PAGE 24

8 Fewer researchers ha ve addressed the reciprocal effect of landscape pattern o n disturbance (Fr an kl in 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 t o the amount and connectivity of patches of habitat susceptible to disturbance. Disturbance and landscape structure, therefore, have a comple x 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 w hich landscapes change. Scientific methods for the study of single species or communities within ec o s ystems 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. Theref o r e lan dscapes 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 (g ap) models (Shugart

PAGE 25

9 and N o ble 1 98 1, S hu g a rt an d We s t 1977) and g r adient models (Kessel 1979). The term landscape m o d e l" is sometimes used to refer to mode l s which simulate proc e sses that determine where entire landscapes of different t y p e s are f o und. These m o dels operate at a higher hierarchical le v el tha n the m o dels referred to in this work Landscape mod e l is us e d here to indicate models which simulate d y namics within a landscape. Appropriate scaling in m o d e ls Ecological c o mputer m o d e ls can be packed with informati o n and simulate a great numb er o f pr oce sses, but this comple x it y does no t necessaril y produc e realistic behavior o r unders t andable results A m o del w hich limits inf o rmation and processes to the temporal and spatial scales pertinent t o the model s subject is m o re c o mprehensible and efficient than o ne wh i ch attempts to e x plicitly model as man y processes as possible In gen e ral m o d e l s are b e st r e stricted to v ariabl e s ranging over n o m o re than t h re e orde rs of magni t ud e in s p a ce or time (Gunderson 199 2 ). Th is l i mit s a m odel t o vari ables w i thin one or two scale d o mains The scale of a var i able o r process can be identified by its frequenc y b y i t s a r ea of i nf lue n ce or b y its energy Areas b e st m eas u red i n centimeters to tens of meters and time frequenci e s o f m inutes to decades fall into the domain termed the micr o sc a l e Areas of tens of meters to hundreds of meters and time frequ e n c i es of decades to cent u ries fall into the mesoscale domain. Area s of hun dreds of meters to t h ousands of

PAGE 26

10 ki lometers and time frequencies of centuries to millennia fall into the "macr oscale 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.

PAGE 27

Spatial models Most ecol o gical m o dels simulate p o pulati o n d y namics or e cosystem functi o ns: th e fl ows of energ y materials, species, or individuals in the system. The majority o f these models are spatially aggregated. This m e ans that while the y calculate the quantities and fl o ws of variables within a system, they ignore the locati o n o f th o se v ariabl es (Sklar and C o stan z a 1991, Turner and Dale 1991). Spatial m o d e ls differ fr o m spatially aggregated m o dels in that they e x plicitl y 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 e ffects are critical. A landscape model must be spatial if landscape pattern is to be analyzed. Individual cell m o dels. One approach to spatial modeling is to create a function m o del for each individual cell within a spatially-referenced gr i d (Richardson 1988). Each individual cell model uses informati o n fr o m neighboring cells to simulate flows between it and other c e lls and to change the value o f wha t is stored within the cell. Ra tes of chang e are theref o re a function of spatial relati o nships. The cell model method can perform c o mple x c o mputati o ns for each cell and handles continu o us variabl e s w e ll. It is often used in models in which an im por t a nt c o m po n e n t is t h e s i mul ation of water flow (C os tan z a e t a l. 1 99 0, W a l ters et a l. 1 992 Fennema e t al. 1 994 ). H o w ever t h e computing requirements to run a model of t h is t y p e are very hig h if numero u s cells are

PAGE 28

12 used. For example, a cell model of the Atchafalaya delta (Costanza et al. 1990) covered 2479 km 2 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 m 2 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. 1990). The number of gridcells in a model can be reduced by modeling small areas (less than 20 km 2 ) or using a coarse resolution (greater than 1 km 2 ). 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 1:h e Everglades using a 4 km 2 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).

PAGE 29

13 Rather than using a single grid, with a model and multiple storages 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 comple x 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

PAGE 30

14 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. ( 19 94) 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. both natural and human in origin. The History of Water Conservation Area 1 These factors are 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

PAGE 31

15 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, Gleas on and Stone 1994) The Hillsborough Lake area was first affected by human hydrologic manipulations in the 1880's, when the Caloosahatchee River canal lin ked 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 flo od 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

PAGE 32

16 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). South Florida Gulf of Mexico 20 mi 32km Numbers indicate Water Conservation Areas Canals: a = St. Lucie b = West Palm Beach c = Hillsboro d = Caloosahatchee River Atlantic Ocean Figure 1-3. Map of South Florida showing the Water Conservation Areas

PAGE 33

17 Historic Everglades 20 mi I 32 km Fisheating Creek Big Cypress Swamp Mangroves and Coasta l Marshes I L o cati o n o f Water T h e Eve r g l a d es Figure 1 4 Map of the hist oric Everglades system

PAGE 34

18 Today flood control str u ct ur es restrict the hydrologic connectivity o f Water Conservation Area 1 to the rest of the Everglades ecosystem. Flood control and water management have allowed rapid growth of the agric ulture and urban development now surrounding the area. It remains a northern refuge for Everglades species threatened by habitat destruction. The Substrate of Water C o nservation Area 1 Bedrock Water Conservati o n Area 1 lies over limestone bedrock known as the F o rt Th o mpson F o rmati o n. Originally formed on the flat floor of a shallow sea, the F o rt Thompson Formation has since been eroding and dissolving. At a fine scale the bedrock surface is uneven, etched with a t opo graphy of depressions and ridges. At a larger scale, a wide trough in the bedrock surface, kn ow n as the Loxahatchee Channel, runs roughly north south from Lake Okeechobee to the so uthern end of Water C o nservation Area 1 (Jones 1948). Water ove rflowing from Lake Okeechobee followed the tr o ugh, w hich 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 ''Hillsb oro ugh Lake." The earliest Everglades peat was f or med in the Lo x ahatchee trough (Gleason and Stone 1994).

PAGE 35

LAKE OKEECHOBEE HENDRY 19 r------+ I ~MARTIN CO WEST PALM ~-==.:.:..~~.....aJBEACH DATUMISMEANSEALEVEL Figure 1 5 Bedrock topography of the Northern Everglades Source : Jones 1948

PAGE 36

20 Peat 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

PAGE 37

21 vegetation communities at Water Conservation Area 1 is correlated with this peat surface topography (Pope 1989). Several forces shape the micro-t opo graphy 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). Peri o ds o f 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 ove r 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 c o nnected with the Everglades water body to create a continuous inclined hydrologic gradient extending from La ke 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 "P a-hay-okee (grassy waters) and river of grass" (Douglas 19 88) Un like a river, however, water flow was

PAGE 38

22 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 relativel y small amount of water actually reached Water Conservation Area 1 via overland flow from the north, the role played by "u pstream 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 veget ation. Evidence of this phenomenon lies in the shape of Water Conservation Area l's larger tree islands (Gleason and St one 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

PAGE 39

23 than the usual sheet flow. 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). T od a y the h y drol ogic 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) Inches Stage Level Feet s-----------------------------------r JAN FEB MAR APR W.Y JUN JLY AUG SEP OCT NOV DEC 18 17 16 15 14 13 12 11 10 Figure 1-6. Rainfall and water regulation schedule of Water Conservation Area 1 Source : Richardson et al 1990

PAGE 40

24 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 ne x t 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 Okeechobee. 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.

PAGE 41

25 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 on ly 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 "uniqu e 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 "wat er tracks", wide treeless fen channels dominated by sedges and Sphagnum species (Glaser 1987). The wat er tracks formed a network similar to a braid ed stream, then expanded and merged,

PAGE 42

26 leaving islands of trees in the interstices. Some water tracks are also patterned with ridg e s ("strings") and sloughs ("flarks") oriented perpendicular to the direction of water flow. Tree islands in Water Cons e rvation 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" (fl o ating rafts o f peat and live vegetation) In peat prairies of the O k efen o kee Swamp, an ecosystem similar to Water Conservati o n Area 1 but without the influence of sheet flow, a connection between p e at 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 f o rmed und e r w et c o nditions when large batteries of peat rise to the surface of the water. Plant succession occurs on the fl o ating batteries, further consolidating the peat structure until w oo dy vegetation is established. While the mechanisms o f peat batter y formation at Water Conservation Area 1 ha v e n ot b ee n d o cumented, newly formed batteries ha v e been o bs e r ved d u r in g high w ater f o llowing a ver y dry period (W. M. Kitch e ns, pe rs o nal c o mmunication). Trapped gasses (Rich 1979) o r bu oy an t plant rhiz o mes (Gleason and Stone 1994) may be involved in fl o ating the batteries.

PAGE 43

27 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 195 2 ; 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

PAGE 44

28 expansion of existing islands in addition to the creation of new islands. 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 work

PAGE 45

29 Fire in Water Cons e rvat i on 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 o f fuel that bec o mes e x plosive during the 6to 8-m o nth dr y season. There are more days with lightning rec o rded in south Florida than elsewhere in the Nation. Over 6,000 lightning strikes were rec o rded in inland s o uth Florida during one 6-hour period in th e summer o f 1976. (Wade et al. 1980:13) 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 fir e s in the Everglades since its genesis, although human-ignited fires were probably less frequent before Eur o peans arri v ed (R o bertson 1953). Fires not caused by humans are ignited by lightning, which is most common during the summ er st or m s eas o n. W hile th e tw o t y pes o f fires in Everglades Nati o nal P ar k g e n erally oc cur a t diff e r e nt times of the ye ar, b o th t ypes b u r n the greatest area i n the month of May (T ay l o r 1 98 1 ) beca u se of factors determ in ing fire propagation and int e n s i ty

PAGE 46

30 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. grass and dead herbaceous stems are fine fuel. Needles, leaves, 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

PAGE 47

31 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 f o r severe fires. Minimal rain falls during summer storms and winter weather fronts. Water evaporates and transpires without replenishm e nt until it stands only in sloughs and alligator holes. Vegetation desiccates, and the peat surface itself may bec ome 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 incr eases fuel densit y Fires ignited under these dense, dry fuel conditions are intense, and spread quickly over large areas (Wade et al. 1980, Du ever 1984). Fire severity The terms fire intensity and fire severity are sometimes used interch angeably More specifically intensity refers t o th e energy expended by a fire while severity refers to the effect of a fire on the landscape The severity of fires in th e Everglad es landscape varies depending upon the intensity of the fire. Th e mort ality of foliage caused by fire is a

PAGE 48

32 function of plant moisture content, the temperature of the fire and the time the plant is e x p o sed 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 hydrology. Fires are low-intensit y 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 vegetati o n is adapted to frequent, lowintensity fires. Unless a low-intensity fire is followed by a long period of deep flo o ding, 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 le v el. 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 e x pend energy reserves to resprout. While sawgrass resprouts quickl y after a m o d e rate fire (F o rthman 1973), small woody seedlings with little e nergy stored in their root systems may not survive. This ser v es t o set back the vegetation

PAGE 49

33 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 hydroperi od 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 mortalit y 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 occ urrence. They performed a Fourier analysis on the log of fire sizes over time and found that th e dominant temporal cycle was an annual one A 10 to 14

PAGE 50

34 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 burns.

PAGE 51

35 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 als o burned in Everglades National Par k (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, 21 1 The most recent severe fire in Water Conservat ion Area 1 burned in April of 198 9 at the end of a dry season following eight months of dr o ught. A winter freeze had heightened the fire potential by increasin g 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 porti o ns 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 m e dium si z ed tree

PAGE 52

36 islands, brush, sawgrass and wet prairie were severely burned (Loxahatchee Nati o nal Wildlif e Refuge, unpublished data). No green vegetation was visible in the burned area, with one exception--the photographs sh o wed that tree crowns in the centers of large tree islands were green and undamaged. It appears that the 19 89 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 envir o nment e x tingu i shed the fire. This resulted in the fire's energ y being released differentl y on large and small islands. If this phen o menon o ccurs in all severe fires, then fire is an important fact o r determining the sizes of tree islands, and thus the landscape pattern in Water Conservation Area 1. (This hypothesis is e x plored in detail in Chapter 4) The interiors o f large islands ma y ha v e escaped the fire because their understory vegetation and microclimate are quite different from other vegetati o n communities in Water Conservation Area 1. The le a f area inde x o f a tree island overstory is significantl y higher than sawgrass or wet prairie vegetation (Gunders o n 1 9 9 2) Theref o re tree island canopies

PAGE 53

37 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 fact or in fire potential (Schroeder and Buck 1970). Figure 1-7. Area 1 Satellite image of 1989 fir e in Water Conservation Transpirati on o n tree islands is also higher than in sawgrass or wet prairies (Gund erson 1992) Tree roots reach

PAGE 54

38 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. 1980). Vegetation 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.

PAGE 55

39 M o re d e t a il ed classifications of Everglades plant c o mmunities were m a de later (Craighead 1971 Wade et al 1980 O lms te d and Loope 1984) Gunders o n ( 1994 : 3 2 7) divided Everglades f re sh w at er wetlands int o forested communities graminoid ass oci ati ons and areas with little or no em er g en t vegetation." F o res ted communities include b a y h ea ds, wi ll ow heads and c y p re ss f orests Gramin o id ass o ciations incl u de sawgrass marshe s p e a t wet prairies and marl wet pr a iries Little or n o em e rg e n t vegetati o n includes ponds cr ee k s and sloughs. Using thi s community classification the co mm o n pl ant species of Water C o ns erv ati on Area 1 are l i st e d in Table 12 Table 1-2. Comm o n pla nt species o f Water Conserv a tio n A r ea 1, by vegetation c o mmuni ty Vegetation Spec ies Commo n N ame Community Bayhead Persea borbonia Red Bay (Tree Island ) Il ex cassine Dahoo n H o ll y B l echnum serrulatum Sw a mp Fe rn Acrostichum danaeifolium Leather Fern Osmunda regalis Royal Fern Lygodium japonicum J a pa n ese C l i mbing F ern* S milax laurifolia Bamboo Vine Myrica cerifera Wax Myrtle C h rysobalanus .1 ca co Coco Pl u m Cephalanthus Buttonbush occidentalis Will o w Head Sali x caroliniana Willow Ludwigia octova li s Water Primrose Mikania s candens Climbing Hemp vine P olygon um Smart weed hydropiperoide s Phragmi tes commun i s C o mm o n (Giant) Reed Sawgrass Marsh Cladium jamaic e ns e Sawg r ass ind i cates exotic species Sources: Rich ardson et al 19 9 0 Si l veira 1 988 data), Th o mp son 1970 A l ex an der and Crook 1975 al. 1994 (unpublish e d a n d B ro wd er et

PAGE 56

40 Table 1-2--continued Veget at ion Species Common Name Community Peat Wet Rhynchospora traceyi Tracy 's Beakrush Prairie Eleocharis montevidensis Spikerush Eleocharis quadrangulata Spikerus h Panicum hemitomon Maidencane Eriocaulon compress um Hat-pins Fuirena scirpoides Rush Fuirena Xyris spp. Yellow -e yed Grass Hypericum spp. St John 's Wort Micro-algae: Periphyto n Mats Myxophyceae, Bacillophyceae and Chlorophyceae groups Pontederia cordata Pickerel weed Sagi ttaria latifolia Arrowhead Slough Utricularia spp Bladderwort s Potamogeton illinoensis? Pondweed Najas guadalupensis Southern Naiad Nuphar advena (luteum?) Spatterdock Nymphoides aqua ti cum? Floating Heart Nymphaea odorata White Water Lily Eichhornia crassipes* Water Hyacinth* Hydrocot yle umbellata Water Pennywort Salvinia spp. Water Fern Invasive Typha domingensis Cattail Species Melaleuca quinquenervia* Punk Tree, Cajeput* indicates e xotic species Sources: Richardson et al. 1990, Silveira 1988 (unpublish ed data), Th ompson 1970, Alexander and Crook 1975, and Browder et al. 1994 Environmental fact ors 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).

PAGE 57

41 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 slo u gh but tolerate longer hydroperiods than sawgrass T ree islands a re h igher 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 1969).

PAGE 58

42 Nutrient concentrations Before drainage and impoundment, Water Conservation Area l'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 197 5) 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. seral stage, as it matures, alters the conditions of its Each

PAGE 59

43 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 e x pl o itation. As time goes on, succeeding stages have more slower-growing, resource conserving species. Climax species are adapted for equilibrium conditions climax community Each geographic region has its own Clements theory was widely accepted, but some later ecologists notably Henry A. Gleason, objected to the rigid concept of a fixed seral progression and climax. They proposed that species composition at a site was determined only by the summation of a variety of st o chastic 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 pla ye d a critical role. In these systems a regional clima x was never achieved. Terms like ''fire climax" were coined to keep these systems under the umbrella of succession theory. The concept of successi on is useful for und ersta nding the spatial pattern of vegetation i n Water Conservation Area 1 Vegetation community change in accorda n ce with Clements theory can be observed in p eat w etland systems (Vogl 1969). In Water Conservation Area 1, w ater lilies in sloughs and alligator holes build up peat raising e l evation and shortening hydroperiod

PAGE 60

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). SHRUBS SAWGRASS PRAIRIE ---------1SERAL STAGES-.--------Figure 1-8. Successional sequence in Water Conservation Area 1

PAGE 61

45 However, Clements' emphasis o n a linear progression through the seral sequence, ending in equilibrium at a stable clima x is not applicable to the Everglades ec o system. 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, disturbanceadapted ecosystem like the Everglades, the c o ncept of a stable clima x at equilibrium is not appropriate. Tropical hardwood hammock forest has been identified as the clima x community of the Everglades (Olmsted and Loo pe 19 8 4 ) but it is not found anywhere in Water C o nser v ati o n Area 1. A more useful non-linear concept of succession and disturbance is found in C.S. Holling's c y clical theory of ecosystem dynamics (Holling 1992a). In this theory, disturbance is part of a continuum with successi o n. An ec o s y stem progresses from an exploitation phase (c o rresp o nding t o early seral stages ) to a conservation phase (lat e seral stages), as in the succession concept. Howev e r, while the s y stem becomes more stable and organized as it appr o aches clima x it also becomes less resilient. Alth o ugh m o r e able t o abs o rb small perturbations, it i s m o r e v uln era bl e t o a maj o r disturban ce Inevitably disturbanc e r e l eases t h e stored e n er g y an d m a t erial in the system and dis o rg a ni zes i t. The system goe s thro u g h a release phase and th e n a reorga n ization phase from w h ich it may follow the same ex pl o i ta ti o n and conservation series it followed before. If condit io n s h ave changed however it may follow a

PAGE 62

different path. This last c o ncept is especially applicable to the Everglades ecos y stem. As humans change the hydrology and introduce e x otic species t o the Everglades, v egetation communities have a greater p o tential t o change in ways not predicted by the o ur current understanding of succession in the Everglades landscape.

PAGE 63

CHAPTER 2 LANDSCAPE DATA COLLECTION Introduction 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 acc o unts of the vegetation of the area were used as clues to th e landscape in the past (Davis 1943, Thompson 1972). Litera t ur e o n the natural history of the Everglades, other Florida w e tl a nds and the Okefenokee swamp als o was used to provide additi o n a l in fo rmati o n on the species f o und in the area (Dressler et al. 1 9 87, C raighead 1971, Glasser 1986, Kushlan 1990, Olmsted and Loope 1984, Pesnell and Brown 1977, Wade et al. 1980). Using the knowledg e g ai n ed from t h e a b ove in for m atio n, a map of Water Conservati o n Area 1 vegetatio n wa s m a d e fro m a 1987 satellite image. The i m a g e processing require d to create th is 47

PAGE 64

48 map is described in the latter part of this chapter Vegetation was computer-classified fr o m the satellite data using field observations, aerial ph o tographs and a vegetation class i fication of the same image by J.R Richardson (Richardson et al. 1 990). Vegetation classes were based on a successional sequence a nd wil l be referred to as s uccessional c lasses" to di s t ingu i s h them from the classes in the Richardson et al. classifi ca tion. The creation of this successional vegetation classificat i on required a synthesis of all the landscape and community ecology information gathered in this w o rk. 1952 Aerial Photographs 1990 Multispectral & 1990 Panchromatic Satellite Images 1990 Merged Image 1989 Raw Panchromatic Image 1987 Multispectral & 1 987 Pan chr o matic Satellite Images Field Data and Eco l ogical Literature 1992 Aerial Photographs 1987 Merged Image _,,. ---------------------, i 18-Class Vegetation 1 2 4 -Class V egetation : ; Map (J .R. Richardson ) =~~~=~ s ~~~ ~ ~~ --------j D ------"""' : I l ______ ; Figure 2 1. Diagram of landscape data s o urces Data Source Classified Image

PAGE 65

49 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 Methods 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

PAGE 66

50 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.

PAGE 67

Figure 2-2 successional Map of the vegetation 1989 map 51 fire, used in with the the location of model in Chapter the 4

PAGE 68

52 Post-fire Field Observations Methods 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 26 30' and 26 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 consuming. 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

PAGE 69

53 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. Results 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 wid e surrounded the islands

PAGE 70

54 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.

PAGE 71

r N II [I 25 m 21 m Small unburned island 55 [ Small burned island Approx. scale: 1:1700 \ 540 m Approx. direction of fire movement Trees, not burned Trees, burned Shrubs, burned Sawgrass: burned on large island 240 m Large burned island Approx. scale: 1:8600 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 lan dscape pattern, the matching of the photographs to map coordinates or

PAGE 72

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. Methods 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 m 2 in the photographs were searched for on the satellite image to detect island degradation and disappearance. Twenty-five of the 1952 aerial photographs were of sufficient quality to sample for tree islands. A quadrat 1 km 2 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 m 2 ), medium to large (between 160 and 2500 m2) and very large (greater than 2500 m 2 )

PAGE 73

57 Results 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 ( islands/ km 2 ) Small islands 2 93.81 (less than 160 m) Medium to large islands 2 5.62 (between 160 and 2500 m) Very large islands 0.48 (greater than 2500 m 2 ) All islands 99.91 A careful comparison between the 1952 photographs and the 1990 satellite image showed that all tree islands greater than 1000 m 2 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)).

PAGE 74

58 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 vis ible 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.

PAGE 75

4 6 7 75 3 59 lilltifil Davis et al plots ~ Zones of change Richardson et al plots Figure 2-4. Areas of vegetati on change in Water Conservation Area 1 between 1952 and 1990, and locati o ns o f ph o toplots f ro m two vegetation studies Sources : Richardson et al 1990 a n d D av i s et a l 1992 Discussion The 1952 ph o t o g rap h s were analyzed visually for c h ange in a general non-qu a n titative fashion A conscious decision was

PAGE 76

60 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 flu x 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, t o 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 to o 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

PAGE 77

61 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

PAGE 78

62 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 22). 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 recovery. 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 22, 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

PAGE 79

63 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 ( Th ompson 1972) When the lower portions of sawgrass plants are submerged, sawgrass and wet prairie appear similar in aerial photographs. The difference in plot compositi o n 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 interpretati on s 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/sl ough sawgrass-tree island mosaic," (Davis et al. 1994:43 2) However, the different results of photoplot studies in Wat er Conservation Area 1 illustrate the

PAGE 80

64 limitations of using a small sample of photographic interpretations to characterize vegetation community change within an Everglades landscap e 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 necessar y f or 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. Methods 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. photographs were taken. Both near-vertical and oblique color 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

PAGE 81

65 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. Results 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 surve y 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.

PAGE 82

66 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 side. 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

PAGE 83

67 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 visi ble light (.50-.59 micrometers). Band 2 is the reflectance of red visible light (.61-.6 8 micrometers). Band 3 is the reflectance of near infrar ed radiation (.79 -. 89 micrometers) The images used in thi s 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 multi spectral 12 X62329 7870404155912 April 4, 1987 panchromatic 11P6 2329 7870404155910 April 11, 1989 panchromatic 11P623297890411155903 April 5, 1990 multispectral 11 X6232979 00405160335 April 5, 1990 panchromatic 1 2P6232979 00405160333 A SPOT image dating fr om April 11, 1989 was taken during the severe fire that burned over 1 8 000 ha o f Water Conservation Area 1. Smoke obscured part of Water Conservation Area 1, and created spectral variation in the image data preventing vegetation interpretati o n or classification of the multispectral image. The panchromatic imag e was nonetheless useful for visual analysis. It was used as a documentation of fire behavior during the 19 89 fire, particularly as it affected tree islands. The multis pectral 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

PAGE 84

68 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

PAGE 85

69 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

PAGE 86

70 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. Re sampling 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 25. This increased the size of the data file by a factor of 4. A 20-meter pixel column (x) = 1 row (y) =1 data (d)= 5 The same 20-meter pixel resampled to 10-meter pixels x=l x=2 y=l y=l d=S d=S x=l x=2 y=2 y=2 d=S d=S 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

PAGE 87

71 of the multispectral and pan c h rom ati c images. Twenty-five corresponding points were id e ntified o n th e tw o images (control points) These points were man-mad e f e atures that could be clearly identified on both images. F o r each c o ntrol point, x and y coordinates on each of the images were rec o rded and paired together. The pairs of coordinates f o r all the c o ntrol 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 r oo t m e an square o f the distance between the actual coordinat e s o f the c o ntr o l p o ints and the coordinates predicted for e a c h p o int b y the transf o rmation matrix. The transformation matri x f o r b o th the 1987 and 1990 image pairs had a root mean square distance less than one half of a pixel (5 m). While the multispectral pi x els were spatially redefined during the resampling and registering process, the multispectral data values were not altered. Ne ar e st n e i g hb o r" resampling was used, so the spectral data values o f th e o riginal pi x els were simply reassigned to the near es t sp at ially redef in e d pi x el. Image transformation Once the two images w ere matched together wit h 10 m resolution, the panchromati c data were incorporated into t h e multispectral data via a c o l or transformation procedure (Richardson et al. 1990).

PAGE 88

72 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 Oto 255. Band 1 data are assigned as blue va lues, 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 Ot o 2 55. In addition to red-green-blue (RGB) color space, this pi xel color can be defined by an alternate colorspacea polar c oordi nate s y stem known as intensity, hue and saturati on (IHS) To merge the panchr omatic and multispectral data, the multispectral pi x el values in RGB colorspace were transformed to IHS values. Next the intensit y va lue f or each multispectral pixel was deleted and replace d by the value found at that pixel location on the panchromatic image. Finally, the IHS coordinates of the pi xels were converted back to RGB colorspace. The band 3 (red), band 2 (green) and band 1 (b lue ) 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 multispectr al data. It is not possible to increase the spatial resolution of multispectral data itself. The addition of panchromatic data to the

PAGE 89

73 multispectral image greatl y enhances the definition of s mall v egetati o n patch e s, p o n ds sloughs roads and buildings original Pixels P 24 RGB to IHS R 66 G 17 B 30 p = R = G = B = I = H = s = Substitute P for I I 150 p 24 H 73 H 73 s 41 s 41 panchromatic brightness red (band 3) green (band 2) blue (band 1) intensity hue saturation Transformed Pixel IHS to RGB R 38 G 9 B 17 Figure 2-6 Diagram of merg ing of images using color transformation pr o cess The spectral res o luti on of the multispectral data was n o t increased by th e mergin g The panchromatic sensor records data in a single ban d tha t lumps together reflectance from the green and red porti o ns of t h e spectrum already sensed by bands 1 and 2 of the multisp e c tra l sensor Some spectral information was actuall y l o s t ; the intensity values from the infrared portion of the spe c trum (b a nd 3) were replaced by panchromatic data that did not include in fr a re d reflectance However the infrared hue

PAGE 90

74 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 vegeta ti on inde x 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 t o 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 denom inator, 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 bands.

PAGE 91

75 24 2 5 24 24 24 26 26 23 26 27 27 25 26 2 7 27 26 Panchr o matic v alu es (band 1) of pixels in mesic f o r e st 6 6 66 64 64 17 1 7 18 18 30 30 30 30 66 66 64 64 17 1 7 18 18 30 30 30 30 67 67 6 7 67 17 17 17 17 31 3 1 30 30 6 7 67 67 67 17 1 7 17 17 31 31 30 30 Multispectral red, g reen and blue values (bands 3 2 and 1 ) of p i xe l s registered to same the locations 38 39 37 37 9 10 10 10 17 18 17 17 38 41 40 35 9 10 11 10 17 18 19 16 4 1 43 43 39 1 0 10 10 10 1 9 19 19 17 4 1 43 43 41 1 0 10 10 10 1 9 19 19 18 Multispectral r e d, g r een and blue values (bands 3 2 and 1) after merging Figur e 2 -7 S am ple pixel values before and after the tran s f o rmati on process

PAGE 92

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. 4 6 0.46 0.44 0.45 0.44 0.45 0.44 Multispectral band ratios before and after merging 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 Vegetation index (NDVI) calculated from multispectral 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) Vegeta tion index (NDVI) calculated from multispectral data for pi xels in Figure 2 -7 before and after merging

PAGE 93

77 Rectification 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 Registrati on 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

PAGE 94

78 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

PAGE 95

79 transformation matri x. The error contributed by each control point was also identifi ed 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 th e transformation matrix the image file coordinates were converted to map coordinates Spectral values were not altered, but were assigned to the nearest rectified pi x el. It was unnecessar y t o adjust for elevation in this rectification, due to the height of the satellite and the flat terrain. The SPOT satel lite has a fixed array of sensors aligned in the x directi on and the data are corrected for the curvature of the earth. Theref ore SPOT images do not ha ve the displacement and distortion found in aerial photographs Distortion in the imag e can occ ur in they direction if the speed of the satellite varies while the image is be ing sensed. A rectificati on 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 m ap-based rectification is appropriate to the 10-m resolution of the data wh en the control points are numerous, well distributed and checked for error.

PAGE 96

80 Image Classificati on 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 classificati on the analyst first defines known areas of vegetati on on the image. Statistics on the spectral data of the pi x els in each area are used to define spectral signatures for each class. An algorithm such as maximum likelihood is used t o 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 suitabl e for modeling the least-changed vegetation of Water Conser vation Area 1, a successional class map was classified from a subset of the 1987 merged image using

PAGE 97

81 an unsupervis e d proc e du re The unsupervised classifi ca t i o n f oc used on id e ntifyin g vegetation in the i nterior of W a ter Conservation Area 1, an d o n defi n ing classes w h ic h el uc i dated t he transitions betw ee n vegetation communities Unsupervised Classificat io n A subset was e x tracted from the 1987 me rg ed i mage which contained a 5000 ha p o rti o n of the interior o f W a t er Conservation Area 1 (s ee Fig ure 2 2) This s u b s et w as run through an unsupervised clu steri n g r outine to p rod uce a 2 8-class classification (ERDAS 1991). Th e clust eri ng w as based on the values in the four spectral da ta band s of each pi x el: green reflectance (band 1), r e d r eflectance (band 2) ne ar i nfrared reflectance (band 3) an d th e NDVI ( a vegetat i o n i nde x o f the relationship between ban d s 2 and 3 ) The main f a ctor s influencing spectral refl e c tance in Water C o n s ervatio n Area 1 are the vegetati o n sp ecies the densit y o f v egeta ti o n, and th e amount of water present. C onveniently these f a c t or s als o define vegetation c o mmunit ies and the gradatio n s bet ween t h e m. It was e x pected, for example that the c o mmunity of we t pr a irie species would be divi ded b etween several classes separ at ed by their proportions of open water, wet pra irie v eg et ati o n, an d sawgrass. Th e s e cla sses repr e s e nt stag es alo ng a successi o nal progression fr o m sl o u g h t o w e t pr a iri e t o sa wgrass Class Identif ica ti o n s The 2 8 un supervis e d cl asses had to be ide n tifie d a nd rank e d into a seral progr e ss io n Thi s w as acco mpl is h e d b y comparing th e classes with g eorefere n ced ground observatio n s

PAGE 98

82 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 classificati on (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 vegeta ti on map (Ri chardson et

PAGE 99

83 al. 1990 : 46) a set of rules for identifying the unsupervised classes based on the summar y results was defined. The rules specif i ed the s u ccessional class of the unsupervised classes based on the percentage of each class in the vegetation map cross refere n ced to it (see Table 2-6) T he 28 c la sses of the unsupervised classification were iden t if i ed according to the rules, renamed, and ranked along a successional continuum (see Table 2 -7) This created a succes si o n al class map suitable for use as a base map for the spatia l model of landscape pattern (see Chapter 4). Tab l e 2 4. Cl a sses in the 1987 veget ation map of Water Conserva t ion Area 1 found in less than 1 % of the successional cl a ss m a p area C l a ss Description N u mber 1 H i g h density sawgrass with some fern tussocks 2 Sawgrass with invasi on of cattail near canal 1 0 C a ttail community close t o canal 11 O p e n deep water mostly along Hillsborough canal 1 2 Sl oug h deeper or less vegetated than class 9 15 Sa w g ra ss with . of wa x myrtle close to canal invasion 18 Ca ttail community Source : Richardson et al 1990 : 40-41

PAGE 100

84 Table 2 5. Description of classes in the 1987 vegetation map of Wat e r Con s ervation Area 1 found in the successional c lass ma p are a C las s De s cription Number 3 Sa wgr a ss ... primary sawgrass class occurring on al l parts of the refuge including the vast s a wgr as s area s on the west side of the refug e 4 B ru sh/ s aw g rass . primarily sawgrass with la rg e a moun ts o f wax myrtle. Some tree islands whic h m a y h ave been burned out previously are made up e nt ire l y o f t he class, particularly in the s o u thern part of the refuge 5 Tree is land .. lower stature tree island community made up of a mix of wax myrtle, dahoo n hol l y a nd red bay. Occurs along edges of larg e tree i s lands with some small islands composed e n tirel y o f thi s c l ass 6 We t p ra i r ie ... l argest area of the refuge Th i s c la ss is t he denser wet prairies occurring ov er al l o f the refuge but the primary community typ e o f t he c en t ral portion of the refuge This cl ass m a y contain small tree islands smaller than 30 feet acr o ss and areas with small sawgrass str ands 7 T r e e island ... core of larger tree islands larg er stature t r ees made up primarily of dahoo n hol l y and red bay with lesser amounts of way my r tl e than class 5 8 Brush ... smaller brush clumps primarily in wet prai r i e 9 We t pr a irie .. sparser wet prairie community o ft en wi th sparse sawgrass 1 3 Wi ll ow/brush . predominantly willow but wit h s o me mi xed c l assification whit wax myrtle bru s h are as 1 4 B ru s h / t ree is l and .. class of many of the smal l e r tree islands with mostly wax myrtle, some s a wgr a ss occasionally dahoon holly and red b a y 16 S a wgrass . slightly higher elevation sawgra ss th a n c lass 3 Core of sawgrass ridges 1 7 Wil l ow ... wil l ow along canal edge, some misclassified floating aquatics along Hillsboroug h canal So ur ce : Richardson et al. 1990 : 40-41

PAGE 101

85 Table 2-6. Rules for succ e ssiona l class identifications based on the 1987 vegetati o n m ap of W a t er Co ns e r v ation Area 1 Vegetation Type R u le for I dent i fic ati o n Based on Cross Reference with Vegetation Map b y Richardson et al. 1. S l o u gh T ype is > = 50 % sparse we t prairie (class 9 ) and >=99 % sparse wet pr a i rie and wet prairie classes combi ned ( classes 6 + 9) Classes are ranked in d escending order by percentage of sparse w et prairie (cl as s 9) 2 Sp ar s e wet T ype is <50 % sparse w et prairie (c l ass 9 ) prairi e but >=85 % sparse we t p rairie and wet prairie c la sses combi ned (classes 6 + 9) C la s ses are ranked i n descending order by percentage of sparse we t prairie (cla ss 9) 3 D en s e wet Type is >=50 % but <85% sparse wet pr a irie pra irie with and wet prairie classe s c o mbined (classe s 6 sawgra ss and/or + 9) Spa r se we t pr airie ( class 9) is < 5 % spars e s h r u bs Sawgrass (classes 3 + 4 + 1 6) and brush (c l asses 8 + 14) are < 30 % Classes a r e ranked i n descending order by percen t age of wet pr airie (class 6) 4 Sawgrass and T y pe is > = 25 % but <50 % wet pr a ir ie ( class shrubs wit h some 6) wet prai r ie Sa wgrass (clas s e s 3 + 4 + 16) and bru s h (c l asses 8 + 14) ar e present. No tree s ( cla ss es 5 + 7 + 13 + 17 ) are pre s e nt. C l asses are ranked in de scending order by percentage of wet prai rie ( class 6) 5 Shrub s Type is <25 % wet prair ie ( class 6) Bru s h is >=40 % and trees are pr ese n t but <10 % (classes 5 + 7 + 13 + 1 7) Classes are ranked in d escending order by percentage of wet prairi e ( class 6) 6 D e n se shr u bs and T ype lS >=10 % and <50 % t r ee s ( classes 5 + 7 t r e e s + 13 + 1 7) Classes are ranked in de scending o rder by percentage of wet prairie (c lass 6) 7 L o w s t a tu re Type is >=50 % non core t r ee s ( classes 5 + trees 13 + 1 7) 8 Is land c o r e Type is >=50 % core trees (c l a ss 7) Ranked tre es in ascending order by per c e ntage of core trees (class 7) So u r c e : R i chards o n et al 1990 : 40 41

PAGE 102

86 Table 2-7. Classes in the successional map of Water Conservation Area 1 Classes Vegetation type 1-3 Slough 4 -9 Sparse wet prairie 10-14 Dense wet prairie with sawgrass and/or sparse shrubs 15-19 Sawgrass and shrubs with some wet prairie 20 -23 Shrubs 24-25 Dense shrubs and trees 26 Low stature trees 2 7-28 Island core trees

PAGE 103

CHAPTER 3 ANALYSIS OF T REE ISL AND DISTRIBUTION In trod ucti on This chapter describes a set of anal yse s o f the distribution of tree islands, w hich led t o the development of a theory of landscape pattern dynamic s in Wate r Conservation Area 1. Rather than attempt an analysis of the pattern of all vegetation types in Water Conservation Area 1 at once tree islands were chosen as indic ators of the landscape pattern. There were several reasons for this choice. First, trees appear clearly in imagery and phot ography The probability of confusing trees with slough, wet prairie o r sawgrass is low. Second, because tree islands have abrupt b o undaries and the highest elevation relati ve t o other vegetation they are less subject to change in si ze due to water level v ariati o n. Finally, the striking visual effect of the pat tern of tree islands in Water Conser vation Area 1 was the inspiration for this body of work. The analyses t ook two different approaches (see Figure 31). In the first approach, the distribution of different-sized tree islands found in a satellite image of the landscape of Water Conservation Area 1 was examined The number and si zes of tree islands in an 18-clas s vegetation map classified from a 1987 satellite imag e (descr ibed in Chapter 2) were tallied 87

PAGE 104

88 1987 Multi. and Pan. Images Field and Remotely Sensed Data, Ecological Literature 1987 Transformed Image ,... i 18-Class Vegetation 1 l Map (J.R. Richardson) J GIS Analysis of Tree Islands Analysis of Tree Island Distribution Using Basic Models D ----'"'\ : I l __ ____ ; D Data Classified Image Modelling and Analysis Figure 3-1. Diagram of GIS landscape analysis using a geographic information system (GIS) The size distribution of the islands was evaluated in light of expected distributions according to hierarchy and other ecological theories. In the second approach several simple population models written in Basic programming language were used to generate size distributi ons using different general processes. The model-generated distributions were compared with the tree island size distributi on to identify which mathematical functions generated a similar pattern

PAGE 105

89 The informati o n gain ed from the GIS analysis and basic models was synthesized with the la nd scape data described in Chapter 2, to develop a conceptual model of vegetatio n succession and tree island growth. described in Chapter 4. This conceptual model is GIS Analysis of Tree Island Distribution Analysis of Size Distributi on As described in Chapter 2 a merged 1987 SPOT satellite image of Water Conservati o n Area 1 was studied at length, both in the field and the laborat ory A print of the image was taken into the field and used as a navigation tool and a guide for making ground observati o ns. A vegetation map classified from this image, ( Richardson et al. 19 90) was analyzed using a geographic information system (ERDAS 1991) to quantify the distribution of tree island s in Water Conserv ati on Area 1. The vegetation map was c omposed of 18 vegetation classes and was rectified to State Plane coordinates using a 30 ft by 30 ft pixel resolution. Methods A 4,440 ha area o f the n orth central portion of the vegetation map was analyzed for tree island distribution The area was chosen becaus e it contained a high concentration of islands and was free from th e influence of the perimeter canals Three of the 18 vegetation classes in the map were used to delineate tree islands: clas s 5 ("low stature tree ) class 7

PAGE 106

90 ( "high stature tree" ) and cl a s s 8 ( brush ) Wi t h in this area all contiguous tree islan d pixels in classes 5 7 and 8 were grouped as a singl e class and clumped together into indi v idual patches using ERDAS softwa re s (version 7 5) CL U M P program ( E RDAS 1 9 91 ) Pi x el c o unts were made f or e ach patch, and these data were imp o rted int o a spread sheet Because the islands numbered in the t ho usands, every island could not be graphed. Islands were grouped by size, and the data were gra p hed acc ording to size class B ec ause of a scaling phenomenon, small-si ze cla ss e s cont aine d m an y islands, while large-si z e classes contained just one island each. Smallsize classes were one pi xe l dif f eren t from e a ch o ther, while the largest size classes w ere hundreds of pixels a p art in size. F o r this reason, the data we re graphed in rank order b y si z e, rather as a histograms with pr e defined size intervals on the x a x is. Histograms can o bscure th e pattern of data smoo th i ng o ut or artificially generating discontinuities depend ing o n the interval's size in relati on t o t he r eso luti o n o f the data (Holling 1992, Gunderson 19 92) Howeve r, i t is important to note that all satellite dat a ar e gro u ped i n t o p re-defined si z e intervals, since th ey a r e measured in discrete p i xe ls. f o r e x ample, it is ver y u n l ikely any 10 pixel i s lan d s wo uld measure e x actly 1000 m 2 if su rv e yed using more precise m e thod s on the ground. In effec t t he rank order graphs in thi s wo r k are still histograms, with a si ze c a tegory increment of o ne pi x el, and with empty categ o ries no t s hown

PAGE 107

Results The frequency of the counted islands are graphically displayed by size in Figure 3-2. The frequency distrubution is shown in more detail, with only islands greater than 10 pixels (836 m 2 ) in Figure 3-3. A total of 3623 islands were counted. An unknown number of islands smaller than the resolution of the satellite image could not be counted. The island density was 81 islands/km 2 less than the 99 islands/km 2 sampled in the 1952 aerial photographs (see Chapter 2) The smallest islands detected were a single pixel in size (90 ft 2 or 83.6 m 2 ), while the largest island detected was 7522 pixels (nearly 63 hectares). The sizes of islands found within the counted area, ranked from smallest to largest, are shown in Figure 3-4. The mean island area was 43 pixels (3595 m 2 ). The median island area was 5 pixels (418 m 2 ) At the lower end of the range of island sizes, islands were found at every possible size increment. In other words, size classes of islands were separated by one pixel. However, above the 366-pixel size class (306 hectares), the gaps between the size classes became larger than one pixel, increasing along with island size. Above the 2238 -pi xel size class (1872 hectares), the gaps between size classes became much larger. A portion of this distribution is shown in more detail in Figure 3-5. The total area of all the islands found within each size class is shown in Figure 3-6. This graph combines both

PAGE 108

92 frequency and size informati on For example the 465 islands in the 2-pixel size class have a total area of 930 pixe ls. Number of islands in each size class 800 700 600 500 Number of 400 islands 300 200 100 0 1 100 Size rank Figure 3-2. Tree island distribution by size 90 80 70 60 Number of 50 islands 40 30 20 10 0 10 Number of islands in each size class (detailclasses 10 pixels and above) 100 200 Size rank Figure 3-3. Number of islands in each size class (detail) \ \ 268 268

PAGE 109

93 Island size classes 8000 7000 6000 Area of 5000 Discontinuity island size class, in pixels 4000 3000 2000 1000 0 1 100 Size rank Figure 3-4. Island size class distribution Area of island size class, in pixels 8000 7000 6000 5000 4000 3000 2000 1000 0 186 196 206 Island size classes (detail) 216 226 236 Size rank Figure 3-5. Island size classes (detail) 200 268 246 256 266

PAGE 110

94 Total area of all islands in each size class 8000 7000 6000 5000 Area 4000 in pixels 3000 2000 1000 0 1 100 200 268 Size rank Figure 3-6. Total area of islands within each size class Discussion The number of tree islands in this portion of Water Conservation Area 1 declined exponentially with increasing size (see Figure 3-2). Approximately half of the islands were 4 pixels or less, and one third of the islands were 2 pixels or less. Numbers stabilized with one island per size class at 70 pixels and larger (see Figure 3-3). The frequency distributi on of islands by size class did not resemble a Poisson distribution. A Poisson distribution would indicate a random process determining island size, as would be found in raindrops splattered on a surface. It instead resembled a frequenc y distribution typically found in a biotic population with high juvenile mortality Discontinuities in the size distribution o f entities in an ecosystem may indicate the influence of different processes on

PAGE 111

the landscape (Holling 199 2 Gund e rs o n 1 9 9 2) Changes in the slope or mathematical functi o n o f th e distributi o n at some size threshold(s) may mark the domain b o undaries o f different processes. Similarly, the age or size distribution of a biotic population may show discontinuities at p o ints in the life cycle where different processes (such as predati o n or reproduction) begin or cease to be influential. If small islands and large islands in Water Conservation Area 1 are f o rmed by different processes, (as suggested b y L o veless 1959 and Gleason and Stone, 1991:185), we would expect the frequenc y o r size distribution of small and large islands to e x hibit different characteristics. Such a discontinuity was suggested in the size class distribution of tree islands (see Figure 3-3). This discontinuity is not necessaril y e v idenc e o f tw o separate processes of island genesis; o ther processes influencing island structure which are operating at different scales o r were activ e during different periods in the past could have had same effect. Some processes of island e x pansi o n or reducti o n that might affect islands of different si z es differentl y are v egetation succession, peat accretion, e r o si o n b y s he et fl o w o f water, faunal activity, and vegetati o n disturbanc e b y wind and fire. Some of these processes op e rate m o st actively a t t he edges of islands (succession, peat ac c r e ti o n, and erosio n). B e cause the ratio of edge to island ar e a d e cr e a ses wit h i n c reasing island size, the effects of th e s e p rocesses may change w it h isl a nds size. However, this ch a n ge should be gradual a n d n ot ca u se a discontinuity in the s ize distribution Other processes ( w i n d

PAGE 112

96 fire, and faunal activity), may have different effects above and below some threshold related to island size. A threshold triggered process such as fire could produce a discontinuity in the distribution of island sizes. One theory that addresses the relationship between processes and spatial extent is hierarchy theory. A key principle of hierarchy theory (O'Neill et al. 1986, Alan and Starr 1982) states that higher order processes operate at larger spatial scales and over longer time scales than lower level processes. Higher order processes control or organize lower level subsystems. While ecologists have defined hierarchical levels in terms of temporal and spatial scales, H. T. Odum (1983) described hierarchies based on ecosystem energetics. Hierarchical levels are defined by energy quality. The quality of energy is measured by the amount of solar energy embodied in it. For example, in a simple hierarchy of plant, herbivore and carnivore, energy quality increases at each higher trophic level. A single calorie of plant material requires 10 calories of solar energy for its manufacture. To make a single calorie within an herbivore requires 10 calories of plant energy, or the equivalent of 100 calories of solar energy. The herbivore therefore has a higher quality energy than a plant. Odum's exponential theory of energy hierarchy predicts that while energy quality increases at each higher hierarchical level, the amount of calories present (not embodied) in each level decreases (See Figure 3-7).

PAGE 113

97 Since each energy transformation step requires utilization and dispersion of potential energy the total energy flows decrease as one passes work from step to step. Conversely, the embodied energy per unit of actual energy increases down the chain. The less actual energy remains, the more embodied energy has been involved. (Odum 1983:270) Energy quality and hierarchical level also relate to the spatial scale of processes or entities: Flows of low-quality energy are abundant and widely dispersed, and individual units [receiving that energy] are small in size. Higher-quality units and their flows, although less in total energy flow, are more concentrated and each unit is larger in size with a larger territory from which it receives energy and feeds back its actions. (Odum 1983:15) Examples of this phenomenon are the trophic structure of ecosystems, where primary producers outnumber consumers; the age structure of stable populations, where small, young individuals outnumber larger adults; and the occurrence of earthquakes, where weak quakes outnumber strong ones Tree islands in Water Conservation Area 1 are distributed exponentially by size, with small islands outnumbering large islands. energy hierarchy in the island population? Does this indicate an An energy hierarchy would not apply to island patches unless larger islands received inputs of higher quality energy than smaller islands. Unlike mice versus bobcats, or nestlings versus adults, large islands do not have a different strategy for obtaining energy than small islands. If the amount and quality of energy required to create one unit area of island is the same regardless of island size, then all islands are at the same hierarchical level. Th ere is no evidence that large islands are receiving higher quality energy than smaller islands

PAGE 114

98 at the present time. It is more likely that islands are at the same hierarchical level, and a higher order disturbance process is organizing them into the sizes we observe. Energy Quantity Number of Individuals Figure 3-7. /~ 1 2 Hierarchical Levels 3 4 I 5 I Energy Quality Size of Individuals Size of Territory Diagram of energy hierarchy I Island Population Model Analysis 6 I / To better understand the observed size distribution of tree islands in Water Conservation Area 1, simple population models were written to generate island distributions using different growth functions. The models were based on the assumption that islands appear and grow in size over time and do not "die." The models were used to illustrate the curves

PAGE 115

99 produced by different growth functions when the data were graphed, and to explore the interaction of the scales of island growth and the unit used to measure growth. Methods Two simple programs were written in Basic programming language. The first program generated a population distribution based on either a constant growth model or an area-dependent growth model, (see Figure 3-8). The program generated a size distribution by "growing" an island and recording its size at each iteration (time step). It assumed that conditions for growth remain constant over time; i.e. a small island in the present grows the same way a large island did in the past when it was small. It also was assumed that new islands were created by peat batteries as described in Chapter 1. The number of new islands introduced to the population could be altered by the program, but most model runs were made with a constant number of islands added to the population each time period, to narrow the number of factors influencing the distribution. were 1 m 2 in size. New islands The second program generated a population distribution using a growth function based on island edge (see Figure 3-9) Otherwise, the program was similar to the one described above. The model assumed islands were circular and calculated island edge based on the circumference of a circle Both programs rounded the output data into integer values to simulate the effect of measuring island areas in pixels. Pixels in a satellite image are classified based on their

PAGE 116

100 spectral values. The spectral values of a pixel integrate the reflectance of all the vegetation in the scanned unit. The dominant vegetation therefore has the greatest influence on pixel spectral values. If the amount of woody vegetation within a scanned unit on the ground reaches a certain threshold, a pixel will be classified as an island class rather than a nonisland class. The area of woody vegetation cannot be measured in fractions of a pixel. In the same way, the area values for islands in these basic models are rounded to the nearest integer. Models were run using different coefficient values, generally for 1000 iterations. Frequency distributions and graphs were generated from the output data using spreadsheet software. Results The constant growth model was first run with the simplest scenario: constant recruitment of one new island per iteration (n=l), and constant incremental island growth of one pixel per time period (g=l). This function is analogous to the following: every island has one alligator hole next to it, regardless of island size. The activity at every alligator hole generates the same amount of disturbed peat, which drifts to the adjacent island, adding to its area.

PAGE 117

101 REM MODEL TO GENERATE ISLAND DISTRIBUTIONINCREMENTAL OR REM AREA-DEPENDENT GROWTH REM i iteration (or age) REM n = number of new islands per iteration REM g = fi xe d area of growth added per iteration REM L gr owth coefficient as a function of island area REM s si ze class REM a t ot al area of islands in size class CLS OPEN "c:\treebas\tree.dat" FOR OUTPUT AS #1 INPUT "enter ma x iterati on and number of islands "; k n INPUT "enter island growth increment (g) for arithmetic growth, or enter 0"; g INPUT "enter island growth coefficient (L) for size -dependent growth, or enter 1"; L PRINT #1, "ma x iteration PRINT #1, ''number of new PRINT #1, "island growth PRINT #1, "island gr ow th PRINT #1, "Iterati on IF g = 0 THEN s = 1 FOR i = 1 TO k s = (s L) + g == ", k isl ands per iteration (n) = ", n increment (g) =", g factor (L) = ", L s n a II PRINT #1, CINT(i), CINT(s) CINT(n) (CINT(n) CINT(s)) PRINT, CINT(i), CINT(s) CINT(n) (CINT(n) CINT(s)) NEXT i CLOSE #1 Figure 3-8. Basic model code to generate island distributions: incremental and area-dependent growth

PAGE 118

102 REM Model to generate island distributionedge-dependent growth REM i iteration (or age) REM n number of new islands per iteration REM s size class REM C circumference, REM a total area of islan ds in size class CLS OPEN "c:\treebas\edgear.t xt FOR OUTPU T AS #1 INPUT "what is the max iteration"; k INPUT "what is the number of new islands per iteration"; n INPUT "what is the initial size class"; s INPUT "growth factor (x)"; x PRINT #1, "ma x iterati on k I PRINT #1, "number of new islands per itera ti on PRINT #1, "island growth fact or (x) ", PRINT #1, "Iteration s PRINT II II II S II II C II lln", (n s) ; l f f c = (SQR(s / 3.141592)) 2 3.141592 FOR i = 1 TO k X n " I (n) = II n a PRINT #1, CINT(i), CINT(s), CINT(n) CINT(n s) PRINT i, s, c, n, ( n s ) s s + ( x c) c = (SQR(s / 3.141592)) 2 3.141592 NEXT i CLOSE #1 II Figure 3-9. Basic model code to generate island distributions: edge-dependent growth

PAGE 119

100 90 80 70 60 Area uni ts 5 0 40 30 20 10 103 Island size classes n=l, g=l 0 .+--==-----1-------+-----+-----+-------t Figure 3-10. classes 0 20 40 60 80 100 Size rank Distribution from incremental growth: island size Island size classes and total area in each size class n=l0, g=l 1000 900 800 700 Number 600 or area 500 units 400 300 200 Size class 100 0 0 20 40 60 80 100 Size rank Figure 3-11. Distribution from incremental growth with high constant island recruitment

PAGE 120

104 This function generated a linear size distributio n (see Figure 3-10). There was just one island per size class, so if the total area per size class were graphed, the graph would be identical to Figure 3-10. If the growth increment (g) was increased, the function generated a line with a steeper slope. When more islands were added per iteration ( n=l0), the size distribution remained th e same, but the total area within size class increased at a ste eper slope (Figure 3-11) When growth is incremental, but new islands are added in changing numbers rather than at a constant rate, the total area in each size class is affected. The model was run to simulate island recruitment that was high when older islands were formed and decreased in a linear fashi on over time (n = n+l). Such a situation might occur if battery formation decreased ov er time as the proliferating islands occupied more and more of the wet prairie area where batteries form. Figure 3 -1 2 shows the results of this model run. The total area per size class was an e x ponential function (see Figur e 3 -13 ) The model was ne x t run to simulate island recruitment that was low when older islands were f ormed and increased for younger age classes, (n = n-1). Such a situation might o ccur if longterm peat buildup, changing h y droperiods, or some other factor increased the rate of battery formation over time. Figure 3-14 shows the results. In this case the total area per size class was quadratic functi on (Figure 3-15).

PAGE 121

105 Island size c l asses and number in each size class n=(n+l) g=2 2 00 180 160 140 1 2 0 Number or 100 area units 80 60 40 2 0 0 0 20 40 60 80 Size rank 100 Figure 3-12. Distribu tion from incremental growth with increasing island recru itment: island size classe s a n d number s Area units Total area of islands in each size clas s n=(n+l) g=2 20000 10000 0 20 40 60 Size rank 80 100 Figure 3-13. Distribut io n from incremental growt h wi th increasing island recru itment : total area in each size class

PAGE 122

106 Island size classes and number in each size class n=(n-1), g=2 200 150 Number or 100 area units Area of size 50 Number of islands 0 0 20 40 60 80 100 Size rank Figure 3-14. Distribution fr om incremental growth with decreasing island recruitment: island size classes and numbers Total area of islands in each size class n=(n-1), g=2 5000 4000 Area 3000 units 2000 1000 0 0 20 40 60 80 100 Size ran k Figure 3-15. Distribution fr om incremental growth with decreasing island recruitment: total area in each size class

PAGE 123

107 The model was next run using size-dependent growth instead of constant growth. This growth function would occur if the rate of island growth increased as the island grew larger. Size-dependent growth could occur if the probability of an island intercepting floating peat increased with increasing size, or a greater proportion of core tree area speeded the rate of island growth. The model was run with the size-dependent growth function, constant island recruitment (n=l) and size-dependent growth equal to 1% of the island's area per iteration (L=l.01). With the size-dependent growth function, a scaling effect emerged, as the model generated islands with areas that were not an integer value. These values were rounded off and put into integer bins. While the resolution of the analysis was fixed at integer intervals, the scale of island growth changed between large and small islands. When islands were small, the amount of growth added each iteration was small enough that several iterations in a row produced islands of the same integer size. When plotted, islands from multiple iterations were grouped into the same integer size class. This scaling effect is apparent in Figure 3-16. Large numbers of islands are grouped into the smallest size classes even though only one new island was added to each iteration. While the growth function of this model produced an exponential curve (Figure 3-17), the curve was forced into a linear distribution at the small scale end of the scale due to the integer constraint (Figure 3-18) When these effects are

PAGE 124

Number of islands Figure 3-16. numbers Area units 60 50 40 30 20 10 0 108 Number of islands in each size class n=l L=l.01 1 51 101 151 201 251 301 351 401 451 501 551 Size rank Distribution from area-dependent gr ow th: island 10000 9000 8000 7000 6000 5000 4000 3000 2000 1000 0 1 Island size classes n=l, L=l 01 51 101 151 201 251 301 351 401 451 501 551 Size rank Figure 3-17. Distribution from area-dependent growth: island size classes

PAGE 125

Area units Figure 3-18. size classes Area units 109 Island size classes (detail) n=l L=l.01 250 200 150 100 so 0 1 51 101 151 Size rank Distributi on fr om area-dependent growth: iland (detail) 10000 9000 8000 7000 6000 5000 4000 3000 2000 1000 0 1 Total area of islands in each size class n=l L=l 01 51 101 151 201 251 301 351 401 451 501 551 Size rank Figure 3-19. Distributi on from area-dependent growth : total area in each si ze clas s

PAGE 126

110 Total area of islands in each size class (detail) n=l, L=l.01 Area units 200 150 100 50 0 1 11 21 31 41 51 61 71 81 91 Size rank Figure 3-20. Distribution from area-dependent growth: total area in each size class (detail) combined in the calculation of the total area per size class, the curve is clearly perturbed at the small scale (Figures 3-19 and 3-20). Finally, a model using growth based on the edge length of the island was run, using constant island recruitment. This growth function could occur if the peat-building and vegetative reproduction of sawgrass at the perimeter of islands were the main factor in island growth rather than the accumulation of floating peat and the litter contributed by trees and other species in the interior of islands. The growth coefficient (x) was the unit area added per unit of edge length. The model assumed islands were circular, and estimated edge length each iteration by calculating the circumference of a circle with the same area as the island. A run of the model using .14 units of area added per unit of edge length (x = .14) generated Figure 3-21. The curve of

PAGE 127

111 the distribution is exponential, but more gradual than the areadependent curves. The scaling effect is less using this function; islands were grouped only in the five smallest classes. 8000 7000 6000 5000 Area 4000 units 3000 2000 1000 0 1 Island size classes n=l, x= .14 51 101 151 201 Si ze ran k 251 301 Figure 3-21. Distributi o n from edge -d epende nt growth: island size classes Discussion Of all the distributi ons generated by the basic models, the size-dependent growth model most resembles t he data from the GIS analysis described earlier in this chapter The scaling phenomenon illustrat ed by this size dependent growth model also appears to be affecting the size class distribution in Richardson et al.' s map, (s ee Figures 3-3 and 3-6) In both cases there are high numb ers of islands in the smaller size classes, (Figur es 3-3 and 3-16) o nly o n e island apiece in the larger classes, and the differenc e b et w ee n th e island size

PAGE 128

112 classes increases as the islands get larger (Figures 3-4 and 31 7) This scaling phenomenon is important to consider when using the sizes and numbers of entities as indicators of ecosystem process domains, especially if satellite imagery is used. The basic model of ar ea -d ependent growth illustrates how a single spatially-dependent function producing a flat frequency distribution (one island per iteration), appears to produce an exponential frequency distribution when measured at too coarse a resolution. This frequenc y distribution, when combined with the island size distribution, exhibited noisy var iation at the low end of the scale. This variation could be misinterpreted as a discontinuity caused by the influence of a second process (Figures 3-19 and 3-20). The distribution of the islands in Richardson et al.'s map shows similar variation at the low end of the scale (Figure 3-6). phenomenon. This is also likely to be a scaling

PAGE 129

CHAPTER 4 LANDSCAPE MODEL Introduction This chapter describes two models of the landscape of Water Conservation Area 1: a conceptual model of island formation and vegetation community change, and a spatial model of landscape pattern. The conceptual model is a synthesis of the information presented in the first three chapters of this work, made to identify the most important factors in island formation and vegetation community change. From the conceptual model, a spatial model with a narrower scope was developed. The spatial model was used to project the relationship between landscape pattern and disturbance by fire in Water Conservation Area 1 over time. The model simulated vegetation succession, the effect of fire on vegetation, and post-disturbance vegetation succession. The landscape patterns generated by the model expressed the interaction of these processes. Conceptual Model of Tree Island Growth and Landscape Pattern A conceptual model of island growth and vegetation community change in Water Conservation Area 1 was developed as a precursor to a spatial model of landscape pattern. The conceptual model united the many pieces of information gathered in the first three chapters of this work: ecological data 113

PAGE 130

114 gathered and analyzed by other researchers landscape data collected both remotely and in th e field and data analyses aimed at a theoretical understanding of landscape pattern (see Fig ure 4-1). The synthesis of this information led to the conceptualization of a system that could generate and maintain the landscape pattern we observe in Water Conservation Area 1. 1987 Transformed Image Field Data and Ecological Literatur e I : 18 Class Veget ation !Map (J.R Richardson) \ GIS Analysis of Tree Island Sizes GIS Analysis of Tree Island Clumping Analysis of Tree Island Distribution Using Basic Models 1 992 Ae rial Photographs 1 952 A e rial Photogra ph s D r-~ ~-~ D 1989 Raw Panchromatic Image D ata C lassified Image Modelling and Analysis Figure 4-1. Information us ed to develop the conceptual model of veget ation change and island dynamics in Water Conserva tion Area 1

PAGE 131

115 This conceptual model is described below. The description summarizes information already presented and referenced earlier in this work. For more complete information, the reader is referred to these earlier sections. Processes Contributing t o Island Growth The conceptual model was limited to the main processes contributing to island growth and to the pattern of all vegetation communities in the interior of Water Conservation Area 1. These processes are fire, peat accumulation, and vegetation succession. These three processes are in turn affected by a number of factors, the majority of which are driven by climate and water management (see Figure 4-2). Vegetation succession Vegetation succession is a positive force in the formation and growth of tree islands. Vegetation succession is the process by which land passes through a succession of vegetation community types over time (seral stages) (Chapter 1:42). Each seral stage, as it matures, alters the conditions of its environment until the environment is more favorable to the next seral stage. Since tree islands in Water Conservation Area 1 are defined by the presence of the bayhead vegetation community (Chapter 1:25), vegetation succession is critical to the dynamics of tree islands. In Water Conservation Area 1, the environmental conditions driving succession are water level and hydroperiod (Chapter 1:41). Water level and hydroperiod have

PAGE 132

116 CLIMATE PATTERN drought cycle li9htning \ ISLAND weather ,~ conditions----~ human ignition water le.vel~'lFIIRE fuel load microclimate WATER MANAGEMENT seed sources / sheet flow battery --------=-, PEAT ACCUMULATION alligator activity litter production Figure 4-2. growth Diagram of processes contributing to tree island positive or negative effects on succession. Lower water level and shorter hydroperiod move succession forward from herbaceous vegetation to woody vegetation. Higher water level and longer hydroperiod move succession backward from woody vegetation to herbaceous vegetation. Vegetati on itself influences water level and hydroperiod, mainly by building peat (Chapter 1:20). Peat accumulation moves succession forward by raising elevation, which reduces water depth and hydroperiod. Hurricanes and fires move succession backwards by destroying or damaging vegetation. Plant reproduction is essential to the process of succession, whether by seed or vegetative reproduction. Changes in seed sources, nutrient le vels and the ability of plants to

PAGE 133

117 reproduce vegetatively can have a positive or negative effect on succession, changing the species composition of plant communities and the competiti ve balance between species. These changes may redefine the series of vegetation commun ities in a successional sequence. Peat accumulation Peat accumulation is a positive force in the formation and growth of tree islands. Peat accumulates on islands via litter production or drifts in from outside the island. When litter from island vegetation accumulates faster than it decomposes, it builds peat. Alligator acti vity disturbs peat, which can drift and collect at island margins (Chapter 1:28). Peat batteries may float and resettle, creating mounds that may become islands (Chapter 1:26). Sheet flow of water across the landscape has a positive or negative effect on island growth: eroding, transporting or depositing peat (Chapter 1: 22) Fire Fire is a negative force in the formation and growth of tree islands. Fire destr oys or damages vege tati o n and can reduce elevation by burning peat moving succession backwards (Chapter 1:33). The pr obability of a fire and the intensity of fire is determined by fir e potential. Fire potential is determined by a variety of factors, including the density and moisture content of the fuel load, atmospheric humidity, wind speed, and soil m o istur e (Chapter 1 : 30) Fuel moisture and soil moisture are a functi on of water level which is determined by water management and the dr oug ht cycle Atmospheric humidity

PAGE 134

118 and wind speed are a function of current weather conditions and the microclimate of a sit e (Chapter 1:36). Fire potential combines with a source of ignition to produce a fire. Ignition may come from a human source or from lightning. If fire potential is low, ignition will produce only a limited fire affecting less than a hectare. If fire potential is very high, fires may pr opagate quickly affecting tens of thousands of hectares (Chapter 1:31). Fire potential in Water Conservation Area 1 is high when the water level is low, particularly at the end of the dry season. During droughts, the conditions in Water Conservation Area 1 are especially favorabl e for the initiation and propagation of fire. With high evapotranspiration and no rainfall, the water level falls below the peat surface. This allows a shift in the species composition and density of the wet prairies. Panicum hemitomon and annual herbs sprout among the sparse cover of Eleocharis and Rhynchospora species, increasing the fuel load to a level capable of carrying a fire. When ignition occurs, fire can then sweep across the landscape (Wade et al. 1980). Vegetation and microclimatic variables that increase fire potential-fuel moisture content, understory vegetation density, wind speed, temperature and humidity-are lower under tree island canopies than in adjacent vegetation communities (Chapter 1:31). As observed during a large severe fire in Water Conservation Area 1, fire can sweep across the landscape, and slow when it reaches the lower fire-potential large tree

PAGE 135

119 islands. The fire front moves around rather than through the island. While the front continues on islands burn from the margins inward until the moister interior condit ions extinguish the fire. While small islands do not have eno ugh interior area to generate a wetter microclimate, the cores of large islands are protected from fire. Tree islands The tree islands in Water Conservation Area 1 are composed of peat, and are built up in the process of vegetatio n succession (Chapter 1: 26) Tree islands may form on peat batteries displaced in wet prairie or they may be created more slowly by litter accumulati on peat buildup and succession from lower seral stage vegetation If the battery-islan d theory is correct initially a peat battery, alligator, or o th er disturbance creates a small topographic high in the wet prairie. The reduced hydroperiod at this higher elevation allo ws colonization b y sawgrass (Cladium jamaicense). Senescent biomass from the sawgrass accumulates and builds up the peat. In the absence of disturbance by fire, colonization by woody plants follows First shrubs (Myrica cerifera and Cephalanthus occidentalis) create a brush island Then trees, (dominated by Persea borbonia and Ilex cassine), become established and create a tree island Tree islands could also develop without a battery as a catalyst. With time and t h e absence of disturbance vegetation continuously builds peat If elevation was lowered by disturbance in one area, adjacent undisturb e d areas would be

PAGE 136

120 left relatively higher and drier. The process of succession would then progress as on a displaced peat battery, forming a tree island. As succession proceeds over time, tree islands spread outward, increasing in area. On established tree islands, dead biomass accumulates from litter originating on the island and from floating debris. This accumulation raises elevation at the island's perimeter. Narrow bands of sawgrass and brush ring the perimeter of tree islands, where sawgrass has spread outward into wet prairie, and brush has colonized the sawgrass (Figure 2-3) This configuration suggests that tree islands grow in area as a function of time. Conceptual Model The many processes contributing to tree island growth listed in Figure 4-2 were narrowed into a simpler and more specific conceptual model (see Figure 4-3). The model is an abstraction of the dynamics of vegeta tion communities in the interior of Water Conservati on Area 1. This conceptual system cycles through three phases: 1) vegetation succession and the growth of tree islands; 2) fire behavior; and 3) post-fire vegetation recovery. These phases are described below. The landscape processes associated with the canals at the perimeter of Water Conservation Area 1 (the increase of nutrients, water depth and hydroperiod) are not addressed in the model.

PAGE 137

battery-forming conditions? PEAT BATTDY + + +++++++ SAWGRASS PATCH peat burn? yes no 121 no yes no no tree peat burn? 111orta.lity? yes yes no peat J burn? yes Figure 4-3. Conceptual model of vegetation and island dynamics within the interior of Water Conservation Area 1 Phase 1: vegetation successi on In the absence of disturbance, vegetation succession follows a sequenc e of seral stages driven by hydrologic conditions, as indicated b y the plus-sign track in Figure 4-3. Succession moves vegetation to the right along this track (from wet prairie to sawgrass, to brush and to trees) and disturbance by fire mov es it to the left The transition from wet prairie to sawgrass and brush may be accelerated by the formation of peat batt eries

PAGE 138

122 Phase 2: fire When conditions are favorable for fire and ignition occurs, fires will burn (indicated by the dashed line in Figure 4-3) The damage to vegetation communities by a fire depends on the fire's severity and the vulnerability of plant species in each community. Damage is portrayed in Figure 4-3 by a series of questions and pathways. As shown on the right-hand side of Figure 4 3, if a tree island is above some size threshold, the trees in its interior escape burning and remain intact. On tree islands below this size threshold, and at the margins large islands, trees are burned. If the burn kills only the above-ground portion of the trees, the trees will resprout in a brushy form, creating a brush island. With the absence of a tree canopy, shrub species will also germinate on the brush island. If the burn is severe enough to cause tree mortality, but the peat itself does not burn below a thin layer at the surface, then a tree island will become a patch of sawgrass and ferns. If the peat burns deeply, reducing the peat elevation, the island will become wet prairie. This is a rare event in the interior of Water Conservation Area 1. Patches of shrubs (brush islands) do not have the moist microclimate of large tree islands, so entire shrub patches will burn in a fire. If only the above-ground portion of the shrubs are killed, then they will resprout, keeping the patch in a shrub community. If the fire causes shrub mortality but the

PAGE 139

123 peat does not burn, then th e patch will become sawgrass. peat burns deeply, the patch will become wet prairie. If the Likewise, if a sawgrass patch burns but the fire does not burn into the peat, the sawgrass will resprout. If it does burn down into the peat, causing sawgrass mortality, then the patch will become wet prairie. Phase 3: post-fire vegetation recovery Vegetation recovers after a fire, either by the regrowth of surviving individuals or by the germination of new individuals. In the interi or of Water Conservation Area 1, where environmental conditions have changed less than at the perimeter of the area, the recovery of vegetatio n follows the same succession track in shown Figure 4-3. At the perimeter of Water C o nservation Area 1 changes in the environment including the invasion of exotic species, nutrient enrichment, and alterati ons of h yd r o period, have changed the successional pathway followed after a fire or other disturbance. This conceptual mod e l does n ot apply to the perimeter area. Fire and landscape pattern If the process of fire disturbance is influenced by the amount of core area of tree islands as t hi s model describes, then fire acts to organize land scape pattern If tree islands above some size thresh old ar e spared the burning of their interiors, their undisturbed cores will promote m ore rapid successional recovery and continued expansion than that occurring on completely burned islands If they expand at a

PAGE 140

124 rate greater than their setback by fire their vulnerability to fire will further diminish in a positive feedb ack. In time, repeated fires will create a population of tree islands divided between islands whose growth is controlled by fire, and larger islands with a core imper vious to fire and a higher growth rate. Such a situati on could produce a discontinu ity in the distribution of tree islands similar to that found in the map of Richardson et al. ( see Figure 3-4 ) This theory cannot be tested in the field in Water Conservation Area 1, because the temporal and spatial scales of the variables are t oo great to measure within a human life-span and a reasonable budget. However, a spatial model can be used to show whether this dynamic system can be reproduced in abstract. Spatial Model of Vegetation Succession and Disturbance by Fire A spatial model of the landscape dynamics of the interior of Water Conservati on Area 1 was developed using a cartographic approach and a geographic inf o rmati on system (see Chapter 1:12). Although cartographic models have usually been used to predict an outcome in a single step this model used a geographic information system iteratively to show landscape change over an extended time under different fire regimes As shown in Figure 4-4, the foundati on of the model was a map o f the vegetation of Water Conser va ti on Area 1 by successional class ( see Chapter 2:81-86). The functi ons of the m od el were based on the conceptual model in Figure 4 3 The model was used to generate

PAGE 141

125 landscapes subjected to fires at different intervals, and at different levels of severity The composition and spatial heterogeneity of the landscap es generated by the model under these different scenarios were compared. An unsuccessful attempt was als o made to compare landscapes generated by the model with independent data from a 1990 SPOT satellite image. ----------------------, 24Class Vegetation j Succession Map : -------------------' GIS Model of Disturbance and Vegetation Succession Model Analysis: Fire Intervals and Fire Severity Hypotheses and Conceptual Model 1990 Multi. and Pan. Images 1990 Attempted Comparison With Independent Data ...._ _____ -t Transformed Image D Data :--1 Classified l __ __ ; D Modelling Analysis Figure 4-4. Flow chart of spatial modeling and analysis Model Design Image and The spatial model simulated succession fire and vegetation recovery using a geographic information system (GIS).

PAGE 142

126 A succession component of the model moved vegetation pixels on a map through a sequence of seral stages by increasing their designated successional class. The spreading of sawgrass into wet prairie via vegetative reproduction also was simulated, based on the juxtaposition of pixels of these two seral stages. A burn component in the model changed pixels from one successional class to another based on their current class and their distance inward from the edge of an island. Vegetation recovery was simulated when the succession component of the model ran after a fire. Hydrologic influences on both fire and succession were not explicit in the model. They were held constant to isolate the effects of fire. The generation of new islands by peat batteries or alligators was not included in the model, because there were insufficient data documenting these sources of island formation in the Everglades. GIS software The model was constructed using ERDAS Inc.'s VGA ERDAS software, version 7.5 (ERDAS 1991). The main GIS programs used to run the model were the SEARCH (spatial analysis) and GISMO (GIS Modeling) programs. The ERDAS SEARCH program calculates the distance from each pixel to other pixels of a specified class or classes. The SEARCH program was used in each model iteration to generate maps of spatial information from the successional vegetation base map. The ERDAS GISMO program uses a set of model rules to produce model output based on conditions found in different

PAGE 143

127 maps. The user specifies logic statements or equations that define the model rules. The GISMO program considers each pixel in a base map, and que ries other map layers to find data values for the same location. These values are then run through the model rules, producing a new value for the pixel The results are saved as a new map. The SEARCH and GISMO programs were both required to produce a single iterati on of the model Each model iteration represented a 10-year time step When the programs were run repeatedly in batch mode, the model simulated landscape change over an e x tended time. BURN NO BURN BORDERS FIRE rgj = SWITCHES IN BURN YEARS BASEMAP GISMO NEXT BASEMAP Figure 4-5. Diagram of model structure Model structure DISTANCE The structure of th e model is shown in Figure 4-5 At the start of a model iteration (the top of Figure 4-5) a base vegetation map w as run through the SEARCH program to create two

PAGE 144

128 spatial information layers. The first was a "borders" map, that defined the boundaries between wet prairie and early successional sawgrass. In the borders map, all pixels in the "slough/sparse wet prairie" and "wet prairie" classes (classes 1-14) that were adjacent to a higher class (classes 15 and higher) were coded as 1. Wet prairie pixels were rarely found adjacent to shrub or tree classes. Pixels in classes 15 and higher were coded as 0. Pi xels in classes 1-14 that were not adjacent to a higher class were coded as 2 If a fire was set to burn during the iteration, a second spatial information layer, "distance," was also created from the base map. This layer showed the distance inward from the edges of tree islands. Slough/wet prairie, wet prairie, and sawgrass classes (classes 1-20) were coded as 0. Pixels in the shrub and tree classes (classes 20 and above) were then re-coded as their distance in pixels from class 0. distance inward from island edges. The result was a map of For example, a pixel coded as 2 was 20 meters into the interior of an island from the island edge. A fire layer was also specified for use by the model. If the iteration included a fire, fire was switched on by setting the fire layer to "burn" (1) If it did not include a fire, the fire layer was switched to "no burn" (0). After generating the spatial information maps and setting the fire switch for an iteration, the GISMO program produced the iteration's outcome. The outcome was a new vegetation map which became the base map for the next iteration. The succession

PAGE 145

129 rules within the GISMO model reassigned the successional class of a pixel in the base map, based upon its present successional class and its location on the "borders" layer. In fire years, the model rules reassigned a pi xel 's successional class to a lower class, based upon its present class and its location on the distance layer. Model base map The base map used for the initial conditions of the model, referred to hereafter as the "starting base map," was produced from a merged data set of April 1987 SPOT panchromatic and multispectral satellite imag ery The processing o f this image is described in detail in Chapter 2. A 5000 hectare subset of the satellite data set (see Figure 2-2) was classified to produce a 28-class unsupervised classification. The 28 classes were then identified as vegetation classes and ranked in order of a successional progression (see Table 2 -6 ) Aerial photography and a vegetation map classified from the same data using 1987 ground truth (Richardson et. al 1990) were used as a reference for the class identificati on s. The staring base map was classified into 28 successional classes rather than just the five basic vegetation types or seral stages shown in Figur e 1-8. This was necessary because the conversion of an entire pixel from one sera l stage to the next higher stage is not compl e t ed in a single decade (the time step of the model). Seral stages were subdivided into a number of successional class es representing gradations between stages

PAGE 146

130 The number of classes in a seral stage depended on the estimated duration of that stage (see Table 4-1). Table 4-1. base map Subdivisions of seral stages in model's starting Seral Stage Successional Duration of Stage Classes Slough/Sparse wet 1-9 90 years prairie Wet prairie 10-14 40 years Sawgrass 15-19 50 years Brush 20-23 40 years Trees 24-28 no limit Model rules and rates The rates built into the model were estimated from hydrologic, vegetation and fire data in the ecological literature. Assumptions regarding vegetation succession and the spatial distribution of vegetation were based on the landscape information collected in this work (see Chapter 2), and on other studies of Everglades vegetation (Loveless 1959, Pesnell and Brown 1977, Olmsted and Loope 1984, Duever 1984, Kushlan 1990), particularly two studies of vegetation community distribution and environmental variables in Water Conservation Area 1 (Pope 1989, Richardson et al. 1990). Rates of vegetation succession were based on studies of the Everglades and the Okefenokee swamp (Craighead 1971, Deuver and Riopelle 1983, Glasser 1986). Ranges of fire frequency were based on fire record in Everglades

PAGE 147

131 National Park (Taylor 1981). Rules for fire behavior were based on observations of a fire in Water Conservation Area 1 (see Chapter 2) and on fire ecology literature (Wade et al. 1980, Anderson 1982, Rothermel 1983, Burgan and Rothermel 1984, Andrews 1986, Andrews and Chases 1989). The initial model rules, before verification and finetuning of the model, are shown in Table 4-2 They were designed to simulate the conceptual model shown in Figure 4-3. If there was not a fire during an iteration (i.e. burn= 0), pixels in each successional class advanced to the next higher successional class. Because sawgrass was assumed to spread vegetatively into wet prairie, successi on was accelerated in wet prairie pixels adjacent to sawgrass pixels. This was accomplished by advancing all wet prairie pi xe ls defined as 1 on the borders map to the lowest sawgrass class (class 15). Table 4-2. Initial landsca pe model rules Succession without fire Condition Result fire is 0 and border is 0 veg= veg+ 1 fire is 0 and border is 1 veg= 15 fire is 0 and border is 2 veg= veg+ 1 Succession with fire Condition Result fire lS 1 and distance >= 3 ve g = veg+ 1 fire is 1 and distance < 3 and v e g = 1 veg < 5) fire is 1 and distance < 3 and ve g = ve g 4 veg >= 5

PAGE 148

13 2 If fire was set to burn during an iteration, succession occurred only in core island pi x els over 30 meters from an island edge (i.e. with a value greater than three on the distance map). All other pi x els w e re reassigned three classes lower, to simulate the net effect o f succession during the iteration and damage from the fire. Model Verification The initial model was run t o check its performance, in order to adjust the rules or rates s o that it reproduced the assumptions that went into its design. Tests must be made to ensure that the model reproduces accurately the data that were used in its construction. This process of verification, as it is sometimes called, is tautologous in principle but nevertheless a necessary check that the mechanisms o f the m o del are in fact doing what the modeller thinks the y are d o ing. (Kitching 1983:42) The model was first run f o r 2 0 iterations, covering a 200year period of time, without fire. The m ov ement of pixels from class to class each iteration was carefull y tracked to verify that the model rules were e x hausti v e and e x clusive, and were functioning as planned. To facilitate the analysis of the m o del's output, the results of each iteration were summari z ed by seral stage. Pixels in the 28 successional classes in each output base map were grouped into the five seral stages using the ERDAS RECODE program. All statistics presented for the model simulations in the remainder of this chapter are cal c ulat e d fr o m base maps

PAGE 149

133 grouped into these sam e ser a l stages. For each iter at io n i n the initial model run with o ut fi re the percentage o f th e base ma p in each seral stage was calculat ed u si n g th e E R DAS BSTATS program and plotted (Figure 4 6) 100 e Slough/sparse wet prairie 90 ---Wet praire Sawgrass )( Shrubs --Trees 80 70 60 50 40 30 20 I I ------/ / I' I I I 0 +----.--......=~;,;;:::$~i>-~-r-----,-.....::::::~-....... ---4~~~"-"41-----"~ 0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 160 Iteration Figure 4-6. Composition o f t he ba se map d uri n g the initial run of the spatial model (percentage o f map in e ach seral stage) As succession in th e m odel progressed during th e simulation, the dominant type in the landscape chan g ed f ro m w et prairie to sawgrass, th e n t o shrubs and finally to trees. By the 160-year iterati o n, tr ees occupied 100 % of t h e base m ap The progressi o n o f s u ccession in the initial model wa s unrealistically rapid, particularly the change from slough and wet prairie to sawgra s s. T he landscape became less heterogeneous as much of the area was converted to a single type, which progress e d t h rough the successional series The source of this problem w as identified as the acceleration of

PAGE 150

134 succession of both wet prairie pixels and slough/sparse wet prairie pixels ne xt to sawgrass. The acceleration of slough/wet prairie succession was not realistic, since sawgrass is not likely to spread into slough. In the improved model, this rule was changed to apply to wet prairie pixels on l y The initial model was next run with fires simulated at different intervals, to test the burn rules. Five model simulations of 190 years each (19 iterations) were run with fire intervals of 30 to 70 years. A burn schedule for the simulations is shown in Table 4-3. The landscape composition of the base maps at the end of each simulation is shown in Figure 4-7. The landscape compositi on was measured by the percentage of the base map in each seral stage, with the slough/sparse wet prairie and wet prairie stages grouped for plot simplicity. Some trends in the dominance of the seral stages at the different fire intervals had been expected Slough and wet prairie were e x pected to dominate the landscape at the shorter fire intervals. intervals. Trees were expected to dominate in the longer The expected trends in seral stage dominance were present in the simulation results in Figure 4 -7. However, the differences between the landscape compositions of the five fireinterval simulations were greater than expected The difference in wet prairie coverage between the 30 year and 40-year intervals, for e xample was unreasonable. This was partly attributable to the accelerated succession of slough/sparse wet prairie to sawgrass, as mentioned above. Another reason was

PAGE 151

Table 4 3. Burn iterations for different fire interval model simulations "B" indicates a burn iteration "* If indicates iteration used Iteration 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 20 -yr B B B B B B B B 30 -yr B B B B B 40 yr B B B B 50 yr B B B 60 yr B B 70 yr B B in 17 * analysis 18 19 B B B f-' w (.J1

PAGE 152

100 90 80 70 60 50 40 30 20 10 136 Landscape composition percentage of area by seral stage o U...E!aL,_1_.b::i.._.~ubo---r-_____ill! Start 30 40 50 60 70 Fire Interval Slough/wet prairie Cll Sawgrass Ill Trees Figure 4-7. Initial landscape model using different fire intervals: percentage of base map in each seral stage after 190 years that vegetation was in a different state of fire recovery at the end of the different the simulation runs. For example, when the 190-year simulation with a 30-year fire interval was over, the base map had been burned during the previous iteration. After 190 years of the 40-year fire interval simulation, a fire had not burned the base map for three iterations. In the first case, vegetation was in early recovery, and in the second case recovery was well advanced. A decision was made to address this problem by comparing the latest iteration in each simulation that was two years post fire, rather than comparing the last iteration in each simulation. These selected iterations are compared in the remainder of this work, and are marked in Table 4-3. When the selected iterations were compared, the differences between the simulations were more realistic (see Figure 4-8). The distributions of the seral stages across the range of fire

PAGE 153

137 intervals followed more gradual curves However none of the simulations produced a lands cape with vegetation as evenly distributed as the starting base map. The landscape spatial h ete r ogeneity produced by the different simulations was measured by the number of patches of vegetation in the base maps. The ERDAS CLUMP program was used to identify vegetation patches in the base maps after they had been re-coded from the 28 successional classes to the five seral stages. The patches were counted using the ERDAS BSTATS program, and are presented in Figure 4-9. The sizes of the largest patch in each base map als o were noted, and are presented in Figure 4-9 as a percentage of the base map area. The greater the size of the largest patch the greater the connectivity of the dominant seral stage in the base map. Landscape heterogeneit y was lowest when fires burned most frequently, increasing the am o unt of wet prairie, and where fires burned least frequentl y all ow ing tree and brush patches to coalesce (see Figures 4-8 and 4 9) This relationship between disturbance frequency and landscape heterogeneity has been noted in landscape ecology literature (Turner 1987). However, in all the simulati ons the numbers of patches were much lower and the si zes of the largest patches were much greater than in the starting base map Thus the model was simplifying the landscap e Some simplification was inevitable because the model was an abstraction of the ecosystem and lacked a mechanism besides succession to generate of new patches (such as peat batteri es or alligators) However it was hoped

PAGE 154

1 38 that removing slou g h-t o -s awgrass succession and impr ov ing the model's fire c o mp o nen t wo ul d a l low the model ba s e maps to retain more spatial heter o g e n ei t y 100 90 80 70 60 50 40 30 20 10 0 Start 30 La nd s cape composition P e rcenta ge of area by seral stage 40 50 60 70 Fire Inter v al D Slough/wet prairie El Sawgrass II Trees Figure 4-8. Initial m o d e l res ul ts u s i ng di fferent fire intervals: percentage o f base map in each se ral stage at selected iterati o ns 400 350 300 250 200 150 100 50 0 +-'----"' Start L a n dscape heterogeneitypatch d e nsit y ( p atches/km 2 ) and largest pa t ch si z e 30 40 50 60 70 patch (percent of map) density Figure 4-9. Init i al m odel results using diffe r en t f i re intervals: patch d e nsit y and the largest patch a t selected iterations

PAGE 155

139 Improved Model Simulations Methods The GIS model rules were revised after verification of the model, and are shown in Table 4-4. As in the initial model, if there was not a fire during an iteration, pixels in each class advanced to the next higher successional class. Succession was still accelerated in wet prairie pixels adjacent to sawgrass pixels, but slough/sparse wet prairie pixels were excluded from the rule. It was assumed that the water depth in slough pixels was too low for sawgrass colonization. To improve the performance of the fire component of the model, new fire rules were written for two levels of fire severity. Using the new rules, the improved model could be run to simulate either a moderate fire without tree and shrub mortality, or a severe fire with tree and shrub mortality. The improved model was first run without fire. As with the initial model, dominance of the landscape progressed from slough and wet prairie to sawgrass, then to brush, then trees (see Figure 4-10). However, the landscape composition was more even throughout the simulati on and thus more realistic than the initial model. Trees occupied all of the base map after 190 years, rather than 160 years. Two separate sets of model rules were written for fire in the improved model. The first set simulated a moderate fire without shrub and tree mortality. If there was a fire in an iteration, succession occurred only in core island pixels over 30 meters from an island edge (i e with a value greater than

PAGE 156

140 Table 4-4. Impro ved landsca pe model rules Succession without fire Condition fire is 0 and border is 0 fire is 0 and border is 1 and veg< 10 fire is 0 and border is 1 and veg>= 10 fire is 0 and border is 2 Succession with fire: moderate Condition fire is 1 and distance >= 3 fire is 1 and distance< 3 and veg < 4) fire is 1 and distance< 3 and veg>= 4 and veg< 20 fire is 1 and distance< 3 and veg>= 20 and veg< 24 fire is 1 and distance< 3 and veg >= 24) Succession with fire: severe Condition fire is 1 and distance>= 3 fire is 1 and distance< 3 and veg< 5) fire is 1 and distance< 3 and veg>= 5 and veg< 24 fire is 1 and distance< 3 and veg>= 24) Result veg= ve g+ 1 veg= veg+ 1 veg = veg+ 3 veg= veg+ 1 Result veg = veg + 1 veg = 1 veg= veg 3 veg= 20 veg= 22 Result veg= veg+ 1 veg= 1 veg= veg 4 ve g= 22

PAGE 157

1 41 Slough/sparse 100 --0 wet prairie 90 /-/ 80 Wet praire / 70 / / 60 / Sawgrass / 50 / I 40 tt Shrubs 30 20 -Trees 10 0 0 20 40 60 80 100 1 20 140 16 0 180 Iteration Figure 4-10 Compositi o n o f the base map duri ng a run of the im pr ov ed la n ds cape model (percentage o f map in each seral stage) three on the distance ma p) S l ough/sparse wet prairi e pi x els, wet prairie pixels, and s awgras s pixels (classes 1 -1 9) were decreased three class e s to s imulate the net effect of successi o n during the iteration f o ll owed by damage from fire Shrub pi x els were set back t o th e l owest shrub class (class 20) to simula t e the net effect of suc ce s sion followed by fire d a ma ge. Tree pixels were set bac k t o the middle shrub class ( c l as s 22 ) t o simulate the effect o f su ccession followed by top k illing o f th e tree by fire It was ass umed that trees would st u m p-spr o ut and reach tree stature aft e r 20 years (two fire free iter at io ns ) The second set o f f ire ru l es simulated a severe f i re allowing for shrub an d t ree mortality but witho u t peat b u rni n g If there was a fir e in an iterati on succession occurred o n ly in core island pi xe l s over 30 meters from an island edge (i e with a value gre ater t h an three on the distance map) Slough/sparse

PAGE 158

142 wet prairie pixels, wet prairie pixels, sawgrass pixels, and shrub pixels (classes 1-23) were decreased four classes to simulate the net effect of succession during the iteration followed by severe damage from fire. to the middle shrub class (class 22) Tree pixels were set back The effect of this last rule was that trees died only if the fire interval was less than thirty years. As with the initial model, the improved model was run with fires simulated at different intervals. A 20 ye ar interval, not included in the initial model verification was added. Simulations of 190 years each (19 iterations) were run with fire intervals of 20, 30, 40, 50, 60 and 70 years. One set of fire interval simulations was run using the moderate fire rules, and one set was run using the severe fire rules. The results of the simulati ons were evaluated by again comparing the latest iteration in each simulation that was two years post-fire, with the exception of simulations using a 20year fire interval (see Table 4 3) Results The compositions of the base maps at the end of the six simulations using the moderate fire rules are shown in Figure 411. The landscapes produced by the improved model were more evenly composed and more heterogeneous than those produced by the initial model (see Figures 4 -11 and 4-13). The 20-year fire interval simulation most retained the qualities of the starting base map made from the 1987 satellite image of Water Conservation Area 1. The 20 year fire interval produced a

PAGE 159

100 90 80 70 60 50 40 30 20 10 0 20 30 143 Landscape composition Percentage of area by seral stage 40 50 60 Fire Inter val 70 prairie Trees Figure 4-11. Improved model results using moderate fire severity and different fire inter vals : percentage of base map in each seral stage at selected iterations landscape with slightl y more wet prairie and shrubs, and fewer trees and sawgrass, than the starting base map. The landscape contained nearl y as man y small patches as the starting base map. The 30-year and 40year interval simulations resulted in the most evenly composed landscapes with similar patch numbers. Sawgrass was dominant in both simulations but the 30-year interval simulation contained more wet prairie and less sawgrass than the 40-year simulation. In the fire inter vals of 50 years and longer the percentage of the base map occ upi ed by trees increased dramatically. The lon ger fire-free period allowed sawgrass to succeed to brush, yet the moderate fire rules did not kill brush and convert it back t o sawgrass The longer fire-free period allowed the increasing brush area to succeed to trees While the 50-year and 60-yea r interval base maps had less sawgrass and more trees than the initial model produced at these intervals

PAGE 160

1 44 they also retained s o m e small patches of s lough t h at preserved more landscape heter o ge neity The compositions o f t he base maps i n e ac h seral stage at the end of the simulati ons using the severe f i r e rules are shown in Figure 4 -12. All the s evere fire simu lat ion s pr o duced less evenl y comp o sed landscap e s than their moderate co unterparts. Brush mortalit y spec i f ied by the severe fire rule s reduced o r eliminated brush at all in t ervals Trees were not eliminated, because the fire rules ga ve t hem more re sili e nce than brush, and island cores were protect e d from fire 100 90 80 70 60 50 40 30 20 10 0 20 30 L a n dscape composition Perce n ta ge of area by s eral stage 40 50 60 F i r e I nterval 70 D Slough/ w et prairi e II Trees Figure 4-1 2 Impr ov ed model results using high fire severit y and different fire inter va l s : percentage of ba s e m ap in each seral stage at selected i terations Severe fires at t he 20 year interval conve rt ed the landscape al mo st en t i rely to wet prairie Only 1 24 small patches o f tre e s an d s a wgrass remained after 170 y ears. Unli k e the m o derate fire s imulation, most of the 30 year f ire inter v a l was occupied b y w et p ra irie The 40 year interv a l simulation

PAGE 161

145 was dominated by sawgrass, as in the moderate simulation In the severe simulati ons with fire intervals greater than 40, sawgrass was more prevalent than in the moderate simulations, owing to the mortality of brush Areas that were brush and trees in the moderate SO-year and 60-year simulations were converted to sawgrass in the severe simulations. 400 350 300 250 200 150 100 50 0+--Start 20 Landscape heterogeneity patch density (patches/km 2 ) 30 40 50 Fire Interval 60 70 severity severity condition Figure 4-13. Improved model results using moderate and severe fires and different fire intervals: landscape heterogeneity at selected iterations Because the severe fire simulations changed the balance between the woody and non-woody seral stages they produced different levels of heterogeneity than their moderate counterparts. Patch density in the moderate simulations was highest in the 20-year interval because this simulation changed the starting base map the least The frequent fires continually returned pi xels that had raised their successional class back to their previous states. Yet the fires did not damage vegetation enough to eliminate many patches

PAGE 162

146 Some heterogeneity was lost in the moderate simulations with fire intervals longer than 20 years because much of the heterogeneity of the starting base map lay in the interspersion of slough/sparse wet prairie and wet prairie vegetation. types decreased with longer fire intervals, so their These contribution to heterogeneity also decreased. At the same time, longer fire intervals allowed succession to progress further between fires. New patches of brush and trees appeared, created by the succession of sawgrass and brush. Heterogeneity increased from the 30-year to the 50-year interval. By the 60year interval, so many patches of brush and trees had appeared that they were consolidating into larger patches, reducing heterogeneity. In the severe fire simulations, these patterns of change were altered by the higher le vel of fire damage. The greater damage reduced all wet prairie pi xels to slough/sparse wet prairie classes. As a result, no wet prairie heterogeneity was retained in any of the simulations. The number of brush patches was reduced at all fire intervals, further eroding heterogeneity. The 40-year interval's heterogeneity was particularly low, because sawgrass was so prevalent that sawgrass patches coalesced. The severe fire rules slowed the spread of trees. The merging of tree patches seen in the moderate simulations with intervals of 50 years and more was delayed in the severe simulations. The simulations with severe fires at these intervals were more heterogeneous than the moderate simulations.

PAGE 163

147 STARTING BASE MAP FROM 1987 SATELLITE IMAGE .. . \ '\ ... .. '. .;.: . ):>{"_: : .. i . : .. : . i < > 1 . 1 . : . ,~ , tt . .... ~ ., ~ \ . . : . : ~. .' .. _ .. ,. . . . t t .~ : l ; : . i :: . ': .. : : . , . '{ .. : ,, . { ~ . . -~ ,J . i .. : M OD ERATE FIRE SEVERE FIRE 20-YEAR INTERVAL I ( \ ) \ I l t f .. ... .. t ' f .J t Figure 4-14. Tr ee islands in landscapes simulated by the spatial model

PAGE 164

148 MODERATE FIRE ... ... ~ . . .. : . 30-YEAR INTERVAL .. ... .. 1 I '/ ... : \ .. .. ~ ~ . . ' .. ' . :~ .. ?{
PAGE 165

149 MODERATE FIRE SEVERE FIRE SO-YEAR INTERVAL Figure 4-14--continued

PAGE 166

150 Tree islands in the spatial model Because tree islands are the focus this work, tree islands in the model simulations were given additional study. The base maps presented in the previous section under the moderate and severe fire rules were again re-coded using the ERDAS RECODE program. This time everything but tree pi xe ls were re-coded to zero. These maps are shown in Figure 4-14, with tree islands appearing in black. Figure 4-14 illustrates the changing patterns of tree islands in the landscape und er different fire regimes. As fires became less frequent, successi on advanced until tree islands merged together. As fires became more frequent or severe, small islands were reduced in size and disappeared. The largest islands in the landscape persisted and maintained an elongate shape even at high fire frequenc y and sever ity. The patches of trees produced by the different simulations were identified and counted using the ERDAS CLUMP and BSTATS programs. These patch data are presented in Figure 4-15. The loss of tree patches at the shortest fire interval is apparent in Figure 4-15, as is the drop in the number of patches from island consolidation at the longer intervals. The Figure suggests that a fire interval between 20 and 30 years would produce a density of tree islands close to the starting base map. While most vegetation, including small islands, was converted to slough/sparse wet prairie by severe fires in the 20-year interval simulation, large tree islands persisted. To

PAGE 167

151 verify that th ese large islands remained bec ause the cores of large is l ands wer e pr otected from fire t he m o del was run with a 2 0-year fire inter v al and severe fire wi t ho u t t he core protection rul e At the end of the simulatio n with o ut the pro t ection rul e only slough/sparse wet prairie w as present (see Tab l es 4 5 and 46) The protection of i s l and cores was the mechanism keeping l ar ge islands in the landsc a p e at this fire fre q uency. 3000 2500 2000 1500 1000 500 0+-Start 20 N u mber of tree patche s 3 0 4 0 5 0 Fire Interval 60 severity severity condition 70 F i g ure 4 -1 5 Impr ove d model res u lts u sing m o derate and severe fi re s and differ e nt fire intervals : number o f tre e islands at selected iterati o n s Table 4 5 Lands c ap e composition after 20 year fi re inter v al sim ul ations, (pe rce n tage of area by seral st a ge ) Seral Fire severity stage Slough Wet pr Sawgrass S h ru bs T ree s Moderate 69 83 3.18 13 22 10 7 3 .0 7 High 97 93 0 49 0 62 0 27 0. 6 8 High with o ut 100 0 0 0 0 core prot e c t i o n

PAGE 168

152 Table 4-6. Number of patches after 20 year fire interval simulations Fire severity Patches Moderate 18368 High 125 High without 1 core protection Comparison of Model with Independent Data Model validation The validation of a model is achieved by comparing the output of the model with independent data not used in the development of the model. A model should be used as a predictive tool only if it is validated. Models are validated by running them on a different site or in a different time period from which they were developed. Assume for example that a forest stand model was developed and verified using data from some time period at a certain site. To validate the model, it could be run over the same time period using input variables from a different site. The output would then be compared with independent stand data collected at the new site. Alternatively, a hydrologic model developed and verified using data from some time period at a certain site could be validated by using climatic variab les from a different period of time to run the model, and comparing the output with water level recordings from the site. The GIS model described in this work could not be validated by running it in a different time period. The rate of

PAGE 169

15 3 succession and the m o d e l' s time step are s o long co m p a r e d t o the period of time during w h ich landscape data have bee n r e co rded that this was impossibl e Although not an o pti ma l strategy an area of Water Conservation Area 1 sepa r a te from the 5000 hec ta re m o del e d area could be used to run th e model for validatio n. Data fr o m all o f Water Conservation Area 1 was us ed in the developme n t o f the conceptual model, bu t on l y the 5000 hectare portio n o f the area was used to verif y the model Data used for v ali d a tion should be independen t f r o m th e data used in model de v e lopment An independ e nt l y p r o duced map with information comparabl e to t he 198 7 mode l base map, o r statistics derived fr o m su c h a map is req u ired fo r v al i dat io n of the spatial model. Her e th e time s cal e o f t h e m o d e l aga i n poses a problem. T o v ali date just o ne time step of th e m o del beginning with the 19 8 7 c lassified satellite image wo uld re q u ir e map data from 1997. An ear l ier map could be used as th e m od el base map, run until 19 8 7, and c ompared with the 1987 cl assi f i ed image, but this w o uld requir e a map fr o m appr ox imately 1947 1957, 1967 or 1977. There are n o ex i sting m a ps o f Wat e r C o ns e r v ation Area 1 that fulfill th e se v al idation requi re m e nts E x isting vegetati o n maps are too gen e r a l in th e ir cla sses d i vi ding the interi o r of the area into on e o r t wo typ e s Pr e -1 982 sa t e llite data are t oo coarse in their res o lu tion t o be c o mpa re d with th e m o del Aerial phot o gra p h s from a s f ar back as 1948 are availa b le although the di fficulti es in i n terpreting and registeri n g old

PAGE 170

154 photography in Water Conservation Area 1 (as discussed at the end of Chapter 2) are great. Because the interpretation of aerial photographs is subjective, the builder of the model could not avoid bias in interpreting a new vegetation map. Therefore vegetation would have to be specially mapped for validation purposes by an independent source. No funds were available for this purpose. Comparison of model and satellite data Because validation of the GIS model of Water Conservation Area 1 was not possible, an attempt was made to compare independent data with the burn function of the model. The area chosen for the comparison was a subset of Water Conservation Area 1 outside the 5000-hectare area used for model verification, and within the area burned by the 1989 fire. Unclassified satellite data were chosen for comparison with the model, to avoid the bias associated with data classification. Unclassified 1987 satellite data from the subset area was to be compared with unclassified 1990 satellite data to detect burn damage from the 1989 fire. The model was to be run in the subset area for a single iteration with fire, using the 1987 base map. The simulated burn damage was to be compared with the real burn damage detected by the raw satellite data. Unfortunately, the change detection procedure run on the 1987 and 1990 satellite images was unsuccessful in showing the effects of the 1989 fire, and could not be compared with burn damage simulated by the model. The difference in water levels

PAGE 171

1 55 in Water Conservation Area 1 b e tw ee n the t wo r aw satellite images caused the failure o f the bu r n damag e detecti o n. Soil moisture and standing wat e r str o ngly influence the spectral reflectance measured by th e SP O T satellite. Rainfall in 1987 was slightly ab o ve n o rmal, wh i l e 1990 was a severe drought year. The change in wat e r l eve l was greatest between the two dates in the area burned b y the fire, since fire spread throughout the driest areas in 19 8 9. The spectral difference between the 1987 image and 1990 image due t o water level was much greater than the spectral difference between the two images due to fire. Although the e x tent of t he fire was clearly apparent on the 1990 image, and t h e r e we re m easurable differences within the image i n re fl e ctance between damaged and undamaged areas, these differ e nc e s w ere ove rwhelmed by the much greater variation in water-deri v ed reflectance between the two images. The first change-detecti o n t ec hniqu e tried o n the two satellite images was simple. With t hi s method a band from o ne image is displayed on the c o mpute r mon i to r in o ne c o l o r with a band from the other image in an o th e r co l o r. T he pi x els that differ between the two images appear in different color from those that do not differ. Pairs of bands fr o m e ach image were displ a ye d, using all four bands (bands 1, 2, 3, and the NOVI band) However n o n e of the band comparisons sh o w ed the effects of fire A second techniqu e for change detection was also used on the images. This m e th od s ub tracts the pixel values of one image

PAGE 172

156 from the pixel values of the other image. The differences are kept as positive numbers by adding a constant to them. Differences are calculated for each band separately. The greater the difference, the greater the change. When an environmental condition that affects the image evenly, such as haze, varies between two images, the difference it introduces can be corrected for. Rather than calculate the difference between the band values for each image, the values in each band are first re-coded as the difference from the mean for that band. The differences from the mean are then used in the difference calculation (ERDAS 1991). Using the ERDAS BSTATS, RECODE, and ALGEBRA programs (ERDAS 1991), the differences between the two images were calculated using the differences from the mean for each image. However, the fire damage was not highlighted by this technique either. The correction using the image mean was not effective, because unlike haze or sun-angle, water level did not change evenly throughout the image. Discussion The design of this spatial model of Water Conservation Area 1 was innovative, because it used existing GIS software to iteratively model landscape change. In creating the model, it was necessary to simplify or ignore some processes in the ecosystem. The model could not be checked against independent data. All available data from Water Conservation Area 1 had to be used in model development, because data were so scarce. This

PAGE 173

157 left no independent data for validation For all of the above reasons, the spatial m odel presented in this chapter is not meant to be, nor should it be, used as a predictive management tool. In spite of its limitati ons the model served to clarify possible interactions betw een fire, succession and landscape pattern in Water Conservati on Area 1. The model was a quantitative and visual expression of a theory. The maps it produced made visualizati on of the theory easier than imagining it in abstract. The process of creating the model also required a thorough review of all the factors that might be influencing the system, even though most o f these factors were n o t included in it. The model is best suited to generate new ideas and questions to be investigated. The simulated landscape s produced by the GIS model sh ow ed that both fire frequenc y and severity strongly influenced the simulated landscape pattern. The simulati on s using a moderate fire severity produced landscapes more similar t o the landscape pattern we currently observe in Water Conservation Area 1. Of the fire intervals simulated with moderate fire the o ptimum frequency for maintaining a vegetation pattern close to the current pattern of Wat er Conservation Area 1 was 20 years A frequency between 20 and 30 years would come even closer This suggests that an optimal frequency range of moderate fires may e x ist for maintaining the historical pattern of the real ecosystem.

PAGE 174

158 No simulation using severe fire produced a vegetation pattern close to the present pattern in Water Conservation Area 1. The entire range of severe fire intervals reduced the evenness of the landscape composition, because severe fire favored sawgrass over brush and reduced the spatial variation in the successional state of wet prairie. Fire controlled the pattern of tree islands produced by the model simulations. Without fire, tree islands in the model grew until they consolidated into a single patch. Fires at intervals of 40 to 60 years resulted in the greatest numbers of tree islands, far more than are found in the current pattern of Water Conservation Area 1. Under the strongest fire regime (severe fires at a 20 year inter val) the protection of island cores built into the model prevented the largest tree islands in the landscape from disappearing.

PAGE 175

CHAPTER 5 SUI::1MAR Y AND CONCLUSIONS As the twentieth century draws to a close, the resources of the planet are becoming ever scarcer. Awareness of human dependence upon natural systems is gr owin g, as the benefits provided by ecosystems disappear. The management of ecosystems only for species or functi ons that benefit humans, rather than for the maintenance of the natural system, has led to the collapse of some ecosystems. In resp onse to a growing urgenc y for conservation, the 1990's began with a new resource management paradigm in the United States: ecosystem management. Implementation of ecos y ste m management, however, has been more difficult than the decision t o pursue that approach. Methods of evaluating whole ecosystems o r landscapes are still being developed and tested. The computer and remote sensing technology that made the inventory of ecosystems possible are recent developments. This technology continues to improve at a rapid rate. Many o f the technological limitations to the work described in this dissertation (barriers of cost, computer disk space, and software) have been resolved during the completion of this document. More geographic data can now be processed faster on less expensive comp u ters to produce a wide variety of spatial statistics. However, the body of theoretical and methodological knowledge to correctly apply these statistics to questions of 159

PAGE 176

160 and methodological knowledge to correctly apply these statistics to questions of ecosystem management is growing at a slower rate. The academic field of landscape ecology is still in its infancy. This doctoral research was an effort to study an ecosystem using many approaches and tools, including remote sensing and computing technology available between 1988 and 1993. The ecosystem under study, South Florida's Water Conservation Area 1, was not well-studied, well-traveled, nor well-understood. As part of the Everglades, it was the object of both human-induced environmental change, and human pressure for restoration and conservation. The e xi sting pattern of its landscape was a critical aspect of wildlife habitat, particularly for endangered wading birds. The ecosystem exemplified both the great need for and the great difficulties facing an ecosystem or landscape approach to management. The work described in this dissertation focused on the pattern of the landscape of Water Conservation Area 1, rather than on the plant and animal species in it. The pattern of the landscape is the pattern of habitats upon which both plants and animals are dependent. This research took a holistic approach to gaining knowledge of the ecosystem at the landscape scale. The investigation of how this ecosystem produced the unique pattern found in its landsca pe required piecing together data from many parts of the system, at different scales, gathered by many people.

PAGE 177

161 Descriptions of vegetati o n communities and v egetation studies by other researchers were used to define the vegetation types making up the landscap e Studies o f the distribution of plant species with respect to hydr o l ogy and theories of vegetation succession and ecosystem evolution led to the description of a successional sequence in Water Conservation Area 1. Studies of peat, including peat formation, oxidation, burning, erosion and depositi on led to the c o ncept of not only vegetation in flux, but the t o pography itself, changing more rapidly than other geologic processes. Obse r v ations and theories of tree island formation suggested several mechanisms responsible for the development of raised areas of peat, although no evidence was c onclusive All of these pieces of informati on t o gether led to the conclusion that islands grow in area over time. If succession is operating, if peat is being f ormed if floating peat and vegetation accumulate, then islands must be growing, unless some other process limits growth. The landscape pattern of Water Conservation Area 1 observed on the ground reinf or ced th e ideas of vegetation succession and island growth. C ore forested areas of tree islands graded into brush and sawgrass a t island edges and island edges were fringed with sawgrass Patches of vegetation appearing to be in vari o us stages of succession w ere abundant : sawgrass patches, sawgr ass patches with several shrub individuals at their c e nt ers brush patches fringed with sawgrass, brush patches fringed with sawgrass with one or two

PAGE 178

162 trees at their centers, etc. Yet the process of succession was too slow to perceive in real time The comparison o f photographs taken in 195 2 with satellite imagery from 1990 sh owed vegetation over a 38 year time span, long enough to detect succession. The veget ati on changes observed over this time span reinforced the existing evidence that hydrology drives succession in the Everglades. Where hydroperiod had increased over the 38 year pe ri od thousands of hectares of sawgrass, brush and trees were lost Where hydroperiod had decreased, hundreds of hectares of brush had appeared. In the areas where hydroperiod had not greatly changed, vegetation changes could not be measured, because the amount of error in matching the photographs and satellite image was as large as the chang es in vegetation Because successi on in the non-degraded parts o f Water Conservation Area 1 could not be measured directly a different approach to the subject of island growth was taken. The distribution of tree islands of different sizes, as recorded in a 1987 classified satelli te image was examined. The satellite image showed many more small islands than large one s. The frequency of islands of each size class decreased exponentially as their sizes increased. When the distribution of the size classes of the islands was e x amined, it was found that the size classes of small islands were separated by the smallest possible increment, one pi xel However, when islands grew above a certain size (1.59 hectar es) the difference between one island's size and the next larger island s size increased with island

PAGE 179

163 size fo ll o w i ng an exponential curve When island sizes reached a n ot her th r esh o ld (2 18 hectares) the increase in the difference b e tw e en one island a n d t h e next larger island followe d a st e eper cur v e, creating a bre ak or discontinuity in the curve o f the di stributi o n. The size and frequ en c y distributions of tree isl a nd s recorded by the sat e lli te image raised three question s. Co u l d the ongoing formati o n an d growth of islands create a s iz e distribution li k e t ha t found in the landscape of Wate r C o nservati o n Ar e al? Why were more small islands t h a n larg e islands f o und in th e satel l ite image of the landscape? Wh a t could cause the disc o ntinu ity in the distribution of islan d sizes? The first questi on was addressed with simple mode ls built to generate island si ze and frequency distributions f or d i fferent gr o wth functi ons. The modeling exercise demon s tr a ted that the creation and growth of islands over time cou l d c r ea te a distribution li k e that f o u nd in the landscape of Water C o nservation Area 1. A growth function dependent upo n isla n d area pr o duced th e distribution most similar to that observed in the landscap e The sec o nd qu estion was also addressed when the models demonstrat e d th a t t h e large numbers of small islands could be the result of a g r owth process that changed w i th island size or s cal e B e cau se t h e metric used for measuring islands has a coarser r e solut ion than the smallest scale of the process, many small island s of different sizes were grouped together in size

PAGE 180

164 classes separated by the smallest unit of the metric. The result was a frequency distribution by size similar to that measured in the satellite image of the landscape. 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. A conceptual model prop osing a mechanism for the discontinuity in the island size distribution was constructed. The hypothetical system was devised from a synthesis of information on the Everglades ecosystem, the Okefenokee swamp ecosystem, and fire behavior. The conceptual model proposed that in the absence of fire, vegetation follows a successional sequence from wet prairie to trees, and that fire damage moves vegetation back along the sequenc e By the mechanism of succession, accelerated by peat accumulation along the margins of incipient and developed islands, tree islands grow in area over time. Fire damage sets back the process of island growth. The cores of large islands develop a wetter microclimate than the rest of the landscape, and so resist fire. Once islands grow large enough for this microclimate effect to operate, their growth rate is boosted by a reduction in fire damage. Island growth and resistance to fire form a positive feedback loop, producing a steep rise in growth rate appearing as a discontinuity in the island size distribution

PAGE 181

165 The conceptual model was converted into a spatial model capable of producing landscape maps. The spatial model was used to simulate landscapes under different fire frequency and severity regimes. Changing the strength and frequency of fire changed the patterns of the modeled landscapes as measured by the vegetation composition and the number of vegetation patches. In the abstract ecosystem of the model fire controlled the number and sizes of tree islands The end product of this investigation of the landscape of Water Conservati on Area 1-the synthesis of existing research pertaining to the ecosystem, the collection and analysis of landscape data, the simulation of different island growth functions, the creation of a conceptual landscape model, and the spatial simulati on of landscape patterns-is a theory of landscape dynamics. model in chapter 4. This theory is depicted in the conceptual The theory has not been proven or disproven here. It may never be possible to test this theory the way a hypothesis of the behavior of a single species or a small set of variables could be tested. The inability to test hypotheses is a common critiqu e of ecosystem and landscape ecology However, theories addressing large-scale questions can only be formulated by thinking at a large scale Communitylevel or species-level hypoth eses are more easily tested, but the answers to individual small-scale questions cannot answer large-scale questions. Many small-scale answers must be fit together to create a larger picture Once a landscape-scale theory is devised, pieces of it can then be tested using

PAGE 182

166 accepted scientific methods, but it cannot be devised without holistic thinking. This investigation has raised questions that need to be addressed to fill in missing pieces of the picture. One of these missing pieces is the actual rotation time of fires in Water Conservation Area 1. This cannot be k n own until detailed fire records are kept for Water Conservation Area 1. It lS recommended that the time, location, and severity of all fires in the area be recorded, so that future researchers can evaluate fire frequency. Another unanswered question concerns the formation of tree islands, particularly the role of peat batteries and alligators in the creation and maintenance of these landforms. A long-term study tracking peat batteries is needed. Studies of alligator behavior with respect to the site-selection and excavation of alligator holes, and the peat-transporting effects of alligator activities are needed. A more difficult missing piece of the picture is the effect of sheet flow on the landscape of Water Conservation Area 1. While we can no longer observe sheet flow in action, perhaps new technology applied to the observation of tree island shapes over time can yield clues to its effect If sheet flow was responsible for the elongate shape of large tree islands, as is frequently assumed, their shapes may be changing in the absence of sheet flow. And finall y the succession of vegetation leading to the expansion of tree island margins (induced and theorized in this work) remains to be tested.

PAGE 183

167 Since science ha s become increasingly reductionist over t i me, holistic appr o ach es h ave been rejected rather than improved. Because eco l ogists have avoided rather th a n developed holistic approaches t o lar ge scale questions resour ce managers are now being forced t o devise ecosystem management strategi e s without the benefit o f an e s tablished scientif i c met h o dol o g y Prejudice against holistic thin k in g amo ng e c o l o gists must be overcome if the needed the or i e s and methods a r e to be de v el o ped. This work is a contributi o n toward the revival o f ho listic thinking, and it is h ope d that this depiction o f the ec o system of Water Conservati o n Ar ea 1 will be useful to the e c o s y stem's managers.

PAGE 184

LIST OF REFERENCES Alexander, T. R., and A. G. Crook. 1974. Recent vegetational changes in Southern Florida. Pp. 67-72 in: Gleason, P. J., ed. Environments of South Florida: Present and Past. Miami Geological Society, Miami, FL. Alexander, T. R., and A.G. Crook. 1975. Recent and Long Term Vegetation Changes and Patterns in South Florida, Part II, Final Report. Appendix Gin: South Florida Ecological Study. University of Miami, Coral Gables, FL. Anderson, H. E. 1982. Aids to Determining Fuel Models for Estimating Fire Behavior. USDA Forest Service General Technical Report INT 122. USDA Forest Service Intermountain Research Station, Ogden, UT. Andrews, P. L. 1986. BEHAVE: Fire Behavior Prediction and Fuel Modeling SystemBURN Subsystem, Part 1. USDA Forest Service General Technical Report INT 194. USDA Forest Service Intermountain Research Station, Ogden, UT. Andrews, P. L., and C. C. Chases. 1989. BEHAVE: Fire Behavior Prediction and Fuel Modeling SystemBURN Subsystem, Part 2. USDA Forest Service General Technical Report INT 260. USDA Forest Service Intermountain Research Station, Ogden, UT. Baker, G. 1952. Narrative Report for Loxahatchee National Wildlife Refuge, Period September 1-December 31 1952. U.S. Fish and Wildlife Service, Boynton Beach, FL. Unpublished. Bond, P., K. M. Campbell, and T. M. Scott 1986. An Overview of Peat Resources in Florida, Special Publication No. 27. Florida Bureau of Geology, Tallahassee, FL. Browder, J. A., P.A. Gleason, and D. R. Swift. Periphyton in the Everglades: spatial variation, environmental correlates, and ecological implications. Pp. 379 418 in: Davis, S. M. and J. C. Ogden, eds. Everglades: The Ecosystem and Restoration. St. Lucie Press, Delray Beach, FL. Burgan, R. E., and R. C. Rothermel. 1984. BEHAVE: Fire Behavior Prediction and Fuel Modeling SystemFuel Subsystem. USDA Forest Service General Technical Report INT 167. USDA Forest Service Intermountain Research Station, Ogden, UT. 168

PAGE 185

169 Clements, F. 1898. The Phytogeography of Nebraska. Ph.D. Dissertation. Universit y of Nebraska, Lincoln, NB. Costanza, R., F. H. Sklar, and M. L. White. 1990. Modeling coastal landscape dynamics. Bioscience 40(2) :91-107. Craighead, F. C., Sr. 1971. The Tr ees of South Florida, Volume I: The Natural Environments and Their Succession. University of Miami Press, Coral Gables, FL. Cypert, E. 1972. The origin of houses in the Okefenokee prairies. American Midland Naturalist 82(2) :448-45 8 Davis, J. H., Jr. 1943. The Natural Features of Southern Florida, Especially the Vegetati on and the Everglades. Bulletin 25. Florida Geological Survey, Tallahassee, FL. Davis, S. M. 1994. Phosphorus inputs and vege tation sensitivity in the Everglades. Pp. 357-378 in: Davis, S. M. and J. C. Ogden, eds. Everglades: The Ecos ys tem and Restoration. St. Lucie Press, Delray Beach, FL. Davis, S. M., L. H. Gunderson, W. A. Park, J. R. Richardson and J. E. Mattson. 1994. Landscape dimension, c om position, and function in a changing Everglades ecosystem. Pp. 419-444 in: Davis, S. M. and J. C. Ogden, eds. Everglades: The Ecosystem and Restoration. St. Lucie Pr ess Delray Beach, FL. DeAngelis, D. L. 1994. Synthesis: spatial and temporal characteristics of the environment. Pp. 307-320 in: Davis, S. M. and J. C. Ogden, eds. Everglades: The Ecosystem and Restoration. St. Lucie Press, Delray Beach, FL. DeAngelis, D. L., and P. S. White. 1994. Ec o systems as products of spatially and temporally varying driving forces, ecological processes, and landscap es : a theoretical perspective. Pp. 9-27 in: Davis, S. M. and J. C. Ogden eds. Everglades: The Ecosystem and Restoration. St. Lucie Press, Delray Beach, FL. Douglas, M. S. edition. 1988. Th e Everglades: River of Grass. Revised Pineapple Press, Sarasota, FL Dressler, R. L., D. W. Hall, K D. Perkins and N H. Williams 1987. Identificati on Manual for Wetland Plant Species of Florida. Institute of Food and Agricultural Sciences University of Florida, Gainesville, FL. Duever, M. J. 1984. Environmental factors controlling plant communities in the Big Cypress Swamp Pp 127-134 in : Gleason, P. J., ed. Environments of South Florida : Present and Past II. Miami Geological Society Miami FL

PAGE 186

170 Duever, M. J., and L. S. Riopelle. 1983. Successional sequences and rates on tree islands in the Okefenokee swamp. American Midland Naturalist 110:186-193. Eisenberg, J. F. 1981. Population Ecology. Techniques for the Study of Primate National Academy Press, Washington, DC. ERDAS, Inc. 1991. Erdas User's Guide, Version 7.5. ERDAS, Inc., Atlanta, GA. Fennema, R. J., C. J. Neidrauer, R. A. Johnson, T. K. Macvicar, and W. A. Perkins. 1994. A computer model to simulate natural Everglades hydrology. Pp. 249-289 in: Davis, S. M. and J. C. Ogden, eds. Everglades: The Ecosystem and Restoration. St. Lucie Press, Delray Beach, FL. Fons, W. L. 1940. Influence of forest cover on wind velocity. Journal of Forestry, 38(6) :481-486. Forman, R. T., and M. Godron. and Sons, New York, NY. 1986. Landscape Ecology. John Wiley Forthman, C. A. 1973. The Effects of Prescribed Burning on Sawgrass, Cladium jamaicense Crantz, in South Florida. M.S. thesis. University of Miami, Coral Gables, FL. Franklin, J. F., and R. T. Forman. 1987. Creating landscape patterns by forest cutting: ecological consequences and principles. Landscape Ecology 1:5-18. Gibson, D. J. 1988. Regeneration and fluctuation of tallgrass prairie vegetation in response to burning frequency. Bulletin of the Torrey Botanical Club 115(1) :1-12. Glaser, P. H. 1987. The Ecology of Patterned Boreal Peatlands of Northern Minnesota: A Community Profile. U.S. Fish and Wildlife Service Report 85(7.14) U.S. Fish and Wildlife Service, Washington, DC. Glasser, J. E. 1986. Pattern, Diversity and Succession of Vegetation in Chase Prairie, Okefenokee Swamp: A Hierarchical Study. Ph.D. dissertation. University of Georgia, Athens, GA. Gleason, H. A. 1926. The individualistic concept of the plant association. Bulletin of the Torrey Botanical Club 44:463-81. Gleason, P. J., D. Piepgras, P.A. Stone, and J. J. Stipp. 1980. Radiometric evidence for involvement of floating islands in the formation of Florida Everglades tree islands. Geology 8:195-199. Gleason, P. J., and P. Stone. 1994. Age, evolution of the Everglades peatland. M. and J. C. Ogden, eds. Everglades: Restoration. St. Lucie Press, Delray origin, and landscape Pp. 149-197 in: Davis, S. The Ecosystem and Beach, FL.

PAGE 187

171 Gunderson, L. H. 1992. Spatial and Temporal Dynamics in the Everglades Ecosystem with Implications for Water Deliveries to Everglades National Park. Ph.D. dissertation. University of Florida, Gainesville, FL. Gunderson, L. H. 1994. Vegetation of the Everglades: determinants of community composition. Pp. 323-340 in: Davis, S. M. and J. c. Ogden, eds. Everglades: The Ecosystem and Restoration. St. Lucie Press, Delray Beach, FL. Gunderson, L. H., and J. R. Snyder. 1994. Fire patterns in the Southern Everglades. Pp. 291 -305 in: Davis, S. M. and J. C. Ogden, eds. Everglades: The Ecosystem and Restoration. St. Lucie Press, Delray Beach, FL. Henderson, R. A., and S. H. Statz. 1995. Bibliography of Fire Effects and Related Literature Applicable to the Ecosystems and Species of Wisconsin. Technical Bulletin No. 187. Department of Natural Resources, Madison, WI. Herndon, A., L. H. Gunderson, and J. Stenberg. survival in fire and flooding. Wetlands, 1991. Sawgrass 11(1) :17-27. Hoffman, W., G. T. Bancroft, and R. J. Sawicki. 1994. Foraging habitat of wading birds in the water conservation areas of the Everglades. Pp. 585-614 in: Davis, S. M. and J. C. Ogden, eds. Everglades: The Ecosystem and Restoration. St. Lucie Press, Delray Beach, FL. Holling, C. S. 1986. The resilience of terrestrial ecosystems: local surprise and global change. Pp. 292-320 in: Clark, W. C. and R. E. Munn, eds. Sustainable Development of the Biosphere. Cambridge University Press, Cambridge, England. Holling, C. S. 1992a. Cross-scale morphology, geometry and dynamics of ecosystems. Ecological Monographs 62(4) :447-502. Holling, C. S. 1992b. The role of forest insects in structuring the boreal landscape. Pp. 170-191 in: Shugart, H. H., R. Leemans, and G. B. Bonan, eds. A Systems Analysis of the Global Boreal Forest. Cambridge University Press, Cambridge England. Jensen, J. R. 1986. Introductory Digital Image Processing: A Remote Sensing Perspectiv e Prentice-Hall, Englewood Cliffs, NJ. Jones, L.A. 1948. Soils, Geology and Water Control in the Everglades Region. University of Florida Agricultural Experiment Station Bulletin, No. 442. University of Florida Gainesville, FL.

PAGE 188

172 Keane, R. E., S. F. Arno, and J. K Brown. 1989. FIRESUMAn Ecological Process Model For Fire Succession in Western Conifer Forests. USDA Forest Service General Technical Report INT 266. USDA Forest Service Intermountain Research Station, Ogden, UT. Kessell, S. R. Management. 1979. Gradient Modelling : Reso urces and Fire SpringerVerlag New York NY. Kessell, S. R., and W. C. Fischer. 1981. Predicting Postfire Plant Succession for Fire Management planning USDA Forest Service General Technical Report INT 94 USDA Forest Service Intermountain Research Station, Ogden UT. Kingsland, S. E. 1985. Population Ecol ogy Modeling Nature: Episodes in the History of University of Chicago Press, Chicago, IL. Kirby, R. E., S. J. Lewis, and T. N. Sexson. 1988. Fire in North American Wetland Ecosystems and Fire-Wildlife Relations: An Annotated Bibliograph y. US Fish and Wildlife Service Biological Report 88(1). US Fish and Wildlife Service, Washington, DC. Kitching, R. L. Modelling. Queensland. 1983. Systems Ecology: Introduction to Ecological Universit y of Q ueen sland Press, St. Lucia, Kushlan, J. A., 1990. Freshwat er marshes. Pp. 324-363 in Myers, R. L. and J. J. Ewel, eds. Ecosystems of Florid a. University of Central Florida Press, Orlando FL. LeBlanc, C. M., and D. J. Leopold. 1992 Demography and age structure of a central New York shrub -carr 94 years after a fire. Bulletin of th e Torrey Botanical Club 119 ( 1) :50-64. Light, S. S., and J. W. Dineen. 1994. Water control in the Everglades: a historical perspective. Pp. 47 -84 in: Davis, S. M. and J. C. Ogden, eds. Everglades: The Ecosystem and Restoration. St. Lucie Press, Delray Beach, FL. Loveless, C. M. Everglades. 1959. A study of the vegeta ti o n in the Florida Ecolog y 40:1-9. Ludwig, J. A., and J. R. Reynolds. 1988. Statistical Ecology. Wiley and Sons, Inc. New York NY John Marston, R. B. 1956. adjacent opening. Air movement under an aspen forest and on an Journal of Forestry 54(7) :468-69. Mattson, J. E., and A. Navarro. 1989. Evaluation o f spatial distribution, sampling precision and sampling accuracy in two populations. Paper written for a course in the quantitative assessment of wildlife habitat, Department of Wildlife and Range Sciences, Universit y of Florida. Unpublished.

PAGE 189

173 M cP h er son, B. F. 197 3 V egetation in Relation to Water De p th in C o nservation Area 3 Florida Florida Open File Repor t 73025 u. s. Geological Su rvey and Central and Sout h ern Florida Flood Control District, T allahassee Florida M e naut, J. C., J. Gign o u x C. Prado and J C l obert 199 0. T r e e community dynamics in a hum i d sava nn a of th e C o te-d'I v oire: modelling the eff e cts of fire and compet i tion w i t h gras s a n d neighbours. J o urn a l of Biogeography 17 : 471 48 1. Odum, H. T. 1983. S y st e ms E cology : An Introd u c ti o n. J o hn Wile y an d Sons, New York, N Y Olmsted, I., and L. L. L oope 1984 Plant commu n it i es of E v erg l a de s National Park. Pp 1 6 7-1 84 in : Gleason P. J ., ed Environments of S o uth Fl orida : Present and Past II. M i ami Geological Societ y Miami FL O'Neill, R. V., D. L. D eA n ge l is J B Waide and T. F. H A l len. 1986. A Hierarchi ca l Concept of Ecosystems P r incet o n University Press, Prin ceton NJ Parker, G. G. 1984. H y dr o l ogy of the pre dra inag e s y stem o f the Everglades in S o ut hern Florida. Pp 28 37 in: Gleas o n, P. J ., ed. Environments o f So ut h Florida : Present and Past II. Miami Geological Societ y Miami, FL Pearlstine, L., H. McKella r a nd W Kitchens 1 98 5. Mo delling the impacts of a river diversi o n o n bot t om l a nd fo r est communities in the Santee River flo o dplain, S o ut h C a r o l i na. Ec o l o gical Modelling 29:281-30 2 Pesnell, G. L., and R. T Brown III. 1977 The Major P l ant Communities of La ke Okeechobee Florida a nd Their Associated Inundation Characteri stics as Determined by Grad i e n t Anal y si s Technical Publicati on 77 -1 Resource Planning Dep a rt m ent S outh Florida Water Manag eme n t District West Palm Beac h, F L. Pope, K. R. 1989. Veg e ta t i on in Relation to Water Q ual i t y and Hydroperiod in th e L oxahatchee National Wildlife Refuge M S. thesis. Universi ty of Florida Gainesville FL Potter, M. W., and S. R Kessell 1980 Predicting mosaics and wildlife di v er si t y resulting from fire disturbance to a forest ecosystem. En v i ronmental Management 4(3) : 247 254 Rich, F. J. 1979 Th e Origin and Development of Tree Islands in the Okefenok e e Sw amp as Determined by Peat Petrography and Pollen Stratigraph y P h. D. dissertation Pennsylvania State University, St ate College PA Richardson, J. R. Ecosystems. Gainesvill e 1988. S p atial Patterns and Maximum Power in Ph.D. dissertation University of Florida, FL

PAGE 190

174 Richardson, J. R., W. L. Bryant, W. E. Kitchens, J. E. Mattson and K. R. Pope. 1990. An Evaluation of Refuge Habitats and Relationships to Water Quality, Quantity and Hydroperiod. A Synthesis Report. November 1990. Florida Cooperative Fish and Wildlife Research Unit, University of Florida, Gainesville, FL. Rothermel, R. C. 1983. How to Predict the Spread and Intensity of Forest and Range Fires. USDA Forest Service General Technical Report INT 143. USDA Forest Service Intermountain Research Station, Ogden, UT. Rothermel, R. C., and J. E. Deeming. 1980. Measuring and Interpreting Fire Behavior for Correlation with Fire Effects. USDA Forest Service General Technical Report INT-93. USDA Forest Service Intermountain Rangeland Experiment Station, Ogden, UT. Schroeder, M. J., and C. C. Buck. 1970. Fire Weathera Guide for the Application of Meteorological Information to Forest Fire Control Operations. USDA Forest Service Agricultural Handbook 360. USDA Forest Service, Washington, DC. Shugart, H. H., and J. R. Noble. 1981. A computer model of succession and fire response of the high altitude Eucalyptus forest of the Brindabella Range, Australian Capital Territory. Australian Journal of Ecology 6:149-164. Shugart, H. H., and D. C. West. 1977. Development of an Appalachian deciduous forest succession model and its application to assessment of the impact of the chestnut blight. Journal of Environmental Management 5:161-179. Silveira, J. E., geographic vegetation Landin, M. Conference LA. South Utica, MS. J. R. Richardson and W. M. Kitchens. 1993. A information systems model of fire damage and recovery in the Northern Everglades. Pp. 455-459 C., ed. Wetlands: Proceedings of the 13th Annual of the Society of Wetland Scientists, New Orleans, Central Chapter, Society of Wetland Scientists, in: Silveira, J. E., J. R. Richardson and W. M. Kitchens. 1992. A GIS model of fire, successi on and landscape pattern in the Everglades. Pp. 88-97 in: ASPRS/ASCM/RT '92 Technical Papers, Volume 5, Resource Technology '9 2 American Society for Photograrnrnetry and Remote Sensing/American Congress on Surveying and Mapping, Washington, DC. Silveira, J. E. 1996. Methods for Determining Change in Florida Wetland Habitats Using SPOT Satellite Data. U.S. National Biological Service Technical Report 52. Florida Cooperative Fish and Wildlife Research Unit, Gainesville, FL.

PAGE 191

175 Sklar, F. H., and R. Costanza. 1991. The development of dynamic spatial models for landscape ecology : a review and prognosis. Pp. 239-288 in: Turn er M G., and R H. Gardner, eds. Quantitative Methods in Landscape Ecology Springer-Verlag, New York, NY. steward, K. K., and W. H. Ornes. in the Florida Everglades. 1975. The autecology of sawgrass Ecology 56 (1) :1 62 Suffling, R., C. Lihou and Y Morand 1988. Control of landscape diversity by catastr ophic disturbance: a theory and a case study of fire in a Canadian boreal forest. Environmental Management 12(1):73-78. Taylor, D. L. 1981. Fire History and Fire Records for Everglades National Park 1948-1979. Report T 619 National Park Service, South Florida Research Center, Homestead, FL. Thompson, R. L. 1970. A Re-evaluation of W ater Level Management, Its Effect on Ecology and Wildlife Resources o f Lo x ahatchee National Wildlife Refuge. Division of Refuges, U.S. Fish and Wildlife Service, Tallahassee, FL. Unpublished. Thompson, R. L. 1972. A Preliminar y Investigation of the Effects of Water Levels on Vegetative Communities o f Loxahatchee National Wildlife Refuge, Fl orida. Bureau of Sport Fisheries and Wildlife, U.S. Fish and Wildlife Service. Unpublished. Tomlin, C. D. Modeling. 1990. Geographic Information Systems and Cartographic Prentice Hall, Englewood Cliffs N J. Turner, M. G., ed. 1987. Landscape Heterogeneity and Disturbance. Springer-Verlag, New York NY Turner, M. G., R. H. Gardner, V. H. Dale and R V O Neill. 1989. Predicting the spread of disturbance across heterogeneous landscapes. Oikos 55:121-1 2 9. Turner, M. G., and V. H. Dale. 1991. Modeling landscape disturbance. Pp. 323-351 in : Turner M G ., and R H Gardner, eds. Quantitative Methods in Landscape Ecology. Springer Verlag, New York, NY. VanArman, J., D. Nealon, S. Burns, B. J ones L. Smith, T Macvicar, M Yansura, A. Federic o J. Bucca M Knapp, and P. Gleason. 198 4 South Florida Water Management D i strict Pp. 1 3 8-15 7 in: Fernald, E. A., and D. A Patt o n e ds Water Res o urces Atlas o f Florida. Institut e of Science a nd Publ ic A f f a ir s, Fl o rida Stat e University, Tallah assee FL Vogl, R. J. 1969. One hundred and th ir t y years of pla n t su c c e ss io n in a Southeast ern Wisconsin l owla n d Ecology 50(2) : 248 255

PAGE 192

176 Wade, D., J. Ewel and R. Hofstetter. 1980. Fire in South Florida Ecosystems. USDA Forest Service Technical Report SE-17. Southeastern Forest Experiment Station, Asheville, NC. Walters, C. J., L. H. Gunderson and C. S. Holling. 1992. Experimental policies for water management in the Everglades. Ecological Applications 2(2) :189-202. Weins, J. A., C. S. Crawford and J. R. Gosz. 1985. Boundary dynamics: a conceptual framework for studying landscape ecosystems. Oikos 45(3): 421-427 White, P. S. 1994. Synthesis: vegetation pattern and process in the Everglades ecosystem. Pp. 445-458 in: Davis, S. M. and J. C. Ogden, eds. Everglades: The Ecosystem and Restoration St. Lucie Press, Delray Beach, FL. Wright, H. A., and A. W. Bailey. 1982. Fire Ecology, United States and Southern Canada. John Wiley and Sons, New York.

PAGE 193

BIOGRAPHICAL SKETCH Jennifer Lynn Enos was born in San Francisco, California, on March 24, 1958. She grew up in Mill Valley, California, attending public schools. In 1982, she earned a Bachelor of Science degree in conservation and resource studies from the University of California, Berkeley, graduating with high honors. She married in 1981, changing her name to Jennifer Enos Mattson. In 1987, she earned a Master of Science degree in wildland resource science from U.C. Berkeley. While working on her master's degree she became interested in remote sensing, and worked as a research assistant at the University of California's Space Sciences Laboratory. In 1988, she came to the University of Florida to pursue a doctoral degree in the Department of Wildlife and Range Sciences. In 1991, she began work as a botanist for the U.S. Fish and Wildlife Service National Wetlands Inventory. After her second marriage in 1992, she changed her name to Jennifer Enos Silveira. 177

PAGE 194

I certify that I have read this study and that in my opinion it conforms to acceptable standards of scholarly presentation and is fully adequate, in scope and quality, as a dissertation for the degree of Doctor of Philosophy. Wiley Kitchens, Chair Associate Professor of Wildlife Ecology and Conservation I certify that I have read this study and that in my opinion it conforms to acceptable standards of scholarly presentation and is fully adequate, in scope and quality, as a dissertation for the degree of Doctor f Philosoph eorge W. Tanner Associate Professor of Wildlife Ecology and Conservation I certify that I have read this study and that in my opinion it conforms to acceptable standards of scholarly presentation and is fully adequate, in scope and quality, as a dissertation for the degree of Docton.~1 clawford S. Holl~ Arthur R. Marshall Jr. Profes5or of Ecological Sciences I certify that I have read this study and that in my opinion it conforms to acceptable standards of scholarly presentation and is fully adequate, in scope and quality, as a dissertation for the degree of Doctor of Philosophy. j A11A. l( ~-< V evi-,v;~ __/r,~kas G. Arvaditis Professor of Forest Resources and Conservation I certif y that I have read this study and that in my opinion it c o nforms to acceptable standards of scholarly presentati o n and is fully adequate, in scope and quality, as a dissertati o n f o r the degree of Doctor of Philosophy. Richardson Sy ms Ecologist The Nature Conservancy

PAGE 195

This dissertation was submitted to the Graduate Faculty of the College of Agriculture and to the Graduate School and was accepted as partial fulfillment of the requirements for the degree of Doct o r of Philosophy. Q~ f. ~/1-'V August, 1996 cfc/ Dean, College of Agriculture Dean, Graduate School


xml version 1.0 encoding UTF-8
REPORT xmlns http:www.fcla.edudlsmddaitss xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.fcla.edudlsmddaitssdaitssReport.xsd
INGEST IEID ECA6528N1_6AWO5D INGEST_TIME 2013-03-21T13:50:17Z PACKAGE AA00013550_00001
AGREEMENT_INFO ACCOUNT UF PROJECT UFDC
FILES