Spatial and temporal patterns of tree islands in the Arthur R. Marshall Loxahatchee National Wildlife Refuge

MISSING IMAGE

Material Information

Title:
Spatial and temporal patterns of tree islands in the Arthur R. Marshall Loxahatchee National Wildlife Refuge
Physical Description:
xii, 140 leaves : ill. ; 29 cm.
Language:
English
Creator:
Brandt, Laura A
Publication Date:

Subjects

Subjects / Keywords:
Wildlife Ecology and Conservation thesis, Ph.D   ( lcsh )
Dissertations, Academic -- Wildlife Ecology and Conservation -- UF   ( lcsh )
Genre:
bibliography   ( marcgt )
non-fiction   ( marcgt )

Notes

Thesis:
Thesis (Ph.D.)--University of Florida, 1997.
Bibliography:
Includes bibliographical references (leaves 130-139).
Statement of Responsibility:
by Laura Ann Brandt.
General Note:
Typescript.
General Note:
Vita.

Record Information

Source Institution:
University of Florida
Rights Management:
All applicable rights reserved by the source institution and holding location.
Resource Identifier:
aleph - 028623480
oclc - 39526158
System ID:
AA00012943:00001


This item is only available as the following downloads:


Full Text











SPATIAL AND TEMPORAL PATTERNS OF TREE ISLANDS IN THE ARTHUR R.
MARSHALL LOXAHATCHEE NATIONAL WILDLIFE REFUGE












By

LAURA ANN BRANDT


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
































Copyright 1997

by

Laura Ann Brandt















ACKNOWLEDGMENTS


I would like to thank my committee members, Drs. Wiley M. Kitchens, Larry

Harris, Buzz Holling, Susan Jacobson, and Ken Portier, for their support and feedback.

Tracey Needle and Randy Van Zee of the South Florida Water Management District

provided access to photography and hydrology data. I thank Burkett Neeley, Su Jewell,

and the other staff at Loxahatchee for assisting with access to the refuge and historic

information. Vicky Drietz provided transportation and assistance on many of my trips to

the field. Barbara Fesler and Debra Hughes helped with numerous administrative

problems. Special thanks go to John Kitchens, Jennifer Swan, Myma Stenberg, and Clare

Stokes for assistance with digitizing, to Christy Steible for answers to my statistical

questions, to Jay Harrison who wrote the S-plus code to calculate tree island orientation,

and to Mike Cherkiss who ran numerous errands for me. Discussions with fellow

graduate students including Jennifer Silveira, Cyndy Loftin, Kim Babbitt, Rob Bennetts,

John Stenberg, and Craig Allen provided insight and amusement throughout the process.

I want to thank (I think) Leonard Pearlstine for introducing me to GIS, providing

feedback and moral support, and putting up with me. Finally, I want to thank Frank

Mazzotti for his friendship and encouragement, for putting up without me, and for

providing me support, when necessary, to complete this work.
















TABLE OF CONTENTS


page


ACKNOWLEDGMENTS ..................... ... ... ............... iii

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

L IST O F FIG U R E S............................................................................... viii

ABSTRACT................................................................ .... xi

CHAPTERS

1 INTRODUCTION ....................... .......... .................................. 1

O overview ................................... ... .. .. .... ...... ... ... ................ 1
L landscape E ecology .................................................. ....................................2
The E verglades ............................................. ... .... ...... ................ 5
The Arthur R. Marshall Loxahatchee National Wildlife Refuge...........................7
Tree Islands ......................... ............ ................................9
D issertation Structure ........................................ .. ........................... ....... 12

2 DISTRIBUTION OF TREE ISLANDS IN LOXAHATCHEE NATIONAL
WILDLIFE REFUGE IN RELATION TO HYDROPATTERNS AND
E L E V A T IO N ........................... ......................... ........... .. ........ .. .......... 15

Introduction ......................... ....... ............ ..... .. ..... ................. 15
M ethods..... .. .............. .. ... .. ............ ..... .. ................................ 19
Study Site.................................... ...................19
H ydrology and Elevation...................................... ................20
Identification and Characterization of Tree Islands....................................21
Analysis........................... ................... .................22
Results.............................. ......... .......................24
H ydrology and E levation ......................................... ................................... 24
T ree Islands....................... ..... ............ 2 5
D iscu ssio n .............................. ..................... ................... 2 8

3 SAMPLING PLOT SIZE AND LOCATION.........................................................45










Introduction .................................... ..... ................45
M methods ................................. ........................... ........................................ 48
Study A rea .................................... 48
Im ag ery ................................. ................. 4 8
S election n o f P lo ts ................................................................... ................ 4 9
Photography ............................. ................... ..................... 51
R e su lts ................................. ................................ .................... ................ 5 1
D discussion .............................. ... ............... 54

4 PATTERNS OF CHANGE IN TREE ISLANDS IN A.R.M.
LOXAHATCHEE NATIONAL WILDLIFE REFUGE FROM 1950 TO
19 9 1 ............................................................7 1

Introduction ............................................ ................... ......................... .......... 7 1
M e th o d s ........................................................................................ ...................7 3
Study A rea .........7.............................. ...................... .................. 73
Photography.......................... ........... ............................. 74
H ydrology and Elevation ...................................................75
Plot Comparisons and Relations to Hydrology ........... .......................... 76
Results.................................... ........ ..................... 78
1950 Plots and N SM Hydrology ................................................78
1991 Plots and W M M Hydrology......................................... ............... 80
Comparison Between Individual Plots .................................. ................82
A ll Plots T together ........................................ ......................... ................ 83
Discussion.................... ........................................ 86

5 SUMMARY AND CONCLUSIONS ................. ...................... .................... 116

S u m m a ry ............................................................. ............................................ 1 1 6
E cological C consequences ........................................................... .... ........ ..... 120
F future D directions .......................................... .... ... ... .............. 124

APPENDICIES

A NATURAL SYSTEMS MODEL AND WATER MANAGEMENT MODEL
ROW AND COLUMN FOR HYDROLOGY ZONES ................. ............... 128

B UTM COORDINATES OF PHOTO PLOTS IN LOXHATCHEE
NATIONAL W ILDLIFE REFUGE................. ........................................ 129

LITERATURE CITED .... ................................ .... ........... .. 130

B IO G R A PH IC A L SK ETC H .......................... ........................ ......................... 140
















LIST OF TABLES


Table page

2-1. Summary statistics for tree islands in Loxahatchee National Wildlife Refuge
identified from satellite im agery. .............................. .... ............ ...................... 34

2-2. Canonical variables for analysis using shape classes. Most significant coefficients are
bold faced. ................. ......... .. .......... .. ..... ......... ............... ... ... ............. 35

2-3. Canonical variables for analysis using length classes. Most significant coefficients are
bold faced. ..................... .... ...... ......... ..... ....... ....... ............ 36

3-1. Number of tree islands per plot based on classification of satellite imagery
(Richardson et al. 1990). Pixels are 9 x 9 m. Statistics are based on 100 resamples
representing the listed percentage of complete plots of that size. True mean,
variance, and standard deviation are for all of the plots of that size. Number of tree
islands on im agery = 2144 .............................................. ....................................... 60

3-2. Mean percent tree island cover per plot based on classification of satellite imagery
(Richardson 1990). Pixels are 9 x 9 m. Statistics are based on 100 resamples
representing the listed percentage of complete plots of that size. True mean,
variance, and standard deviation are for all of the plots of that size. Percent cover of
tree islands on imagery = 1.96 ...................... .. ....................61

3-3. Mean size (pixels) of tree islands per plot based on classification of satellite imagery
(Richardson 1990). Pixels are 9 x 9m. Statistics are based on 100 resamples
representing the listed percentage of complete plots of that size. True mean,
variance, and standard deviation are for all of the plots of that size. Mean size of tree
islands on imagery = 52.6 pixels, variance = 572.3, SD = 23.9............................ 62

3-4. Summary statistics for plots from classified image and photo plots...................... 63

4-1. Variables used, description, transformation, and significance of variable............... 90

4-2. Summary statistics for 1950 photo plots from Loxahatchee National Wildlife Refuge.
Orientation is 0 to 180 degrees with 0 = north and 180 = south..............................96










4-3. Hydrology data for hydrology zones in Loxhatchee National Wildlife Refuge that
contained photo plots. See Appendix A for relation of hydrology zone to NSM and
W M M row s and colum ns. ................... ...... ............................ ..................98

4-4. Correlations between tree island variables and hydrology variables for 1950 photo
plots Top number is correlation coefficient, bottom number is p value. A p value <
0.05 is considered significant.............................. ...... ........... .......... ...... 100

4-5. Summary statistics for 1991 photo plots from Loxahatchee National Wildlife Refuge.
Orientation is 0 to 180 degrees with 0 = north and 180 = south.......................... 101

4-6. Correlations between tree island variables and hydrology variables for 1991 photo
plots. Top number is correlation coefficient, bottom number is p value. A p value <
0.05 is considered significant. ................ ....................... ..... ... ............ 103

4-7. Differences between variables from 1950 to 1991. N.S. indicates no significant
difference. Positive value for density, percent cover and change in ratio indicate an
increase from 1950 to 1991. Positive change in orientation magnitude indicates a
clockwise change in orientation. A negative change indicates a counter clockwise
shift..................................... .... .......... 104

4-8. Spearman correlations between change in tree island variables and change in
hydrology variables for 28 photo plots in Loxahatchee. Top number is correlation
coefficient, bottom number is p value. A p value < 0.05 is considered significant.. 106















LIST OF FIGURES


Figure page

1-1. Framework for examining the relations between landscape patterns and processes
(Left) and how it is applied in this study (Right). Shaded boxes indicate topics of
Chapters in this work. ................... .... ........ .... ........... 13

1-2. Location of Loxahatchee National Wildlife Refuge within the Everglades Ecosystem.
Arrows show general direction of historic sheet flow. Shaded area indicates the
extent of the historic Hillsboro Marsh. Adapted from Light and Dineen 1994 and
Parker 1984. .................... ... .... ..... ........ ......... .. ... .... ..... .......... ... 14

2-1. Hydrology zone boundaries overlain on tree islands identified from satellite image
classification of Loxahatchee National Wildlife Refuge. Classification is from
Richardson et al. 1990. Hydrology zones are numbered left to right from 2 to 89.
See Appendix A for correspondence with overall NSM and WMM row and columns.38

2-2. Relation between hydrologic variables in the Natural Systems Model (Fennema et al.
1994) and the Managed Systems Model (MacVicar et al. 1984). Values are average
values for data from 1965-1990................................................. ....................... 39

2-3. Frequency distribution of tree island size (top), cumulative area (middle), and
contribution of each size class to overall tree island area. Sizes range from 0.01 to
61.97 ha. Eighty-nine percent of the tree island area is from tree islands <= 6.22 ha
(rank 169). ............................................... ............................ 40

2-4. Density per grid cell of tree islands in Loxahatchee National Wildlife Refuge. Data
are from Richardson et al. 1990. Each cell is 1800 x 900 m. X and Y axis are in
UTM coordinates............. ................... .. ....................... 41

2-5. Difference in degrees between mean tree island orientation calculated from satellite
image classification of Loxahatchee National Wildlife and flow orientation from
N S M ........................................ .......................................................... ................. 4 2

2-6. D-values for rank order of tree island area. D-values greater than 0.9 indicate a
potential clump. Small tree islands are < 1.73 ha; medium 1.9 to 3.41 ha; large tree
islands 3.82 to 4.61 ha; extra large tree islands > 5.10 ha..................................... 43

2-7. D-values for rank order of tree island length. D-values greater than 0.9 indicate a
potential clump. Small tree islands are < 40 m in length; medium 45-351 m; large










tree islands 386 to 1173 m. The three largest tree islands represent their own clumps.
.......... .......................... ............. .. ...................... .... ........ ............. .. 44

3-1. Framework developed to relate patterns of tree islands quantified using a classified
satellite image of all of Loxahatchee National Wildlife Refuge and patterns quantified
using aerial photography covering 10% of the refuge....................................... 64

3-2. Location of groups resulting from 5 class cluster analysis from satellite image
classification using the tree island variables number per cell, mean size, SD of size,
and percent cover ...................... ......... ....... .. ...... ....... .........................65

3-3. Relation between number of tree islands in satellite image and photo plots of
Loxahatchee National Wildlife Refuge. Plots are 1800 x 900 m. ........................... 66

3-4. Relation between mean size of tree islands in satellite image and photo plots of
Loxahatchee National Wildlife Refuge. Plots are 1800 x 900 m .......................... 67

3-5. Relation between SD of size of tree islands in satellite image and photo plots of
Loxahatchee National Wildlife Refuge. Plots are 1800 x 900 m............................68

3-6. Relation between percent cover of tree islands in satellite image and photo plots of
Loxahatchee National Wildlife Refuge. Plots are 1800 x 900 m............................69

3-7. Location of groups resulting from 5 class cluster analysis of photo plots using the
tree island variables number per cell, mean size, SD of size, and percent cover. ...... 70

4-1. Location of 1800 x 900 m photo plots in Loxahatchee National Wildlife Refuge.
Large squares are the boundaries of the hydrology zones.................................... 107

4-2. Histograms of mean orientation of tree islands from 1950 photo plots (top) and
orientation of NSM flow for grid cells containing photo plots (bottom)................ 108

4-3. Frequency distribution from all 1950 photo plots of tree island size (top), cumulative
area (middle), and contribution of each size class to overall tree island area. Sizes
range from 0.01 to 21.22 ha. Tree islands < 0.16 ha (rank of 16) make up 50% of
the total tree island area. .............................. ............ ................. .......... 109

4-4. Rank order of tree island area (top) and long axis (bottom) for 1950. Boundaries of
potential clumps are indicated by bars that extend above D >=0.9...................... 110

4-5. Histograms of mean orientation of tree islands from 1991 photo plots (top) and
orientation of WMM flow for grid cells containing photo plots (bottom).............. 111

4-6. Frequency distribution from all 1991 photo plots of tree island size (top), cumulative
area (middle), and contribution of each size class to overall tree island area. Sizes
range from 0.01 to 11.50 ha. Tree islands < 0.13 ha (rank of 13) make up 50% of
the total tree island area. ................................... .................... 112










4-7. Rank order of tree island area (top) and long axis (bottom) for 1991. Boundaries of
potential clumps are indicated by bars that extend above D >=0.9........................ 113

4-8. Location of photo plots showing changes in percent cover. Down arrow indicates a
decrease from 1950 to 1991, up arrow indicates <=5% increase and up arrow +5
indicates > 5% increase ................ ............. ........ 114

4-9, Location of potential area clumps using a criteria of D>= 085 (top) and potential
long axis (bottom) clumps using a criteria of D>=0.9 for 1950 and 1991.............. 115















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

SPATIAL AND TEMPORAL PATTERNS OF TREE ISLANDS IN THE ARTHUR R.
MARSHALL LOXAHATCHEE NATIONAL WILDLIFE REFUGE

By

Laura Ann Brandt

December, 1997


Chairman: Dr. Wiley M. Kitchens
Major Department: Wildlife Ecology and Conservation

A landscape approach was used to examine the relations between patterns of tree

islands and patterns of hydrologic variables in the Arthur R. Marshall Loxahatchee

National Wildlife Refuge, a remnant of northern Everglades wetland. Landscape level

patterns were examined using a classified satellite image. Results from analysis of the

satellite image were used to design a sampling regime for analysis of current (1991) and

historic (1950) patterns of tree islands using aerial photography. Tree island number, size,

shape, and orientation were compared between years and between methods. Patterns of

tree islands were related to the hydrologic variables flow magnitude, flow direction,

hydroperiod, and ponding depth, obtained from two models designed to simulate

hydrologic conditions on the post-managed (Managed Systems Model) and pre-managed

landscape (Natural Systems Model).









Tree island size, shape, and orientation varied considerably throughout the study

area and between methods. Small tree islands made up a larger percent of the total area

covered by tree islands than did large tree islands. The orientation of elliptical tree islands

was not different from the orientation of modeled historic water flows. Historically, range

in hydroperiod and flow magnitudes were related to tree island shape, size, and number.

Current relations between tree islands and hydrologic variables were very different from

historic relations, with current hydroperiod and ponding depth related to tree island size.

Percent cover of tree islands in plots closer to the canals decreased while percent cover of

tree islands in plots in the interior part of the refuge increased. Changes in patterns of tree

islands were correlated with changes in hydrologic variables, with larger changes in the

hydrologic variables correlated with larger changes in tree island variables.

Results from this study illustrate the importance of flow magnitude as well as

hydroperiod and depth in structuring the patterns of tree islands within this peat wetland.

Restoration of historic hydroperiods and depths without historic flow patterns may not be

sufficient to restore or maintain the historic pattern and function of the system.















CHAPTER 1
INTRODUCTION

Overview


Anthropogenic activities have changed landscape patterns world wide. Large scale

human alterations change not only a landscape's appearance, but also it's function

(Forman 1995). The Florida Everglades is an example of how large scale human

alterations have led to changes in the appearance and function of a landscape. Once a

dynamic flow-through system, the Everglades has been transformed into a series of

impoundments and canals. The changes in hydrology are believed to have caused changes

in vegetation communities, water storage capacity, and species diversity (Davis and Ogden

1994). Concern that changes in hydrology, and development of urban and agricultural

areas are contributing to the "death of the Everglades" resulted in multi-agency efforts to

develop Everglades restoration criteria. Successful restoration will not occur without an

understanding of the relations between landscape change and changes in structuring

processes. This study is a step toward understanding such relations. By identifying and

quantifying patterns of tree islands in a remnant part of northern Everglades wetlands and

relating these patterns to changes in hydrologic variables, this study provides a statistical

description of patterns and their associations with process. The results of this study will

provide a better understanding of the importance of hydrologic variables in structuring the









patterns of tree islands In addition, it will provide documentation of changes through

time that can be used to develop hypotheses on the linkages of pattern and process.

The framework followed (Figure 1-1) uses existing data sources available for all of

the Arthur R. Marshall Loxahatchee National Wildlife Refuge (a classified satellite image

and hydrologic models) to answer the questions: Can tree islands be classified by size or

shape? Do tree islands of different size or shape occur in different landscape positions?

Are the differences in distribution related to hydrology and topography? The distribution

of the variables measured using the imagery then are used to determine an appropriate plot

size and shape, and to determine the location of sample plots for the analysis of changes in

tree island patterns over time using aerial photography. Aerial photography is used to

identify tree islands in sample plots that cover 10% of the refuge from 1950 and from

1991. These data are used to answer the questions: Are the 1950 patterns of tree islands

correlated with pre-drainage hydrology? Are the 1991 patterns of tree islands correlated

with post-drainage hydrology? Have the patterns of tree islands changed over the last 40

years? Are the changes in patterns of tree islands correlated with changes in hydrology?

If the patterns of tree islands are correlated with hydrology, what are the most important

hydrologic variables? This framework addresses the structure and change in a landscape,

two of the three major focuses of landscape ecology (Forman and Godron 1986).

Landscape Ecolovg


Landscape ecology is the study of the reciprocal effects of spatial pattern on

ecological processes and has developed as a means of understanding the development and

dynamics of pattern in ecological phenomena, the role of disturbance, and the









characteristic spatial and temporal scales of ecological events (Urban et al 1987).

Knowledge about the relations among the building blocks of the landscape (structure), and

the functioning of the landscape as a system can be used as a basis for planning, managing,

improving, and conserving lands (Zonneveld 1990). Before relations between landscape

patterns and the processes can be made, the elements and the structure of the system must

be ordered and described Description of landscape elements includes the study of

morphology (description of the structure and elements), classification (systematic

ordering), chorology (spatial patterns and variations), and chronology (temporal variation-

e.g. change over time; Zonneveld 1990). These interdependent areas provide the basis

from which the structure, function, and change of landscapes can be studied.

Structure is the spatial relation or pattern of landscape elements including the

distribution of energy, materials, and species in relation to the sizes, shapes, numbers,

kinds, and configuration of the elements. Much research has been conducted recently on

the quantification of landscape patterns (see Turner and Gardner 1991), the identification

of spatial units (e.g. what to quantify), development of measures that relate to ecological

processes (Kepner et al. 1995), and the use of remote sensing in landscape studies

(Quattrochii and Pelletter 1991, Roughgarden et al. 1991). Many studies have focused on

the importance of scale in the identification of landscape patterns (Turner et al. 1989,

Turner et al.1991, O'Neill et al. 1992, Cullinan and Thomas 1992, O'Neill et al.1995).

Scale consists of both extent, the overall bounds of the study, and grain, the size of the

smallest individual unit of observation. The extent and grain of a study must match the

question being addressed and the features being studied or the landscape patterns will be

meaningless. For example, Gardener et al. (1987) have shown that both changing the










grain and the extent can influence measured landscape patterns. In addition, patterns that

appear to be ordered at one scale may appear random when viewed at other scales (Miller

1978). The development of remote sensing techniques makes it possible to conduct studies

of larger extent. To be logistically feasible, studies of larger extent generally have larger

grain. An understanding of the relation between patterns measured over larger extents

and larger grain with patterns observed at smaller extents and finer grain will be important

in our ability to link patterns and processes at multiple scales.

Function is the interaction among the spatial elements, the flows of energy,

materials, and species among the elements. Spatial pattern has been shown to influence

many ecologically important processes such as the spread of fire (Turner and Bratton

1985), the movement of organisms across the landscape (Wegner and Merriam 1979),

species diversity (Milne and Forman 1986), and recolonization (Hardt and Forman 1989).

In addition, driving forces such as hydroperiod, elevation, flow patterns, and disturbance

are known to influence processes such as soil accumulation, erosion, colonization, and

mortality, which in turn, influence spatial patterns such as patch size, shape, and

distribution. The result is a landscape whose pattern reflects the complex interaction

between structure and function.

Change is the alteration in the structure and function of the ecological mosaic over

time (Forman and Godron 1986). Landscapes change naturally through succession,

responses to global climate change, and in response to disturbance. Change in unmanaged

landscapes usually reflects the natural periodicities in the structuring processes such as

rainfall, hydrology, and disturbance. Likewise, changes observed in managed or

developed systems are a reflection of the structuring processes operating under









management guidelines Often systems are managed to reduce variability (Holling 1987)

resulting in processes and patterns unlike those found in natural systems

Understanding the relation between landscape structure, function, and change is

not a simple matter. Unlike smaller scale studies, it is not possible to carry out traditional

experiments where there are controls and replications. We can either conduct experiments

on "landscape microcosms" with the assumption that these can be scaled up to the

landscape or take advantage of "natural experiments." The Everglades is an example of a

large scale "experiment" that may help us to better understand landscape dynamics.

The Everalades


The Everglades, once a vast dynamic flow-through wetland extending from the

Kissimmee chain of lakes (in central Florida) south to Florida Bay (Figure 1-2), has been

severely altered during the last century. Known as the "River of Grass," the Everglades is

unlike any other river system in flow characteristics or vegetation. Historically, water

flowed slowly along the shallow elevation gradient (a change of about 5 m over 200 km)

from Lake Okeechobee to Florida Bay through a mosaic of aquatic slough and wet prairie.

Flow rates reached a maximum of approximately 36 m per h (Holling et al. 1994) during

the June to October rainy season when water upstream created a hydraulic head that

pushed water south. The development of south Florida has led to the conversion of

portions of the historic system to urban and agricultural uses and the conversion of the

once sheet-flow system into a series of highly regulated impoundments and canals. The

vast mosaic of marsh, slough, tree islands, and pinelands has been reduced to half of the

historic 3.6 million ha (Davis and Ogden 1994). In addition, what remains is subject to









hydrologic regimes that, in places, bear little resemblance to pre-drainage patterns (Light

and Dineen 1994).

The historic hydrological patterns were vital in maintaining the heterogeneity of

the landscape which in turn contributed to the persistence and resilience of the system.

The Everglades, like other ecosystems, is the product of forces such as climate, geology,

hydrology, nutrient inputs, and disturbance (DeAngelis and White 1994). These driving

forces operate over a range of spatial and temporal scales that interact to form the current

landscape patterns. Changes in macroscale forces such as geology and climate occur over

long time scales and are generally not noticeable within a human lifetime. At the other

extreme, changes in microscale processes such as nutrient transfer can occur very rapidly

and the consequences to wetland vegetation may be observable in a season. In between

are mesoscale forces and processes that act on a scale of kilometers to tens of thousands

of kilometers and over a time scale of 10 to 100 years. Hydrology is the dominant

mescoscale force acting in the Everglades. It is also the force that has been most seriously

altered and has caused the largest changes to the remaining vegetation in the system.

Changes in hydrologic patterns have contributed to the reduction in plant community

heterogeneity, the compositional change and decrease in wildlife populations, and changes

in the historic functioning of the Everglades. Changes in vegetation structure and

composition have been noted in many regions of the Everglades (Loveless 1959,

Alexander and Crook 1984, Parks 1987, Worth 1988), however, the total extent of these

changes is unknown, nor will the magnitude of the changes ever be completely

understood. Restoration of the Everglades ecosystem requires an historic reference on

what vegetation communities existed and in what spatial arrangement, how historic










processes contributed to the landscape patterns, and wherever possible, what the relation

between anthropogenic changes in structuring processes and current landscape patterns

are.

The Arthur R. Marshall Loxahatchee National Wildlife Refuge


The Arthur R. Marshall Loxahatchee National Wildlife Refuge, or Water

Conservation Area 1, is a 57,234 ha remnant of northern Everglades wetland. The

Loxahatchee NWR was established in 1951 to enhance populations of rare and

endangered species, protect native flora and fauna, and maintain populations of wading

birds and water fowl (Thompson 1970). The U.S. Fish and Wildlife Service has

administrative responsibility for the area which also is used by the South Florida Water

Management District to store water from agricultural fields in the wet season.

Loxhatchee NWR is located within a portion of what was historically known as

Hillsboro Marsh or Hillosboro Lake (Figure 1-2), which is south and west of the

Loxahatchee River and slough. Hillsboro Marsh corresponded roughly to a trough in the

underlying limestone bedrock known as the Loxahatchee Channel and was part of an

uninterrupted wetland extending from Lake Okeechobee to Florida Bay. This channel,

running generally north-west to south-east from Lake Okeechobee to the southern end of

what is now Loxahatchee NWR, was believed to act as an overflow valve for Lake

Okeechobee (Gleason et al. 1984). Unlike many areas of the Everglades where there is

little substrate overlaying the bedrock, the Hillsboro Marsh was an area with a deep peat

base (Stephens and Jones 1951, Stephens 1984). Peat depths in Loxahatchee NWR range

from 1.25 m to over 4.5 m (Richardson and Kitchens unpublished data). Distribution of










vegetation communities in Loxahatchee is correlated with the peat surface topography

(Pope 1991) and includes a mosaic of sawgrass marsh, wet prairie, slough, and tree

islands.

Historically, the hydrologic patterns in Loxahatchee were driven by rainfall (Parker

et al.1955). During most of the year, rain that fell directly on Loxahatchee was the

primary hydrologic input. In wetter periods inputs from upstream, such as overflows from

Lake Okeechobee, also were important. In fact, it may well be that these "pulse" events

were as important in shaping the landscape as were the average conditions. Natural flows

were generally south and east following the topography of the region (Figure 1-1).

Hydrologic changes to Loxahatchee and the rest of the Everglades started in the

1800s with the connection of Lake Okeechobee to the Calossahatchee River to the west.

Additional canals and levees were constructed primarily for flood control and

"reclamation" of the Everglades (Light and Dineen 1994). By 1917 four major canals

(West Palm Beach, Hillsboro, North New River, and Miami) draining water from Lake

Okeechobee to the east were in place and the dynamic sheet-flow from the Lake to

Loxahatchee had been altered. Further hydrologic alterations occurred with the

completion of the St. Lucie canal to the north in 1931. In the early 1950s (1952-1954) the

eastern Levee of Loxahatchee (L-40) was constructed and by 1961 the entire Refuge was

bounded by canals. The result was an impoundment in an area that had been a flow-

through system. Water levels are regulated by the Army Corps of Engineers with water

control structures in the north and south to match a prescribed schedule. Water enters the

Refuge via the surrounding canals and pools behind the dike in the south until it is released









into Water Conservation Area 2A to the south. The result is longer hydroperiods of

deeper depth in the south and shorter hydroperiods in the north.

Tree Islands


Tree islands are a dominant feature of the Everglades and of the Loxahatchee area.

They have been called "...the most striking botanical feature in the Everglades" (Loveless

1959 pp. 4) and "one of the chief scenic features of the landscape" (Davis 1943 pp. 166).

They represent areas of slightly higher elevation where non-wetland plants have been able

to colonize. They are important ecologically as sites of high botanical species richness and

as habitat for species such as wading birds, alligators, turtles, and deer.

Tree islands are an integration of many processes operating over a wide range of

temporal and spatial scales. At the microscale, processes such as the deposition and

removal of sediment around the island, establishment of seedlings, tree growth, and

mortality dominate at time scales ranging from seconds to years. Microscale processes are

influenced by higher level processes/driving forces such as hydroperiod, inundation depth,

and water flow that act over a temporal scale of 1-100 years. These higher level processes

are in turn influenced by events such as hurricanes, floods, and fire and by still higher level

processes such as climate change and geomorphology operating on temporal scales of

hundreds of years. Major differences in the magnitude or frequency of any of these

processes will be expressed by different patterns on the landscape through changes in

physical and biotic mechanisms. The vegetation of a particular tree island is determined by

mechanisms operating on individual plants. For example, whether a tree island has a

particular species will depend on the availability of a seed source and the availability of










appropriate dispersal mechanisms. Dispersal mechanisms include wind, water flow, and

biotic dispersers such as bird, mammals, and reptiles. Once a seed reaches a tree island,

physiological tolerances and life history traits will determine whether or not it germinates.

If the tree island is too wet, does not have the appropriate nutrients or pH the seed may

not be able to germinate before it dies or is moved to another location. If the seed

germinates, competitive interactions among and between species, response to herbivory

and extreme events such as fire, freezes, and floods will determine if the plant survives.

The feedbacks between colonization, growth, competition, death, and decomposition

determine the structure of the tree island.

Tree islands have been described by their general size, shape, and orientation as

either small and circular; "Circular tree islands...are normally quite small in size, ranging

from only about one-quarter acre to five or six acres in extent." (Loveless 1959 pp. 4) or

large and elongated and oriented in the direction of flow (Davis 1943, Jones 1948,

Gleason et al. 1984). It is generally thought that the interaction of topography and surface

water flow resulted in the characteristic shape and orientation of the larger tree islands as

is illustrated by this quote from Davis (1943 pp. 179):

The vegetation of the Tamiami Slough and tree-island area on the western side of
the Everglades is definitely determined by topography and surface water
conditions. On the higher parts, which are low ridges, are island-like groups of
trees, and in the lower parts or sloughs there are aquatic plants and marsh
vegetation. These sloughs and tree-islands are elongated in the direction of flow
of the surface water. In fact the oval shaped tree-islands are "stream-lined" with
their bluntest ends facing the direction of flow of the water, and they seem to have
become this shape as a consequence of the direction of flow of the surface water.

Jones (1948 pp. 48) made a similar reference to tree islands in the Loxhatchee area

as well as for areas farther south.









These islands are usually oval in shape, widest on the upstream end and tapering
toward the lower end. The pattern that is developed therefore appears to indicate
the direction in which water flowed in the past. In Hillsboro Marsh the islands
point southeasterly. In the southwestern section, north of the trail they point
southeasterly and south of the Trail southwesterly.

In these and in more recent writings there is discussion and speculation on the

origin of these tree islands. The discussions focus on two general mechanisms for tree

island formation. The first is that the tree islands have formed on peat ridges or bedrock

outcrops between the sloughs (Loveless 1959). The second is that the tree islands have

formed from mats of floating vegetation or peat "popups." The first is a more plausible

explanation for tree islands in the lower Everglades where limestone outcrops are more

common and a shallow peat layer makes it unlikely that floating peat islands could

develop. It also may be a reasonable explanation for the formation of tree islands in the

eastern portion ofLoxahatchee, where there are more pronounced peat ridges.

The formation of tree islands from peat "popups" or floating vegetation is more

likely in areas such as Loxahatchee where peats are relatively deep (3 to 4 m). There are

three ways that peat "popups" might form: bulges, free floating batteries, or "gator holes".

Bulges are formed when a surface layer of peat becomes separated from the deeper peat

and rises to the water surface, but remains attached to the submerged peat on its periphery

(Cypert 1972). Free-floating batteries occur where the peat breaks loose from the bottom

and rises to the surface. They also may occur with the return of high water following a

period of low water. There is some evidence that this occurs in Loxahatchee (Gleason et

al.1980). Disturbance by alligators also might result in floating peat mats. As an alligator

creates it's wallow the underlying substrate is pushed out of the way. This could result in

either a floating peat mat or a local topographic high, either of which could lead to tree










island formation In Loxahatchee, tree islands are often associated with alligator holes

(Silveira 1996). Once these areas of peat reach the surface they can be colonized first by

aquatic vegetation such as sawgrass (Cladium jamaicensis) and Eleocharis. as these plants

die, the mass increases making it possible for larger more woody plants to become

established. A similar successional pattern from floating mat to tree island probably

occurs on the floating vegetation mats described by Davis (1943) and Jones (1948).

As illustrated above, hydrology is thought to play an important role in the

distribution and form of tree islands. Alterations in the hydrology in Loxahatchee provide

an opportunity to study the effects of changes in landscape patterns in relation to changes

in structuring forces and to develop hypotheses concerning the linkage of pattern and

process.

Dissertation Structure


The objective of this research is to identify and quantify the pattern of tree islands

in Loxahatchee and to relate these patterns to selected driving forces that helped to form

them. Chapter 2 describes the patterns of tree islands in Loxahatchee in relation to

selected potential driving forces using a coarse view that encompasses the entire refuge.

Chapter 3 addresses the problem of subsampling in a heterogeneous environment and

compares the results of a coarse-scale and fine-scale analysis. Chapter 4 addresses the

question of patterns of change over time by using photo plots from 1950 and 1991. These

patterns are discussed in the context of current and historic hydropatterns. Chapter 5

summarizes the results of the previous chapters and discusses directions for future

research.










Identify Patterns Identify Processes

Identify Potential
Spatial and Temporal
Scales

Determine Extent
and Grain of Study


Quantify Patterns Quantify Processes

Microscale Microscale

M esoscale FMeoscale


SMacroscale I


L


Macroscale


Make associations and linkages
between patterns and process
within and between scales


Tree islands are small and Tree islands are a result of
round or large and interactions of topography
elongated and oriented in and hydrology (Loveless
the direction of flow (Davis 1959)
1943, Loveless 1959)
Tree islands, hydrology, and
topography are mesoscale,
but influenced by micro and
macroscale processes.
Plot size. snap. location
and percent of refuge
(Chapter 3)
Quantify Patterns Quantify Processes

Mesoscale- Mesoscale-
Tree Island density H r period
location

shape. orlentaton
(Chapters 2 and 4) ,Fio*


Make associations between panerns
and processes within scales
(Chapters 2 and 4)


Figure 1-1. Framework for examining the relations between landscape patterns and processes (Left) and how it is applied in this study
(Right). Shaded boxes indicate topics of Chapters in this work.


J


I





















































Figure 1-2. Location of Loxahatchee National Wildlife Refuge within the Everglades
Ecosystem. Arrows show general direction of historic sheet flow. Shaded area indicates
the extent of the historic Hillsboro Marsh, Adapted from Light and Dineen 1994 and
Parker 1984.















CHAPTER 2
DISTRIBUTION OF TREE ISLANDS IN LOXAHATCHEE NATIONAL WILDLIFE
REFUGE IN RELATION TO HYDROPATTERNS AND ELEVATION

Introduction


The linkage of spatial patterns and ecological processes at landscape scales has

been identified as a need of both basic research and for solving applied environmental

problems (Turner and Gardner 1991). Many of today's environmental issues are related

to how the landscape has changed in response to changes in environmental processes,

often as a result of human alteration. Evaluation of future consequences of these changes

will require a landscape perspective based on identification and quantification of the

linkages of pattern and process. Successful linkage of pattern and process will include:

identification of the pattern, identification of the processess, identification of appropriate

temporal and spatial scales, quantification of the pattern, quantification of the processess,

identification of associations between pattern and process, and linkage of the patterns and

process through causal mechanisms. Because of interactions between pattern and process,

within and between scales, this is not a linear process. A framework for such

investigations is presented in Figure 1-1. This framework is used in this study to

investigate the potential associations between patterns of tree islands (areas of trees that

occur on elevations slightly higher than the surrounding marsh) in Loxahatchee National

Wildlife Refuge and hydrology.









Landscape patterns are the complex interaction of driving forces operating at

different temporal and spatial scales (DeAngelis 1994, Holling 1986, 1992). Holling

(1992) has proposed that the landscape forms a hierarchy that contains breaks in object

sizes, object proximities, and textures at particular scales. These breaks in geometry in the

landscape occur because structuring processes exert their influence over defined ranges of

scale. Identification of the patterns of discontinuities in landscape structure may give us

insight into the nature and scale of historic processes operating on the landscape and help

us to assess the impacts of changes to these processes.

Tree islands are a predominant landscape element in the Everglades. They provide

habitat for a wide range of species including alligators, wading birds, and deer and have

been called "the most striking botanical feature in the Everglades" (Loveless 1959 pp. 4)

and "one of the chief scenic features of the landscape" (Davis 1943 pp. 179). These and

other authors have noted the general shape and orientation of these islands to be small and

circular; "Circular tree islands.. are normally quite small in size, ranging from only about

one-quarter acre to five or six acres in extent." (Loveless 1959 pp. 4) or large and

elongated and oriented in the direction of flow (Davis 1943, Jones 1948, Gleason et

al. 1984). There is geographic variation in the size and shape of tree islands throughout

the Everglades. Tree islands in the central part of the lower Everglades are large and

teardrop shaped (Wade et al. 1980), while small circular tree islands are prevalent in

Loxahatchee and in the marl prairies of the lower eastern Everglades. Large elongated

strand tree islands occur in Loxahatchee and in the other Water Conservation Areas.

Though the origin of tree islands is not well understood, it is generally thought that the









interaction of topography and surface water flow resulted in the characteristic shape and

orientation of the tree islands (Davis 1943; Jones 1948)

Teardrop-shaped tree islands are thought to have formed on pockets of soil that

accumulates downstream of bedrock outcrops or other obstructions. Their characteristic

tail is formed as a result of soil accumulation as dictated by water flow magnitude and

direction. There are several hypotheses for the formation of strand tree islands including

Egler's (1952) suggestion that the entire Everglades basin was a continuous swamp forest

and that fire restricted tree islands to their present locations. Similar degenerative

processes are believed to have structured the majority of tree islands in the boreal

peatlands, where there is evidence that forested areas have been replaced with wetlands

through flooding and erosion (Glaser 1987). Little evidence exists for such processes

operating in the Everglades. Other hypotheses of tree island development focus on

additive processes rather than degenerative processes. Elongated tree islands are believed

to have developed as a result of a general trend of emergence from long-flooded marsh to

occasionally-flooded bayhead (Gleason and Stone 1994), while smaller more circular tree

islands may have formed in relation to solution features that produced local ponded areas

that developed peat deposits as a result of hydrophytic community structure (Robertson

1953).

It also has been suggested that small round tree islands in Loxahatchee and other

deeper peat areas such as the Okeefenokee Swamp in Georgia, are related to the

development of floating mats (Gleason and Stone 1994). Patterns of tree islands in some

of the prairie regions of Okeefenokee appear similar to those in Loxahatchee, though they

do not have the characteristic alignment that is observed in the Loxahatchee islands. This









difference in patterns may be due to differences in water flow patterns. Okeefenokee

experiences very little flow while Loxahatchee, at least historically, experienced sheet-flow

especially during the wet season.

If tree islands of different types are a consequence of different hydrologic

processes, changes in hydrologic variables may affect the patterns of formation and

development of tree islands of different types. Hagenbuck et al. (1974) suggested that

longer hydroperiods might favor the development of tree islands formed from peat

batteries by allowing detritus to build up as a loose mat. This unattached mat would be

more prone to becoming free-floating and promote the formation of "popup" battery tree

islands. At the same time, longer hydoperiods and deeper depth might drown lower-

elevation tree islands.

To assess the potential impacts of altered hydrology (including flow magnitude and

direction, inundation depth, and length of inundation) on tree islands of different types at

the landscape scale we must first have some way of distinguishing different types of tree

islands. Types may be distinguished by form, position, distribution, or association with

other tree islands. Once types have been identified their occurrence in the landscape can

be related to environmental variables to assess the relations between pattern and process.

The pattern of landscape elements is a result of driving forces operating at a wide

range of temporal and spatial scales. In the case of tree islands it has been hypothesized

that topography and hydrology are the primary driving forces that have shaped the

distribution and form of the tree islands. Here these hypotheses are addressed by

examining the distribution of tree islands in relation to environmental variables in a deep

peat wetland in the Everglades. Specifically, the following questions are addressed: Can









tree islands in Loxahatchee be classified by size or shape? Are there differences in the

distribution of tree islands of different types? Can differences in distribution and type of

tree islands be related to hydrologic and topographic variables?

Methods


Study Site


The Arthur R. Marshall Loxahatchee National Wildlife Refuge (Loxahatchee) is a

57,324 ha area of northern Everglades wetlands. Located in Palm Beach County, Florida,

south and east of Lake Okeechobee (Figurel-1), it was once a connected part of the

historic Everglades system. Loxahatchee is located over the Loxahatchee Channel, a

shallow peat-filled depression extending from Lake Okeechobee south and east that may

have acted as an overflow valve for Lake Okeechobee, shunting water east and south

(Gleason et al. 1984). Changes to the flow patterns in and around Loxahatchee started as

early as the 1800s with the construction of the Caloosahatchee canal, which moved water

to the west from Lake Okeechobee and continued with the completion of the St. Lucie

canal, which moved water to the east from Lake Okeechobee, in 1931. Additional

changes occurred with the completion of the Hillsboro Canal in the 1920s and with the

complete enclosure of the area by the L-7 canal on the west and the L-40 canal on the east

during the 1950s (Light and Dineen 1994). The construction of these canals changed a

once-dynamic flow-through system to an impounded marsh with the majority of non-

rainfall inputs being shunted around the marsh via the exterior canals.

Loxahatchee is a peat-based marsh system consisting of a mosaic of sloughs and

wet prairie to sawgrass, brush, and tree islands. Tree islands are one of the dominant









features of the landscape and provide habitat for a large number of wildlife species

including deer, alligators, and wading birds.

Hydrology and Elevation

Data on hydroperiod, water depth, flow direction, and magnitude were obtained

from the South Florida Water Management District. Data are from two models: the

Natural Systems Model version 4.4 (NSM) which simulates the hydrologic response of

pre-drainage south Florida using climatic data from 1965-1990 (Fennema et al. 1994), and

the South Florida Water Management Model (WMM) which was developed to simulate

the hydrology of the water management system in south Florida (MacVicar et al. 1984).

Both models have a grid cell size of 3.2 km x 3.2 km (2 miles x 2 miles). Grid cells are

refereed to as hydrology zones though out this text. Depth, flow direction and magnitude

are the yearly average values for each hydrology zone calculated over the entire period of

record (1965-1990). Direction was standardized to 0-180 degrees to match the range

used for the tree island orientation and then transformed using tan(theta /2). Two

variables were used to represent hydroperiod: the first was the 90% hydroperiod or the

hydroperiod that was exceeded in 90% of the 26 years. The second hydroperiod variable

was range in hydroperiod. This was used to reflect the variability among the hydrology

zones.

The topographic surface generated by Richardson et al. (1990) for Loxahatchee

NWR was used for this analysis. The grid cell resolution of this surface was 183 m x

183m. The 3.2 x 3.2 km cell boundaries for the hydrology models were overlain on the

elevation surface and average elevation and elevation gradient were calculated for each

zone of the hydrology model.









Identification and Characterization of Tree Islands

Tree islands were identified from an existing land cover classification for

Loxahatchee developed from merged, HIS transformed, 10 m panchromatic data and 20 m

SPOT data (Richardson et al. 1990). Details of the classification methods are available in

Richardson et al. 1990. Two tree island classes described as lower stature tree island

community made up of a mix of wax myrtle (Myia cerifera), dahoon holly (Ilex cassine)

and red bay (Persea borbonia)", and "core of larger tree islands, larger stature trees made

up primarily of dahoon holly and red bay..." (Richardson et al. 1990 pp. 40) were used in

this analysis (Figure 2-1). Fragstats (McGarigal and Marks 1995) was used to calculate

area, perimeter, nearest neighbor (edge to edge), and proximity index for each tree island.

The proximity index quantifies the spatial context of a tree island in relation to its

neighbors. The index equals the sum of tree island area divided by the nearest edge-to-

edge distance squared between the tree island of interest and all tree islands within a 500

m radius (McGarigal and Marks 1995). All other things being equal, a tree island located

in an area with more tree islands will have a larger index value than one located in an area

with fewer tree islands. Tree islands > 0.04 ha were classified as circular, elliptical, or

irregular using a combination of a circularity index (Miller 1953) and ellipse index. These

indices are the ratio of the area of the tree island to the area of a circle or ellipse

respectively given the same long axis and secondary axis. Tree islands with a circularity

index of 0.5 or larger were considered circular. Tree islands with a circularity index < 0.5

were elliptical if their ellipse index was >0.6 and < 2.5. All remaining tree islands were

classified as irregular. Orientation was determined for elliptical tree islands by calculating

the direction of the long axis using a program developed in S-plus (Statistical Sciences,









Inc. 1995). Orientation ranges from 0 to 180 with 0 as north and 180 as south. No

attempt was made to distinguish the leading edge of the tree island from the trailing.

Orientations were transformed using tan(theta/2) for analysis with the linear variables.

Analysis

Because the data did not meet the assumption of normality, the following

transformations were performed prior to analysis: number of tree islands = sqrt(number of

tree islands); long axis = In(long axis).

To test if significant size groups of tree islands (total area and long axis) could be

identified I used methods developed to examine the discontinuities in body mass (Holling

1992). Holling (1992) has proposed that processes that operate at different scales

differentially structure the landscape and that this lumpy structure is reflected throughout

the ecological hierarchy from species body mass distribution to landscape physiography.

To test this, several methods have been developed. The first is the Body Mass Difference

Index (BMDI), the second is the Gap Rarity Index (Marples unpublished). Though these

primarily have been used in the analysis of species body-mass distributions they should be

able to identify clumps in any size-related variable. Tree island area was used because it is

analogous to body mass, reflecting an integration of a range of processes, and long axis

because it may more directly reflect the processes relating to water flow that have shaped

the tree islands.

The BMDI was used on a ranked list of log tree island size (ha) and is calculated

as BMDI=(S,+i-S.,I)/(S,)' where S, is the size of the nth tree island in the ordered list and

x is an exponent sufficient to detrend the data. The results are plotted as a histogram

showing the differences between adjacent sizes. Large differences indicate a relatively









large difference between adjacent sizes. Very large differences suggest gaps between

sizes.

The Gap Rarity Index compares each data set to a random sample drawn

repeatedly from a continuous null distribution bootstrappingg). The output of the Gap

Detector is a discontinuity distribution value (D-value) for each size. Gaps in the size

distribution are indicated by high values. The advantage of this method is that the D-

distribution has known statistical properties that allow for an assignment of a significance

value, in this case 0.05.

Comparisons between tree island variables of different size or shapes were made

using a Wilcoxon Test. Relations among individual variables were examined using

correlation analysis. Comparison of distributions were done using a Kolmogorov-

Smirnov goodness of fit test. Paired-t-tests or Signed-rank tests were used to compare

variables summarized by cell.

The relations between tree island variables (mean size, mean long axis, number of

circular and number of elliptical tree islands per cell, or number of tree islands of different

size by cell) and hydrologic variables (NSM magnitude, NSM hydroperiod, NSM depth,

NSM range in hydroperiod, WMM magnitude, WMM hydroperiod, WMM depth, WMM

range in hydroperiod, mean elevation, and elevation gradient) were examined using

canonical correlation analysis. Canonical correlation analysis is used to investigate the

relationship between two sets of variables by deriving a linear combination of the X

variables (U=aXI + aX2 + ... +aXi) and a linear combination of the Y variables (V=aYI +

aY2 + ... +aYi) such that the correlation between U and V is as large as possible (Manly

1986). A non-significant result indicates that the largest canonical correlation can be









accounted for by sampling variation alone. Canonical correlation combines the multiple

variables into a single index variable (U or V). Examination of the coefficients of the

individual variables that make up U or V indicates the contribution of that variable to the

overall index. Variables with higher coefficients are more important. In these analyses the

hydrologic variables are the X variables and the tree island variables are the Y variables.

Statistical analysis was done using S-plus (Statistical Sciences, Inc. 1995), SAS

(SAS Institute Inc. 1989), or SigmaStat (Fox 1994).

Results


Hydrology and Elevation

Hydrologic variables were significantly different between the NSM and WMM.

Flow direction by grid cell for the NSM and WMM averaged 187.2 a standard

deviation of 26.9 and 200.151.32 respectively and were significantly different (Signed-

rank-test, p=0.028, W=-443.0). Flow magnitude, average hydroperiod and depth, and

range in hydroperiod for the NSM and WMM were significantly different (Paired t-test,

p<0.01, t=23.3; Wilcoxon signed-rank-test, p=0.0001, z=3.8158, p=0.0032, z=2.9525;

p=0.0011, z=-3.2536; Figure 2-2). For both models cell hydroperiod and depth were

highly correlated (p=0.0001, r=0.91 and r=0.96, for NSM and WMM). Hydroperiod and

depth were not correlated between the models. Mean cell elevation was highly correlated

with WMM hydroperiod and depth (p=0.0001, r=-0.93 and r= 0.91), but not with NSM

hydroperiod and depth. NSM hydroperiod range was highly correlated with NSM

hydroperiod and depth (p=0.001, r=-0.89 and r=-0.88), but WMM hydroperiod range and

WMM hydroperiod and depth were not.









Tree Islands

Tree island variables were examined for all tree islands, for tree islands of different

shapes and, for tree islands of different size based first on area and then by length. The

land cover classification revealed two thousand one hundred forty four tree islands that

ranged in size from 0.01 ha to 61.97 ha. Twenty-five percent of the tree islands were <

0.02 ha, 48% <= 0.04 ha and 75% < 0.15 ha. The distribution of tree island sizes showed

many small tree islands and few larger tree islands (Figure 2-3). The distribution of sizes

did not fit any of the common distributions (normal, log normal, poisson, exponential, or

weibul). Fifty percent of the tree island area was made up of tree islands < 6.22 ha. This

is 89%/ of the observed tree islands.

Nearest neighbor distance averaged 103 +152 m and proximity index 80 + 422.

There was a strong geographic trend in the distribution of tree islands. The general trend

reflected more tree islands in the northwest than southwest and more in the east than the

west (Figure 2-4).

Six hundred thirty three of the tree islands larger than 0.04 ha met the criteria for

being elliptical and 335 as being circular. Circular tree islands were significantly smaller

than elliptical tree islands (Wilcoxon rank-sum, p=0, z=-19.2772). No circular tree islands

were > 0.6 ha. Circular tree islands had significantly closer nearest neighbors (mean =

126 150 versus 106 168 m; Wilcoxon rank-sum, p=0, z-5.548) and had a significantly

smaller proximity index than elliptical tree islands (mean 22 168 versus 32 332 ,

Wilcoxon rank-sum, p=0, z=-4.090). There was no difference in the spatial distribution of

circular and elliptical tree islands in either the east to west direction or the north to south

direction. There was a significant difference in the spatial distribution of smaller (< 0.6 ha)









elliptical tree islands and larger elliptical tree islands in the east to west direction

(Kolmogorov-Smirnov goodness-of-fit test; p=0.00, ks = 0.2233), but not in the north to

south direction. Orientation of elliptical tree islands averaged 168.87 degrees with a

circular standard deviation of 59.35 degrees. The distribution was not normal, with the

majority of the orientations between 150 to180 degrees and 0 to 50 degrees. Tree islands

< 0.6 ha had significantly different orientations than those > 0.6 ha (166.7866.23 and

172.37+40.86 respectively, p-0.03, F=4.79). Mean tree island orientation per grid cell

was not significantly different from the flow orientation by grid cell for the NSM;

however, the difference between the mean tree island direction and the model direction

varied by hydrology zone (Figure 2-5). Mean tree island orientation by grid cell was

significantly different than the flow orientation by grid cell for the WMM (Signed-rank-

test p=0.00, W=-769).

Elliptical tree islands were not at significantly different elevations than circular tree

islands. However, elliptical tree islands > 0.6 ha occurred at significantly lower elevations

than smaller elliptical tree islands (one-sided Wilcoxon test, p=0.009, z=2.7055) and

circular tree islands (one-sided Wilcoxon test, p-0, z=-2.7000). Elliptical tree islands <

0.6 ha and circular tree islands were not in significantly different elevations.

The two methods of clump analysis revealed similar results for both area and long

axis. Four potential clumps were identified in the area data (< 1.73 ha, 1.9 to 3.41 ha, 3.82

to 4.61 ha, and > 5.10 ha) and 3 main clumps in the long-axis data (< 40 m, 45-351 m, and

386-1173 m) Figure 2-6 and 2-7.

Summary statistics for the different sized tree islands are presented in Table 2-1.

Nearest neighbor distances and the proximity index were significantly different among the









area size classes (Kruskal-Wallis test, p = 0, X2=62.6269, df= 3; p=0.001, X2=20.3405

respectively) and among length size classes (Kruskal-Wallis test, p = 0, X2=54.81, df= 2;

p=0.001, X2=17.69 respectively).

Orientation of tree islands were not significantly different among area sizes classes,

but were significantly different among length size classes (Kruskal-Wallis test, p=0.0017,

X2=12.735, df-2).

Spatial distribution of tree islands of all size groups, by area, and by length, were

significantly different in the east to west direction (Kruskal-Wallis test, p = 0, X2=28.439,

df= 3; p=0, X2=37.35, df= 2 respectively), but not the north to south direction.

The canonical correlation of the hydrology variables (NSM magnitude, NSM

hydroperiod, NSM depth, NSM range in hydroperiod, WMM magnitude, WMM

hydroperiod, WMM depth, WMM range in hydroperiod, mean elevation, and elevation

gradient) and the tree island variables (mean size, mean long axis, number of circular and

number of elliptical tree islands) yielded two significant correlations between the linear

combination of the hydrology variables (hydrology index) and the linear combination of

the tree island variables (tree island index) (p = 0.0001, r = 0.82; p = 0.01, r = 0.71

respectively). Fifty-eight percent of the variability in the first linear combination was

explained. An additional 28% of the variability was explained by the second correlation.

The hydrology variables most associated with the first canonical index were WMM depth

and mean elevation. The tree island variable mean long axis was most associated with the

first tree island canonical index (Table 2-2). The second hydrology canonical index

represented a contrast between NSM depth and NSM hydroperiod and magnitude while

the second tree island canonical index was associated with tree island area.









Similar results were obtained when the tree island variables number of short,

medium, and long tree islands (number by length class) or number of small, medium, large,

or extra large tree island (number by area class) were used instead of number of circular

or elliptical tree islands in the canonical analysis. In both analyses there were two

significant correlations (p = 0.0001, r = 0.82; p = 0.01, r = 0.71 for number by length

class and p = 0.0003, r = 0.81; p = 0.04, r = 0.74 for number by area class) that explained

76 and 77% of the variability respectively. In both analyses, the first hydrology canonical

index was described by WMM depth and elevation respectively and the second by the

contrast between NSM depth and NSM hydroperiod and magnitude. The first tree island

canonical index for the length class analysis described the number of small tree islands

while the second described the mean length and area of tree islands. The first tree island

canonical index for the area class analysis described the number of small tree islands while

the second described the length and the number of small tree islands.

To summarize, the three analyses indicated basically the same pattern. The

variables of WMM mean water depth and mean elevation are highly associated with the

occurrence of small tree islands, and the contrast between NSM depth, NSM hydroperiod,

and magnitude are highly associated with tree island size.

Discussion


Our ability to manage and restore ecosystems depends in part on our

understanding of the association and linkages between pattern and process. This study is

a first step in the process of linking hydrologic variables to the spatial distribution and size

of tree islands, a dominant ecologically important element in the Everglades landscape.









Tree islands in this area of the Everglades can be grouped by shape or size. This study

indicates that the distribution of tree islands of different types are related to hydrologic

conditions. This provides us with a landscape level tool that can be used to evaluate the

potential effects of alternative management scenarios.

Size rather than shape seems to be a better variable for characterizing the

differences in the relationship of environmental variables and types of tree islands.

Circular tree islands are always small (< 0.6 ha). Elliptical tree islands occur in a range of

sizes. Small elliptical tree islands occur in positionally similar locations in the landscape as

circular tree islands. Tree islands of different lengths show the most separation spatially

with most of the long tree islands found on the east side of the refuge in the area

historically categorized as the "wet prairie, tree island, aquatic slough-lakes" (Hagenbuck

et al. 1974).

Identification of potential clumps in the tree island size distributions provides the

starting point for examining ecological differences among tree islands of different types.

Larger, elongated tree islands may be ecologically different than the smaller tree islands.

Different size or shaped tree islands may have different origins, different vegetation

composition and structure, and may respond differently to changes in water management.

Numerous authors have indicated that large tree islands are lower in elevation and have

vegetation that is more tolerant to flooding than small tree islands. None of these authors

have defined large. Based on the clump analysis performed here, large tree islands in

Loxahatchee are > 350 m in length and/or >3.8 ha in area, while small tree islands are <=

40 m in length and/or <1.73 ha The next step is to test the hypothesis that tree islands in









the different size classes are different in species composition, vegetative structure and

wildlife use by conducting field studies.

Tree islands are an expression of processes operating at a range of temporal and

spatial scales. The current distribution of tree islands is undoubtedly related to historic

conditions and current conditions as has been shown in the canonical correlation analysis.

Current mean ponding depth and elevation are strongly associated with the number of

small tree islands. As ponding depth decreases the number of small tree islands decreases.

This may be a reflection of two ongoing processes. The first is the expansion of tree

islands in the shallower areas. Higer and Kolipinski (1987) noted that in Shark Slough

shorter periods of inundation corresponded to an increase in the growth of woody

vegetation. This appears to be what is happening in the northern portion of Loxahatchee.

The second process is the drowning of tree islands in the lower elevation areas. The

elevation gradient in Loxahatchee is generally north to south. The lower elevation areas in

the south end of the refuge have experienced the greatest changes in hydrology. A

primary result has been the destruction of tree islands due to high water (Richardson et

al.1990). The association of lower elevation areas with lower numbers of small tree

islands may simply be a reflection of the lack of tree islands in these areas in general.

Of secondary importance in explaining the relations between the hydrologic and

tree island variables were the association between NSM volume, hydroperiod, and depth

and the size and number of tree islands. The association of shorter tree islands with areas

of higher historic flow was unexpected. It was expected that longer tree islands would be

found in areas of higher flow. It may be that areas that had higher historic flow also

experienced the most extreme ranges in flow magnitudes and that it is the extremes that









have more of an influence on the tree island size and shape than the average flow. It also

may be that the true length of the tree islands was not measured in this analysis. The

vegetation classes used are the vegetation classes that make up the core area of the tree

island and do not include the lower brush areas that can be associated with tree islands,

particularly the downstream tails (Ward et al.1980). Had those brushy areas been included

in the calculation of tree island long axis the results may have been different.

The orientation of elliptical tree islands gives an indication of historic flow patterns

and is consistent with the predictions of the NSM. This relation, along with the

association between historic flow magnitudes and form and number of tree islands has

implications for the prediction of future hydrologic modifications. Primary focus of

restoration efforts have been on establishing appropriate hydroperiods and depth. In as

little as five years, changes in hydropatterns can cause changes in herbaceous vegetation

(Hagenbuck et al. 1974). Changes in water depth can drown tree islands in a matter of

years. Relatively little research has been conducted on the influence of changes in flow

direction and magnitude in the Everglades system. Flow magnitude and direction have a

significant effect on nutrient, sediment and debris transport which in turn influences tree

island form and function. Though changes may not be evident over the short term,

patterns of tree islands in Loxahatchee are predicted to change in response to the changes

in flow dynamics, becoming less oriented in the direction of historic flow and more

randomly distributed. A study comparing patterns of tree islands in 1950 to patterns of

tree islands in 1991 is currently underway to see what changes have occurred in 40 years.

In interpreting the results of this study several factors relating to the sources and

scales of the data sets must be considered. The NSM predicts pre-drainage conditions









based on data from 1965-1990. By 1965 extensive drainage had already occurred in south

Florida so NSM values do not represent true historic conditions, but rather a "best guess".

The models used for this analysis were designed to show general patterns across the south

Florida ecosystem, and may not be appropriate when analyzed on a cell by cell basis. As

such, the 3.2 x 3.2 km grid cell size is very coarse in relation to the processes operating on

individual tree islands. The assumption has been made that the tree island values of size,

shape, and orientation summarized by cell are related to average hydrology values within

each cell. Additionally, the method used for identifying tree islands has the disadvantage

that it was created for another purpose in which tree islands were not the focus. Satellite

imagery allows for classification of large areas, but has the disadvantage that the elements

identified are a function of the pixel size and shape. Only elements whose sizes are

multiples of pixel sizes can be properly identified. This can obscure the relation between

size of the element and environmental variables. Elements (tree islands) smaller than the

minimum pixel resolution will not be identified, and boundaries of elements larger than the

minimum resolution may be obscured. A study examining the relation between tree island

patterns (size, shape, orientation, and spatial distribution) distinguished from different

media is currently underway.

Despite these methodological problems, this study provides insight and

quantification of the relation between tree island size, shape, and distribution, and

environmental variables. These results support hypotheses on the differentiation of tree

islands by size and provides a rough idea of where to start looking for tree islands of

specific characteristics. The ability to use landscape level tree island characteristics (length,

area, orientation, and location) as indicators of differential ecological characteristics






33


(elevation and vegetation and hence ecological function) provides us with a valuable tool

for evaluating alternative management scenarios.









Table 2-1. Summary statistics for tree islands in Loxahatchee Natioanal Wildlife Refuge identified from satellite imagery.

Size Class n Median nearest Mean nearest Median Mean Mean
neighbor neighbor proximity proximity orientation
distance (m) distance (m) index index SD (degrees)
SD

Small area <= 1.73 ha 983 63.0 116.7157.5 1.2 31.8292.3 145.5
Medium area 1.74-3.82 ha 34 9.0 72.1237.7 2.4 4.25.5 159.4
Large area 3.83-5.10 ha 6 23.1 78.5123.7 0.9 1.641.8 173.5
Extra large area > 5.10 ha 32 9.0 21.430.3 5.5 12.820.2 171.9

Small length <=40 m 336 71.5 109.9135.2 1.1 65.8481.3 133.7
Medium length 41-351 m 670 56.9 119.6+173.0 1.3 13.892.5 150.4
Large length >351 m 49 9.0 25.4+50.3 3.6 10.0+17.1 168.6






35



Table 2-2. Canonical variables for analysis using shape classes. Most significant
coefficients are bold faced.


UI = -0.06 NSM magnitude + -1.12 NSM hydroperiod + 0.71 NSM depth + -0.56 NSM
hydroperiod range + -0.13 WMM magnitude + 0.96 WMM hydroperiod + -3.66 WMM
depth + -0.58 WMM hydroperiod range + -3.07 Mean elevation + 0.39 Elevation
gradient.

VI = 0.11 Mean long axis + 0.01 Mean area + -0.88 Number of circular tree islands + -
0.13 Number of elliptical tree islands.

U2 = -1.03 NSM magnitude + -1.14 NSM hydroperiod + 1.60 NSM depth + -0.22
NSM hydroperiod range + -0.22 WMM magnitude + 0.87 WMM hydroperiod + -0.16
WMM depth + 0.31 WMM hydroperiod range + -0.03 Mean elevation + -0.43 Elevation
gradient.

V2 = 0.42 Mean long axis + 0.64 Mean area + 0.33 Number of circular tree islands +
0.16 Number of elliptical tree islands.






36



Table 2-3. Canonical variables for analysis using length classes, Most significant
coefficients are bold faced.


UI = -0.06 NSM magnitude + -1.03 NSM hydroperiod + 0.57 NSM depth + -0.84 NSM
hydroperiod range + -0.14 WMM magnitude + 0.95 WMM hydroperiod + -3.41 WMM
depth + -0.50 WMM hydroperiod range + -2.94 Mean elevation + 0.45 Elevation
gradient.

VI = -0.001 Mean long axis + 0.05 Mean area + -0.74 Number of short tree islands + -
0.34 Number of medium tree islands + 0.17 Number of long tree islands

U2 = -1.09 NSM magnitude + -1.31 NSM hydroperiod + 1.65 NSM depth + -0.14
NSM hydroperiod range + -0.30 WMM magnitude + 0.80 WMM hydroperiod + -0.18
WMM depth + 0.31 WMM hydroperiod range + -0.34 Mean elevation + -0.42 Elevation
gradient.

V2 = 0.63 Mean long axis + 0.59 Mean area + 0.28 Number of short tree islands + 0.02
Number of medium tree islands + -0.28 Number of long tree islands,






37

Table 2-4. Canonical variables for analysis by area classes. Most significant coefficients
are bold faced.


UI = -0.33 NSM magnitude + -1.16 NSM hydroperiod + 0.91 NSM depth + -0.71 NSM
hydroperiod range + -0.20 WMM magnitude + 0.88 WMM hydroperiod + -3.08 WMM
depth + -0.27 WMM hydroperiod range + -3.03 Mean elevation + 0.29 Elevation
gradient.

VI = 0.14 Mean long axis + 0.15 Mean area + -0.96 Number of small tree islands + -
0.14 Number of medium tree islands + 0.19 Number of large tree islands + 0.01 Number
of extra large tree islands.

U2 = 1.17 NSM magnitude + 1.48 NSM hydroperiod + -1.74 NSM depth + -0.01
NSM hydroperiod range + -0.26 WMM magnitude + -0.16 WMM hydroperiod + -0.65
WMM depth + -0.66 WMM hydroperiod range + 0.15 Mean elevation + 0.51 Elevation
gradient.

V2 = -0.90 Mean long axis + -0.28 Mean area + -0.78 Number of short tree islands +
0.55 Number of medium tree islands + 0.31 Number of long tree islands + 0.09 Number of
extra large tree islands.












3 4 5 6 7 8 9




10

'. :'
18


26,


34


i I t I
42 t '


50


58 "


66


74


82

2 0
Figure 2-1. Hydrology zone boundaries overlain on tree islands identified from satellite
image classification of Loxahatchee National Wildlife Refuge. Classification is from
Richardson et al. 1990. Hydrology zones are numbered left to right from 2 to 89. See
Appendix A for correspondence with overall NSM and WMM row and columns.


















0 100 200 300

NSM hydroperiod (days)


0.0 0.2 0.4 0.6

NSM mean ponding depth (meters)


5 0 40000 80000 120000 3 0 100 200 300

NSM volume (cubic meters x 10"3) NSM hydroperiod range (days)

Figure 2-2. Relation between hydrologic variables in the Natural Systems Model (Fennema et al. 1994) and the Managed
Systems Model (MacVicar et al. 1984). Values are average values for data from 1965-1990.













400






50


Rank of size
2000

100





























Rank of sIze
700






























Figure 2-3. Frequency distribution of tree island size (top), cumulative area (middle), and
contribution of each size class to overall tree island area. Sizes range from 0.01 to 61.97
ha Eighty-nine percent of the tree island area is from tree islands <= 6.22 ha (rank 169).
ha. Eighty-nine percent of the tree island area is from tree islands <-- 6.22 ha (rank 169).















0







0
















00^00 ^










Figure 2-4. Density per grid cell of tree islands in Loxahatchee National Wildlife Refuge.
Data are from Richardson et al. 1990. Each cell is 1800 x 900 m. X and Y axis are in
UTM coordinates.






















































Scal-
= Kilometers
2 0


Figure 2-5. Difference in degrees between mean tree island orientation calculated from
satellite image classification of Loxahatchee National Wildlife and flow orientation from
NSM.










_^ Large
Medium Large
0.9 Small X-large

0.8

0.7

0.6



0.4

0.3

02

01




Area (ha)


Figure 2-6. D-values for rank order of tree island area. D-values greater than 0.9 indicate a potential clump. Small tree islands
are < 1.73 ha; medium 1.9 to 3.41 ha; large tree islands 3.82 to 4.61 ha; extra large tree islands > 5.10 ha.











I Small
0.9
0.87
0.7-


Large

I h._


0.6










Lon axis (m)


Figure 2-7. D-values for rank order of tree island length. D-values greater than 0.9 indicate a potential clump. Small tree
islands are < 40 m in length; medium 45-351 m; large tree islands 386 to 1173 m. The three largest tree islands represent their
own clumps.














CHAPTER 3
SAMPLING PLOT SIZE AND LOCATION

Introduction


Concern over large scale development and changes in land use have led to the

interest in developing methods for examining ecological impacts at broad spatial scales

(O'Neill et al 1995). Promising methods include tools such as remote sensing, that can be

used to map land covers over large areas, and methods of identifying appropriate sampling

regimes and scales for a study.

The availability of remotely sensed data over a variety of spatial and temporal

scales makes it possible to examine landscape patterns at different scales and over time.

Aerial photography provides the longest temporal record while satellite imagery (available

in a variety of formats and scales) provides the largest spatial extent. Both methods have

strengths and weaknesses that are related to the minimum mapping unit, delineation of

boundaries, and costs of processing (Burrough 1986).

The increasing availability of satellite imagery at pixel resolutions of 10 to 30 m

makes it possible to create land cover maps over large areas. While these maps may be

appropriate for illustrating broad scale patterns, they may not be appropriate for capturing

patterns when the feature of interest is close to the minimum mapping unit or if land cover

patterns are needed prior to 1970. In such cases, photointerpretation of aerial

photography may be more appropriate. The ability to use both methods together may









strengthen our ability to study landscape spatial and temporal dynamics Using the

methods together requires that the relations between the landscape features and patterns

be comparable between the two methods. A number of studies have examined how well

the amounts of mapped land cover types agree between different methods (Cushnie 1987,

Gervin et al 1985, Jensen et al. 1978, Latty and Hoffer 1981, Quirk and Scarpace 1982).

Fewer studies have examined how the quantification of landscape patterns varies among

methods (Benson and MacKenzie 1995, O'Neill et al. 1995). A common conclusion of

these studies was that agreement between methods varied considerably between study

sites and land cover types. These inconsistencies emphasize the importance of

understanding the relation between mapping methods used together.

Conducting fine scale analysis across a large landscape often is not logistically

feasible. Extrapolating results from a sample to the entire landscape is a difficult task

(Allen and Hoekstra 1992, Meentemeyer and Box 1987, Wiens 1989) and will depend on

the scale of the patterns being observed and the scale of observation. Determining

appropriate sampling units has been a topic of ecological studies for many years (Greig-

Smith 1961, Kershaw 1957, Turner et al.1991). Different sampling methods emphasize

different population properties, resulting in collection of different kinds of data. Some

methods emphasize estimating species composition or inclusion of representative species.

Others strive to minimize the variance in plot to plot measurements. In all cases, plot size

and shape should be based on the question being addressed and the size and the spatial

distribution of the entity being studied {Green 1979 #4650). Plots that are too large mask

the patterns by averaging the values within the plot, while plots that are too small may be









large inadequate to accurately represent the spatial pattern (Meentemeyer and Box 1987,

Milne 1991).

The position of the plots in the landscape influences the interpretation of the

results. Sample sites are usually selected randomly when it is assumed that the spatial

distribution of the variables of interest are homogeneous across the area of study.

Because landscapes are heterogeneous, use of simple random sample locations often is not

appropriate or effective. Whenever possible the area should be divided into homogeneous

(based on the variables of interest) sub-areas and samples selected at random and in

proportion to sub-area size (Green 1979 #4650}. Broad scale land cover developed from

satellite imagery provides a method for identifying homogeneous areas within a landscape

and is a tool for that can be used to select appropriate sample plot sizes and location.

The objectives of this study were to 1) determine plot size, shape, and locations

that accurately and precisely represents the patterns of tree islands as quantified from

satellite imagery, 2) compare the values from the sample of the imagery to the values

obtained from the imagery for the entire study area, 3) test the hypothesis that tree island

patterns in sample plots quantified using a second method, photointerpretation, are not

different from tree island patterns quantified using the imagery, and 4) if the patterns are

different, determine if simple linear correction factors can be used to relate the patterns in

the two methods (Figure 3-1).









Methods


Study Area


The Arthur R. Marshall Loxahatchee National Wildlife Refuge (Loxahatchee), is a

57,324 ha area of northern Everglades wetlands. It is a peat-based marsh system

consisting of a mosaic of slough, wet prairie, sawgrass (Cladium jamaicensi), brush, and

tree islands. Tree islands are one of the dominant features of this landscape and have been

described as small and round or large and elongated and oriented in the direction of water

flow (Davis 1943, Jones 1948, Gleason et al. 1984). The location, size, shape, and origin

of tree islands is thought to be related to geomorphology, hydroperiod, and water flow,

though the exact mechanisms are not well understood {Loveless 1959 #4620}. Tree

islands are not distributed evenly throughout the refuge (see Chapter 2).

Imagery

Tree islands were identified from an existing land cover classification for

Loxahatchee developed from merged, IHS transformed 10 m panchromatic data and 20 m

SPOT data (Richardson et al. 1990) and referenced to State Plane coordinates. Details of

the classification methods are available in Richardson et al. 1990. The classified image

was converted to Universal Transverse Mercator (UTM) resulting in a cell size of 9 x 9 m.

Two tree island classes described as: "lower stature tree island community made up of a

mix of wax myrtle (Myrica cerifera, dahoon holly (Ilex cassine) and red bay (Persea

borbonia)", and "core of larger tree islands, larger stature trees made up primarily of

dahoon holly and red bay..." (Richardson et al. 1990 pp. 40) were used in this analysis

(Figure 2-1). All other classes were considered background. Two tree island layers were









generated for this analysis; the first was an image with tree islands coded as 1, all other

vegetation classes in Loxahatchee as 2, and the area outside of Loxahatchee as 3. In the

second layer each tree island was labeled with a unique number using the Clump command

in Erdas Imagine {Erdas Inc. 1995 #4660) which identifies contiguous groups of pixels of

a class.

Selection of Plots

The selection of plot size, shape, number and location was sequential process in

which the results from the previous analysis determined the starting point for the next

analysis. First an appropriate plot size was selected. Next the percentage of the refuge to

sample, and hence the number of plots was determined. Finally, the plot locations were

selected using stratified random sampling based on the patterns observed in the first

analysis.

Plot size was determined by comparing values from a series of regular grids.

Regular grids of 50 x 50 pixels, 100 x 100 pixels, 200 x 100, and 200 x 200 pixels (length

x width) were generated in ARC/INFO. These grids were imported into Erdas Imagine

and overlain on the three class image (background, tree islands, not tree islands) and the

clumped image (layer with individual tree islands labeled) using the Erdas Imagine

Summary command. Output from these procedures was three data sets for each grid size:

1) the percent cover of tree islands in each grid cell, 2) the number of different tree islands

in each grid cell, and 3) the size of each tree island in each grid cell. Mean, median,

variance, and standard deviation were calculated for percent tree island cover, number of

tree islands and size of tree islands for all grid cells combined at each cell size. These data

were used to determine an appropriate plot size.









To examine the effects of different sampling intensity on the accuracy and

precision of the results, estimates of tree island size, variance in size, number, and percent

cover were calculated 100 times using random samples of grid cells to represent 100%,

20%, 30%, 40%, and 50% coverage of the study area.

Results were compared to the values computed for the entire area (all grid cells of

that size) to determine which plot size and coverage percentage were most accurate and

precise in representing the size, percent cover, and number of refuge tree islands.

Estimates also were compared to the parameter values calculated for the entire refuge

without using the grids. Based on the results from the analysis described in the previous

paragraphs, and the knowledge that longer tree islands tend to orient north-south, a plot

size of 200 x 100 pixels was selected.

An objective of the resampling analysis at different sampling intensities was to

determine the sampling intensity using photography that was both most accurate and

precise; however, because of the amount of effort involved in identifying and digitizing

tree islands from photography (approximately 15 hours per plot) it was determined that

despite the improved precision of a larger sample (see results) only 10% of the refuge

(twenty-seven plots) would be sampled.

To ensure that samples selected for photographic analysis represented the range of

tree island densities and size within the refuge, cluster analysis using Ward's Cluster

Method (SAS Institute Inc. 1989) was used to group grid cells based on percent cover of

tree islands, number of tree islands, mean size of tree islands, and the variance of tree

island size. Grid cells with no tree islands were not included in the cluster analysis, but

were assigned to a cluster of their own. Twenty-seven grid cells (plot locations) were









selected using random samples from each strata based on the proportion of grid cells in

that strata.

Photography

Image sources for this analysis were twenty 1:40,000 color-infrared diapositives

flown in December 1990 or January 1991. Each photo was photogrammetrically scanned

at a scanning resolution to result in 2 m ground resolution. The images were referenced to

a 10 m geocoded SPOT satellite image and mosaicked using Orthoengine from PCI

Remote Sensing Corporation. The result was a complete image of the refuge with

average residual error of approximately 2 m,

Each of the twenty-seven plots was subsetted from the overall mosaic and used as

the background in ARC/INFO. Boundaries of tree islands were screen digitized using the

original photography as reference. Tree island identification was verified by field sampling.

Number of tree islands, area of each tree island and percent cover of tree islands

were calculated for each plot. Comparisons were made between the variables from the

photo plots and the image plots on a plot by plot basis and for all plots combined. A

cluster analysis was used on the results from the photo plots to see if the grouping of the

plots was similar to those from the imagery.

Results


Changing plot size and percentage resulted in larger changes in precision than in

accuracy (Tables 3-1 to 3-3). Sampling more of the refuge resulted in lower variance

(higher precision) for all variables and all plot sizes. The smallest plot size (50 x 50) had

the smallest variance for all variables followed by the 200 x 100 plot size. Only compete









grid cells were used in the above analyses so the amount of the refuge sampled varied

slightly among grid sizes.

Two thousand one hundred forty-four tree islands were identified on the classified

image without the overlaying grid cells. Average tree island size ( SD) was 52.6 23.9

pixels (0.4 ha) Tree islands comprised 1.96 percent of the refuge.

The cluster analysis indicated that grid cells could be divided into five groups

based on the number, average size, and percent tree islands within the cell (Figure 3-2).

The number of cells and the proportion of the total cells in each group were:

Group 1- (low number of tree islands, medium average size, low percent cover)-

30 cells, 12%;

Group 2- (high number of tree islands, medium average size, high percent cover)-

185 cells, 73%;

Group 3- (low number of tree islands, small average size, low percent cover)- 5

cells, 2%;

Group 4- (low number of tree islands, large average size, high percent cover)-6

cells, 3%;

Group 5- (no tree islands)-26 cells, 10%.

Samples for photographic analysis were selected proportionally from each group

so that 10% of the grid cells were represented. The result was random selection of 4 plots

from group 1, 19 plots from group 2, 1 plot from groups 3 and 4, and 2 plots from group

5.

The number of tree islands identified in the plots ranged from 0 to 24 in the image

plots and 0 to 400 in the photo plots (Table 3-4, Figure 3-3). The average values for the









samples using the image plots were similar to the values calculated using the entire refuge

(Table 3-4). Average size and percent cover were higher in the photo plots than in the

image plots (12065 m2 compared to 2033 m2 and 29% compared to 11% respectively;

Figure 3-4 to 3-6). There were significant correlations between mean size of tree islands

(r=0.42, p=0.03), percent cover of tree islands (r=0.54, p=0.003), and standard deviation

of size of tree islands (r-0.70, p=0.00) for the two media. There were significant linear

relations between the two methods for percent cover (F=5.19, p=0.03, r2=0.17; percent

cover in photo = 8.46 + 1.05 x percent cover in imagery) and standard deviation of tree

island size (F=23.9, p<0.0001, r2=0.50; standard deviation from photo = 1848.5 + (1.42 x

standard deviation from imagery), but not for number or size of tree islands. Number of

tree islands and median size of tree islands in the imagery plots and photo plots were not

correlated. The five class cluster analysis performed on the results from the photo plots

yielded groups with the following characteristics (Figure 3-6):

Group 1- no tree islands (1 plot);

Group 2- medium number of tree islands, small average size, low percent cover (7

plots);

Group 3- low number of tree islands, large average size, low percent cover (5

plots);

Group 4- high number of tree islands, medium average size, medium percent cover

(12 plots);

Group 5- large number of tree islands, large average size, high percent cover (2


plots).









Discussion


Increased interest in landscape analysis presents new challenges in sampling design.

In many cases it is not possible to measure variables in detail over the entire landscape or

to obtain data from the same source over the time frame of interest. The availability and

increasing use of satellite imagery for describing large scale vegetation patterns facilitates

a landscape perspective. Presented here is one method for selecting sampling locations

based on satellite image classification. Using a resampling technique, the accuracy and

precision of different combinations of size, shape, and percentage samples was evaluated.

Results from the sample compared favorably to the results for the entire study area. The

hypothesis that aerial photo plots and image photo plots would show the same tree island

patterns was tested and the relation between the variables quantified using the two

methods examined. There were significant differences between the two methods, and only

weak relations between some of the tree island variables. The rationale for selection of

sampling size and location, reasons the imagery and photography show little relation to

each other, and ways to improve agreement between patterns identified from different

methods are discussed below.

In unbiased sampling the mean compiled from successive resampling should

exactly equal the true mean of the population. The mean and standard deviation of the

mean can be used to determine which sampling regime provides the most accurate (closest

to the true value) and precise (repeatedly close to the original value) estimates. In this

study, mean values were similar for each variable at all percentage samples of Loxahatchee

indicating similar accuracy at all percentage samples. As expected sampling a greater









percentage of the refuge resulted in a more precise estimate of the mean. Because of

logistical constraints (the effort and cost of a larger sample) sampling percentage was

limited to 10% of the refuge. Had a greater proportion of the refuge been sampled the

results would be 20-40% more precise. This suggests that some caution should be used in

extrapolating results from a 10% sample to the entire refuge.

Estimates of the total number of tree islands in Loxahatchee were closer to the true

number when the smaller plot sizes were used, probably because the amount of area

sampled was closer to the actual area of the refuge (only plots completely within the

refuge boundaries were used). Estimates from the 200 x 100 plots were better than the

200 x 200 or 100 x 100. Likewise, estimates of percent cover using the 200 x 100 plot

size were closer to the mean percent cover and had a lower standard deviation than other

plot sizes.

As was found by O'Neill et al. 1995, sample plots of different size and shape

resulted in different values of the patch (tree island) statistics. Mean tree island size was

underestimated in the small plots because of the plots inability to include tree islands of the

largest sizes. On the other hand, large plots tended to over estimate the mean size. An

explanation for this is that larger plots are more likely to include all of a large tree island at

the exclusion of more smaller tree islands, resulting in an increase in the average tree

island size in each plot. An advantage of smaller plots is that more plots are needed to

sample the same area of the refuge than are needed with larger plots; therefore, the plots

can be spread throughout the refuge. The disadvantage is that smaller plots may not

represent the spatial distribution of the tree islands because the scale is too fine.









Based on the above analysis, the fact that the spatial patterning of the tree islands

seems to be more dominant in the north/south direction, and the added constraint that

plots should be large enough to contain the largest tree islands, a plot size of 200 x 100

was selected.

The hypothesis that patterns of tree islands measured from the satellite imagery

were not different than the patterns of tree islands identified in the photography was

rejected for all of the tree island variables. In addition, there only were weak relations

between the results from two methods. The number of tree islands in the imagery were

not correlated with the number of tree islands in the photo plots. Mean tree island size

and percent cover were weakly correlated, while standard deviation of size had the highest

correlation (r=0.70). These results indicate that the general patterns of variance may be

represented in the imagery, but the true values are not. Only percent cover and standard

deviation showed significant linear relation that might allow predictions to be made from

this imagery; however, neither predictive relationships were very strong.

Studies comparing land covers developed from different remote sensing methods

have shown similar inconsistencies between methods in the amount of a given cover type

delineated using each method (Jensen et al. 1978, Quick and Scarpace 1982). Spatial

patterns also have been shown to vary among classification methods (Benson and 1995,

Moody and Woodcock 1994, O'Neill et al. 1995). There are a number of possible

explanations for this lack of agreement between methods including differences inherent in

the classification methods (resolution, sensor type, manual versus computer classification),

purposes, and level of detail, and no direct aggregation from the vegetation classes in one

method to the classes delineated in another.









Satellite image classification and photo interpretation are two very different

techniques that both have strengths and weaknesses (Burrough 1986). The strengths of

image classification include objective rapid class identification over a large area based on

spectral reflectance values. Weaknesses included the blurring of boundaries between

classes due to mixed pixel effects and the difficulties associated with spectral

differentiation of specific vegetation types. Photo interpretation is much more time

intensive and subjective (due to interpreter bias), but can provide clearer boundaries of

vegetation types and often a smaller effective minimum mapping unit. In both cases a

decision must be made of how much detail (how many classes to use) is required to

address the issue for which the land cover map or classification is being developed.

The classified image used in this study was developed primarily to examine the

expansion of cattail (Typh latiflia) in Loxahatchee (Richardson et al. 1990). The focus

was on the separation of vegetation types not on the delineation of tree islands. The tree

island classes used in this study only were mature tree island classes and thus represent a

conservative estimate of tree island coverage. Originally, a third class representing

"brush/tree island" was included as a tree island class; however, the resulting image under

represented the shape of tree islands and over represented the tree island coverage (W.

Kitchens, Personal communication). Comparison of the tree islands identified in the photo

plots with the tree islands identified on the imagery show that much of what was identified

as a tree island in the photo plots was classified as brush in the imagery; however, there

also were many areas labeled as brush on the image classification that were not identified

as tree islands on the photos so adding the brush category would not have necessarily

improved the agreement. Additionally, smaller tree islands in a wet prairie matrix were









classified as wet prairie. This resulted from the mixed pixel effect in satellite imagery,

where the reflectance value is the average value for that cell. If the reflectance from a

vegetation type does not dominate the cell that feature do not appear on the classification.

This raises two issues for consideration when using vegetation classifications for

delineating landscape features. What is the feature of interest and how do you describe it?

And, which classes of an existing classification can be combined to represent that feature?

The first sounds fairly straightforward; however, what theoretically sounds like a good

definition of a tree island is often hard to translate to photography and the field. Tree

islands have been described as slightly elevated sites with definitive species components

and distinct physiognomic structure {Loveless 1959 #4620). Loveless (Loveless 1959

#4620) also noted that there may be an abrupt demarcation between tree islands and

surrounding communities, or that there may be a transition zone with species such as

sawgrass, willow (Salix caroliniana), buttonbush (Cephalanthus occidentalis), elderberry

(Sambucus spp.), and wax myrtle (Mvrica cerifera). There are several practical challenges

to this definition. Is the transition zone part of the tree island? At what point does a brush

area become a tree island? Do tree islands have a distinctive shape or can they be any area

of woody vegetation within a wetland?

For purposes of photo interpretation I defined a tree island as a consolidated group

of trees covering at least 100 m2. Tree islands could contain brush (and often did) if the

brush was part of the consolidated mass. This definition was subject to interpreter

judgment as to what was a group of trees. Often a decision was made based on the shape

and structure of the vegetation as well as the presence of trees. No such pattern matching

was done in the satellite image classification, rather the classification delineated areas of









trees and brush. Often the result was that only the densest group of red bays and dahoon

holly were classified as tree islands. Thus very different things were represented by the

two methods. The imagery delineated the vegetation type while the photo interpretation

delineated tree islands based on the presence of trees and on the association of the trees

with other vegetation classes.

This lack of agreement between variables measured by the two methods makes it

difficult to expand the patterns observed from a sample to the entire study area. Some of

the differences are inherent in the methods (differences in effective minimum mapping unit,

boundary definition) and studies have been done that examine these issues. Often the

awareness of these differences can be incorporated into the interpretation of the results.

Other problems resulting from differences between vegetation classes and landscape

features are more problematic. If remote sensing methods are to be used to map

landscape features, the features must be described in terms that can be identified by these

methods. It is not always possible to select the individual vegetation classes that represent

a feature as seen in this analysis. It may be necessary to add spatial factors such as the

juxtaposition of vegetation types into the selection of classes. Because of these potential

problems, it is recommended that when land covers developed for one purpose are used to

address another purpose, careful consideration be given to the applicability of original

vegetation class definitions to the purpose of the second study. A pilot comparison

between the patterns of interest from a small area of all the methods being used will assist

in class definition. Consideration of these problems will increase our ability to correctly

interpret patterns measured using multiple methods.











Table 3-1. Number of tree islands per plot based on classification of satellite imagery
(Richardson et al. 1990). Pixels are 9 x 9 m. Statistics are based on 100 resamples
representing the listed percentage of complete plots of that size. True mean, variance, and
standard deviation are for all of the plots of that size. Number of tree islands on imagery
=2144.


Plot size
(pixels)


Sample size


Percent Mean Variance SD


200x200 100 samples of 11
100 samples of 22
100 samples of 33
100 samples of 44
100 samples of 55


True values


200x100 100 samples of 25
100 samples of 50
100 samples of 75
100 samples of 100
100 samples of 126


14.19 21.36 4.62
14.36 8.53 2.92
14.25 4.39 2.09
14.84 2.26 1.5
14.19 1.74 1.32


100 14.46 226.12 15.03


7.59 2.86 1.69
7.36 1.14 1.07
7.38 0.62 0.79
7.39 0.52 0.72
7.46 0.25 0.5


100 7.42 86.57 9.3


100x100 100 samples of 45
100 samples of 90
100 samples of 135
100 samples of 180
100 samples of 226


True values


50x50 100 samples of 188
100 samples of 376
100 samples of 565
100 samples of 753
100 samples of 941


4.04 0.59 0.77
4.04 0.3 0.55
3.96 0.15 0.38
3.97 0.09 0.31
3.97 0.06 0.26


100 3.96 30.2 5.5


1.14 0.019 0.14
1.11 0.008 0.09
1.14 0.003 0.06
1.13 0.003 0.05
1.12 0.002 0.04


100 1.13 3.64 1.91


True values


True values










Table 3-2. Mean percent tree island cover per plot based on classification of satellite
imagery (Richardson 1990). Pixels are 9 x 9 m. Statistics are based on 100 resamples
representing the listed percentage of complete plots of that size. True mean, variance, and
standard deviation are for all of the plots of that size. Percent cover of tree islands on
imagery = 1.96.

Plot size Sample size Percent Mean Variance SD
(pixels)

200x200 100 samples of 11 10 1.98 0.3 0,55
100 samples of 22 20 1.9 0.2 0.44
100 samples of 33 30 1.89 0.09 0.29
100 samples of 44 40 1.9 0.06 0.24
100 samples of 55 50 1.91 0.04 0.19

True values 111 100 1.91 4.01 2.00

200x100 100 samples of 25 10 1.88 0.21 0.46
100 samples of 50 20 1.89 0.15 0.38
100 samples of 75 30 1.85 0.08 0.27
100 samples of 100 40 1.85 0.05 0.22
100 samples of 126 50 1.83 0.03 0.18

True values 252 100 1.85 7.99 2.83

100x100 100 samples of 45 10 1.86 0.42 0.65
100 samples of 90 20 1.82 0.15 0.39
100 samples of 135 30 1.88 0.09 0.3
100 samples of 180 40 1.82 0.06 0.24
100 samples of 226 50 1.84 0.04 0.20

True values 451 100 1.85 16.23 4.03

50x50 100 samples of 188 10 2.07 0.15 0.38
100 samples of 376 20 2.03 0.1 0.31
100 samples of 565 30 1.96 0.05 0.23
100 samples of 753 40 1.97 0.03 0.17
100 samples of 941 50 1.97 0.02 0.16


1882 100 2 41.03 6.40


True values











Table 3-3. Mean size (pixels) of tree islands per plot based on classification of satellite
imagery (Richardson 1990). Pixels are 9 x 9m. Statistics are based on 100 resamples
representing the listed percentage of complete plots of that size. True mean, variance, and
standard deviation are for all of the plots of that size. Mean size of tree islands on imagery
= 52.6 pixels, variance = 572.3, SD = 23.9.


Plot size
(pixels)


Sample size


200x200 100 samples of 11
100 samples of 22
100 samples of 33
100 samples of 44
100 samples of 55

True values 111

200x100 100 samples of 25
100 samples of 50
100 samples of 75
100 samples of 100
100 samples of 126


True values


252


100x100 100 samples of 45
100 samples of 90
100 samples of 135
100 samples of 180
100 samples of 226


True values


50x50 100 samples of 188
100 samples of 376
100 samples of 565
100 samples of 753
100 samples of 941


True values


Percent Mean Variance SD


10 71.68 1052.99 32.45
20 70.57 572.07 23.92
30 74.44 407.28 20.18
40 72.93 188.67 13.74
50 73.53 132.36 11.5

100 73.41 14485.1 120.35

10 55.66 370.14 19.24
20 61.58 243.3 15.6
30 57.76 130.08 11.4
40 59.61 69.03 8.31
50 59.9 42.91 6.55

100 59.85 12626.47 112.37

10 47.04 497.88 22.31
20 52.88 278.377 16.68
30 51.81 158.88 12.6
40 53.67 89.91 9.48
50 52.61 70.42 8.39

100 52.57 28449.05 16.867

10 28.56 82.13 9.06
20 28.01 30.94 5.56
30 27.9 15.48 3.93
40 27.76 11.08 3.33
50 27.97 8.78 2.96

100 27.73 15750.02 125.5











Table 3-4. Summary statistics for plots from classified image and photo plots.

Imagery Photo plots
Plot Number Percent Average SD of Number Percent Average SD of
of tree cover size of tree of tree cover size of tree
islands oftree tree islands islands oftree tree islands
islands islands size islands islands size
(m2) (m2)


0 0
10 0.16
18 1.01
230 10.89
47 2.28
39 4.9
90 10.91
159 10.55
202 14.71
256 10.11
19 2.37
117 5.61
332 22.12
166 10.39
39 4
124 6.24
135 6.08
400 15.9
57 5.93
137 7.18
174 16.54
321 29.36
394 19.04
267 13.96
269 16.14
345 16.32
304 24.59


0
0 0
9230 3830.5
1701 0
567 0
162 0
20682 3964.66
2766 617.00
1458 282.03
405 0
1118 109.39
999 86.48
1932 357.57
1597 402.21
2128 413.02
608 105.34
405 37.29
4278 1322.37
1981 313.29
474 26.20
3377 1183.97
14823 4878.00
1215 198.00
1134 234,00
34587 5373.00
6156 1449.00
16362 4221.00


36
55
62
110
124
144
147
159
173
205
210
215
227
238
257
260
306
310
325
328
377
c1911e
c1911w
jk9l11e
jk911w
191 1e
191 1w

Min
Max
Average
SD

Entire
area


7.42 1.85 4261


0 0.00
0 0.00
19 10.83
1 0.11
1 0.04
3 0.03
3 3.83
7 1.20
6 0.54
1 0.03
10 0.69
3 0.19
7 0.84
7 0.69
11 1.45
4 0.15
13 0.33
11 2.91
24 2.94
7 0.21
16 3.34
10 9.12
10 0.76
9 0.65
2 4.27
5 1.89
9 9.11


0
259 227.64
910 1044.38
767 1790.98
786 853.32
2034 8532.71
1964 12065.58
1075 6010.95
1180 2237.48
640 660.53
2018 1973.87
777 1378.92
1079 2260.46
1014 2554.68
1662 3266.14
815 1080.21
729 526.19
644 2124.33
1686 2351.85
848 744.51
1540 3899.51
1482 10558.95
783 1166.55
847 1632.10
972 6674.24
766 3338.15
1310 10862.89


0 0 0 0 0 0 0 227.64
24 10.83 34587 5373.00 400 29.36 2034 12065.58
7.37 2.08 4820 1130.94 172.26 10.64 1059 3454.5
5.90 3.03 8016 1720.62 125.30 7.67 509 3460.18

































Figure 3-1. Framework developed to relate patterns of tree islands quantified using a classified satellite image of all of Loxahatchee
National Wildlife Refuge and patterns quantified using aerial photography covering 10%/ of the refuge.
























































Scale
--mm Kilomreers
2 0


Figure 3-2. Location of groups resulting from 5 class cluster analysis from satellite image
classification using the tree island variables number per cell, mean size, SD of size, and
percent cover


















20




15



1i *





5 4

**



0 50 1o0 150 200 250 300 350 400
Number of tree Islands In photography


Figure 3-3. Relation between number of tree islands in satellite image and photo plots of Loxahatchee National Wildlife Refuge. Plots
are 1800 x 900 m.




















25000











I?
e 15000 *)


10000


500


o .: w. *
0 500 1000 1500 2000 2500
Mean tree Island size In photography (sq n)


Figure 3-4. Relation between mean size of tree islands in satellite image and photo plots of Loxahatchee National Wildlife Refuge.
Plots are 1800 x 900 m.













6OOD



5000





















SD area photography
Plots are 1800 x 900 m.
0 200 400 O 8= 10000 12000 1 4000
SD area photography


Figure 3-5. Relation between SD of size of tree islands in satellite image and photo plots of Loxahatchee National Wildlife Refuge.
Plots are 1800 x 900 m.


















10











4



*
0 ** *
0. e -
0 5 10 15 20 25 30
%tree Island cover photography


Figure 3-6. Relation between percent cover of tree islands in satellite image and photo plots of Loxahatchee National Wildlife Refuge.
Plots are 1800 x 900 m.






















































Scale
CmL Kilometers
2 0


Figure 3-7. Location of groups resulting from 5 class cluster analysis of photo plots using
the tree island variables number per cell, mean size, SD of size, and percent cover.















CHAPTER 4
PATTERNS OF CHANGE IN TREE ISLANDS IN A.R.M. LOXAHATCHEE
NATIONAL WILDLIFE REFUGE FROM 1950 TO 1991

Introduction


The Everglades, once a vast dynamic flow-through wetland extending from the

Kissimmee chain of lakes south to Florida Bay, has been severely altered in the last

century. The development of south Florida has led to the conversion of portions of the

historic system to urban and agriculture and the conversion of the once sheet-flow system

into a series of highly regulated impoundments and canals. The vast mosaic of marsh,

slough, tree islands, and pinelands has been reduced to a fraction of the historic 3.6 million

ha (Davis and Ogden 1994). In addition, what remains is subject to hydrologic regimes

that, in places, bear little resemblance to pre-drainage patterns (Fennema et al. 1994).

The historic hydrological patterns were vital in creating and maintaining the

heterogeneity of the landscape, which in turn, contributed to the persistence and resilience

of the system (Holling et al.1994). The Everglades, like other ecosystems, is the product

of the interaction of biota and forces such as climate, hydrology, nutrient inputs, and

disturbance (DeAngelis and White 1994). These processes, operating over a range of

spatial and temporal scales, interact to form the current landscape patterns. Changes in

macroscale processes such as geology and climate occur over long time scales and are

generally not noticeable within a human lifetime. At the other extreme, changes in

microscale processes such as nutrient transfer occur very rapidly and the consequences to

71









wetland vegetation can be observed in a season. In between are mesoscale processes that

act on a scale of kilometers to tens of thousands of kilometers and over time scales of

decades to centuries. Hydrology is the dominant natural mesoscale process acting in the

Everglades. It is also the process that has been most seriously altered and has caused the

largest changes to the remaining vegetation in the system. Changes in hydrologic patterns

have contributed to the reduction in plant community heterogeneity, the change and

decrease in wildlife populations, and changes in the historic functioning of the Everglades

(see Davis and Ogden 1994). Changes in vegetation structure and composition have been

noted in many regions of the Everglades (Loveless 1959, Alexander and Crook 1984,

Higer and Kolipinski 1987, Parks 1987, Worth 1988) and Davis and Ogden (1994) have

documented massive loss or conversion of entire landscape types. However, the total

extent of these changes is unknown, and it is unlikely that the magnitude of the changes

will be completely understood.

Restoration of the Everglades ecosystem will require an historical reference on

what vegetation communities existed and in what spatial arrangement, an understanding of

how historical processes contributed to the landscape patterns, and wherever possible, an

understanding of the relation between anthropogenic changes in structuring processes and

current landscape patterns.

This study examines the historic patterns (1950) of tree islands in a remnant of the

northern Everglades and relates these patterns to pre-drainage hydrologic processes. The

current patterns (1991) of tree islands, the relationship of those patterns to post-drainage

hydrology, and the relation between the magnitude of change in tree island patterns and









hydrologic patterns then are examined in an attempt to understand better the influence of

hydrologic processes on landscape patterns.

Methods


Study Area


The Arthur R. Marshall Loxahatchee National Wildlife Refuge (Loxahatchee), is a

57,324 ha area of northern Everglades wetlands. Located in Palm Beach County, Florida,

south and east of Lake Okeechobee, (Figure 2-1) it was once a connected part of the

historic Everglades system. Loxahatchee is located over the Loxahatchee Channel, a

shallow peat filled depression extending from Lake Okeechobee south and east that may

have acted as an overflow valve for Lake Okeechobee shunting water east and south

(Gleason et al. 1984). Changes to the flow patterns in and around Loxahatchee started as

early as the 1800s with the construction of the Caloosahatchee canal, which moved water

from Lake Okeechobee to the west. Additional changes occurred with the completion of

the Hillsboro Canal in the 1920s, with the completion of the St. Lucie canal in 1931, and

the with the complete enclosure of the area by the L-7 canal on the west and the L-40

canal on the east during the 1950s (Light and Dineen 1994). The construction of these

canals effectively isolated Loxahatchee from it's original watershed. This changed the

system from a dynamic flow-through, sheet-flow driven fluvial system to an impounded

marsh with the majority of overland flow inputs being shunted around the marsh via the

exterior canals.

Loxahatchee is a peat-based wetland system consisting of a mosaic of aquatic

sloughs and expanses of wet prairie strands of sawgrass, patches of brush, and tree islands.









Tree islands are one of the most conspicuous and dominant features of the landscape and

provide habitat for a large number of wildlife species including deer, alligators, and wading

birds.

Photography

Image sources for this analysis were twelve 1:60,000 panchromatic diapositives

flown in November or December 1950 and twenty 1:40,000 color-infrared diapositives

flown in December 1990 or January 1991. Each set of photos was photogrammetrically

scanned at the appropriate scanning resolution to result in 2 m ground resolution. The

images were referenced to a 10 m geocoded SPOT satellite image and mosaiced using

Orthoengine from PCI Remote Sensing Corporation. The result was two complete images

of the Refuge with average residual errors of approximately 4 m for 1950 and 2 m for

1991. Because plots not individual tree islands were the unit being compared this level of

error was deemed acceptable.

Twenty-seven 1800 x 900 m photo plots, approximately 10% of the refuge, were

selected for this analysis (Figure 4-1). Selection of photo plot location was designed to

sample the range of tree island density, tree island size and variation in tree island size

found on the refuge and was based on a stratified random sampling regime (see Chapter 3

for details).

Boundaries of tree islands larger than 100 m2were screen digitized in ARC/INFO

using scanned images of the original photography. The original photography was used for

reference for both years. Additional 1:24,000 black and white photography was used as

reference for the 1950 plots. Tree island identification on the 1991 photography was

verified by field sampling.









Number, area, perimeter, and centroid locations (X and Y coordinates) for all tree

islands were obtained directly from ARC/INFO. Length of long axis and secondary axis

were calculated using a program developed in S-plus (Statistical Sciences, Inc. 1995).

Tree islands were classified as circular, elliptical (shaped) or irregular (no shape) using a

combination of a circularity index (Miller 1953) and ellipse index. These indices are the

ratio of the area of the tree island to the area of a circle, or ellipse, respectively given the

same long axis and secondary axis. An index of 1 indicates complete agreement (an exact

circle or ellipse). Values above or below 1 indicate deviation from the ideal shape. Index

thresholds were determined by randomly selecting tree islands and visually categorizing

them as circular, elliptical, or irregular. The calculated index values were compared and

cutoff criteria developed based on the number of tree islands in that range identified as a

particular type. Tree islands with a circularity index of 0.85 or larger were considered

circular. Tree islands with a circularity index < 0.85 were elliptical if their ellipse index

was <6. All remaining tree islands were classified as irregular. Orientation was determined

for elliptical tree islands by calculating the direction of the long axis using a program

developed in S-plus (Statistical Sciences, Inc. 1995). Orientation ranges from 0 to 180

with 0 as north and 180 as south. No attempt was made to distinguish between the

leading and trailing edge of the tree island.

Hydrology and Elevation

Data on hydroperiod, water depth, flow direction and magnitude were obtained

from the South Florida Water Management District. Data are from two models 1) the

South Florida Water Management Model Version 2.10(WMM) developed to simulate the

hydrology of the water management system in south Florida (MacVicar et al. 1984)and 2)









the Natural Systems Model version 4.4 (NSM), adapted from the WMM to simulate the

hydrologic response of pre-drainage south Florida using climatic data from 1965-1990

(Fennema et al. 1994). Both models have a grid cell size of 3.2 x 3.2 km (2 miles x 2

miles). Grid cells are referred to as hydrology zones though out this text. Depth, flow

direction and magnitude are the yearly average values for each grid cell calculated over the

entire period of record (1965-1990). Direction of flow was standardized to 0-180 degrees

to match the range used for the tree island orientation and then transformed using

tan(theta /2). Two variables were used to represent hydroperiod: the first was the 90%

hydroperiod or the hydroperiod that was exceeded in 90% of the 26 years. The second

hydroperiod variable was range in hydroperiod. This was used to reflect the variability

among the hydrology zones. The assumption is that tree islands are the resultant of

processes operating over time and that tree islands in 1950 and the NSM output reflect

pre-drainage hydrology. Likewise, the assumption is made that 1991 tree islands and

WMM output reflect the last 40 years of post drainage patterns.

The topographic surface generated by Richardson et al. 1990 for Loxahatchee was

used for this analysis. The grid cell resolution of this surface was 183 x 183 m. The 3.2 x

3.2 km cell boundaries for the hydrology models were overlain on the elevation surface

and average elevation and elevation gradient calculated for each cell of the hydrology

model. Because the relation between current and historic elevation is unknown the

topographic information was used only with the 1991 photo plots.

Plot Comparisons and Relationship to Hydrology

Because initial plot selection was done independent of the hydrological analysis, all

of the plots did not fall completely in one hydrology zone. In cases where the plots were









in more than one zone, analysis was done for each zone separately. At least one third of

the plot had to be in the hydrology zone for it to be considered. Tree island numbers were

converted to densities to allow comparison among plots of different sizes. Median area

and median long axis were used as size variables for the plot comparisons because

individual area and long axis of tree islands within the plots were not distributed normally,

and could not be normalized using simple transformations. Partial plots did not have

significantly different median tree island size or densities than full plots. The following

analysis was done on the resulting 28 plots.

Density, percent cover of tree islands, median area, median long axis, ratio of

elliptical to circular tree islands, ratio of non-shaped to shaped tree islands, and mean

orientation of tree islands were compared between the two years. Data that were non

normally distributed were transformed as indicated in Table 4-1 prior to analysis.

Hydroperiod, mean ponding depth, range in hydroperiod, flow magnitude, and flow

orientation data from the NSM and WMM also were compared.

The relations between tree island variables and hydrologic variables were examined

using canonical correlation analysis for the 1950 and NSM data, the 1991 and WMM data,

the 1950 tree island data, the 1991 tree island data, and the NSM and WMM data.

Canonical correlation analysis is used to investigate the relations between two sets of

variables by deriving a linear combination of the X variables (U=aX1 + aX2 + ... +aXi) and

a linear combination of the Y variables (V=aYi + aY2 + ... +aYi) such that the correlation

between U and V is as large as possible (Manly 1986). A non-significant result indicates

that the largest canonical correlation can be accounted for by sampling variation alone.

Canonical correlation combines the multiple variables into a single index variable (U or V).









Examination of the coefficients of the individual variables that make up U or V indicates

the contribution of that variable to the overall index. Variables with higher coefficients are

more important.

The size structure of the tree islands (both long axis and area) was examined using

a clump analysis (Holling 1992, Marples unpublished) designed to identify significant

breaks in size distributions, A significant clump indicates the presence of a process

functioning to structure the landscape. Changes in the clumping pattern over time indicate

a potential change in the processes structuring the landscape. These methods are similar

to cluster analysis in that they find natural groupings in the data. The first step is to test if

randomly sampled data can explain the clumping pattern in the data. The second step is to

measure the discontinuities between the rank-size-order data. The result is a D-value

(discontinuity value) that indicates whether a clump is significant or not at a specified level

of significance (0.05 in this case).

Results


1950 Plots and NSM Hydrology

Density of tree islands per plot ranged from 0.08 tree islands/ha to 2.16 tree

islands/ha. Percent cover ranged from 0.2 to 26.7%. Median tree island size ranged from

187 m2 to 755 m2and median long axis from 20 to 46 m. The ratio of elliptical to circular

tree islands varied from 0.3 to 25 (Table 4-2) and the ratio of non-shaped to shaped from

0.15 to 8.12. Mean orientation of elliptical tree islands ranged from 1 to 180 degrees with

most between 1 to 30 or 150 to 180 degrees (Figure 4-2), Density of tree islands was

correlated with location (X-Y coordinates of plot centroid). Higher densities of tree









islands occurred in the northern and eastern plots. There was no correlation between the

area or long axis of tree islands and location. The ratio of elliptical to circular tree islands

increased from north to south indicating relatively fewer elliptical tree islands in the more

northern plots.

The distribution of tree island sizes from all plots combined showed many small

tree islands and few larger tree islands (Figure 4-3). Fifty percent of the tree island area

was made up of tree islands < 0.16 ha. Ninety-four percent of the 3769 observed tree

islands were < 0.16 ha.

Clump analysis for the 1950 data, using a D-value of 0.9 indicated that there were

three potential clumps in tree island area: < 0.81 ha, 0.96-1.61 ha, > 2.15 ha and three

potential clumps for long axis: < 247 m, 271-303 m, > 357 m (Figure 4-4).

Mean hydroperiod for the NSM in the cells in which photo plots occurred ranged

from 240 to 329 days and mean ponding depth from 0.2 m to 0.4 m. Range in

hydroperiod was as little as 47 days and as great as 272 days (Table 4-3). Density of all

tree islands and circular and elliptical tree islands were positively correlated with

hydroperiod and negatively correlated with hydroperiod range (Table 4-4). Overall

density of tree islands and density of elliptical tree islands were correlated with mean

depth. Median area of tree islands was positively correlated with mean ponding depth and

negatively correlated with range in hydroperiod. Median long axis was not correlated with

any of the hydrology variables. Mean tree island orientation was not significantly different

from orientation of NSM flow.

Canonical correlation analysis was performed with the tree island variables density,

area, and ratio of elliptical to circular tree islands and the hydrology variables NSM









magnitude, hydroperiod, mean ponding depth, and hydroperiod range. It was performed

again replacing area with long axis. In both instances only the first canonical correlation

was significant (p=0.0002, r=0.85, p=0.0003, r=0.85 respectively). In both analyses the

hydrology variables NSM hydroperiod range and NSM magnitude (coefficients of-0.82

and -0.62 and 0.78 and 0.61 for hydroperiod range and magnitude in the two analyses

respectively) were the most important, as was the tree island variable ratio of elliptical to

circular tree islands (coefficients of-0.74 and 0.77). Eighty-eight and eighty-seven

percent of the variation between the linear combinations were explained.

1991 Plots and WMM Hydrologv

Density of tree islands per plot ranged from 0.01 tree islands/ha to 2.31 tree

islands/ha. Percent cover ranged from 0.03 to 30.2%. Median tree island size ranged

from 132 m2 to 1429 m2 and median long axis from 18 m to 59 m. Ratio of elliptical to

circular tree islands from 0 to 6.7 (Table 4-5) and ratio of non-shaped to shaped from 0.16

to 17.00. Mean orientation of elliptical tree islands ranged from 1 to 180 degrees with

most between 1 to 30 or 150 to 180 degrees (Figure 4-5). Density of tree islands was

correlated with location, with a trend of higher densities in the east than in the west.

Median tree island area and long axis were correlated with location, with larger tree

islands in the north and tree islands with shorter long axis in the south and east. The ratio

of elliptical to circular tree islands was not significantly correlated with location.

As with the 1950 photo plots, the distribution of tree island sizes showed many

small tree islands and few larger tree islands (Figure 4-6). Fifty percent of the tree island

area was made up of tree islands < 0.13 ha. Eighty-seven percent of the 4392 identified

tree islands fell into this category.









Clump analysis for the 1991 area data indicated that there were no significant

clumps at D >= 0.9. But, if a D-value of 0.8 was used, three clumps emerged: < 1.15 ha,

1.37-1.44 ha, > 1.84 ha. There were four potential clumps in long axis at D-value > 0.9:

< 150 m, 150-218 m, 251-329 m, >418 m (Figure 4-7).

Mean WMM hydroperiod in the cells within which photo plots occurred ranged

from 183 to 344 days with a range between 47 and 272 days. Mean ponding depth ranged

between 0.07 m and 0.73 m. Density of tree islands, and density of circular and elliptical

tree islands were correlated with elevation gradient (Table 4-6). Density of circular tree

islands also was correlated with mean cell elevation. Median tree island area and median

long axis were negatively correlated with hydroperiod, mean ponding depth, and elevation

gradient and positively correlated with hydroperiod range and mean elevation. Mean tree

island orientation was not significantly different from WMM flow orientation or NSM

flow orientation.

Canonical correlation analysis was performed with the tree island variables density,

area, and ratio of elliptical to circular tree islands and the hydrology variables WMM

magnitude, hydroperiod, mean ponding depth, and hydroperiod range. The analysis was

performed again replacing area with long axis and a third and fourth time with the addition

of mean elevation and elevation gradient. In all cases, there were two significant canonical

correlations.

In the first two analyses (without elevation), the first significant correlation

(p=0.001, r=0.85 and p=0.0003, r=0.79) was between the hydrology variables WMM

hydroperiod range and mean ponding depth (coefficients of 0.49 and 0.43 and 0.58 and -

0.41 respectively) and the tree island variable representing size (coefficients of-0.98 and -









0.96 for area and long axis respectively). Seventy-seven and sixty-seven percent of the

variability was explained by the first significant correlation. The second significant

correlations explained an additional 21 and 20% of the variability respectively. The

important variables in the second correlations were mean ponding depth and hydroperiod

(coefficients of 4.96 and -4.56 and 4.93 and -4.60 for analysis using area and long axis

respectively). In both cases, the important tree island variable was tree island density

(coefficients of-0.98 and -0.94 respectively).

When the elevation variables were added to the analysis, mean elevation and mean

ponding depth (coefficients of 1.54 and 0.73) were the most important hydrology variables

in the first significant correlations (p=0.0001, r=0.91 and p=0.0001, r=0.86). Median area

was the most important tree island variable when size was represented as area (coefficient

of 0.82). Both tree island density and long axis (coefficients of-0.77 and -0.70) were

important when length was used. Seventy-six and sixty-nine percent of the variability was

explained by the first sets of correlations. An additional 20 and 27% of the variability was

explained by the second significant correlations. The most important hydrology variables

in the second set of correlations were hydroperiod (coefficients of-3.10 and -3.29 for

analysis using area and long axis respectively). Tree island density (coefficient of-0.84)

was the most important variable when area was used to represent size. Both long axis and

density (coefficients of 0.72 and -0.63 respectively) were important when long axis was

used to represent size.

Comparison Between Individual Plots

Twenty of the twenty eight plots showed significant differences between the

median size of tree islands between years (Table 4-7). Seventeen increased in percent









cover and eleven decreased (Figure 4-8) Density of tree islands in ten plots decreased, in

three plots stayed the same and increased in fifteen plots. Eleven plots had significant

changes in mean tree island orientation, fourteen did not. Sample size in the remaining

plots was too small for comparison. The ratio of number of elliptical to circular tree

islands in 1991 was larger in fourteen plots, smaller in twelve plots and the same in two

plots. The ratio of non-shaped to shaped tree islands was larger in 20 plots and smaller in

eight plots.

All Plots Together

When analyzed together in a pair-wise analysis, median tree island area, median

long axis, and ratio of non-shaped to shaped tree islands were significantly different

between years (Wilcoxon signed-rank W=240, p=0.006; W=225, p=0.01; W=248,

p=0.0001 respectively). Density of tree islands were not significantly different, nor was the

number of elliptical and circular tree islands per plots, ratio of elliptical to circular tree

islands, or the orientation of the tree islands.

Canonical correlation analysis was performed using the tree island variables

density, area, and ratio of elliptical to circular tree islands for 1950 and 1991 to examine

changes in the relation of the variables between the time periods. Only the first canonical

correlation was significant (p=0.0001 r=0.94). Ninety percent of the variation was

explained by this analysis. Density and area from 1950 and 1991 had similar coefficients

(0.82 and 0.90 for density and 0.22 and 0.29 for area respectively). The coefficient for

ratio of elliptical to circular tree islands was slightly more important in the 1950 data than

in the 1991 data (-0.37 and 0.14 respectively).









A similar analysis was performed using the hydrology data for the two sample

periods. In this analysis the first two correlations were significant (p=0.0001, r=0.88 and

p=0.020, r=0.67 for the first and second correlations respectively). Unlike the pattern for

the tree island variables, the coefficients between the two years were not similar. In the

first correlation, hydroperiod range (coefficient of 1.10) was the most important NSM

hydrology variable followed by NSM depth (coefficient of 0.71), NSM magnitude

(coefficient of 0.61), and hydroperiod (coefficient of 0.18). WMM depth (coefficient of

1.06) was the most important WMM hydrology variable with hydroperiod range

(coefficient of 0.06), hydroperiod (coefficient of-0.03), and magnitude (coefficient of

0.01) contributing little. The order of importance for the second correlation was similar to

the first for the NSM data (coefficients of 1.28, 1.05, 0.91, and -0.75 for hydroperiod

range, depth, hydroperiod, and magnitude respectively). The importance of the WMM

variables was not similar to that seen in the first correlation and was not similar to that of

the NSM data (coefficients of 2.56, -1.96, 0.87, -0.29 for hydroperiod, depth, hydroperiod

range, and magnitude).

All hydrology variables except for flow orientation were significantly different

between the two time periods (hydroperiod- t=-2.8, p=0.01; hydroperiod range- W=176.0,

p=0.02; mean ponding depth- W=-169.0, p=0.02; flow magnitude- t=14.0, p<0.0001).

Mean water depths were generally shallower and hydroperiods shorter for the WMM than

for the NSM. Depth and hydroperiod were more variable under WMM, while

hydroperiod range and flow magnitude were less variable (Table 4-3).

A decrease in flow magnitude was correlated with a decrease in tree island area

and long axis. Decreases in hydroperiod and depth were correlated with increases in area.









Decreases in hydroperiod range were correlated with decreases in the number of tree

islands (Table 4-8).

Change in density was positively correlated with location, with a greater change in

density in the east than the west. Change in area and change in long axis were not

correlated with location. The change in the ratio of elliptical to circular tree islands

increased from north to south.

Comparison of the location of the clumps in tree island size show differences

(Figure 4-9). For area, the smallest and largest size clumps are much wider than in 1950

and the middle clump is much narrower. For the long axis data the opposite pattern is

seen with the width of the smallest and largest size clumps larger in 1950 than in 1991 and

the middle clump narrower in 1950 than in 1991.

Discussion


Historically, in Loxahatchee, there were more tree islands in areas of longer

hydroperiod and greater depth. These areas also were less variable in hydroperiod range.

The multivariate analysis demonstrated that hydroperiod range and flow magnitudes are

important in explaining the ratio of elliptical to circular tree islands. Areas that were less

variable with lower flow had lower ratios of elliptical to circular tree islands (e.g. more

circular than elliptical tree islands). In addition, areas with less variation in hydroperiod

had more tree islands of larger size that covered a larger area. These results support the

hypotheses that flow is important in shaping tree islands and that battery tree islands form

under conditions of greater hydroperiod and depth.









The relation between the tree island variables and hydrology variables in 1991 is

very different from 1950. There were no significant correlations between the individual

tree island variables and hydrology as there were with the 1950 data. The multivariate

analysis shows that of the variables used here, hydroperiod and depth were the most

important in explaining tree island size. Areas of longer hydroperiod and depth had

smaller tree islands.

Loxahatchee can be grouped into three zones according to the patterns of change

of tree islands observed from 1950 to 1991. 1) the edge of the refuge adjacent to the

canals, 2) the eastern interior of the refuge, and 3) the western interior of the refuge. The

general trend is for tree islands along the edge of the refuge to have decreased in size,

number, and percent cover, while those on the interior increased in size, number, and

percent cover.

The patterns of change are what is expected given the changes in hydrology.

Areas of extreme hydroperiod and increased ponding depth should have a decrease in tree

island area due to drowning of the vegetation on the tree islands. This is seen primarily

along the southern boundary of the refuge. The plots along the Hillsboro Canal have

obvious remnants of drowned tree islands. These plots also are the plots that did not

show a significant difference in median tree island area between 1950 and 1991, possibly

because the greatest changes in tree islands in these areas occurred after the construction

of the canal in 1915. Areas of shorter hydroperiod and shallower ponding depth should

show an increase in tree island area as woody vegetation colonizes the drier sites. In this

analysis, decreases in hydroperiod and depth were correlated with increases in tree island









area. The ultimate consequence of these changes are to change a heterogeneous wetland

into a more homogeneous upland.

Hagenbuck et al. (1974) expressed concern that an increase in the aquatic setting

in Loxahatchee (e.g. extended hydroperiod and greater depth) might lead to an increase in

the formation of battery islands. These islands form in areas where hydroperiods are

longer and detritus settles to the bottom and forms loose peat mats. These then can

become dislodged and become floating peat mats that are colonized by woody vegetation.

The cumulative effect of this is to diminish the total water storage capacity of the area and

reduce the amount of available wetland habitat (Hagenbuck et al. 1974). This may be

happening in portions of Loxahatchee; however, because of the coarseness of the

hydrology data, it is not possible to determine if the increase in the number of small

circular tree islands thought to be battery tree islands within a plot, is due to increases in

hydroperiod or other factors. In some of the plots, especially those along the edge of the

refuge, where there has been an increase in the number of circular tree islands there has

been a decrease in the median area of the tree islands. It may be that tree islands that were

not circular in 1950 are circular in 1991 due to loss of area.

In other areas of the refuge, particularly in the northern interior, an increase in the

percent cover and size of tree islands indicates that tree islands are expanding in size.

Much of this area has had shorter hydroperiods and lower mean ponding depths in the

recent past, both of which would seem to promote tree island growth. This pattern of

increase in percent cover of tree islands in the northern areas was noted by Richardson et

al. (1990), Silveira 1996, and Hagenbuck et al. in 1974. Hagenbuck et al.(1974)

hypothesized that this expansion would continue to occur in areas of shorter hydroperiod.









These data support that. It may not be the changes in the mean hydrology variables

themselves that result in the changes, but the loss of the dynamic pulsing nature of the

historic processes.

An indicator that historic structuring processes have been lost in this system is seen

when examining the results of the clump analysis. In 1950 there are potential clumps in

tree island size that may correspond to processes operating at particular scales. Using the

same criteria (D >= 0.9) these clumps are absent in the 1991 data. Additionally, the

relation between the hydrology variables is very different now than it was historically.

Flow, which was important in describing the historic patterns of tree islands, now is

virtually non existent. Processes such as nutrient transport, seed dispersal, and soil

accretion and deccretion are influenced by flow magnitudes. The results of removal of this

structuring force from the system may be more subtle than changing hydroperiods, but just

as important.

These analyses provide evidence that changes in flow magnitudes as well as other

hydrologic variables contribute to the changes in the nature of tree islands in Loxahatchee.

Several assumptions made in these analyses now should be considered. It has been

assumed that tree islands are the resultant of processes operating in previous years and

that these processes are related to the average values of structuring forces of flow,

hydroperiod, and depth. Another assumption is that the relationship between the

hydrologic variables in the 26 years averaged by the NSM are similar to the hydrologic

variables that the landscape experienced in the 26 years prior to 1950. Because historic

data do not exist, this represents the best available information. Because of the magnitude

of changes that have occurred in terms of hydrology, the model outputs probably give a




Full Text
xml version 1.0 encoding UTF-8
REPORT xmlns http:www.fcla.edudlsmddaitss xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.fcla.edudlsmddaitssdaitssReport.xsd
INGEST IEID EGPE1A3L8_F8J72W INGEST_TIME 2013-01-23T16:14:56Z PACKAGE AA00012943_00001
AGREEMENT_INFO ACCOUNT UF PROJECT UFDC
FILES