<%BANNER%>

Meta-Stable States of Vegetative Habitats in Water Conservation Area 3A, Everglades


PAGE 1

META-STABLE STATES OF VEG ETATIVE HABITATS IN WATER CONSERVATION AREA 3A, EVERGLADES By ERIK POWERS A THESIS PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLOR IDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF SCIENCE UNIVERSITY OF FLORIDA 2005

PAGE 2

Copyright 2005 by Erik Powers

PAGE 3

This document is dedicated to my father, Dr. Lawrence W. Powers, who inspired my fascination with science early in life.

PAGE 4

iv ACKNOWLEDGMENTS I must first thank Dr. Wiley Kitche ns for his unwavering support and encouragement throughout my graduate career I thank my committee members Dr. Paul Wetzel and Dr. Ted Schuur for their advi ce and guidance. Instrumental in the experimental design, Paul Wetzel has pr ovided support from the beginning. Paul Conrads of the USGS performed the neural ne twork analysis for the hydrologic data set. His assistance was paramount to the completion of this thesis. Logistic support, including airboats and lodging, was provided by the Florida Cooperative Fish and Wildlife Unit, University of Florida, Gainesville. The following University of Florida graduate students and staff helped with field sampling and data processing: Stephen Brooks Janell Brush, Melissa DeSa, Jamie Duberstein, Joey Largay, Kristianna Lindgren, Julien Marti n, Ann Marie Muench, Alison Pevler, Laura Pfenninger, Derek Piotrowicz, Zach Welch, a nd Christa Zweig. Lastly, I thank my wife and best friend, Kristy Powers, for her undying patience and compassion despite my propensity for tracking mud into the house.

PAGE 5

v TABLE OF CONTENTS Page ACKNOWLEDGMENTS.................................................................................................iv LIST OF TABLES...........................................................................................................viii LIST OF FIGURES...........................................................................................................ix ABSTRACT.....................................................................................................................xiii CHAPTER 1 INTRODUCTION........................................................................................................1 What Are Meta-stable States?......................................................................................2 How Community Subtypes and Dr iving Forces Are Determined................................4 Communities of the Everglades....................................................................................6 Project Objectives.........................................................................................................8 2 DETERMINING COMMUNITY STRUCTURE........................................................9 Description of Study Site............................................................................................11 Methods and Materials...............................................................................................14 Sampling Regime................................................................................................14 Sampling Methodology.......................................................................................15 Processing Methodology.....................................................................................16 Data preparation and Relativization....................................................................16 3 CLASSIFICATION OF META-STABLE STATES.................................................20 Hierarchical Agglomerative Cluster Analysis............................................................21 Testing Importance Value Assumptions.....................................................................21 Indicator Species Analysis..........................................................................................27 Matching Similar Community Descriptions Between Sampling Events....................28 Distribution of Meta-Stable States Across the Landscape.........................................30 4 MULTIVARIATE ANALYSIS AND RESULTS.....................................................33 Hydrology...................................................................................................................33 Selecting Hydrologic Variables...........................................................................33

PAGE 6

vi Calculating Hydrologic Variables.......................................................................35 Hindcasting using neural networks..............................................................35 Extrapolating from well data to sample unit data........................................38 Nonmetric Multidimen sional Scaling.........................................................................39 Classification Trees and Char acterization of Meta-Stable States by Environmental Variables................................................................................................................45 Slough Physiognomic Type.................................................................................46 Deep sloughs................................................................................................47 Eleocharis elongata sloughs........................................................................48 Panicum sloughs..........................................................................................48 Shallow sloughs............................................................................................48 Wet Prairie Physiognomic Type..........................................................................48 Eleocharis sp. prairie...........................................................................................49 E. elongata prairie........................................................................................50 Panicum/Paspalidium/Eleocharis prairies...................................................50 Sawgrass prairies..........................................................................................50 Sawgrass Physiognomic Types...........................................................................51 Sawgrass monoculture (heavy sawgrass).....................................................51 Sawgrass with Bacopa and Ludwigia ...........................................................52 Sawgrass with Eleocharis sp. and Panicum .................................................52 Sawgrass with E. elongata and Crinum .......................................................52 Island Physiognomic Types.................................................................................53 Ghost islands................................................................................................54 Sawgrass ghost islands.................................................................................54 Tree islands..................................................................................................54 5 SUMMARY AND CONCLUSIONS.........................................................................55 Discussion...................................................................................................................55 Comparing NMS and Classi fication Tree Techniques...............................................63 Review of Methodology and Fu ture Tracks of Research...........................................64 APPENDIX A INDICATOR SPECIES ANALYSIS TABLES AND FIGURES..............................67 November 2002 Indicator Species Analysis Graphs..................................................67 June 2003 Indicator Species Analysis Graphs............................................................68 November 2003 Indicator Species Analysis Graphs..................................................69 June 2004 Indicator Species Analysis Graphs............................................................70 B COMMUNITY STATES AND THEIR STRUCTURAL SIGNATURES................71 C RESULTS OF THE NONMETRIC MULTIDIMENSIONAL SCALING ANALYSES...............................................................................................................80

PAGE 7

vii D IMPORTANCE CHARTS OF ENVIROMENTAL VARIABLES FROM CLASSIFICATION TREE ANALYSIS....................................................................84 LIST OF REFERENCES...................................................................................................87 BIOGRAPHICAL SKETCH.............................................................................................90

PAGE 8

viii LIST OF TABLES Table page 2-1 Complete species list for vegeta tive study in Water Conservation Area 3A. Authority for plant names and status is from Wunderlin, R.P. 1998 Guide to the Vascular Plants of Florida. University Press of Florida, Gainesville. Includes unknown species that occur in more than one sample.............................................17 2-2 An abridged species data matrix w ith importance values for species in each community unit........................................................................................................19 3-1 Meta-stable community states and their frequency at each sample event...............30 4-1 Hydrological variab les with abbreviations..............................................................34 4-2 Neural network model statistics for e ach station hindcasted. PME (percent model error) = RMSE (root mean-square er ror) / (range of measured data)......................38 4-3 Environmental variables used in the multivariate analyses and how they were relativized if a transfor mation was appropriate........................................................39 C-1 Stress in relation to dimensionality for slough NMS. A two-dimensional solution was chosen................................................................................................................81 C-2 Stress in relation to dimensionality fo r prairie NMS. A two-dimensional solution was chosen................................................................................................................81 C-3 Stress in relation to dimensionality for sawgrass NMS. A three-dimensional solution was chosen..................................................................................................82 C-4 Stress in relation to dimensionality fo r island NMS. A three-dimensional solution was chosen................................................................................................................83

PAGE 9

ix LIST OF FIGURES Figure page 1-1 A community can shift to an alternate state if the perturbati on is strong enough, or conditions change steadily over time. Note that two community states can exist in the same environmental conditions. For instance, two meta-s table states can operate under the same hydrological condi tions, but have different hydrologic thresholds (Scheffer 2001).........................................................................................3 2-1 Shaded area is the location of the st udy area. Water Conservation Area 3A is designated as section 9 on this map.........................................................................11 2-2 Satellite composite of the study area in Water Conservation Area 3A. Twenty plots were distributed with a stratified rand om design for the sampling procedures........12 2-3 An overlay of a square kilometer plot on satellite imagery. The blue dots signify reference poles aligned with belt transects wi thin the plot. Each transect crosses at least one community boundary................................................................................13 2-4 A diagram of a belt transect consisting of three traversable subtransects. Each sub transect can be sampled on four differe nt occasions twice on each side..............14 3-1 Cluster dendrogram from November 2002 sampling event. Community units are listed on the left and color coded with respect to their a priori designation............22 3-2 Cluster dendrogram from June 2003 sa mpling event. Community units are listed on the left and color coded with resp ect to their a priori designation......................23 3-3 Cluster dendrogram from November 2003 sampling event. Community units are listed on the left and color coded with respect to their a priori designation............24 3-4 Cluster dendrogram from June 2004 sa mpling event. Community units are listed on the left and color coded with resp ect to their a priori designation......................25 3-5 A scatterplot of sawgrass comm unities sampled in November 2002. Axes correspond to percent relative biomass and percent relative de nsity. Each point represents one sawgrass unit. Each sa wgrass type resulting from the cluster analysis is coded in the legend.................................................................................26 3-6 Shows the distribution of meta-sta ble states by physiognomic type into four quadrants of the study landscape..............................................................................31

PAGE 10

x 4-1 Green triangles represent the monitori ng stations set up by vari ous agencies. These stations upload real-time data to the web daily. Yellow ci rcles indicate the temporary stations that were established in December 2002...................................36 4-2 A whole scale ordination plot of th e community sample units. Triangles represent individual sample units and crosses repr esent species. The key to the legend: 1 prairie physiognomic type; 2sloug h physiognomic type; 3sawgrass physiognomic type; 4island physiognomic type. The environmental gradients (minimum depth and mean depth) are repr esented with red vect ors closely aligned with axis 2................................................................................................................41 4-3 Island-type ordination plots. Triangles represent i ndividual sample units and crosses represent species. The key to the legend: 1sawgrass ghost island; 2 ghost island; 3tree island......................................................................................42 4-4 Slough-type ordination plots. Triangles represent individual sample units and crosses represent species. The key to the legend: 4deep slough; 6 Panicum slough; 7 E. elongata slough; 8shallow slough.................................................43 4-5 Sawgrass-type ordination plots. Tria ngles represent individual sample units and crosses represent species. The ke y to the legend: 9sawgrass with Eleocharis sp./ Panicum ; 10sawgrass with Bacopa/Ludwigia ; 11sawgrass with E. elongata/Crinum ; 12sawgrass monoculture.........................................................43 4-6 Prairie-type ordinati on plots. Triangles represent individual sample units and crosses represent species. The key to the legend: 1 Eleocharis sp. prairie; 2 Panicum/Paspalidium/Eleocharis sp. prairie; 3 E. elongata prairie; 5sawgrass prairie. Environmental gradients shown as red vectors closely aligned with axis 2.44 4-7 Classification tree for the meta-stable states on 10 environmental variables. The number of sample units in each leaf are shown in parentheses below each bar graph, which shows the compositions of communities within each leaf.................46 4-8 Classification tree for 4 slough community states on 10 environmental variables. This model was pruned from a tree size of 7 leaves to five, based on a cost complexity pruning curve, selecting the sma llest tree within one standard error of the best. The number of sample units in each leaf is shown in parentheses below each bar graph, which shows the compositions of communities within each leaf...47 4-9 Classification tree for 4 wet prai rie community states on 10 environmental variables. This model was pruned from a tr ee size of 11 leaves to eight, based on a cost complexity pruning curve, selecting th e smallest tree within one standard error of the best.................................................................................................................49 4-10 A classification tree for 4 sawgra ss community states on 10 environmental variables. This model was pruned from a tr ee size of 14 leaves to eight, based on a cost complexity pruning curve, selecting th e smallest tree within one standard error of the best.................................................................................................................51

PAGE 11

xi 4-11 Classification tree for 3 island-t ype community states on 10 environmental variables. This model was pruned from a tr ee size of 10 leaves to eight, based on a cost complexity pruning curve, selecting th e smallest tree within one standard error of the best.................................................................................................................53 5-1 Distribution of sample units of each community state along a hydrologic variable (mean annual water depth). Communities are grouped by physiognomic type: squares=prairies, triangles=sloughs, ci rcles=sawgrass, crosses=islands.................57 5-2 Distribution of sample units of each community state along a peat depth gradient. Communities are grouped by physiogno mic type: squares=prairies, triangles=sloughs, circles=sa wgrass, crosses=islands..............................................58 5-3 A time-series graph of water stage at a monitoring station within Plot 4. Note the extreme highs and lows of the second water year compared to the first water year.59 A-1 Change in p-value from the randomiza tion tests, averaged across species at each step in the clustering.................................................................................................67 A-2 Number of species with p 0.05 for each step of clustering...................................67 A-3 Change in p-value from the randomiza tion tests, averaged across species at each step in the clustering.................................................................................................68 A-4 Number of species with p 0.05 for each step of clustering...................................68 A-5 Change in p-value from the randomiza tion tests, averaged across species at each step in the clustering.................................................................................................69 A-6 Number of species with p 0.05 for each step of clustering...................................69 A-7 Change in p-value from the randomiza tion tests, averaged across species at each step in the clustering.................................................................................................70 A-8 Number of species with p 0.05 for each step of clustering...................................70 B-1 Structural signature of the Panicum/Paspalidium/Eleocharis Prairie......................72 B-2 Structural signature of the Shallow Slough..............................................................72 B-3 Structural signatu re of the Ghost Island...................................................................73 B-4 Structural signature of the Deep Slough..................................................................74 B-5 Structural signature of the Eleocharis elongata Slough...........................................74 B-6 Structural signature of the Eleocharis elongata Prairie...........................................74 B-7 Structural signature of the Sawgrass Prairie............................................................75

PAGE 12

xii B-8 Structural signature of the Eleocharis Prairie..........................................................75 B-9 Structural signature of the Panicum Slough.............................................................76 B-10 Structural signature of the Tree Island.....................................................................76 B-11 Structural signatu re of Sawgrass with Bacopa and Ludwigia ..................................77 B-12 Structural signature of the Sawgrass Ghost Island...................................................77 B-13 Structural signatu re of Sawgrass with E. elongata and Crinum ..............................78 B-14 Structural signature of Heavy Sawgrass...................................................................78 B-15 Structural signatu re of Sawgrass with Eleocharis and Panicum ..............................79 C-1 A scree plot for th e slough-type ordination..............................................................80 C-2 A scree plot for th e prairie-type ordination..............................................................81 C-3 A scree plot for th e sawgrass-type ordination..........................................................82 C-4 A scree plot for th e island-type ordination...............................................................83 D-1 Importance rankings of predictor variables for the slough physiognomic type.......84 D-2 Importance rankings of predictor va riables for the prairie physiognomic type.......85 D-3 Importance rankings of predictor vari ables for the sawgrass physiognomic type...85 D-4 Importance rankings of predictor va riables for the island physiognomic type........86

PAGE 13

xiii Abstract of Thesis Presen ted to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Master of Science META-STABLE STATES OF VEG ETATIVE HABITATS IN WATER CONSERVATION AREA 3A, EVERGLADES By Erik Powers December 2005 Chair: Wiley Kitchens Major Department: Interdisciplinary Ecology The Everglades consists of a consta ntly dynamic patchwork of vegetative communities, confined by a matrix of lev ees and canals into impoundments. Water Conservation Area (WCA) 3A is a cent rally located impoundment, relatively far downstream from the nutrient-laden waters of the Everglades Agricultural Area. The major determinants of community struct ure within WCA 3A are hydrology and soil characteristics. This study monitors plan t community structure over two years, in transects randomly stratified across the landscape, to determine what community states manifest between marsh physiognomic types. These communities are constantly shifting to alternate states, thus described as meta -stable states. They are identified through unique, but related, vegetative st ructure, and characterized by a combination of particular environmental conditions. The transects were sampled semiannually for species biomass and density within ecotonal boundaries of approximately 140 communities identified a priori resulting in a data set of 513 community sample units. Hydrology was monitored

PAGE 14

xiv with surface water data loggers and levels were hindcast 10 years prior to the beginning of the study with neural network models. Pe at depths were recorded for each of the community units. A hierarchical cluster analysis on the sa mple units for each sample event produced distinct groups that, following an indicator species analysis, were interpreted as metastable states of the four physiognomic types of communities of WCA 3A: slough, wet prairie, sawgrass, and islandtype. The fifteen meta-stabl e states include deep slough, shallow slough, Panicum slough, E. elongata slough, E. elongata prairie, Eleocharis sp. prairie, Panicum/Paspalidium/Eleocharis prairie, sawgrass prairie, sawgrass with Eleocharis/Panicum sawgrass with E. elongata/Crinum heavy sawgrass, sawgrass with Bacopa/Ludwigia sawgrass ghost island, ghost island, a nd tree island. A classification tree analysis of each physiognomic type determined that both hydrology and peat depths were major determinants of community composition. The meta-stable states had unique envir onmental characteristics when accounting for multiple variables. However, when environmental variables are examined individually between community states, substantial overlap of environmental thresholds is evident. It can be concluded that the state of a community in the Everglades is dynamic due to overlap of individual threshol ds, but can potentially be predicted through multivariate modeling. The capability to model community dynamics of Everglades habitats is crucial to hydrologi cal management strategies. As restoration efforts proceed, models incorporating how communities respond to management regimes can be essential tools in scenario analysis.

PAGE 15

1 CHAPTER 1 INTRODUCTION In ecology, a biological community consis ts of coexisting organisms that are linked to one another through unique interactions and asso ciations, thus forming a complex whole. Plant communities can easily be observed in the field, as they are relatively sessile and, given a sharp physical boundary, have welldefined ecotones. These ecotones are usually comprised of a combination of species of the bounding communities and some unique species as well (Kent and Coker 1992). Therefore, communities can be identified as physiognomic types. Physiognomic types are defined by their species structure, or by what sp ecies exist and their relative densities and biomass. The four physiognomic community types of the central Everglades, as described by Davis (1943) and Loveless (1959), are sawgra ss, wet prairie, slough, and tree islands. These are easily recognized and usually ha ve sharp boundaries corresponding to only a slight change in elevation (McPherson 1973). Water covers the Everglades landscape the vast majority of the time, leading to a widely believed theory that communities are driven by hydrologic variables (White 1994). This study attempts to determine vegetative community subtypes, or meta-stable states, within sawgrass, wet prairie, and slough. This includes the exploration of methods that will enable scientists to document shifts between community states and between majo r physiognomic community types over time. In doing so, a physical hydrologic threshol d can be associated with specific physiognomic community types.

PAGE 16

2 What Are Meta-stable States? Sawgrass, slough, and wet prairie physiognomic community types exhibit multiple assemblages and representations of plant species or multiple meta-stable steady states. These different representation s of the same community are alternate representations and states of that community, and reflect th e environmental conditions at that point in space and time (Gunderson and Pritchard 2002). Transitions between these within-type community states (tall sawgrass into short sawgrass), or between the major physiognomic community types (e.g., sawgrass into wet prairie) are indicative of responses to environmental change. One example of the existence of multip le steady states is the well-documented process of eutrophication of lakes. Two alte rnative states can be characterized as (a) clear water and rooted macrophyt es or (b) turbid water with planktonic algae. These states are relatively stable, but can slip into the other due to a pertur bation of a keystone process, or the removal or addition of a ke ystone species (Carpent er et al. 2001). If environmental conditions change slowly, a sh ift in community state can occur given the conditions continue to change over time. Alternative states may even share some of the range of environmental conditions that they could potentially exist in. Figure 1-1 shows how a community state can shift given perturbations in the environment. In the context of community state theory, meta-stable states can be defined as an alternative state of a physiognomic type of community that occurs under predictable environmental conditions, yet those condi tions are dynamic by nature resulting in constantly shifting community states of that physiognomic type. These meta-stable states can be witnessed throughout a landscap e with multiple physiognomic types and fluctuating environmental conditions. Meta-sta ble states imply that community structure

PAGE 17

3 is changing in the direction that conditions are driving it, and there are no equilibriums associated with them. Such is the case in the Everglades, where elevational gradients are slight, yet hydrology fluctuates considerably on a semiannual basis. These communities are continually stresse d. The structure of the community state is a representation of various environmental conditions presently, prev iously, and historically. In summary, a meta-stable state is a representation of the trajectory of environmental conditions in the species structure of the physiognomic type of community. The identification of metastable states, as determined in this study, will be addressed in Chapter 3. Figure 1-1. A community can shift to an alte rnate state if the perturbation is strong enough, or conditions change steadily ove r time. Note that two community states can exist in the same environmen tal conditions. For instance, two metastable states can operate under the sa me hydrological conditions, but have different hydrologic threshol ds (Scheffer et al. 2001).

PAGE 18

4 How Community Subtypes and Driving Forces Are Determined Scientists define ecological resilience as the property that mediates transition among stability domains (Holling 1973). If the stable states of a specific community can be defined by their relative composition (species present, relative dens ity of each species, relative biomass of each species), then the environmental conditions that the community state tends to persist in indi cate a potential driving force. A record of historical and present conditions of th e driving force(s) at that site can lead to clues as to where environmental thresholds lie for the current community type. Determining community states requires identification of external (abiotic) and internal (biological) driving forces. All biological communities have several driving forces, some of them working in concert. However, depending on the temporal and spatial scale of interest, some of these forces can be ignored as ha ving negligible effect (DeAngelis and White 1994). A study of community level pr ocesses over the course of two years can rule out slow processes such as interglacial sea level rise, tectonic movements, and global climate change, as well as intermediate processes such as weathering, and soil accretion. The focus of a study such as this should be on processes that will affect the st ructure of communities within the time frame of the study. In a hydrologically driven system such as the Ever glades, hydropattern is a driving force that will have one of the strongest effects on vegetation composition and structure. For example, recruitment of many wetland sp ecies through their respective seed and propagule bank is dependent on meeting cert ain hydrologic and othe r criteria (van der Valk 1990). Sawgrass ( Cladium jamaicense ), the archetypical plant species of the Everglades, requires occasional drying events in order to germinate (Smith et al. 2002), as is generally the case with emergent wetland plant species (G erritson and Greening

PAGE 19

5 1989). This is especially true in low nutrient regions of the Everglades where sawgrass stands may neglect important biological functions when exposed to extremely long hydroperiods (Weisner and Miao 2004). Emerge nt plants tend to allocate biomass to shoot length and blade growth in deeper hydrologic conditions, and allocate less energy to developing belowground rhizome biomass (Weisner and Strand 1996) In regions of the Everglades such as southern WCA 3A th at exhibit perpetually long hydroperiods and low nutrient concentrations, sawgrass co mmunities, though persistent, are unable to recruit resulting in sparse, pa tchy distributions, rather than thick, conti nuous landscapes that were present prior to drainage and impoundment (Wood and Tanner 1990). The resulting communities are composed of emergents such as Pontederia cordata or Sagittaria lancifolia and woody vegetation such as Cephalanthus occidentalis interspersed with floatin g leaf aquatics such as Nymphaea odorata usually associated with deep water marshes. Peat accretion is a slow pr ocess that is an important driving force in determining topography and hence hydroperiod. Competi ng with this process is decomposition, which may occur at a much faster rate duri ng drought effects through oxidation or fire. As these changes in peat depth occur, be drock topography continues to exert a strong influence on vegetation through its influence on the patterns of water depth and flow. Autogenic succession may occur over long periods of time, but is probably rare (Gleason and Stone 1994). Other forces that could have major effects on community composition in the Everglades and within the time frame of the study are fire, va riation in nutrient supply, freezes and wind (DeAngelis 1994). During the pe riod of study, fire did not occur in the

PAGE 20

6 monitoring transects, and hence was not a fa ctor in determining community composition. However, the history of fire for each of th e transects is unknown. For the purposes of this study, it is assumed that the transects had not been burned for a considerable time prior to the study. Intense or repeated fr eezes were also not issues. A series of hurricanes did strike Florid a in the summer of 2004, how ever Miami-Dade County was not in the path of any of those disturbances. The study area being investigated is far from any upstream point source of nutrients and ex otic species invasions (canals, urbanized areas, etc.). The Everglades, historically, is an oligotrophic system, so nutrient loading will be assumed to be constant at low levels. Communities of the Everglades Of the major physiognomic community types of the Everglades, I intend to focus on three that are both naturally and an thropogenically influenced by hydrology sawgrass marshes, peat-based wet prairies ( Eleocharis flats), and sloughs. These three herbaceous communities all occur in southern and central WCA-3A, and often adjacent to each other. They occur in areas with slightly different relative hydroperiods, with sawgrass being the driest followed by wet pr airie and finally slough as typically the wettest of the physiognomic types (White 1994). Sawgrass is the characteristic plant species of the freshwater Everglades. It is well adapted to flooding, drought, and burning bu t is killed if high water levels are prolonged (Herndon et al. 1991). Sawgrass domi nates the oligotrophic fresh waters of the Everglades due to its low nutrient re quirements (Gunderson 1994). Sawgrass occurs in strands that run longitudinall y (the historical direction of water flow) in WCA 3A. It also persists in patches of deep water in th e southern extent of 3A, as well as on floating peat mats and the outer edges of tree islands. Occasionally shrub islands appear in place

PAGE 21

7 of burned or drowned tree islands and sawgrass strands. Cephalanthus occidentalis and Pontederia cordata are common associates of sa wgrass in these transitional physiognomic types. Islands and their transitio nal states will also be examined in this study. Wet prairies can be classified into tw o groupspeat-based and marl-based. Marlbased wet prairies are confined to the marl wetlands situated in the Everglades National Park, and were not included in this study. Peat -based prairies can fu rther be divided into three types Eleocharis Rhynchospora and Panicum flats. Of these types, only Eleocharis or spikerush, flats, and Panicum or maidencane flats, are present in the study area. Rhynchospora prairies are relatively rare after the impoundment of the Everglades. Wet prairies are typically more diverse than sawgrass or slough communities and occur often as transitional communities in deeper areas where slough communities are prevalent (Gunderson 1994) or between slough and sawgrass community as a transitional community. Slough communities consist of associations of floating-leafed aquatic plants and are generally the wettest of the communities in WCA-3A. Submerged aquatics are also associated with sloughs and provide structure for periphyton, the main source of primary production in the freshwater Everglades (Gunderson 1994). Each of these communities has been observe d in different forms and structure, yet they are documented in the general body of scie ntific literature as single communities. This project presents evidence that these alte rnative community states are characteristic of the environmental conditions at that si te. More importantly, transitions between alternative states of one physiognomic type ma y occur more readily than a shift between

PAGE 22

8 major physiognomic types. In other word s, the resilience of a major physiognomic community type is greater than the resilience of one of its alternate states. This is tested through the identification of the hydrologic ranges of each meta-sta ble state. If there is a substantial overlap of hydrol ogy between meta-stable states of communities, then it can be concluded that a shift to an alternate community state while maintaining its basic physiognomic community type is a possible re sponse to extended exposure to threshold conditions (see Figure 1-1). Project Objectives With this research, I intend to descri be the multiple meta-stable states of physiognomic marsh types of Water Conserva tion Area 3A in terms of community structure and their respective environmental tolerances. Fi rst, through tabul ating relative densities and relative biomass of species pr esent on established transects during wet and dry seasons, the current state of a community at any given point in time during the study was identified. Vegetative community monitoring efforts will continue for two years, sampling at a rate of twice a year for a tota l of four sampling events. Hydrologic ranges for the various meta-stable states of sawgra ss, slough, and wet prairie are identified and each community state is charac terized by its environmental variables using classification trees. Inferences, based on the range of conditions that a state may tolerate, can be made on the dynamics and resiliency of va rious vegetative meta-stable states. Although the study monitors communities over time, tracking the change of specific sites over time was not included in the scope of this project. The temporal scope of this study is limited to observing commun ities under different seasons and water years to capture the various comm unity states that might ma nifest under those conditions.

PAGE 23

9 CHAPTER 2 DETERMINING COMMUNITY STRUCTURE Everglades communities were originally identified by the dominant species associated with a congregati on of smaller or less preval ent species. Loveless first documented vegetative assemblages in the Ev erglades (Loveless 1959). Several types of sawgrass, wet prairie, and slough commun ities were identified through the abundance and densities of dominant and associative sp ecies. His descriptions of vegetative communities serve as an introduction and as the basis of comparison for the communities revealed in the following analyses. The following community types were identified by Loveless: Cladium Sagittaria Panicum hemitomon : This sawgrass community can occur in sparse, dense, or monotypic stands of sawg rass. It is associated with duck potato and maidencane, as well as a suite of other sp ecies depending on the density of sawgrass. Species composition tends to vary between the dry and wet seasons. Cladium Myrica Ilex : A drier community of sawgrass, this congregation of species includes woody thickets of bu ttonbush, wax myrtle, and dahoon holly. Cladium Panicum hemitomon : Similar to the duck potato/maidencane sawgrass community, but occupies drier si tes. Densities range from sp arse to moderately thick. Rhynchospora Flats : This community was more prevalent during predrainage conditions. The wettest of th e Loveless communities except for sloughs, this assemblage includes beakrush as the domina nt species and spikerush as the common associate. These communities are typically f ound adjacent to sawgrass and shrub island communities.

PAGE 24

10 Panicum hemitomon Flats : Maidencane is the dominant member of this community and usually occupies drier sites. This community is resilient to fire and can withstand long periods of flooding while main taining its basic species configuration. Associative species usually include spikerush and spider lilies. Eleocharis Flats : Easily recognizable as monotypi c stands of spikerush. This community is usually found along the s outhern and western reaches of Water Conservation Area 3A. Sloughs : The wettest of the communities, slough s are usually filled with water year round. Species associated with sloughs ar e floating water lily bladderwort, and spatterdock. Sloughs comprise the drainage vectors of the Everglades, running generally longitudinally along the landscap e in a north-south direction. The communities described by Loveless, while useful from a naturalists perspective, are outdated with respect to the decades of impoundment effects on Everglades ecology and irrelevant to studyi ng short-term succession. Communities of the Everglades are dynamic on two time scales: seasonal and long term (multiannual). Subtropical south Florida has two distinct hydrologic seasons a wet season during the summer and fall months, and a dry season duri ng the winter and spring months. Changes in hydrology imposed by both seasonal fluctu ations and water regimes managed by state agencies will have subtle if not pr ofound effects on community composition. If restoration agents mean to influence the eco logy area from the bottom-up, that is get the water right, then the scientific lens must focus on the immediate responses of plant communities that provide wildlife ha bitat to variations in hydrology.

PAGE 25

11 Description of Study Site The study was conducted in the southern half of Water Conservation Area 3A located in Dade and Broward counties (see Figure 2-1). Bounded by Tamiami Trail to the south, Holiday Trail (a heavily trafficked airboat trail) to the north, Big Cypress National Preserve to the west, and Water Cons ervation Area 3B to the east, the study site is made up of a matrix of freshwater ha bitats ranging from short hydroperiod bay and willow tree islands to deep water sloughs. Strands of sawgrass run longitudinally, divided by wet prairie and slough. This area was chosen because of the smattering of distinct communities, abundance of ecotones, and noticeable elevational gradients on a landscape scale as well as community scale. The total area of the study site is 62,000 hectares. Figure 2-1. Shaded area is the location of the study area. Wa ter Conservation Area 3A is designated as section 9 on this map.

PAGE 26

12 A comparative observational study was determ ined to be the best scientific method to investigate response of natural communities to environmental variations. Studies of this type have a wide domain of infere nce and are conducive to the confirmational hypothesis that Everglades plant communitie s are dynamic with respect to hydrology. Twenty study plots were established based on a stratified random design, using landscape-scale elevational (longitudinal) a nd peat depth (latitudi nal) gradients (see Figure 2-2). The square plot s are one kilometer on a side, a scale that sufficiently includes variety of communities and ecotones. Figure 2-2. Satellite compos ite of the study area in Water Conservation Area 3A. Twenty plots were distributed with a stratified random desi gn for the sampling procedures.

PAGE 27

13 Each plot contains two or three belt tran sects that crosses at least one community boundary (see Figure 2-3). The design of the belt transects allows for the repeated destructive sampling of each transect while a voiding the issues associ ated with repeated measures. Every sample event allows for th e removal of plant material from the field under the assumption that previous sampling efforts have neglig ible effects on the following sample. Each transect was establis hed at a random location within a plot. The number of samples within the transect vary from 10 to 34, depending on the length of the transect. Figure 2-3. An overlay of a square kilometer plot on satellite imagery. The blue dots signify reference poles aligned with be lt transects within the plot. Each transect crosses at least one community boundary.

PAGE 28

14 Methods and Materials Sampling Regime Destructive sampling along the belt transect was scheduled twi ce a year once at the peak of the dry season (June), and on ce at the peak of the wet season (November) which corresponds to the growing season. Sampling along the belt transects was organized to avoid removal of plant materi al from the same place at any given time during the study. The belt transects consisted of three parallel subtransects spaced four meters apart (see Figure 2-4). Each subtra nsect could be sampled four times twice on each side with staggered placement of sample locations. Sample locations were randomly selected for each sample event. For example, November 2002 sample event was randomly determined to be sample G, which corresponds to the right side of the middle subtransect and starts from the zero me ter point. Samples are spaced three meters along the transect. Sample H would be stagge red and correspond to the right side of the middle subtransect and start from the 1.5 meter point. N L M K J H I G F D E C Figure 2-4. A diagram of a belt transect cons isting of three traversable subtransects. Each sub transect can be sampled on f our different occasions twice on each side.

PAGE 29

15 Sampling Methodology The area of each sample is determined by a .25 square-meter circular hoop with its center around a dowel placed at the samp le point offset from the subtransect by a meter. The dowel marks the sample point and allows for a reference when the hoop has a tendency to float or deviate from its original placement. Floating vegetation is collected from the sample first. After the floating vegetation is removed any material that may subsequently drift into the sample area is di sregarded. The rooted vegetation is then cut at the soil surface and collected. All vegetation is collected in burlap sacks to allow some air exchange for the evaporation of excess moisture. The vegetation remained more resistant to disintegration and mold when stor ed in burlap rather th an being stored in plastic. During the sample harvest, rotten material determined by its structural integrity, was discarded. For example, if the material when given a gentle shake did not maintain any rigidity, than the material was deemed to be rotten and associated with the peat substrate. Rotten vegetation was difficult to identify and proved almost impossible to quantify. This structural integrity test provi ded consistent and comprehensive criteria for determining viable plant material. Samplers remained within the one meter wide subtransect path to avoid walking on sample locations. Water depths were measured at each sample point and at the transect start and end poles for reference point s to be tied back in to the monitoring wells for continuous hydrologic data for each samp le point. Samples for the transect were labeled and loaded into an airboat for transport back to a refrigerated storage unit. A total of 1190 samples were collected from the study area per sample event.

PAGE 30

16 Processing Methodology Each sample was sorted by species and th e numbers of individuals were tabulated for each species. Counts for Eleocharis Pontederia Nymphaea Bacopa Crinum and woody vegetation were determined by the number of emergent stems. Cladium and Typha counts were determined by the number of emergent culms. Utricularia and Chara counts were impossible to determine in the laboratory and were tabulated as either present or absent. Species for each sample were separated into paper bags, labeled, and dried for at least two weeks in walk-in ovens set at 140 F. Dried plant material avoids the inclusion of water weights that can vary considerably between species. After the samples were dried, the dry biomass for each species was measured on digital scales to the nearest hundredth of a gram. The drie d plant material was then discarded into compost. Biomass and count data were tr anscribed into an Ex cel spreadsheet in accordance with appropriate quality control me asures. See Table 2-1 for a complete species list. Data Preparation and Relativization Community units are heretofore defined as the conglomeration of the samples within one of the physiognomic community un its represented in a transect. Each community unit is designated with a plot number, transect number, a priori physiognomic type, and sample event. A priori physiognomic types include cattail (C), sawgrass (G), ghost island (I), prairie (P), slough (S), and tree island (T). These habitats are important to aquatic macrofauna and are used differently by vari ous suites of species (e.g., Loftus and Kushlan 1987, Gunderson a nd Loftus 1993, Jordan et al. 1994, 1996). The sample event is designated by where within the belt transect the community was sampled for that sample collection. Samp le events include: G November 2002, E

PAGE 31

17 Table 2-1. Complete species list for vegetative study in Water Conservation Area 3A. Authority for plant names and status is from Wunderlin, R.P. 1998 Guide to the Vascular Plants of Florida. University Press of Florida, Gainesville. Includes unknown species that occur in more than one sample. Scientific name Code Family Bacopa caroliniana BAC Scrophulariaceae Blechnum serrulatum BLS Blechnaceae Cephalanthus occidentalis CEO Rubiaceae Chara spp. CHsp Characeae Cladium jamaicense Alive CLA Cyperaceae Cladium jamaicense Dead CLD Cyperaceae Crinum americanum CRA Amaryllidaceae Cyperus haspan CYH Cyperaceae Dryopteris ludoviciana DRY Dryopteridaceae Eleocharis elongata ELG Cyperaceae Eleocharis spp. Elsp Cyperaceae Fuirena breviseta FUB Cyperaceae Hymenocallis sp. HYsp Amaryllidaceae Leersia hexandra LEH Poaceae Ludwigia spp. Lusp Onagraceae Nymphaea odorata NYO Nymphaeaceae Nymphoides aquatica NMA Menyanthaceae Osmunda regalis OSR Osmundaceae Panicum hemitomon PAH Poaceae Paspalidium geminatum PDG Poaceae Peltandra virginica PEV Araceae Polygonum spp. POsp Polygonaceae Pontederia cordata PNC Pontederiaceae Potamogeton spp. PTsp Potamogetonaceae Rhynchospora tracyi RHT Cyperaceae Sagittaria lancifolia SAL Alismataceae Salix caroliniana SAC Salicaceae Typha domingensis Dead TYD Typhaceae Typha domingensis Alive TYA Typhaceae Unk. Jointed stem UnkJS Unk. Segmented rush UnkSR Unk. Triangular stem UnkTS Unk. Sawgrass-like grass UnkSG Utricularia spp. UTsp Lentibulariaceae Vallisneria sp. VAsp Hydrocharitaceae Vine Unkn VIN Woodwardia virginica WOV Blechnaceae

PAGE 32

18 June 2003, D November 2003, and J June 2 004. For example, P18E2 refers to the prairie community in plot 18, transect 2, sampled on event E (June 2003). Prior to relativizing the data, I dele ted samples that were missing count or biomass data. This amounted to approximately 1% of the total sample. I also removed samples that occurred adjacent to the ecotone. Locations of ecotones were determined in the field by noting the samples between whic h dominant species appear and disappear, indicating a different physiognomic type. The definitions of the a priori communities were used to determine physiognomic types. Ecotones in the conservation area are typically sharp and distinguishable allowi ng for minimal observer error in designating ecotone location. This was done to remove samples that may be considered to be transitional or not a typical representation of that commun ity unit. Approximately 85% of the samples remained in the analysis a nd are assumed to be re presentative of the community units sampled. The community data was converted into relative proportions for each community unit sampled. Counts for each species in ev ery sampled community were expressed as the relative density of th at species. For example, the relative density of Eleocharis in community unit P18E2 equa ls the total count of Eleocharis stems in P18E2 divided by the total count of all species in community unit P18E2. Relative biomass was calculated in similar fashion equaling the proportion of the total biomass of a species in that community unit to the total biomass of all species in that unit. Averaging the relative density and relative biomass results in an importance value for each species in each community unit. Th e advantage of using importance values in ecological community analysis is that they are equally influenced by large biomasses and

PAGE 33

19 large stem densities, so that species that di ffer in size and density ca n be compared within the same sample unit. The disadvantage of im portance values is that a species that has large biomass values and sparse densities can have the same importance value as a species with small biomass values and high densities (McCune and Grace 2002). Later I will discuss how I tested the assumption that importance values can distinguish different community stands, regardless of the vulnerabi lity associated with importance values. The resulting importance values for each species in each community unit were transcribed into a data matrix for analysis (s ee Table 2-2). The data matrix is then ready to be processed for multivariate analysis including clustering, indicator species analysis, and ordination. Table 2-2. An abridged species data matrix with importance values for species in each community unit. Units Species BAC CEO CLA CLD CRA ELG ELsp C2D1 0.512563 0.00000 11.09662 5.36581 0.00000 0.31718 45.34719 C2D2 3.045374 0.00000 20.70394 15.23677 1.30300 0.00000 0.00000 G0D1 13.694759 0.00000 38.46195 6.61118 0.00000 22.69860 0.25166 G0D2 0.000000 0.00000 70.29885 18.51327 5.63622 0.00000 0.00000 G0D3 1.566028 0.00000 60.90582 17.21278 0.00000 0.00000 0.00000 G10D1 31.561145 0.39661 27.15960 4.80673 3.86429 0.00000 0.25813 G10D2 0.000000 2.26382 79.35584 6.85827 0.00000 0.00000 0.00000 G10D3 0.000000 22.46180 59.52005 13.52616 0.00000 0.00000 0.00000 G11D1 4.807172 4.24228 22.93135 0.91426 6.15084 44.02458 0.35548 G11D2 0.000000 1.99904 76.09261 18.43122 0.00000 0.00000 0.00000 G11D3 17.612373 6.71708 36.94830 4.02826 15.79072 11.58430 0.00000 G12D1 51.287816 0.00000 14.06211 0.00000 0.00000 20.81183 0.60776 G12D2 0.000000 5.57459 60.60979 23.03208 0.00000 0.00000 0.00000 G13D1 3.280178 16.32098 34.92181 11.40519 0.00000 0.00000 0.00000 G13D2 0.223056 0.00000 32.10193 5.04516 0.00000 0.52885 30.94700 G13D3 22.684489 0.00000 57.68911 8.67561 0.73124 0.00000 0.00000 G14D1 0.000000 0.00000 5.43556 0.00000 5.95042 0.00000 0.00000 G14D2 35.224940 0.00000 2.88301 0.00000 0.00000 0.00000 2.29568

PAGE 34

20 CHAPTER 3 CLASSIFICATION OF M ETA-STABLE STATES In order to determine how the vegetative habitats of the Everglades change in response to continuously varying environmen tal conditions, identification of the metastable states in which they are observed is required. Informal observation of plant communities in Water Conservation Area 3A yields four physiognomic community types: sawgrass, wet prairie, slough, and sh rub/tree island. Ghost islands are also a distinguishable community as old sawgrass ri dges or islands that have experienced a disturbance such as extreme flooding or fire. Ghost islands generally have some sparse sawgrass, pickerelweed and buttonbus h associated with them. Cattail ( Typha spp.) communities can also be observed, however there were only two community units sampled that had cattail as a major component. Subtle differences in the composition w ithin these physiognomic types require the statistical analysis of hier archical classification. Cla ssification through hierarchical cluster analyses is necessary to identify these meta-stable community states and recognize the subtle differences in community structure between these states. Metastable states will be identified as discerna ble subunits of physiognomic types. In the Everglades, these meta-stable states w ill be represented through the range of environmental conditions that physiognomic type s in WCA 3A exhibit. After the cluster analyses, environmental conditi ons at those sites are invest igated to produce profiles of environmental conditions and thresholds.

PAGE 35

21 Hierarchical Agglomerative Cluster Analysis I used the species matrices referenced in Chapter 2 to apply a hierarchical agglomerative cluster analysis. Each matrix contains importance values for every species in each community unit for a particular sampling event. In all, there were four sample events yielding four matrices with the same community units in each matrix. Agglomerative clustering methods build groups hierarchically from the bottom up, forming groups by fusing similar subgroups together (McCune and Grace 2002). The optimal number of groups is calculated thr ough an indicator species analysis. Cluster analyses first calculate a matrix of distances between each pair of entities. Groups that meet the minimum distance criteria are merg ed and their attributes combined. The merging process continues until there is only one group. The result is a dendrogram complete with a distance measure (from the di stance matrix). The distance measure is a function of the information lost at each clustering step (Wishart 1969). The cluster analysis was performed on th e PC-Ord software using a Euclidian (Pythagorean) distance measure. Wards linkage method was chosen for its combinatorial compatibility. Wards method al so conserves the proper ties of the original space as group attributes merge, keeping th e Euclidian distances consistent throughout the analysis (Wishart 1969). Comm unity units were color coded by a priori classification of physiognomic community types based on obse rvation in the field. See Figures 3-1, 32, 3-3, and 3-4 for the resulting dendrograms for each sampling event. Testing Importance Value Assumptions As mentioned previously, im portance values have one major disadvantage in that a large, sparse stand has the same value as a small, dense stand. Because the purpose of this study is to discriminate between vegetative community stat es by their structure, those

PAGE 36

22 Figure 3-1. Cluster dendrogram from Nove mber 2002 sampling event. Community units are listed on the left and color coded w ith respect to their a priori designation.

PAGE 37

23 Figure 3-2. Cluster dendrogram from June 2003 sampling event. Community units are listed on the left and color coded with respect to their a priori designation.

PAGE 38

24 Figure 3-3. Cluster dendrogram from Nove mber 2003 sampling event. Community units are listed on the left and color coded w ith respect to their a priori designation.

PAGE 39

25 Figure 3-4. Cluster dendrogram from June 2004 sampling event. Community units are listed on the left and color coded with respect to their a priori designation.

PAGE 40

26 habitats must be recognized as distinct states in th e analysis. To test the assumption that importance values will discriminate between these community assemblages, I plotted sawgrass stands from one sample event using the two components of importance values (relative biomass and relative density) on each axis (Figure 3-5). A ll sawgr0 10 20 30 40 50 60 70 80 90 100 0102030405060708090100 Relative De n 2 3 4 5 9 11 1 3 1 4 1 5 1 9 Figure 3-5. A scatterplot of sawgrass co mmunities sampled in November 2002. Axes correspond to percent relative biomass a nd percent relative density. Each point represents one sawgrass unit. E ach sawgrass type resulting from the cluster analysis is coded in the legend. The points are color coded to match the gr oups that they were assigned to via the cluster analysis discussed in the followi ng section. If the importance values were distinguishing the differences between large/sparse and small/dense stands, then the points of the same group would be cluste red together on the bi-plot. While the assumption does not hold up perfectly, there is a definite clustering effect. Obviously, the importance values are not distinguishing the difference between the large/sparse and small/dense stands due to the previously stated disadvantages involving importance values, but rather certain associative species ma y be more prevalent in one or the other. I conclude that it is safe to assume, for the purpose of this study, that the disadvantage of

PAGE 41

27 using importance values in this study is not relevant due to mitigating variables such as species associates. Indicator Species Analysis Following the generation of cluster dendrogr ams, an indicator species analysis provides a subjective determin ation of the optimal number of groups based on how well any of the species acts as a significant i ndicator of a group. Dufrne and Legendres (1997) method of calculating species indicator values combines information on the concentration of species importance values and the faithfulness, or endemism, of a species to a particular group. I ndicator values are tested for statistical sign ificance using 1000 Monte Carlo randomizations. Each sample event was subjected to an indicator species analysis on the PC-Ord software 29 times, testing statistical signifi cance of every species from 30 groups to 2 groups. The program provided a table for each sp ecies and a p-value, or the proportion of randomized trials with an indicator value equal to or exceeding the observed indicator value. Average p-values for each run and number of significant species (p<0.05) were plotted in a spreadsheet (see Appendix A). Bo th plots were used to determine the optimal number of groups to prune the cluster de ndrogram. Low average p-values across the suite of species, and high numbers of si gnificant species determined the number of groups. Thirteen groups of community units developed from the November 2002, June 2003, and June 2004 sampling data. Fourteen groups of community units developed from the November 2003 sampling data. The indicator species analysis also produ ces a table of indicator values, or the percent of perfect indication based on combin ing the values for relative magnitude and

PAGE 42

28 relative frequency of importance values, for each species. These tables were translated into graphical signatures for each community state (See Appendix B). The resulting structural signature was an important guide to describing the groups, or meta-stable community states. Species with high indicato r values (>15%) were si gnificant species of that community. The resulting community stat es and their descriptions are shown on the cluster dendrograms in Fi gures 3-1, 3-2, 3-3, and 3-4. Matching Similar Community Descriptions Between Sampling Events The groups resulting from the cluster analyses were translated into meta-stable state entities, as defined in Chapter 1. In this st udy, the meta-stable states are groups of similar clusters that were consolidated across sample events and in some cases within events. As a result, the community states identified repr esent the range of states that occurred throughout the landscape thr ough different seasonal and annual environmental conditions. Some of these states were persistent th rough the study period, and some occur infrequently. The following is a descri ption of the methods used to group clusters together across and within sample events, and define them as meta-stable community states. A general description of each group arising from the cluster and indicator species analyses was constructed for each sampling ev ent. In order to facilitate matching community states between events, a multi-re sponse permutation procedure (MRPP) was utilized. MRPP is a nonparametric procedur e for testing the hypothesis of no difference between entities (Biondini et al. 1985). Each community unit was tested for heterogeneity over time. In other words, if a community unit was similar over the four sampling events, or minimal change had occurred over the course of monitoring, then

PAGE 43

29 that unit would receive a low p-value. Units th at exhibit little cha nge can be associated with the same meta-stable community state de scribed in the indicat or species analysis. The MRPP procedure helped with establis hing only four of the community state descriptions and was insufficient in finding similar states between sample events. Another approach, used as a complementary an alysis to the MRPP, was an agglomerative clustering process applied to all of the groups resulting from the cluster analyses of the four sample events. Similar community stat es should cluster together. Most of the resulting clusters of groups included only one group from each sample event, corroborating that those states are unique within sample even ts and indicating that they are common throughout the study period. Th ese were designated as the community states of the major physiognomic types. Tw o clusters included groups from the same sample event that were similar: the hea vy sawgrass group consists of two combined clusters in June 2003 and November 2003, a nd the E. elongata slough consists of two combined clusters in November 2002. Other community groups may have been represented only two or three times over the four sampling events. This method of defining groups over multiple sampling events provided an objective approach to comparing group structural signatures, and was a potential check on the first set of cluster analyses. If the differences of groups within sample events are less than the differences among groups then those clusters can be essent ially combined. See Table 3-1 for a list of meta-stable community states and the freque ncy of community units in each event. Discussion of these results continues in chapter 5.

PAGE 44

30 Table 3-1. Meta-stable community states a nd their frequency at each sample event. Nov. 2002 June 2003 Nov. 2003 June 2004 Eleocharis Prairie 7 4 4 8 Panicum/Paspalidium/Eleocharis Prairie 9 12 5 11 Eleocharis elongata Prairie 6 13 23 13 Deep Slough 9 16 7 14 Sawgrass Prairie 0 5 11 0 Panicum Slough 4 0 0 5 Eleocharis elongata Slough 31 22 15 0 Shallow Slough 0 0 6 14 Sawgrass with Eleocharis/Panicum 8 0 0 7 Sawgrass with Bacopa/Ludwigia 11 5 9 11 Sawgrass with Eleocharis elongata/Crinum 8 10 6 5 Sawgrass monoculture 19 25 27 17 Sawgrass Ghost Island 13 8 1 1 Ghost Island 2 4 4 18 Tree Island 0 7 9 3 Distribution of Meta-Stable States Across the Landscape In an impounded hydroscape like Wate r Conservation Area 3A, hydrologic conditions across the landscape can differ from on e end of the drainage basin to the other. Environmental gradients in WCA 3A such as substrate type and peat thickness in conjunction with varying hydroperiods should distribute plant communities throughout the landscape accordingly. Therefore, some insight as to the hydrology of the metastable states that were identified in the cl uster analysis may be gained by mapping their locations. Figure 3-6 splits the study area into four qua drants and identifies the proportion of each meta-stable state by physiognomic type in each quadrant. Each quadrant roughly represents the high and low ends of the hydrologic and substrate depth gradients (i.e. the northwest quadrant is short hydroperiod/shallow peats; southwest quadrant is long hydroperiod/shallow peats; nor theast quadrant is sh ort hydroperiod/deep peats; southeast quadrant is long hydroperiod/d eep peats). The rationale for dividing the

PAGE 45

31 Figure 3-6. Shows the distri bution of meta-stable states by physiognomic type into four quadrants of the study landscape.

PAGE 46

32 landscape into square quadra nts is simply for conveni ence since the extent of impoundment effects on the hydroscape have not been determined and a detailed survey of peat depths in the conser vation have yet to be produced. Examination of the distribution of meta -stable states shows several insightful trends. Sawgrass communities showed little di fference in change across the landscape. There was a slight increase in the share of Panicum and Bacopa associated sawgrass communities towards the western extents of the impoundment area Among prairie communities the Eleocharis cellulosa state was confined strictly to the west whereas the Eleocharis elongata states incr eased dramatically further ea st, which resembles the peat depth and substrate type gradients. This s uggests that the state th at a wet prairie might exhibit has a lot to do with substrate prope rties. Not surprisingly, deep sloughs dominated by water lilies were more prev alent in the longer h ydroperiod south. Tree island types of islands dominated by woody vegetation and shade tolerant herbaceous species were virtually none xistent in the southeast.

PAGE 47

33 CHAPTER 4 MULTIVARIATE ANALYSIS AND RESULTS Hydrology Selecting Hydrologic Variables The multivariate approach to statistical analys is of data allows the researcher to test simultaneously for significance among variables. There are many factors that may have a hand in the makeup of community assemblages, but hydrology is the main environmental force driving Everglades community structur e. In order to include hydrology in the statistical analysis of data, it was necessary to first determine which metrics can best represent the hydrology of the Everglades. A set of hydrological variables was selected a priori and calculated for each sample unit. Table 4-1 lists the hydrologic variables selected along with the ecological rationale underlying their use. Hydrology can be described to reflect one or several different as pects of flooding: depth, duration, time, and the magnitude of extreme events of flooding and drought (Richter et al. 1997). The frac tion of the year that a given site is inundated was chosen due to the prominence in the literature, especially regarding the Everglades and ease of calculation (Toner and Keddy 1997). The mean de pth of flooding, whether it be above or below the ground surface was selected to repres ent the depth of water that species must adapt to. Range of depths defines the amount of water level fluc tuation that occurs during a year. Flooding and dr ought event legacies are a f unction of the length of time used for hydrological records to calculate hydrology. Therefore, inundation times were calculated using one, three, five, and ten-year time records. If there is a consistent period

PAGE 48

34 Table 4-1. Hydrological vari ables with abbreviations. Abbreviation Variable Units Ecological implications (T)yr.inun* Number of days per year during which flooding occurred (inundated) days Reproduction of some species. Exposure of soils to oxidation processes. MeanDepth Mean depth of flooding over a ten year period feet Establishment of aquatic vs. emergent species. max Average 7-day maximum water depths over 10 years feet Anaerobic stress in plants. Spatial extent of extreme conditions. datemax Average Julien date of maximum water levels day of year Coordination of hydrologic factors with temperature and photic factors. highdurat Duration of high water levels days Anaerobic stress in emergent wetland species. min Average 7-day minimum water depths over 10 years feet Indication of pot ential oxidation of soils. Reproduction opportunities. datemin Average Julien date of minimum water levels day of year Coordination of hydrologic factors with temperature and photic factors. lowdurat Duration of low water levels days Opportunities for emergents to develop and compete against floating leaf aquatics. Exposure of soils to oxidation processes. variation Average annual range of depth feet Amount of va riation in the environment that must be tolerated. (T) denotes the length of record used to ca lculate the metric. Periods of time used to measure inundation time are 1, 3, 5, and 10 y ears. All of the other metrics are calculated using a 10-year time span. of time that hydrology affects into the futu re then it would be revealed in the classification tree analysis. Duration of t ypical high and low water level events were calculated using the algorithms in the IHA (I ndicators of Hydrologic Analysis) software (Richter et al 1996). Finally, extremes during the average year were calculated as sevenday highs and lows. Timing of these extrem es was considered. These metrics were chosen due to literature citing the importanc e of extreme, stochastic events in the

PAGE 49

35 Everglades that occur periodi cally and play an important role in community dynamics by offering opportunities for species recruitment, movement, and nutrient availability that otherwise are unavailable (K itchens et al. 2002). Calculating Hydrologic Variables The hydrologic data collected was processed extensively in order to be applied to the multivariate analysis as the metrics th at relevantly described the hydrology of the sites for the community sample units. The monitoring wells were set up shortly following the first sample event. Water data had to be extrapolated up to ten years prior to November 2002. The well data also needed to be applied to the various sample units that it was monitoring. The community units re quired classification of their elevations in order to get depths that were relative to the well monitoring that plot. The following is a discussion of the methodology used to calculate those hydrologic profiles. Hindcasting using neural networks To calculate the hydrologic variables me ntioned previously, precise water data dating back ten years was needed for the study plots. Prior to the study there were three permanent gauging stations, established by vari ous state and federal agencies, within the study area (See Figure 4-1). Two of these three stations, 3-64 and 3-65, had been established and collecting da ta longer than ten years prior to 2002. The agency monitoring stations and their data was not su fficient for producing hydropatterns at each of the 20 study plots. A netw ork of monitoring stations need ed to be established that could provide accurate hydrologic data at the community scale. Although the vast majority of WCA 3A is flooded most of the time, a flat pool of water cannot be assumed over such an expans ive landscape. Since the plot size was a

PAGE 50

36 Figure 4-1. Green triangles represent the mon itoring stations set up by various agencies. These stations upload realtime data to the web daily. Yellow circles indicate the temporary stations that we re established in December 2002.

PAGE 51

37 kilometer squared, a flat pool was assumed w ithin a kilometer radius. Temporary data loggers, designed to monitor water depth were installed in December 2002 at each plot with a couple of exceptions. Plot 7 and plot 4 each were within one kilometer of an agency monitoring station. Plots 13 and 15 shar ed a station, as well as plots 10 and 11. The wells are driven through the peat subs trate to the limestone. The peat soils usually provide enough stabilization to prevent the wells from leaning even in tropical storm force winds. In the few areas where th e substrate is insufficiently thick, the wells are stabilized by makeshift tripods. The data loggers are attached to wells that measure surface water depths from its base. The dept h from the substrate is simply calculated by subtracting the amount of the we ll that is buried in the peat The data loggers measure the water depth at their respect ive stations twice a day. Ev ery month, data is downloaded from the data logger to a laptop. Neural networks are a pattern recognition statistical a pplication that search for patterns over time with the use of multiple model runs. They are especially useful in situations that have static as well as dynamic properties (Bishop 1995). The landscape position remains static between the monitoring stations, yet th e data are dependent with time. After a significant amount of data wa s collected from the ne w monitoring stations, they were joined with the agency stations in a neural network to produce a constructedtopographical model of the water surface. Using neural networks, models were applied to hindcast, or extrapolate, the depth of each of the new stations to produce hydrologic data for all of the study plots. Table 4-2 contains the results of the neural network analysis.

PAGE 52

38 Table 4-2. Neural network model statistics for each station hindcasted. PME (percent model error) = RMSE (root mean-square error) / (range of measured data). Plot n R R2 Mean Error (ft) RMSE (ft) PME (%) 0 694 0.995 0.990 0.028 0.072 2.6% 1 352 0.999 0.997 0.001 0.041 1.5% 2 594 0.992 0.985 0.026 0.067 3.0% 3 301 0.996 0.992 0.009 0.066 2.7% 5 563 0.999 0.997 0.005 0.044 1.5% 6 682 0.994 0.988 0.034 0.075 3.0% 7 222 0.997 0.994 0.004 0.038 1.8% 8 690 0.997 0.995 0.011 0.049 1.8% 9 567 0.993 0.986 0.007 0.082 3.1% 11 658 0.996 0.992 0.026 0.070 2.3% 12 603 0.998 0.996 0.025 0.055 1.6% 14 674 0.998 0.996 0.052 0.072 2.2% 15 659 0.997 0.994 0.013 0.064 1.9% 16 377 0.997 0.994 0.004 0.054 1.8% 17 392 0.991 0.982 0.001 0.089 2.8% 18 613 0.988 0.976 0.033 0.111 3.8% 19 426 0.994 0.988 0.020 0.080 2.7% Extrapolating from well data to sample unit data The continuous well data that dated back at least 10 years, produced by the neural network models, corresponded to a point within each plot. Data for each community sample unit on transects within the plots n eeded to be calculated. During each sample event, water depths were taken at each vege tation sample. For every community sample unit, those depths were averaged to get the average depth of that community unit for that date. From that date, the wate r depth at that site was extr apolated from its corresponding well for ten years into the pa st by subtracting the differen ce between the well depth and the average community unit depth and applying it as a constant differential. As a result, a 10-year historical hydrologic record for each community sample unit was created. It was assumed that the water within th e square kilometer area of a plot is a flat pool in order to

PAGE 53

39 make this extrapolation. The hydrologic vari ables mentioned previous ly were calculated using the resulting records. Nonmetric Multidimensional Scaling After the hydrologic variables were calcula ted for each community unit, the matrix of potential environmental drivers was comp lete. Table 4-3 lists the environmental factors used in the multivariate analysis of the community data. Some of the variables were relativized to vary within two orders of magnitude (0-10). Transformations such as this are imperative in producing rational re sults in multivariate community analysis (McCune and Grace 2002). Table 4-3. Environmental variables used in the multivariate analyses and how they were relativized if a transformation was appropriate. Environmental variable Abbreviation Range Relativization (if any) Peat Depth PeatDepth 0.07-5.74 feet no relativization Mean Depth MeanDepth 0.09-3.49 feet no relativization Minimum Water Depth min 0.97-2.38 feet no relativization Maximum Water Depth max 0.824.29 feet no relativization Timing of Minimum Water Depth datemin day 147-153 Range from 0 (January 1) to 1 (December 31) Timing of Maximum Water Depth datemax day 278-311 Range from 0 (January 1) to 1 (December 31) Duration of High Water Levels highdurat 3.25-89 days Proportion of year (range from 0-1) Duration of Low Water Levels lowdurat 2.25-115.8 days Proportion of year (range from 0-1) Average Range of Water Depths in a year variation 0.70-3.11 feet no relativization Inundation times (T)yr.inun 51%-100% of year Proportion of year (range from 0-1) Ordination of community data organizes the structural composition of the sample units into a multidimensional space using nonpara metric scaling techniques. The output of sample units as points in space is the re sult of multiple runs using various numbers of dimensions to find the best fit of the data on a hypothetical landscape that minimizes the

PAGE 54

40 stress of a solution. Stress is a measur e of departure from monotonicity in the relationship between the distance in the community data and th e distance in the resulting multidimensional space. This particular method of ordination is well suited to nonnormal, arbitrary scales that are comm onplace in community ecology due to its distance-preserving properties (Clarke and Ainsworth 1993). A Sorenson distance measure was used fo r the ordinations because community analyses in ecology require a metric, city -block distance calculation to handle the intricacies and occasionally l ong distances that can occur be tween species. The program PC-ORD provided the algorithms for the NM S procedures. See Mather (1976) and Kruskal (1964) for the methods. The program supplied a random star ting configuration. Each analysis was run 15 times with the real data and a Monte Carlo test was performed 30 times for comparison. Appendix C provides the results of the Monte Carlo tests and the probabilities that a similar final stress could have been obtained by chance. Scree plots in Appendix C show the stress reduced per dimension added and provide a visual check on the stability criterion (.0001 st. de v. in stress over last 10 iterations). The ordination procedure was first performe d using all of the community data and grouped by the physiognomic groups that resulted from the cluster analyses. The wholescale community ordination grouped the phys iognomic types distinctly in a twodimensional space. Figure 4-2 shows the ordination plot and the vectors that represent significant environmental gradients (r2 > 0.2) that correspond to (in this case) axis 2. Minimum depth and mean depth were the e nvironmental variable s that explain the distinction between sample units and comm unities in Water Conservation Area 3A.

PAGE 55

41 Figure 4-2. A whole scale ordi nation plot of the community sample units. Triangles represent individual sample units and crosses repr esent species. The key to the legend: 1prairie physiognomic type; 2slough physiognomic type; 3 sawgrass physiognomic type; 4island physiognomic type. The environmental gradients (minimum dept h and mean depth) are represented with red vectors closely aligned with axis 2. The ordination procedure was repeated using each of the physiognomic types separately to attempt to discover the rela tionship between states of each physiognomic type and the environment. Figure 4-3 is an ordination plot of island-type communities. The three meta-stable states of island-types ar e best fit into a three-dimensional space. Shown are ordination plots of ax is 1 and axis 2 and a plot of axes 2 and 3, providing a less than clear distinction between states of island-types. There is no correlation of environmental variables to the ordination axes.

PAGE 56

42 Figure 4-3. Island-type ordinati on plots. Triangles represent individual sample units and crosses represent species. The key to the legend: 1sawgrass ghost island; 2ghost island; 3tree island. Figure 4-4 shows the ordination plot of slough communities grouped by their respective meta-stable states. The procedure resulted in a fit within two dimensions, but did not recognize an overall significant environmental va riable that explains the distinction between states No significant environmental gradients were found to explain the position of community units in ordination space. An NMS ordination plot shows the four states of sawgrass-type communities in multivariate space (Figure 4-5). A three-dimensional plot was determined as the most stress-reducing solution. The gr oups are clearly distinct in a plot of axes 2 and 3, though no environmental variable met the criteria to be established as a si gnificant gradient in ordination space. The wet prairie physiognomic type is plotte d in ordination space in Figure 4-6. The peat depth gradient, which is strongly correlate d with longitude, is a ligned with axis two

PAGE 57

43 Figure 4-4. Slough-type ordinati on plots. Triangles represent individual sample units and crosses represent species. The key to the legend: 4deep slough; 6 Panicum slough; 7 E. elongata slough; 8shallow slough. Figure 4-5. Sawgrass-type ordina tion plots. Triangles repres ent individual sample units and crosses represent species. The key to the legend: 9sawgrass with Eleocharis sp./ Panicum ; 10sawgrass with Bacopa/Ludwigia ; 11sawgrass with E. elongata/Crinum ; 12sawgrass monoculture.

PAGE 58

44 Figure 4-6. Prairie-type ordination plots. Triangles repres ent individual sample units and crosses represent species. The key to the legend: 1 Eleocharis sp. prairie; 2 Panicum/Paspalidium/Eleocharis sp. prairie; 3 E. elongata prairie; 5 sawgrass prairie. Environmental gradients shown as red vectors closely aligned with axis 2. (r2=0.386 along axis two). Eleocharis sp. prairies and sawgra ss prairies are clearly situated in shallow peat subs trates. In the middle of th e plot along axis two is the Panicum/Eleocharis/Rhynchospora prairie located intermediate ly with respect to peat depths. Eleocharis elongata prairies are characterized by de ep peat depths and located at the top of the plot along axis two. It can be concluded th at the species of Eleocharis are distributed along the peat depth gradient, de termining what state of wet prairie will manifest when the conditions prevail for a wet prairie. Inspection of the ordination results can give insight into the strength or weakness of the distinction between meta-sta ble states determined by the cluster analyses. The results also provide indications of pr evailing driving forces determining the community structure of such states.

PAGE 59

45 Classification Trees and Characterization of Meta-Stable States by Environmental Variables The use of classification tr ees can determine the envi ronmental variables that distinguish different groups ba sed on community structure. In the case of this study classification trees define the groups based on the designation they were assigned by the cluster analyses. The environmental variable s are analyzed for each group and branches are created in the classification tree where appropriate to predict what community state will result under significant environmental conditions. The result is a dichotomous key classifying states by the variab les that distinguish the groups quantitatively (Urban 2002). More specifically, environmenta l thresholds can be determined for each meta-stable state at the leaf of the classificati on tree. For the example of th is study the dependent variable is the community state, wher eas the predictor variables ar e the environmental variables listed in Table 4-3. A complete interpre tation of the classifi cation trees for each community state is included as separa te sections. The importance of various environmental variables is graphed in Appe ndix D and indicates th e significance of a specific variable in indicating a meta-stable st ate. A discussion of the classification trees and their implications is included in Chapter 5. Figure 4-7 is a classification tree of all of the identified meta-stable states. A combination of environmental variables dis tinguishes each community state. This particular classification tree is meant to be only an overview and should be a general guide to understanding the relationshi p between physiognomic types and their community states. The explanatory power of this tree is minimal (variation explained = 23%) due to the number of states it was meant to classify. Classification by

PAGE 60

46 physiognomic type is a more useful method of determining the factors that drive community state structure. Figure 4-7. Classification tree for the meta-sta ble states on 10 environmental variables. The number of sample units in each leaf are shown in parentheses below each bar graph, which shows the compositions of communities within each leaf. Slough Physiognomic Type Figure 4-8 is the classification tree fo r the slough physiognomic type. The first node in the classification tree is the timing of the maximum water depth during the year (before or after October 21st). Variation of mean depth and the depth of maximum water levels are the second level of branching in the tree. Finally, ot her hydrologic factors work in concert to determine slough meta-stabl e states. The misclassification rate of the models was 36% and the amount of variati on explained was 54% (1-Relative Error).

PAGE 61

47 Figure 4-8. Classification tree for 4 sl ough community states on 10 environmental variables. This model was pruned from a tree size of 7 leaves to five, based on a cost complexity pruning curve, se lecting the smallest tree within one standard error of the best. The number of sample units in each leaf is shown in parentheses below each bar graph, which shows the compositions of communities within each leaf. Deep sloughs Deep sloughs are typically lo cated in the southern exte nt of the study area, but found throughout the basin. A ssociate species include Nymphoides aquatica Nymphaea odorata and Utricularia sp. According to the classifi cation tree analysis deep sloughs occur in areas where the timing of maximu m water depths is earlier than the 293rd Julien day of the year (October 21st) and maximum depths are at least 2.91 feet. The misclassification rate of d eep sloughs is very low.

PAGE 62

48 Eleocharis elongata sloughs This particular slough community stat e occurs everywhe re except in the southeastern portion of the water conservation area, which is typified by deep peats, long hydroperiods, and floating mats. Associat e species of this community include Eleocharis geniculata, Nymphaea odorata, Utricularia, and Hymenocallis According to the classification tree they occur wh ere variation in water depths is less than 2.29 feet during the year. Panicum sloughs Panicum hemitomon sloughs are mostly lim ited to the northwestern reaches of the study area. Associate species of this community include Panicum hemitomon, Paspalidium geminatum and some of the typical slough species ( N. odorata, N. aquatica, and Utricularia ). According to the classification tree this community state occurs in relatively shallow site, requiring maximum wa ter depths of less than 2.91 feet and minimum water depths of less than 0.735 feet. Shallow sloughs Shallow sloughs are evenly distributed across the landscape. They consist of mainly Utricularia but associates may include P. hemitomon, P. geminatum, N. odorata, and N. aquatica According to the classification tree they prefer maximum depths late in the year and occur in sites with greater va riation in water depths throughout the year. Wet Prairie Physiognomic Type Figure 4-8 is the classificati on tree for the wet prairie physiognomic type. The first node and most important explanatory variable distinguishing conditions in wet prairie communities is peat depth. Duration of high water levels, hydroperiod, timing of maximum water levels, duration of low wate r levels, hydroperiod and annual average of

PAGE 63

49 minimum water levels are the hydrologic factor s that determine community state of wet prairies. The misclassification rate of the models was 28% and the amount of variation explained was 54% (1-Relative Error). Figure 4-9. Classification tr ee for 4 wet prairie community states on 10 environmental variables. This model was pruned from a tree size of 11 leaves to eight, based on a cost complexity pruning curve, se lecting the smallest tree within one standard error of the best. Eleocharis sp. prairie This particular state of wet prairie is confined to the we stern half of the study area, concentrated mostly in the southwest. The dominant species is Eleocharis cellulosa with some Eleocharis equisifoides occurring on occasion. Accordi ng to the classification tree, this community state flourishes in areas that have short durations of low water levels, and

PAGE 64

50 low peat depths. Other criteria include mi nimum water levels great er than 0.75 feet and long hydroperiods. E. elongata prairie This community state occurs throughout th e study area, however it tends to be concentrated in the eastern reaches and is th e only type of prairie in the northeastern quadrant. This prairie state is dominated by Eleocharis elongata but can also include associate species such as Bacopa caroliniana, Hymenocallis, N. odorata, P. hemitomon, P. geminatum, and Utricularia According to the classification tree, E. elongata prairies occur in areas with a long duration of low water levels and intermediate peat depths. Panicum/Paspalidium/Eleocharis prairies These prairies occur mainly in the wester n reaches of the study area. Dominant species include P. hemitomon, P. geminatum, and E. cellulosa Associate species include B. caroliniana, N. odorata, N. aquatica, Hymenocallis, and Utricularia According to the classification tree these prairi es tend toward peat depths less than 2.67 feet, shorter hydroperiod sites than sawgrass prairies, and lower minimum water levels than E. cellulosa prairies. Sawgrass prairies Sawgrass prairies occur almost exclusivel y in the southeast portion of the study area. They are similar in structure to Eleocharis sp. prairies except for the inclusion of sawgrass. According to the classification tree, sawgrass prairies occur in shorter hydroperiod sites than the other prairie comm unities and in peat depths less than 1.21 feet.

PAGE 65

51 Sawgrass Physiognomic Types Figure 4-10 is the classification tree for th e sawgrass physiognomic type. The first node in the classification tree is peat dept h. Ten-year hydroperiod, timing of extremes and maximum water levels are also cons idered in the classification scheme. Misclassification rates for this community stat e were low. The misclassification rate of the models was 28% and the amount of varia tion explained was 54% (1-Relative Error). Figure 4-10. A classification tree for 4 sawgrass community states on 10 environmental variables. This model was pruned from a tree size of 14 leaves to eight, based on a cost complexity pruning curve, se lecting the smallest tree within one standard error of the best. Sawgrass monoculture (heavy sawgrass) This sawgrass community state is evenly distributed throughout WCA 3A. It is dominated by Cladium jamaicense but associate species include Crinum americana and

PAGE 66

52 Ludwigia spp. According to the classification tree, heavy sawgrass communities are well distributed across all conditions. Heavy sa wgrass is generally found in deeper peat. Sawgrass with Bacopa and Ludwigia This state of sawgrass occurs throughout th e study area, but is concentrated in the southwest reaches. Typical species include sawgrass and B. caroliniana Associate species include Ludwigia spp. and C. Americana According to the classification tree, this community state occurs in peat depths ranging from 1.2 ft to 2.6 ft. Maximum water levels should remain under 2.9 ft. Sawgrass with Eleocharis sp. and Panicum The sawgrass with Eleocharis and Panicum state occurs throughout the study area except for the southeastern quadrant. This community state is dominated by sawgrass and Eleocharis sp. Associate species include P. hemitomon, P. geminatum, Typha sp. and Leersia hexandra According to the classification tree, this state occurs in peat depths of less than 1.2 ft and maximu m water levels that remain under 2.9 ft. Sawgrass with E. elongata and Crinum This community occurs throughout the st udy area, yet is rarely found in the southeastern quadrant. A di verse community state, dominant species include sawgrass, E. elongata and C. americana Associate species include B. caroliniana, Cephalanthus occidentalis, Hymenocallis, L. hexandra, Peltandra virg inica, Pontederia cordata, Ludwigia sp., and E. equistifoides According to the classifi cation tree, this community state occurs in peat depths that are between 2.6 ft and 4.2 ft, and relatively long hydroperiods compared to heavy sawgrass. Ti ming of extreme water levels should occur later in the year.

PAGE 67

53 Island Physiognomic Types Figure 4-11 is the classification tree for the island physiognomic type. The term island is being used loosely, and typically is associated with woody vegetation and can occur as fringe marshes adjacent to true tree is lands. Islands can also exist as strands of sawgrass that include woody vegetation and typica l tree island associates. The first node in the classification tree is duration of hi gh water levels, followed by a second branch representing duration of low water levels. Classification of the island-type communities broke out simply and evenly with only minor instances of mi sclassification. The misclassification rate of the models was 54% and the amount of va riation explained was 63% (1-Relative Error). Figure 4-11. Classification tree for 3 island -type community states on 10 environmental variables. This model was pruned from a tree size of 10 leaves to eight, based on a cost complexity pruning curve, se lecting the smallest tree within one standard error of the best.

PAGE 68

54 Ghost islands The ghost island community state occurs throughout the study area. Dominant species include P. virginica, P. cordata, and C. occidentalis Associate species include Crinum, Ludwigia, Sagittaria lancifolia and sawgrass. According to the classification tree, ghost island communities re quire durations of high water levels for over 20% of the year. Sawgrass ghost islands This community state occurs throughout the study area except in the northwest quadrant. Sawgrass ghost islands are dominated by sawgrass, L. hexandra, P. virginica, S. lancifolia, and C. occidentalis Associate species include P. cordata, Blechnum serrulatum and Ludwigia sp. According to the classi fication tree, these communities prefer durations of high water less than 20% of the year and durations of low water greater than 10% of the year. Tree islands This community state occurs throughout th e study area except in the southeast. Due to the low number of samples however this community probably occurs in all quadrants of the study area. The dominant species are the ferns Osmunda regalis and B. serrulatum Typical associates are Ludwigia, Typha, S. lancifolia, P. cordata, C. occidentalis and sawgrass. According to the classification tree, tree island communities occur in areas where duration of high water levels is less than 20% of the year and duration of low water levels is less than 10% of the year.

PAGE 69

55 CHAPTER 5 SUMMARY AND CONCLUSIONS Discussion This thesis examines community-scale vegetation dynamics across the landscapelevel processes and drivers of Water Conservation Area 3A, of the Everglades. Hydrologic processes vary seas onally and annually, but leave le gacies that last years or decades. The other processes that were examined operate on the temporal scale of decades, such as soil accretion and erosion. Some of the processes examined are anthropogenically influenced as well as dr iven by weather patterns. Vegetative communities respond differentially to thes e drivers, and comprehension of how communities respond is essential to mana gement and restoration efforts. Decompartmentalization and reorganization of the hydroscape for restoration purposes illustrates the need to grasp the intricacies of community dynamics in the Everglades. Only through the understanding of how the ma jor drivers of the landscape operate, can decision-makers rationalize a truly bottom-up a pproach to management of such a unique and complex system of habita t and wildlife. Models based on a few key environmental variables can be valuable tools in conser vation management of dynamic wetlands (Toner and Keddy 1997). This study provides the next step in the identification of those variables and how they infl uence community structure. Identification of meta-stable states of communities is essential for examination of whole-scale community dynamics. Proce sses may have only subtle effects on community structure over a shor t period of time. These subtleties can provide clues as to

PAGE 70

56 the drivers that are causing pot ential shifts in landscap e-level configurations of vegetation. Chapter 3 identifie s the meta-stable states of marsh physiognomic types of WCA 3A, using a hierarchical clustering t echnique to analyze the community data monitored semiannually over two years. Append ix B provides the structural signature of each of these states. Four slough states we re identified: deep slough, shallow slough, Panicum slough, and Eleocharis elongata slough. Prairie communities manifest as four different states: Eleocharis sp. prairie, Panicum/Paspalidium/Eleocharis prairie, E. elongata prairie, and sawgrass prairie. Four states of the sawgrass physiognomic type were identified: sawgrass with Eleocharis/Panicum sawgrass with Bacopa/Ludwigia sawgrass with E. elongata/Crinum and heavy sawgrass. Finally, shrubby marsh communities, noted in this thesis as island-t ypes, manifest as three different states: sawgrass ghost island, ghost island, and tree island. The majority of analyses performed on the data collected were multivariate in approach due to the complex interactions and combinations of environmental variables working in concert to create a unique, dense matrix of conditions in the study area. Community ecology is best studied in a multivariate framework because of the complexities of ecological interactions. Univariate comparisons, however, can be useful in discovering basic trends in distributions of communities across a single gradient. Figure 5-1 plots the mean water depths of th e meta-stable states identified by the cluster analyses. The chart illustra tes both how slight the diffe rentiation, as well as the distinctive differences of hydrologic f actors are between physiognomic types and community states are. Deep slough and E. elongata slough states are th e deepest of the communities, though they also persist in a re latively wide range of hydroperiods. The E.

PAGE 71

57 elongata prairie is the next deepest followed by the Panicum prairie, E. cellulosa prairie, sawgrass prairie, and shallow slough. Th e sawgrass communities and sawgrass ghost island are the next shallowe st communities, though the h eavy sawgrass community is evenly distributed along a relatively wide hydr ologic range. Finall y, the ghost island and tree island communities have the shortest hydr operiods of the community states. The Panicum slough community did not clus ter around a definable range, probably due to the low number of sample units. 0 0.5 1 1.5 2 2.5 3 3.5 4Mean Water Depth (ft) Eleocharis prairie PAN/PDG/Elsp prairie ELG prairie Sawgrass prairie Deep Slough Panicum slough ELG slough Shallow slough SG with Elsp/PAN SG with BAC/LUD SG with ELG/CRA Heavy sawgrass SG ghost island Ghost island Tree island Figure 5-1. Distribution of sample units of each community state along a hydrologic variable (mean annual water depth). Communities are grouped by physiognomic type: squares=prairies, triangles=sloughs, circles=sawgrass, crosses=islands. Peat depths were plotted for each samp le unit grouped by community state on the x-axis in Figure 5-2. The soil gradient is distributed among the physiognomic types more than the hydrologic gradient. Peat depths are high among the island-types, the heavy sawgrass community, sawgrass with E. elongata/Crinum sawgrass with Bacopa/Ludwigia shallow slough, E. elongata slough, deep slough, and E. elongata prairies. Shallow peat depths ar e characteristics of sawgrass with E. cellulosa/Panicum

PAGE 72

58 sawgrass prairies, Panicum prairies, and E. cellulosa prairies. Again, the Panicum slough could not be located on the soil gradient becau se of the small number of samples. 0 1 2 3 4 5 6 7Peat Depth (ft) Eleocharis prairie PAN/PDG/Elsp prairie ELG prairie Sawgrass prairie Deep Slough Panicum slough ELG slough Shallow slough SG with Elsp/PAN SG with BAC/LUD SG with ELG/CRA Heavy sawgrass SG ghost island Ghost island Tree island Figure 5-2. Distribution of sample units of each community state along a peat depth gradient. Communities are grouped by physiognomic type: squares=prairies, triangles=sloughs, circles=sa wgrass, crosses=islands. The amount of overlap between comm unity states along the individual environmental variable illustrates why multivar iate statistics are the preferred method of analysis in community ecology. Combinations of environmental conditions create unique conditions that are favorable for specific commun ities to develop. It is interesting to note that communities that are not differentia ted by soil characteristics, typically are differentiated by hydrology. For this reason, it can be concluded that these are the two major driving factors determining community structure. Table 3-1 shows the frequency of the comm unity states for each of the sampling events. Many of the community states we re present in steady numbers throughout the study, such as the heavy sawgrass community. However, there were some states that occur in large numbers at one event, only to virtually disappear in other sample events. The Panicum/Paspalidium/Eleocharis prairie fluctuates with respect to the wet and dry

PAGE 73

59 season variation. Communities such as the Eleocharis elongata slough are common at the beginning of the study, only to drop out completely by the end of the study. One explanation for this phenomenon could be th e differences between the water years of 2003 and 2004. Figure 5-3 is a time series gr aph of water stage during the time of sampling. The Everglades was subjected to a moderate fluctuation of water levels from November 2002 to June 2003. From that tim e forward extreme levels were reached during the wet and dry seasons, potentially infl uencing the state of physiognomic types in the study area. It can be speculated that these hydrologic extremes discouraged the persistence of E. elongata cau sing a shift to the shallow slough state, a less diverse community. Although actual shifts in stat e were not documented, the fluctuation in frequencies of community stat es can be interpreted as sh ifting meta-stable states. Figure 5-3. A time-series graph of water stag e at a monitoring station within Plot 4. Note the extreme highs and lows of th e second water year compared to the first water year. The distribution of community states acro ss the landscape provide s evidence of the environmental thresholds intrinsic to those states. Compartmentalization of the water

PAGE 74

60 conservation areas creates a rela tively flat hydroscape to the elevational gradient. The same physiognomic types are exposed to di fferential hydroperiods, depending on their position along a north-south axis. Soil types an d characteristics vary along a longitudinal axis, with shallow peat depths in the west and deep peat depths to the east. Figure 3-6 is a graphical representation of the distribution of community st ates across the landscape. It is apparent through examination of the commun ity distribution figure that species such as Panicum hemitomon and E. cellulosa are endemic to areas within the region that parallel environmental requirements for those species. P. hemitomon is almost exclusively located in the northern reaches of the study area where s horter hydroperiods persist. Prairies dominated by E. cellulosa are confined to the west ern reaches where shallow peats are typical. The scale of the study allo ws observation of the range of communities and environmental conditions within WCA 3A. An investigation of the environmental condi tions of each of these community states offers insight as to how they are related in terms of successional order and environmental thresholds. When variables that influence the state of a physiognomic type are identified, they can be quantified using classification tr ees. Thresholds of hydrologic variation and extremes are discovered and a picture of wher e meta-stable states li e on the continuum of a multidimensional environmental space begi ns to emerge. Relative positions of physiognomic types in this space are already co mmon knowledge to ecologists. In order of short to long hydroperiod communities, the sequence goes: island, sawgrass, prairie, slough. The evidence presented through this study shows some overlap in hydrology between these physiognomic types, t hough the basic relationships remain.

PAGE 75

61 The classification tree analyses in Chapte r 4 describe how the community states identified by the clustering techniques break out in terms of envi ronmental conditions. Slough communities were determined by hydrologic factors such as timing of maximum water depths and the magnitude of extreme water levels. Deep slough communities require early timing of maximum water levels Early timing of water levels suggests location further to the north wher e there is little time lag comp ared to the southern extent of the conservation area wh ere water tends to pile up later along Tamiami Trail. However, high maximum water levels are requ ired to drown out t ypical slough associates like Panicum hemitomon Panicum sloughs favor shorter hydrope riod sites, as revealed by the classification tree. E. elongata sloughs favor late timing and minimal variation of water depths, whereas shallow sloughs tolera te a wide range of water depths. The implications for this could mean that Eleocharis elongata as a species, may be less resilient and more sensitive to hydrologic disturbance. Sh allow sloughs tend to be less diverse, so in the event of ma ssive variation between seasons, E. elongata sloughs could convert into a shallow slough state. Peat depth was the major determinant for wet prairie meta-stable states. A close inspection of the classification tree reveals that E. elongata prairies tend to persist in deep peat depths and relatively long duration of lower water levels. Confined mostly toward the eastern half of the stu dy area, this prairie state o ccurs where soil conditions are conducive for its establishment. Eleocharis sp prairies, on the other hand, prefer shallow peats, which are typical of the western reache s of the study area. Sawgrass prairies have similar requirements as Eleocharis prairies, but tend towards shorter hydrope riod sites.

PAGE 76

62 Finally Panicum/Paspalidium/Eleocharis prairies lack a restric tive soil requirement, but prefer even shorter hydroperiod sites than sawgrass prairies. In the cluster analyses sawgrass units we re grouped mainly by diversity. Heavy sawgrass states were fairly common. The clas sification tree analysis reveals that heavy sawgrass is ubiquitous throughout the study area and occupies the range of environmental conditions that are characteristic of th e sawgrass physiognomic type. Sawgrass with Bacopa/Ludwigia occur in shorter hydroperiod sites. Also occupying short hydroperiod sites is the sawgrass with Eleocharis/Panicum state, yet it is confined to areas of shallow peats, as are most Eleocharis cellulosa associated communities. Sawgrass with E. elongata/Crinum communities prefer areas of long hydroperiods and deep peat depths. This community state is probably the sparse sawgrass communities of the southeastern WCA 3A. These tall sawgrass communities are exposed to continuous inundation and recruitment of sawgrass individuals is nonexist ent. The resilience of these communities is evident, as these conditions have prevailed for years. However, they are slowly being replaced by some prairie and primarily sl ough associates in the absence of drawdown events. Island-type communities were determined entirely by duration of high and low water events. The duration of annual extrem e water level conditions has implications on the balance of competitive and stress-toleran t organisms, as well as anaerobic stress in plants. Ghost islands characterized by arrowheads and woody shrubs prefer high duration of water levels. This may suggest they are or have been subject to soil subsidence or oxidation. L ong duration of low water levels is required for sawgrass ghost islands, which may be old tree islands wi th slightly higher water levels, or sawgrass

PAGE 77

63 ridges that have been colonized by woody shr ubs for one reason or another. The tree island community state is classified as a community that prefers short durations of extreme water levels in general. This may suggest that tree islands are not resilient to shifts in hydrologic regime and require mini mal annual variation of water levels. Comparing NMS and Classification Tree Techniques One interesting observation worth discussi on is the fact that the NMS ordination joint plots for each of the physiognomic t ypes did not reveal relationships to environmental variables (except for the prai ries). The ordination of the physiognomic types confirmed what we know about the re lationships between them. Hydrology is clearly the factor determin ing physiognomic type. When the NMS ordination did not reveal the same relationship between meta-s table states, it was somewhat surprising. However, the ordination plots did show distin ct patterns that distinguish the meta-stable states from each othe r in species space. The classification tree analyses interpret groups using the environmental variables given. In this case, the meta-stable states ar e classified by the vari ables that distinguish them. Misclassification is common and overfitt ing the data could be an issue. The interpretability of the cla ssification trees, however, provide insight into the conditions that may be distinguishing the community stat es. The diversity of variables that were determined to be associated with each phys iognomic type outlines the complexity of environmental variation between the meta-stabl e states. This may e xplain why individual variables were not determined to be explanat ory to structural vari ation in the NMS. Perhaps the creation of an alte rnative metric that accounts fo r the variables determined by the classification tree would be useful including in the ordination procedure.

PAGE 78

64 Overall, the NMS technique was helpful in visualizing the differences between the meta-stable states in terms of community stru cture. The classificat ion trees provided the environmental context of each physiognomi c type, by distinguishing particular combinations of conditions that determine a meta-stable state. In the case of this study, neither method was more correct than the othe r. Both analyses helped to provide the whole explanation behind what the meta-stable states are. Review of Methodology and Future Tracks of Research The analytical techniques of clustering and classificati on trees were successful at identifying subtle differences in comm unity structure and their corresponding environmental characteristics. These specific multivariate analyses are geared toward the nonparametric nature of community ecology. Sampling was sufficient for establishing environmental criteria for each of the physi ognomic types, with probable exception to island-types, although even these communities we re distinct in the classification tree analysis. Importance values are a useful tool in summarizing the complex structure of community data. However, the addition of relative frequency to the computation of importance values should be considered. This metric would include clustering and evenness of species within a community to the equation, providing a more descriptive index of community structure. In the case of sparse sawgrass communities such as the sawgrass with E. elongata/Crinum state, the unevenness of sawgrass would be accounted for, rather than assumed due to the types of associates. Although the set of environmen tal variables analyzed to differentiate the metastable states were comprehensive in rega rd to hydrology, many soil parameters were lacking. Soil nutrient con centrations, acidity, and bul k density, although shown in

PAGE 79

65 previous studies to be uniform on a landscape scale in WCA 3A (given sufficient distance from point sources i.e. canals, etc.), could be a factor in hot spots of wildlife activity. Alligator holes and bird roosts tend to a lter local soil chemistry characteristics. Expansion of community dynamic studies in the Everglades should consider landscape geometry and proximity to seed banks. It was not necessary to include these landscape metrics in this study, because all of the rese arch was located in one impounded section of the Everglades that is relatively distant from disturbed upland habitat that would provide a source of ruderal recruitment. Future research should include the inves tigation of the movement of individual sample units over time. NMS is a useful tool for quantifying and visualizing the trajectory of a sample in species space. C oupled with changes in seasonal environmental conditions, additional insights could be made into landscape level trends of community shifts. Inferences into the resilience of community states could be made through NMS or Multiresponse Permutational Procedures (MRPP) of individual sample units. Long-term studies should monitor the movement of ecotones relative to hydrologic dynamics. Reduction of the Everglades ecosystem to ha lf of its size, loss of sheetflow through the system, and loss of habita t diversity are some of the functional losses due to human engineering efforts of the mid-20th century (Davis et al. 1994). Monitoring and predicting landscape-scale vegetation dynamics is essential to the restoration efforts. The potential for community dynamic research usi ng the concepts of meta-stable states is considerable. An adaptive management appr oach to Everglades restoration means that hydrologic regimes will continue to change alon g with management strategies. The need for greater understanding of the implications of future scenarios, will at the same time

PAGE 80

66 allow researchers to further examination of the intricacies of Everglades community dynamics.

PAGE 81

67 APPENDIX A INDICATOR SPECIES ANALYSIS TABLES AND FIGURES November 2002 Indicator Species Analysis Graphs 0.08 0.09 0.1 0.11 0.12 0.13 0.14 0.15 0.16 0.17 0.18 0.1930 28 26 24 22 2 0 18 16 14 12 10 8 6 4 2Number of clusters Figure A-1. Change in p-value from the random ization tests, averaged across species at each step in the clustering. 0 2 4 6 8 10 12 14 16 1830 28 26 24 22 20 18 16 1 4 1 2 1 0 8 6 4 2Number of clusters Figure A-2. Number of species with p 0.05 for each step of clustering.

PAGE 82

68 June 2003 Indicator Species Analysis Graphs 0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.230 28 26 24 22 2 0 18 16 14 12 10 8 6 4 2Number of clusters Figure A-3. Change in p-value from the random ization tests, averaged across species at each step in the clustering. 0 5 10 15 20 2530 28 26 24 22 20 18 16 14 12 10 8 6 4 2Number of clusters Figure A-4. Number of species with p 0.05 for each step of clustering.

PAGE 83

69 November 2003 Indicator Species Analysis Graphs 0.1 0.11 0.12 0.13 0.14 0.15 0.16 0.17 0.18 0.1930 28 26 24 22 20 18 16 14 12 10 8 6 4 2 Figure A-5. Change in p-value from the random ization tests, averaged across species at each step in the clustering. 0 2 4 6 8 10 12 14 16 18 2030 29 2 8 2 7 26 25 24 2 3 22 21 20 1 9 18 17 16 1 5 1 4 13 12 11 1 0 9 8 7 6 5 4 3 2Number of clusters Figure A-6. Number of species with p 0.05 for each step of clustering.

PAGE 84

70 June 2004 Indicator Species Analysis Graphs 0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.1830 28 2 6 2 4 22 20 18 16 14 1 2 1 0 8 6 4 2Number of clusters Figure A-7. Change in p-value from the random ization tests, averaged across species at each step in the clustering. 0 5 10 15 20 2530 28 2 6 24 22 20 18 16 14 12 10 8 6 4 2Number of clusters Figure A-8. Number of species with p 0.05 for each step of clustering.

PAGE 85

71 APPENDIX B COMMUNITY STATES AND THEIR STRUCTURAL SIGNATURES The groups from the dendrograms in Fi gures 3-2, 3-3, 3-4, and 3-5 were run through another cluster analysis to match si milar groups between sampling events. Those groups from the second clusteri ng run were designated as th e community states of Water Conservation Area 3A. Two groups from the same sampling event were occasionally pooled together due to the similarity be tween community structures. The following figures are the structural signatures of th e community states resulting from cluster analyses of community units. There were a total of fifteen community states. Species codes align the x-axis (see Table 2-1 for species names and codes). The yaxis corresponds to the percent of perfect indication relating each species to that particular community state. The percent of perfect indication combines relative abundance (the average abundance of a give n species in a given group of communities over the average abundance of that species in all communities) and th e relative frequency (percent of community units in given group wher e given species is present). This value is how well that species is an indicator of th at community state. When all species are included on the same graph a signature develops that is unique to that community state. The signatures for each community state at each sampling event (sometimes there were two in a sampling event, sometimes there were none) are aligned on the z-axis.

PAGE 86

72 BAC BLS CEO CLA CLD CRA ELG Elsp Hysp LEH NMA NYO OSR PAH PDG PEV PNC SAL TYA TYD UNKJS UNKSR Utsp 0 5 10 15 20 25 30 35 40% of Perfect IndicationSpeciesPanicum/Paspalidium/Eleocharis Prairie Nov-02 Jun-03 Nov-03 Jun-04 Figure B-1. Structural signature of the Panicum/Paspalidium/Eleocharis Prairie. BAC BLS CEO CLA CLD CRA ELG Elsp Hysp LEH NMA NYO OSR PAH PDG PEV PNC SAL TYA TYD UNKJS UNKSR Utsp 0 10 20 30 40 50% of perfect indicationSpeciesShallow Slough Nov-03 Jun-04 Figure B-2. Structural signa ture of the Shallow Slough.

PAGE 87

73 BAC BLS CEO CLA CLD CRA ELG Elsp Hysp LEH NMA NYO OSR PAH PDG PEV PNC SAL TYA TYD UNKJS UNKSR Utsp 0 10 20 30 40 50 60 70 80 90% of perfect indicationSpeciesGhost Island Nov-02 Jun-03 Nov-03 Jun-04 Figure B-3. Structural signa ture of the Ghost Island. BAC BLS CEO CLA CLD CRA ELG Elsp Hysp LEH NMA NYO OSR PAH PDG PEV PNC SAL TYA TYD UNKJS UNKSR Utsp 0 10 20 30 40 50 60% of perfect indicationSpeciesDeep Slough Nov-02 Jun-03 Nov-03 Jun-04 Figure B-4. Structural si gnature of the Deep Slough.

PAGE 88

74 BAC BLS CEO CLA CLD CRA ELG Elsp Hysp LEH NMA NYO OSR PAH PDG PEV PNC SAL TYA TYD UNKJS UNKSR Utsp 0 5 10 15 20 25 30% of perfect indicationSpeciesEleocharis elongata Slough Nov-02 Nov-02 Jun-03 Nov-03 Figure B-5. Structural signature of the Eleocharis elongata Slough. BAC BLS CEO CLA CLD CRA ELG Elsp Hysp LEH NMA NYO OSR PAH PDG PEV PNC SAL TYA TYD UNKJS UNKSR Utsp 0 5 10 15 20 25 30 35 40% of perfect indicationSpeciesEleocharis elongata Prairie Nov-02 Jun-03 Nov-03 Jun-04 Figure B-6. Structural signature of the Eleocharis elongata Prairie.

PAGE 89

75 BAC BLS CEO CLA CLD CRA ELG Elsp Hysp LEH NMA NYO OSR PAH PDG PEV PNC SAL TYA TYD UNKJS UNKSR Utsp 0 5 10 15 20 25 30% of perfect indicationSpeciesSawgrass Prairie Jun-03 Nov-03 Figure B-7. Structural signatu re of the Sawgrass Prairie. BAC BLS CEO CLA CLD CRA ELG Elsp Hysp LEH NMA NYO OSR PAH PDG PEV PNC SAL TYA TYD UNKJS UNKSR Utsp 0 5 10 15 20 25 30 35 40 45 50% of perfect indicationSpeciesEleocharis Prairie Nov-02 Jun-03 Nov-03 Jun-04 Figure B-8. Structural signature of the Eleocharis Prairie.

PAGE 90

76 BAC BLS CEO CLA CLD CRA ELG Elsp Hysp LEH NMA NYO OSR PAH PDG PEV PNC SAL TYA TYD UNKJS UNKSR Utsp 0 5 10 15 20 25 30 35 40 45% of perfect indicationSpeciesPanicum Slough Nov-02 Jun-04 Figure B-9. Structural signature of the Panicum Slough. BAC BLS CEO CLA CLD CRA ELG Elsp Hysp LEH NMA NYO OSR PAH PDG PEV PNC SAL TYA TYD UNKJS UNKSR Utsp 0 10 20 30 40 50 60 70 80 90 100% of perfect indicationSpeciesTree Island Jun-03 Nov-03 Jun-04 Figure B-10. Structural si gnature of the Tree Island.

PAGE 91

77 BAC BLS CEO CLA CLD CRA ELG Elsp Hysp LEH NMA NYO OSR PAH PDG PEV PNC SAL TYA TYD UNKJS UNKSR Utsp 0 10 20 30 40 50 60% of perfect indicationSpeciesSawgrass with Bacopa and Ludwigia Nov-02 Jun-03 Nov-03 Jun-04 Figure B-11. Structural si gnature of Sawgrass with Bacopa and Ludwigia BAC BLS CEO CLA CLD CRA ELG Elsp Hysp LEH NMA NYO OSR PAH PDG PEV PNC SAL TYA TYD UNKJS UNKSR Utsp 0 5 10 15 20 25 30 35% of perfect indicationSpeciesSawgrass Ghost Island Nov-02 Jun-03 Figure B-12. Structural signature of the Sa wgrass Ghost Island.

PAGE 92

78 BAC BLS CEO CLA CLD CRA ELG Elsp Hysp LEH NMA NYO OSR PAH PDG PEV PNC SAL TYA TYD UNKJS UNKSR Utsp 0 5 10 15 20 25% of perfect indicationSpeciesSawgrass with E. elongata and Crinum Nov-02 Jun-03 Nov-03 Jun-04 Figure B-13. Structural si gnature of Sawgrass with E. elongata and Crinum BAC BLS CEO CLA CLD CRA ELG Elsp Hysp LEH NMA NYO OSR PAH PDG PEV PNC SAL TYA TYD UNKJS UNKSR Utsp 0 5 10 15 20 25 30% of perfect indicationSpeciesHeavy Sawgrass Nov-02 Jun-03 Jun-03 Nov-03 Nov-03 Jun-04 Figure B-14. Structural si gnature of Heavy Sawgrass.

PAGE 93

79 BAC BLS CEO CLA CLD CRA ELG Elsp Hysp LEH NMA NYO OSR PAH PDG PEV PNC SAL TYA TYD UNKJS UNKSR Utsp 0 5 10 15 20 25% of perfect indicationSpeciesSawgrass with Eleocharis and Panicum Nov-02 Jun-04 Figure B-15. Structural si gnature of Sawgrass with Eleocharis and Panicum

PAGE 94

80 APPENDIX C RESULTS OF THE NONMETRIC MULTI DIMENSIONAL SCALING ANALYSES The following are results of the NMS ordi nation analyses used to plot the community sample units in multidimensional ordination space. Included are scree plots used to assess the dimensionality of the data se t. The figures plot th e final stress vs. the number of dimensions. Stress is an inverse measure of fit to the data. The randomized data from a Monte Carlo te st are analyzed as a null model for comparison. The dimension selected is prior to which additional dimensions provided only small reductions in stress. Also in cluded are tables comparing th e solution to the Monte Carlo result. Finally, the stress and stability of the solution are included. Stress and stability were listed in the numerical output of the NMS. Figure C-1. A scree plot for the sloughtype ordination.

PAGE 95

81 Table C-1. Stress in relation to dimensi onality for slough NMS. A two-dimensional solution was chosen. Stress in real data Stress in randomized test Axes Minimum Mean MaximumMinimum Mean Maximump 1 28.676 44.690 57.313 38.592 48.260 57.336 0.0323 2 13.422 16.720 41.513 16.868 21.001 41.495 0.0323 3 8.469 10.355 32.522 11.537 12.579 14.161 0.0323 4 6.675 12.480 26.998 9.128 10.829 26.956 0.0323 Final stress for two-dimensional solu tion = 15.10441. Final instability = 0.0001. Figure C-2. A scree plot for the prairi e-type ordination. Table C-2. Stress in relation to dimensionality for prairie NMS. A two-dimensional solution was chosen. Stress in real data Stress in randomized test Axes Minimum Mean MaximumMinimum Mean Maximump 1 24.387 38.624 57.243 42.741 48.798 57.302 0.0323 2 14.463 15.606 18.013 21.872 25.496 29.070 0.0323 3 9.502 14.279 32.404 14.924 16.221 17.261 0.0323

PAGE 96

82 4 7.149 20.353 26.866 11.636 16.581 26.869 0.0323 Final stress for two-dimensional solu tion = 14.71455. Final instability = 0.00007. Figure C-3. A scree plot for the sawgrass-type ordination. Table C-3. Stress in relation to dimensionality for sawgrass NMS. A three-dimensional solution was chosen. Stress in real data Stress in randomized test Axes Minimum Mean MaximumMinimum Mean Maximump 1 36.107 45.100 57.495 43.616 49.797 57.390 0.0323 2 21.343 22.866 24.880 24.024 25.749 27.922 0.0323 3 13.965 14.658 15.821 15.967 17.322 18.811 0.0323 4 12.419 16.130 23.565 11.609 17.810 27.082 0.2903 Final stress for three-dimensional so lution = 16.53583. Fina l instability = 0.01579.

PAGE 97

83 Figure C-4. A scree plot for the island -type ordination. Table C-4. Stress in relation to dimensionality for island NMS. A three-dimensional solution was chosen. Stress in real data Stress in randomized test Axes Minimum Mean MaximumMinimum Mean Maximump 1 34.018 46.591 56.898 38.144 48.194 56.891 0.0323 2 21.074 22.272 23.640 22.478 25.339 40.804 0.0323 3 14.394 14.962 17.061 15.748 17.117 18.931 0.0323 4 10.780 13.211 26.054 12.075 13.028 13.898 0.0323 Final stress for three-dimensional so lution = 14.70141. Fina l instability = 0.00283.

PAGE 98

84 APPENDIX D IMPORTANCE CHARTS OF ENVI ROMENTAL VARIABLES FROM CLASSIFICATION TREE ANALYSIS The following charts rank the importance of the environmental variables in explaining the differences betw een groups in the classificati on tree analysis. The groups, or meta-stable states of communities, were analyzed as physiognomic types. The importance scale is based on rankings of e xplanatory power for each physiognomic type. Interpretations and discussion of thes e charts can be found in Chapter 5. Figure D-1. Importance rankings of pred ictor variables for the slough physiognomic type. Slough physiognomic type predictor variables

PAGE 99

85 Figure D-2. Importance rankings of predic tor variables for th e prairie physiognomic type. Figure D-3. Importance rankings of predicto r variables for the sawgrass physiognomic type. Prairie physiognomic type predictor variables Sawgrass physiognomic type predictor variables

PAGE 100

86 Figure D-4. Importance rankings of predicto r variables for the is land physiognomic type. Island physiognomic type predictor variables

PAGE 101

87 LIST OF REFERENCES Biondini M, Bonham C, Redente E, 1985, Seconda ry successional patterns in a sagebrush ( Artemisia tridentata ) community as they relate to soil disturbance and soil biological activity. Vegetatio 60:25-36. Bishop C, 1995, Neural Network for Pattern Re cognition. Oxford University Press, New York City, New York. Carpenter S, Walker B, Anderies J, Ab el N, 2001, From metaphor to measurement: resilience of what to wh at? Ecosystems 4:765-781. Clarke K, and Ainsworth M, 1993, A method of linking multivariate community structure to environmental variables. Mari ne Ecology Progress Series 92:205-219. Davis J, 1943, The natural features of southern Florida. Fla. Geol Surv. Biol. Bull. No. 25. Tallahassee, FL, USA. Davis S, Gunderson L, Park W, Richards on J, Mattson J, 1994, Landscape dimension, composition, and function in a changing Everglades ecosystem. In Everglades: The Ecosystem and Its Restoration, Davis S and Ogden J (Eds.), St. Lucie Press, Delray Beach, FL, pp. 419-444. DeAngelis D, 1994, Synthesis: Sp atial and temporal characteri stics of the environment. In Everglades: The Ecosystem and Its Rest oration, Davis S and Ogden J (Eds.), St. Lucie Press, Delray Beach, FL, pp. 307-321. DeAngelis D, and White P, 1994, Ecosystems as products of spatially and temporarily varying driving forces, ecological pro cesses, and landscapes: a theoretical perspective. In Everglades : The Ecosystem and Its Restoration, Davis S and Ogden J (Eds.), St. Lucie Press, Delray Beach, FL, pp. 9-28. Dufrene M, and Legendre P, 1997, Species asse mblages and indicator species: the need for a flexible asymmetrical appr oach. Ecological Monographs 61:53-73. Gerritsen J, and Greening H, 1989. Marsh seed banks of the Okefenokee swamp: effects of hydrologic regime and nutrients. Ecology 70:750. Gleason P, and Stone P, 1994, Age, origin, a nd landscape evolution of the Everglades peatland. In Everglades: The Ecosystem a nd Its Restoration, Davis S and Ogden J (Eds.), St. Lucie Press, Delray Beach, FL, pp. 149-198.

PAGE 102

88 Gunderson L, and Pritchard L, 2002, Resilien ce and the Behavior of Large Scale Ecosystems. SCOPE volume, Island Press, Washington, DC. Gunderson L, 1994, Vegetation of the Ever glades: determinants of community composition. In Everglades: The Ecosystem and Its Restoration, Davis S and Ogden J (Eds.), St. Lucie Press, Delray Beach, FL, pp. 323-340. Gunderson L, Loftus W, 1993, The Everglades: competing land uses imperil the biotic communities of a vast wetland. In Biotic Communities of the Southeastern United States, Martin W, Boyce S, and Estern acht A (Eds.), John Wiley and Sons, New York, NY, pp. 123-134. Herndon A, Gunderson L, Stenberg J, 1991, Sawgrass ( Cladium jamaicense ) survival in a regime of fire and flooding. Wetlands 11:17-27. Holling C, 1973, Resilience and stability of ecological systems. Annual Review of Ecology and Systematics 4:1-24. Jordan F, Babbitt K, McIvor C, Miller S, 1996, Spatial ecology of the crayfish Procambarus alleni in a Florida wetland mosaic. Wetlands 16:134-142. Jordan F, Jelks H, Kitchens W, 1994, Habitat use by the fishing spider Dolomedes triton in a northern Everglades wetland. Wetlands 14:239-242. Kent M, and Coker P, 1992, Vegetation Descri ption and Analysis: A Practical Approach. CRC Press Inc., Boca Raton, FL. Kitchens W, Bennetts R, DeAngelis D, 2002, Linkages between the snail kite population and wetland dynamics in a highly fragmented South Florida landscape. In The Everglades, Florida Bay, and Coral Reef s of the Florida Keys: An Ecosystem Sourcebook, Porter J and Porter K (Eds .), CRC Press, Boca Raton, FL, chapter 6. Kruskal J, 1964, Nonmetric multidimensional scaling: a numerical method. Psychometrika 29:115-129. Loftus W, Kushlan J, 1987, Freshwater fishes of southern Florida. Bulletin of the Florida State Museum, Biologica l Sciences 31:147-344. Loveless C, 1959, A study of the vegetation in the Florida Evergl ades. Ecology 40:1-9. Mather P, 1976, Computational methods of mu ltivariate analysis in physical geography. J. Wiley & Sons, London. McCune B, and Grace J, 2002, Analysis of Ecological Communities. MJM Software Design, Gleneden Beach, OR. McPherson B, 1973, Vegetation in Relation to Water Depth in Conservation Area 3, Florida. U.S. Geological Survey Open-File Report No. 73025, Tallahassee, FL.

PAGE 103

89 Richter B, Baumgartner J, Wigington R, Braun D, 1997, How much water does a river need? Freshwater Biology 37:231-249. Richter B, Baumgartner J, Powell J, Br aun D, 1996, A method for assessing hydrologic alteration within ecosystems. Conservation Biology 10:1163-1174. Scheffer M, Carpenter S, Foley J, Folke C, Walker B, 2001, Catastrophic shifts in ecosystems. Nature 413:591-596. Smith S, McCormick P, Leeds J, Garrett P, 2002, Constraints of seed banks and water depth for restoring vegetation in the Fl orida Everglades, U.S.A. Restoration Ecology 10:138-145. Toner M, and Keddy P, 1997, River hydrology a nd riparian wetlands: a predictive model for ecological assembly. Ec ological Applications 7:236-246. Urban D, 2002, Classification and regression trees. In Analysis of Ecological Communities. MJM Software Desi gn, Gleneden Beach, OR, pp. 222-232. van der Valk A, 1991, Response of wetland vegeta tion to a change in water level. In Wetland Management and Restoration, C Finlayson and T Larson (Eds.), Proc. Workshop, Solna, Sweden 1990, Swedish Environmental Protection Agency Report, pp. 7-15. Weisner S, and Miao S, 2004, Use of morphological variability in Cladium jamaicense and Typha domingensis to understand vege tation changes in an Everglades marsh. Aquatic Botany 78:318-335. Weisner S, and Strand J, 1996, Rhizome architecture in Phragmites australis in relation to water depth: implications for within -plant oxygen transport distances. Folia Geobot. Phytotax 31:91. White P, 1994, Synthesis: Vegetation pattern an d process in the Everglades ecosystem. In Everglades: The Ecosystem and Its Restor ation, S Davis and J Ogden (Eds.), St Lucie Press, Delray Beach, FL, chapter 18. Wishart D, 1969, An algorithm for hierarchi cal classifications. Biometrics 25:165-170. Wood J, and Tanner G, 1990, Graminoid commun ity composition and structure within four Everglades management areas. Wetlands 10:127-149. Wunderlin R, 1998, Guide to the Va scular Plants of Florida. Un iversity Press of Florida, Gainesville.

PAGE 104

90 BIOGRAPHICAL SKETCH Erik Norman Powers was born in Corpus Christi, Texas, on September 2, 1976. He grew up in Destin, Florida, after having resi ded briefly outside Mobile, Alabama. From 1994-1996 he attended the Oregon Institute of Technology. He transferred to the University of Florida in 1997 and receive d a Bachelor of Science degree in environmental science. Following a two-y ear employment stint at C&N Environmental Consulting, Inc., he began the graduate program at the University of Florida through the School of Natural Resources under Wiley Kitche ns. He married Kristy Shreve on August 21, 2004, in Destin. Kristy and Erik are expe cting their first child in March 2006.


Permanent Link: http://ufdc.ufl.edu/UFE0013333/00001

Material Information

Title: Meta-Stable States of Vegetative Habitats in Water Conservation Area 3A, Everglades
Physical Description: Mixed Material
Copyright Date: 2008

Record Information

Source Institution: University of Florida
Holding Location: University of Florida
Rights Management: All rights reserved by the source institution and holding location.
System ID: UFE0013333:00001

Permanent Link: http://ufdc.ufl.edu/UFE0013333/00001

Material Information

Title: Meta-Stable States of Vegetative Habitats in Water Conservation Area 3A, Everglades
Physical Description: Mixed Material
Copyright Date: 2008

Record Information

Source Institution: University of Florida
Holding Location: University of Florida
Rights Management: All rights reserved by the source institution and holding location.
System ID: UFE0013333:00001


This item has the following downloads:


Full Text












META-STABLE STATES OF VEGETATIVE HABITATS IN WATER
CONSERVATION AREA 3A, EVERGLADES















By

ERIK POWERS


A THESIS PRESENTED TO THE GRADUATE SCHOOL
OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OF
MASTER OF SCIENCE

UNIVERSITY OF FLORIDA


2005

































Copyright 2005

by

Erik Powers

































This document is dedicated to my father, Dr. Lawrence W. Powers, who inspired my
fascination with science early in life.















ACKNOWLEDGMENTS

I must first thank Dr. Wiley Kitchens for his unwavering support and

encouragement throughout my graduate career. I thank my committee members Dr. Paul

Wetzel and Dr. Ted Schuur for their advice and guidance. Instrumental in the

experimental design, Paul Wetzel has provided support from the beginning. Paul

Conrads of the USGS performed the neural network analysis for the hydrologic data set.

His assistance was paramount to the completion of this thesis.

Logistic support, including airboats and lodging, was provided by the Florida

Cooperative Fish and Wildlife Unit, University of Florida, Gainesville. The following

University of Florida graduate students and staff helped with field sampling and data

processing: Stephen Brooks, Janell Brush, Melissa DeSa, Jamie Duberstein, Joey

Largay, Kristianna Lindgren, Julien Martin, Ann Marie Muench, Alison Pevler, Laura

Pfenninger, Derek Piotrowicz, Zach Welch, and Christa Zweig. Lastly, I thank my wife

and best friend, Kristy Powers, for her undying patience and compassion despite my

propensity for tracking mud into the house.
















TABLE OF CONTENTS

Page

A C K N O W L E D G M E N T S ................................................................................................. iv

L IST O F T A B L E S ........................ .. ............................. .. .. ..... ........... .. viii

LIST OF FIGURES ......... ........................................... ............ ix

A B S T R A C T .............................................. ..........................................x iii

CHAPTER

1 IN TR OD U CTION ............................................... .. ......................... ..

W hat Are M eta-stable States? ................................................ .......................2
How Community Subtypes and Driving Forces Are Determined.............................4
Com m unities of the Everglades.............................. .......................... ...............6.
P project O bj ectiv es ............... ................................................... ................ 8

2 DETERMINING COMMUNITY STRUCTURE...................................................9

D description of Study Site ................................................................. ....................11
M methods and M materials .................................... ..... .. ...... .............14
Sam pling R egim e .......................................... .. .. .... .. ....... ........ 14
Sam pling M ethodology ................................................................. ............... 15
Processing M ethodology .................................. .....................................16
Data preparation and Relativization ........................................................16

3 CLASSIFICATION OF META-STABLE STATES ...........................................20

Hierarchical Agglomerative Cluster Analysis...................................................21
Testing Importance Value Assumptions.......................... ........................21
Indicator Species Analysis..................... .......................................27
Matching Similar Community Descriptions Between Sampling Events....................28
Distribution of Meta-Stable States Across the Landscape .................................30

4 MULTIVARIATE ANALYSIS AND RESULTS ............................................. 33

H y d ro lo g y ......................................................... .............. ................ 3 3
Selecting H ydrologic V ariables...................................... ......................... 33









Calculating H ydrologic V ariables ............................................ ............... 35
Hindcasting using neural netw orks ........................................ ...... .... 35
Extrapolating from well data to sample unit data .....................................38
N onm etric M ultidim ensional Scaling........................................ ......... ... ............... 39
Classification Trees and Characterization of Meta-Stable States by Environmental
V ariables ........................................45
Slough Physiognom ic Type........................................... .......................... 46
D eep sloughs ...........................................................47
Eleocharis elongata sloughs ............................................. ............... 48
P anicum sloughs ................................... ................... ......... 48
Shallow sloughs.............. ..... ............. ....... ........ 48
W et Prairie Physiognom ic Type.................................... ........................ 48
E leocharis sp. prairie.......... .................................. .......... .... .... ............... 49
E. elongata prairie ........... ..... .... .. ...... .... ... ....... ....50
Panicum/Paspalidium/Eleocharis prairies .............. ....................... 50
Saw grass prairies.......... ..................................... ........ ........ ... ......... 50
Sawgrass Physiognomic Types ....................................... ........................ 51
Sawgrass monoculture (heavy sawgrass).................. ................................51
Sawgrass with Bacopa and Ludwigia............................ .. ...............52
Sawgrass with Eleocharis sp. and Panicum ........................................52
Sawgrass with E. elongata and Crinum .................... ................... .......... 52
Island Physiognom ic Types........................................... .......................... 53
G h o st islan d s ............................................................ 54
Sawgrass ghost islands ............ .. ............................... ...................54
T ree islan d s .............................................................5 4

5 SUMMARY AND CONCLUSIONS.........................................................55

D iscu ssion ........................... ........ ........ .............................. .................. 55
Comparing NMS and Classification Tree Techniques ............................................63
Review of Methodology and Future Tracks of Research............... .................64

APPENDIX

A INDICATOR SPECIES ANALYSIS TABLES AND FIGURES............................67

November 2002 Indicator Species Analysis Graphs ...............................................67
June 2003 Indicator Species Analysis Graphs .............................. ................68
November 2003 Indicator Species Analysis Graphs ..............................................69
June 2004 Indicator Species Analysis Graphs .............................. ................70

B COMMUNITY STATES AND THEIR STRUCTURAL SIGNATURES ................71

C RESULTS OF THE NONMETRIC MULTIDIMENSIONAL SCALING
A N A L Y S E S ...............................................................................................................8 0









D IMPORTANCE CHARTS OF ENVIRONMENTAL VARIABLES FROM
CLASSIFICATION TREE ANALYSIS ........................................ .....................84

L IST O F R E F E R E N C E S ........................................................................ .....................87

B IO G R A PH IC A L SK E TCH ..................................................................... ..................90
















LIST OF TABLES


Table page

2-1 Complete species list for vegetative study in Water Conservation Area 3A.
Authority for plant names and status is from Wunderlin, R.P. 1998 Guide to the
Vascular Plants of Florida. University Press of Florida, Gainesville. Includes
unknown species that occur in more than one sample. .........................................17

2-2 An abridged species data matrix with importance values for species in each
com m unity unit. .................................................... ................. 19

3-1 Meta-stable community states and their frequency at each sample event .............30

4-1 Hydrological variables with abbreviations............... ............................................ 34

4-2 Neural network model statistics for each station hindcasted. PME (percent model
error) = RMSE (root mean-square error) / (range of measured data). ....... .......... 38

4-3 Environmental variables used in the multivariate analyses and how they were
relativized if a transformation was appropriate......................................................39

C-l Stress in relation to dimensionality for slough NMS. A two-dimensional solution
w a s ch o sen ...................................................................... 8 1

C-2 Stress in relation to dimensionality for prairie NMS. A two-dimensional solution
w a s ch o sen ...................................................................... 8 1

C-3 Stress in relation to dimensionality for sawgrass NMS. A three-dimensional
solution w as chosen .................. ............................... ...... .. ............ 82

C-4 Stress in relation to dimensionality for island NMS. A three-dimensional solution
w a s ch o sen ...................................................................... 8 3















LIST OF FIGURES


Figure page

1-1 A community can shift to an alternate state if the perturbation is strong enough, or
conditions change steadily over time. Note that two community states can exist in
the same environmental conditions. For instance, two meta-stable states can
operate under the same hydrological conditions, but have different hydrologic
thresholds (Scheffer 2001). ............................................................................ 3

2-1 Shaded area is the location of the study area. Water Conservation Area 3A is
designated as section 9 on this m ap. ............................................... ............... 11

2-2 Satellite composite of the study area in Water Conservation Area 3A. Twenty plots
were distributed with a stratified random design for the sampling procedures........12

2-3 An overlay of a square kilometer plot on satellite imagery. The blue dots signify
reference poles aligned with belt transects within the plot. Each transect crosses at
least one com m unity boundary. ........................................ .......................... 13

2-4 A diagram of a belt transect consisting of three traversable subtransects. Each sub
transect can be sampled on four different occasions twice on each side .............14

3-1 Cluster dendrogram from November 2002 sampling event. Community units are
listed on the left and color coded with respect to their a priori designation. ...........22

3-2 Cluster dendrogram from June 2003 sampling event. Community units are listed
on the left and color coded with respect to their a priori designation ....................23

3-3 Cluster dendrogram from November 2003 sampling event. Community units are
listed on the left and color coded with respect to their a priori designation. ...........24

3-4 Cluster dendrogram from June 2004 sampling event. Community units are listed
on the left and color coded with respect to their a priori designation ....................25

3-5 A scatterplot of sawgrass communities sampled in November 2002. Axes
correspond to percent relative biomass and percent relative density. Each point
represents one sawgrass unit. Each sawgrass type resulting from the cluster
analysis is coded in the legend. ..... .....................................................................26

3-6 Shows the distribution of meta-stable states by physiognomic type into four
quadrants of the study landscape ............................ ......... ........ ......... 31









4-1 Green triangles represent the monitoring stations set up by various agencies. These
stations upload real-time data to the web daily. Yellow circles indicate the
temporary stations that were established in December 2002 ................................36

4-2 A whole scale ordination plot of the community sample units. Triangles represent
individual sample units and crosses represent species. The key to the legend: 1-
prairie physiognomic type; 2-slough physiognomic type; 3-sawgrass
physiognomic type; 4-island physiognomic type. The environmental gradients
(minimum depth and mean depth) are represented with red vectors closely aligned
w ith ax is 2 ........................................................ ................ 4 1

4-3 Island-type ordination plots. Triangles represent individual sample units and
crosses represent species. The key to the legend: 1-sawgrass ghost island; 2-
ghost island; 3- tree island. ..... ........................... ......................................42

4-4 Slough-type ordination plots. Triangles represent individual sample units and
crosses represent species. The key to the legend: 4-deep slough; 6-Panicum
slough; 7-E. elongata slough; 8-shallow slough....................... .............. 43

4-5 Sawgrass-type ordination plots. Triangles represent individual sample units and
crosses represent species. The key to the legend: 9-sawgrass with Eleocharis
sp./Panicum; 10-sawgrass with Bacopa/Ludwigia; 11-sawgrass with E.
elongata Crinum; 12- sawgrass monoculture...... .... ......................................... 43

4-6 Prairie-type ordination plots. Triangles represent individual sample units and
crosses represent species. The key to the legend: 1-Eleocharis sp. prairie; 2-
Panicum/Paspalidium/Eleocharis sp. prairie; 3-E. elongata prairie; 5-sawgrass
prairie. Environmental gradients shown as red vectors closely aligned with axis 2.44

4-7 Classification tree for the meta-stable states on 10 environmental variables. The
number of sample units in each leaf are shown in parentheses below each bar
graph, which shows the compositions of communities within each leaf. ...............46

4-8 Classification tree for 4 slough community states on 10 environmental variables.
This model was pruned from a tree size of 7 leaves to five, based on a cost
complexity pruning curve, selecting the smallest tree within one standard error of
the best. The number of sample units in each leaf is shown in parentheses below
each bar graph, which shows the compositions of communities within each leaf...47

4-9 Classification tree for 4 wet prairie community states on 10 environmental
variables. This model was pruned from a tree size of 11 leaves to eight, based on a
cost complexity pruning curve, selecting the smallest tree within one standard error
of the best. ...........................................................................49

4-10 A classification tree for 4 sawgrass community states on 10 environmental
variables. This model was pruned from a tree size of 14 leaves to eight, based on a
cost complexity pruning curve, selecting the smallest tree within one standard error
of the best. ...........................................................................5 1









4-11 Classification tree for 3 island-type community states on 10 environmental
variables. This model was pruned from a tree size of 10 leaves to eight, based on a
cost complexity pruning curve, selecting the smallest tree within one standard error
of the best. ...........................................................................53

5-1 Distribution of sample units of each community state along a hydrologic variable
(mean annual water depth). Communities are grouped by physiognomic type:
squares=prairies, triangles=sloughs, circles=sawgrass, crosses=islands................57

5-2 Distribution of sample units of each community state along a peat depth gradient.
Communities are grouped by physiognomic type: squares=prairies,
triangles=sloughs, circles=sawgrass, crosses=islands............................................58

5-3 A time-series graph of water stage at a monitoring station within Plot 4. Note the
extreme highs and lows of the second water year compared to the first water year.59

A-i Change in p-value from the randomization tests, averaged across species at each
step in the clustering ....... ......... ... ........ ...............................67

A-2 Number of species with p < 0.05 for each step of clustering................................67

A-3 Change in p-value from the randomization tests, averaged across species at each
step in the clustering ....... ......... ... ........ ...............................68

A-4 Number of species with p < 0.05 for each step of clustering................................68

A-5 Change in p-value from the randomization tests, averaged across species at each
step in the clustering ....... ......... ... ........ ...............................69

A-6 Number of species with p < 0.05 for each step of clustering................................69

A-7 Change in p-value from the randomization tests, averaged across species at each
step in the clustering ....... ......... ... ........ ...............................70

A-8 Number of species with p < 0.05 for each step of clustering................................70

B-1 Structural signature of the Panicum/Paspalidium/Eleocharis Prairie....................72

B-2 Structural signature of the Shallow Slough.....................................................72

B-3 Structural signature of the Ghost Island ........................................................73

B-4 Structural signature of the Deep Slough ...................................... ............... 74

B-5 Structural signature of the Eleocharis elongata Slough.....................................74

B-6 Structural signature of the Eleocharis elongata Prairie. .............. .... ........... 74

B-7 Structural signature of the Sawgrass Prairie. ........................................ ................75









B-8 Structural signature of the Eleocharis Prairie. ........ ....................................75

B-9 Structural signature of the Panicum Slough ................... .............................76

B-10 Structural signature of the Tree Island. ....................................... ............... 76

B-11 Structural signature of Sawgrass with Bacopa and Ludwigia.............................77

B-12 Structural signature of the Sawgrass Ghost Island................................. ..........77

B-13 Structural signature of Sawgrass with E. elongata and Crinum. ........................... 78

B-14 Structural signature of Heavy Sawgrass........ ....... ...... .............. .............. 78

B-15 Structural signature of Sawgrass with Eleocharis and Panicum ...........................79

C-l A scree plot for the slough-type ordination.............. ..... .................. 80

C-2 A scree plot for the prairie-type ordination ........................................ ...................81

C-3 A scree plot for the sawgrass-type ordination.................. ................ ............... 82

C-4 A scree plot for the island-type ordination ................. ....................................83

D-l Importance rankings of predictor variables for the slough physiognomic type.......84

D-2 Importance rankings of predictor variables for the prairie physiognomic type ......85

D-3 Importance rankings of predictor variables for the sawgrass physiognomic type...85

D-4 Importance rankings of predictor variables for the island physiognomic type........86















Abstract of Thesis Presented to the Graduate School
of the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Master of Science

META-STABLE STATES OF VEGETATIVE HABITATS IN WATER
CONSERVATION AREA 3A, EVERGLADES

By

Erik Powers

December 2005

Chair: Wiley Kitchens
Major Department: Interdisciplinary Ecology

The Everglades consists of a constantly dynamic patchwork of vegetative

communities, confined by a matrix of levees and canals into impoundments. Water

Conservation Area (WCA) 3A is a centrally located impoundment, relatively far

downstream from the nutrient-laden waters of the Everglades Agricultural Area. The

major determinants of community structure within WCA 3A are hydrology and soil

characteristics. This study monitors plant community structure over two years, in

transects randomly stratified across the landscape, to determine what community states

manifest between marsh physiognomic types. These communities are constantly shifting

to alternate states, thus described as meta-stable states. They are identified through

unique, but related, vegetative structure, and characterized by a combination of particular

environmental conditions. The transects were sampled semiannually for species biomass

and density within ecotonal boundaries of approximately 140 communities identified a

priori, resulting in a data set of 513 community sample units. Hydrology was monitored









with surface water data loggers, and levels were hindcast 10 years prior to the beginning

of the study with neural network models. Peat depths were recorded for each of the

community units.

A hierarchical cluster analysis on the sample units for each sample event produced

distinct groups that, following an indicator species analysis, were interpreted as meta-

stable states of the four physiognomic types of communities of WCA 3A: slough, wet

prairie, sawgrass, and island-type. The fifteen meta-stable states include deep slough,

shallow slough, Panicum slough, E. elongata slough, E. elongata prairie, Eleocharis sp.

prairie, Panicum Paspalidium Eleocharis prairie, sawgrass prairie, sawgrass with

Eleocharis/Panicum, sawgrass with E. elongata/Crinum, heavy sawgrass, sawgrass with

Bacopa/Ludwigia, sawgrass ghost island, ghost island, and tree island. A classification

tree analysis of each physiognomic type determined that both hydrology and peat depths

were major determinants of community composition.

The meta-stable states had unique environmental characteristics when accounting

for multiple variables. However, when environmental variables are examined

individually between community states, substantial overlap of environmental thresholds

is evident. It can be concluded that the state of a community in the Everglades is

dynamic due to overlap of individual thresholds, but can potentially be predicted through

multivariate modeling. The capability to model community dynamics of Everglades

habitats is crucial to hydrological management strategies. As restoration efforts proceed,

models incorporating how communities respond to management regimes can be essential

tools in scenario analysis.














CHAPTER 1
INTRODUCTION

In ecology, a biological community consists of coexisting organisms that are

linked to one another through unique interactions and associations, thus forming a

complex whole. Plant communities can easily be observed in the field, as they are

relatively sessile and, given a sharp physical boundary, have well-defined ecotones.

These ecotones are usually comprised of a combination of species of the bounding

communities and some unique species as well (Kent and Coker 1992). Therefore,

communities can be identified as physiognomic types. Physiognomic types are defined

by their "species structure," or by what species exist and their relative densities and

biomass.

The four physiognomic community types of the central Everglades, as described

by Davis (1943) and Loveless (1959), are sawgrass, wet prairie, slough, and tree islands.

These are easily recognized and usually have sharp boundaries corresponding to only a

slight change in elevation (McPherson 1973). Water covers the Everglades landscape the

vast majority of the time, leading to a widely believed theory that communities are driven

by hydrologic variables (White 1994). This study attempts to determine vegetative

community subtypes, or "meta-stable" states, within sawgrass, wet prairie, and slough.

This includes the exploration of methods that will enable scientists to document shifts

between community states and between major physiognomic community types over time.

In doing so, a physical hydrologic threshold can be associated with specific

physiognomic community types.









What Are Meta-stable States?

Sawgrass, slough, and wet prairie physiognomic community types exhibit

multiple assemblages and representations of plant species or multiple meta-stable steady

states. These different representations of the same community are alternate

representations and states of that community, and reflect the environmental conditions at

that point in space and time (Gunderson and Pritchard 2002). Transitions between these

"within-type" community states (tall sawgrass into short sawgrass), or between the major

physiognomic community types (e.g., sawgrass into wet prairie) are indicative of

responses to environmental change.

One example of the existence of multiple steady states is the well-documented

process of eutrophication of lakes. Two alternative states can be characterized as (a)

clear water and rooted macrophytes or (b) turbid water with planktonic algae. These

states are relatively stable, but can slip into the other due to a perturbation of a keystone

process, or the removal or addition of a keystone species (Carpenter et al. 2001). If

environmental conditions change slowly, a shift in community state can occur given the

conditions continue to change over time. Alternative states may even share some of the

range of environmental conditions that they could potentially exist in. Figure 1-1 shows

how a community state can shift given perturbations in the environment.

In the context of community state theory, meta-stable states can be defined as an

alternative state of a physiognomic type of community that occurs under predictable

environmental conditions, yet those conditions are dynamic by nature resulting in

constantly shifting community states of that physiognomic type. These meta-stable states

can be witnessed throughout a landscape with multiple physiognomic types and

fluctuating environmental conditions. Meta-stable states imply that community structure









is changing in the direction that conditions are driving it, and there are no equilibriumss"

associated with them. Such is the case in the Everglades, where elevational gradients are

slight, yet hydrology fluctuates considerably on a semiannual basis. These communities

are continually stressed. The structure of the community state is a representation of

various environmental conditions presently, previously, and historically. In summary, a

meta-stable state is a representation of the trajectory of environmental conditions in the

species structure of the physiognomic type of community. The identification of meta-

stable states, as determined in this study, will be addressed in Chapter 3.


Ecosystem state

Figure 1-1. A community can shift to an alternate state if the perturbation is strong
enough, or conditions change steadily over time. Note that two community
states can exist in the same environmental conditions. For instance, two meta-
stable states can operate under the same hydrological conditions, but have
different hydrologic thresholds (Scheffer et al. 2001).









How Community Subtypes and Driving Forces Are Determined

Scientists define ecological resilience as the property that mediates transition

among stability domains (Holling 1973). If the stable states of a specific community can

be defined by their relative composition (species present, relative density of each species,

relative biomass of each species), then the environmental conditions that the community

state tends to persist in indicate a potential driving force. A record of historical and

present conditions of the driving forces) at that site can lead to clues as to where

environmental thresholds lie for the current community type.

Determining community states requires identification of external (abiotic) and

internal (biological) driving forces. All biological communities have several driving

forces, some of them working in concert. However, depending on the temporal and

spatial scale of interest, some of these forces can be ignored as having negligible effect

(DeAngelis and White 1994). A study of community level processes over the course of

two years can rule out slow processes such as interglacial sea level rise, tectonic

movements, and global climate change, as well as intermediate processes such as

weathering, and soil accretion. The focus of a study such as this should be on processes

that will affect the structure of communities within the time frame of the study. In a

hydrologically driven system such as the Everglades, hydropattern is a driving force that

will have one of the strongest effects on vegetation composition and structure. For

example, recruitment of many wetland species through their respective seed and

propagule bank is dependent on meeting certain hydrologic and other criteria (van der

Valk 1990). Sawgrass (Cladium jamaicense), the archetypical plant species of the

Everglades, requires occasional drying events in order to germinate (Smith et al. 2002),

as is generally the case with emergent wetland plant species (Gerritson and Greening









1989). This is especially true in low nutrient regions of the Everglades where sawgrass

stands may neglect important biological functions when exposed to extremely long

hydroperiods (Weisner and Miao 2004). Emergent plants tend to allocate biomass to

shoot length and blade growth in deeper hydrologic conditions, and allocate less energy

to developing belowground rhizome biomass (Weisner and Strand 1996). In regions of

the Everglades such as southern WCA 3A that exhibit perpetually long hydroperiods and

low nutrient concentrations, sawgrass communities, though persistent, are unable to

recruit resulting in sparse, patchy distributions, rather than thick, continuous landscapes

that were present prior to drainage and impoundment (Wood and Tanner 1990). The

resulting communities are composed of emergents such as Pontederia cordata or

Sagittaria lancifolia and woody vegetation such as Cephalanthus occidentalis

interspersed with floating leaf aquatics such as Nymphaea odorata, usually associated

with deep water marshes.

Peat accretion is a slow process that is an important driving force in determining

topography and hence hydroperiod. Competing with this process is decomposition,

which may occur at a much faster rate during drought effects through oxidation or fire.

As these changes in peat depth occur, bedrock topography continues to exert a strong

influence on vegetation through its influence on the patterns of water depth and flow.

Autogenic succession may occur over long periods of time, but is probably rare (Gleason

and Stone 1994).

Other forces that could have major effects on community composition in the

Everglades and within the time frame of the study are fire, variation in nutrient supply,

freezes and wind (DeAngelis 1994). During the period of study, fire did not occur in the









monitoring transects, and hence was not a factor in determining community composition.

However, the history of fire for each of the transects is unknown. For the purposes of

this study, it is assumed that the transects had not been burned for a considerable time

prior to the study. Intense or repeated freezes were also not issues. A series of

hurricanes did strike Florida in the summer of 2004, however Miami-Dade County was

not in the path of any of those disturbances. The study area being investigated is far from

any upstream point source of nutrients and exotic species invasions (canals, urbanized

areas, etc.). The Everglades, historically, is an oligotrophic system, so nutrient loading

will be assumed to be constant at low levels.

Communities of the Everglades

Of the major physiognomic community types of the Everglades, I intend to focus

on three that are both naturally and anthropogenically influenced by hydrology -

sawgrass marshes, peat-based wet prairies (Eleocharis flats), and sloughs. These three

herbaceous communities all occur in southern and central WCA-3A, and often adjacent

to each other. They occur in areas with slightly different relative hydroperiods, with

sawgrass being the driest followed by wet prairie and finally slough as typically the

wettest of the physiognomic types (White 1994).

Sawgrass is the characteristic plant species of the freshwater Everglades. It is

well adapted to flooding, drought, and burning but is killed if high water levels are

prolonged (Herndon et al. 1991). Sawgrass dominates the oligotrophic fresh waters of

the Everglades due to its low nutrient requirements (Gunderson 1994). Sawgrass occurs

in strands that run longitudinally (the historical direction of water flow) in WCA 3A. It

also persists in patches of deep water in the southern extent of 3A, as well as on floating

peat mats and the outer edges of tree islands. Occasionally shrub islands appear in place









of burned or drowned tree islands and sawgrass strands. Cephalanthus occidentalis and

Pontederia cordata are common associates of sawgrass in these transitional

physiognomic types. Islands and their transitional states will also be examined in this

study.

Wet prairies can be classified into two groups-peat-based and marl-based. Marl-

based wet prairies are confined to the marl wetlands situated in the Everglades National

Park, and were not included in this study. Peat-based prairies can further be divided into

three types-Eleocharis, Rhynchospora, and Panicum flats. Of these types, only

Eleocharis, or spikerush, flats, and Panicum, or maidencane flats, are present in the study

area. Rhynchospora prairies are relatively rare after the impoundment of the Everglades.

Wet prairies are typically more diverse than sawgrass or slough communities and occur

often as transitional communities in deeper areas where slough communities are

prevalent (Gunderson 1994) or between slough and sawgrass community as a transitional

community.

Slough communities consist of associations of floating-leafed aquatic plants and

are generally the wettest of the communities in WCA-3A. Submerged aquatics are also

associated with sloughs and provide structure for periphyton, the main source of primary

production in the freshwater Everglades (Gunderson 1994).

Each of these communities has been observed in different forms and structure, yet

they are documented in the general body of scientific literature as single communities.

This project presents evidence that these alternative community states are characteristic

of the environmental conditions at that site. More importantly, transitions between

alternative states of one physiognomic type may occur more readily than a shift between









major physiognomic types. In other words, the resilience of a major physiognomic

community type is greater than the resilience of one of its alternate states. This is tested

through the identification of the hydrologic ranges of each meta-stable state. If there is a

substantial overlap of hydrology between meta-stable states of communities, then it can

be concluded that a shift to an alternate community state while maintaining its basic

physiognomic community type is a possible response to extended exposure to threshold

conditions (see Figure 1-1).

Project Objectives

With this research, I intend to describe the multiple meta-stable states of

physiognomic marsh types of Water Conservation Area 3A in terms of community

structure and their respective environmental tolerances. First, through tabulating relative

densities and relative biomass of species present on established transects during wet and

dry seasons, the current state of a community at any given point in time during the study

was identified. Vegetative community monitoring efforts will continue for two years,

sampling at a rate of twice a year for a total of four sampling events. Hydrologic ranges

for the various meta-stable states of sawgrass, slough, and wet prairie are identified and

each community state is characterized by its environmental variables using classification

trees. Inferences, based on the range of conditions that a state may tolerate, can be made

on the dynamics and resiliency of various vegetative meta-stable states.

Although the study monitors communities over time, tracking the change of

specific sites over time was not included in the scope of this project. The temporal scope

of this study is limited to observing communities under different seasons and water years

to capture the various community states that might manifest under those conditions.














CHAPTER 2
DETERMINING COMMUNITY STRUCTURE

Everglades communities were originally identified by the dominant species

associated with a congregation of smaller or less prevalent species. Loveless first

documented vegetative assemblages in the Everglades (Loveless 1959). Several types of

sawgrass, wet prairie, and slough communities were identified through the abundance

and densities of dominant and associative species. His descriptions of vegetative

communities serve as an introduction and as the basis of comparison for the communities

revealed in the following analyses. The following community types were identified by

Loveless:

Cladium Sagittaria Panicum hemitomon: This sawgrass community can occur

in sparse, dense, or monotypic stands of sawgrass. It is associated with duck potato and

maidencane, as well as a suite of other species depending on the density of sawgrass.

Species composition tends to vary between the dry and wet seasons.

Cladium Myrica Ilex: A drier community of sawgrass, this congregation of

species includes woody thickets of buttonbush, wax myrtle, and dahoon holly.

Cladium Panicum hemitomon: Similar to the duck potato/maidencane sawgrass

community, but occupies drier sites. Densities range from sparse to moderately thick.

Rhynchospora Flats: This community was more prevalent during predrainage

conditions. The wettest of the Loveless communities except for sloughs, this assemblage

includes beakrush as the dominant species and spikerush as the common associate. These

communities are typically found adjacent to sawgrass and shrub island communities.









Panicum hemitomon Flats: Maidencane is the dominant member of this

community and usually occupies drier sites. This community is resilient to fire and can

withstand long periods of flooding while maintaining its basic species configuration.

Associative species usually include spikerush and spider lilies.

Eleocharis Flats: Easily recognizable as monotypic stands of spikerush. This

community is usually found along the southern and western reaches of Water

Conservation Area 3A.

Sloughs: The wettest of the communities, sloughs are usually filled with water year

round. Species associated with sloughs are floating water lily, bladderwort, and

spatterdock. Sloughs comprise the drainage vectors of the Everglades, running generally

longitudinally along the landscape in a north-south direction.

The communities described by Loveless, while useful from a naturalist's

perspective, are outdated with respect to the decades of impoundment effects on

Everglades ecology and irrelevant to studying short-term succession. Communities of the

Everglades are dynamic on two time scales: seasonal and long term (multiannual).

Subtropical south Florida has two distinct hydrologic seasons a wet season during the

summer and fall months, and a dry season during the winter and spring months. Changes

in hydrology imposed by both seasonal fluctuations and water regimes managed by state

agencies will have subtle if not profound effects on community composition. If

restoration agents mean to influence the ecology area from the bottom-up, that is "get the

water right", then the scientific lens must focus on the immediate responses of plant

communities that provide wildlife habitat to variations in hydrology.









Description of Study Site

The study was conducted in the southern half of Water Conservation Area 3A

located in Dade and Broward counties (see Figure 2-1). Bounded by Tamiami Trail to

the south, Holiday Trail (a heavily trafficked airboat trail) to the north, Big Cypress

National Preserve to the west, and Water Conservation Area 3B to the east, the study site

is made up of a matrix of freshwater habitats ranging from short hydroperiod bay and

willow tree islands to deep water sloughs. Strands of sawgrass run longitudinally,

divided by wet prairie and slough. This area was chosen because of the smattering of

distinct communities, abundance of ecotones, and noticeable elevational gradients on a

landscape scale as well as community scale. The total area of the study site is 62,000

hectares.







n n i









12^






Figure 2-1. Shaded area is the location of the study area. Water Conservation Area 3A is
designated as section 9 on this map.









A comparative observational study was determined to be the best scientific method

to investigate response of natural communities to environmental variations. Studies of

this type have a wide domain of inference and are conducive to the confirmational

hypothesis that Everglades plant communities are dynamic with respect to hydrology.

Twenty study plots were established based on a stratified random design, using

landscape-scale elevational (longitudinal) and peat depth latitudinall) gradients (see

Figure 2-2). The square plots are one kilometer on a side, a scale that sufficiently

includes variety of communities and ecotones.


Figure 2-2. Satellite composite of the study area in Water Conservation Area 3A.
Twenty plots were distributed with a stratified random design for the sampling
procedures.









Each plot contains two or three belt transects that crosses at least one community

boundary (see Figure 2-3). The design of the belt transects allows for the repeated

destructive sampling of each transect while avoiding the issues associated with repeated

measures. Every sample event allows for the removal of plant material from the field

under the assumption that previous sampling efforts have negligible effects on the

following sample. Each transect was established at a random location within a plot. The

number of samples within the transect vary from 10 to 34, depending on the length of the

transect.


Figure 2-3. An overlay of a square kilometer plot on satellite imagery. The blue dots
signify reference poles aligned with belt transects within the plot. Each
transect crosses at least one community boundary.










Methods and Materials

Sampling Regime

Destructive sampling along the belt transect was scheduled twice a year once at

the peak of the dry season (June), and once at the peak of the wet season (November)

which corresponds to the growing season. Sampling along the belt transects was

organized to avoid removal of plant material from the same place at any given time

during the study. The belt transects consisted of three parallel subtransects spaced four

meters apart (see Figure 2-4). Each subtransect could be sampled four times twice on

each side with staggered placement of sample locations. Sample locations were

randomly selected for each sample event. For example, November 2002 sample event

was randomly determined to be sample G, which corresponds to the right side of the

middle subtransect and starts from the zero meter point. Samples are spaced three meters

along the transect. Sample H would be staggered and correspond to the right side of the

middle subtransect and start from the 1.5 meter point.


Eotone
N L
D00000000 0000
m* amu m a Mma nE
3m"
3m * IM mK
S_ Transect Pales
O 0 U 0 U 0 0 0 Il OG



I I
3m
] 0.25 rm2san-le plts


Figure 2-4. A diagram of a belt transect consisting of three traversable subtransects.
Each sub transect can be sampled on four different occasions twice on each
side.









Sampling Methodology

The area of each sample is determined by a .25 square-meter circular hoop with

its center around a dowel placed at the sample point offset from the subtransect by a

meter. The dowel marks the sample point and allows for a reference when the hoop has a

tendency to float or deviate from its original placement. Floating vegetation is collected

from the sample first. After the floating vegetation is removed any material that may

subsequently drift into the sample area is disregarded. The rooted vegetation is then cut

at the soil surface and collected. All vegetation is collected in burlap sacks to allow some

air exchange for the evaporation of excess moisture. The vegetation remained more

resistant to disintegration and mold when stored in burlap rather than being stored in

plastic.

During the sample harvest, rotten material, determined by its structural integrity,

was discarded. For example, if the material when given a gentle shake did not maintain

any rigidity, than the material was deemed to be rotten and associated with the peat

substrate. Rotten vegetation was difficult to identify and proved almost impossible to

quantify. This structural integrity test provided consistent and comprehensive criteria for

determining viable plant material.

Samplers remained within the one meter wide subtransect path to avoid walking

on sample locations. Water depths were measured at each sample point and at the

transect start and end poles for reference points to be tied back in to the monitoring wells

for continuous hydrologic data for each sample point. Samples for the transect were

labeled and loaded into an airboat for transport back to a refrigerated storage unit. A total

of 1190 samples were collected from the study area per sample event.









Processing Methodology

Each sample was sorted by species and the numbers of individuals were tabulated

for each species. Counts for Eleocharis, Pontederia, Nymphaea, Bacopa, Crinum, and

woody vegetation were determined by the number of emergent stems. Cladium and

Typha counts were determined by the number of emergent culms. Utricularia and Chara

counts were impossible to determine in the laboratory and were tabulated as either

present or absent. Species for each sample were separated into paper bags, labeled, and

dried for at least two weeks in walk-in ovens set at 1400F. Dried plant material avoids

the inclusion of water weights that can vary considerably between species. After the

samples were dried, the dry biomass for each species was measured on digital scales to

the nearest hundredth of a gram. The dried plant material was then discarded into

compost. Biomass and count data were transcribed into an Excel spreadsheet in

accordance with appropriate quality control measures. See Table 2-1 for a complete

species list.

Data Preparation and Relativization

Community units are heretofore defined as the conglomeration of the samples

within one of the physiognomic community units represented in a transect. Each

community unit is designated with a plot number, transect number, apriori

physiognomic type, and sample event. A priori physiognomic types include cattail (C),

sawgrass (G), ghost island (I), prairie (P), slough (S), and tree island (T). These habitats

are important to aquatic macrofauna and are used differently by various suites of species

(e.g., Loftus and Kushlan 1987, Gunderson and Loftus 1993, Jordan et al. 1994, 1996).

The sample event is designated by where within the belt transect the community was

sampled for that sample collection. Sample events include: G November 2002, E -









Table 2-1. Complete species list for vegetative study in Water Conservation Area 3A.
Authority for plant names and status is from Wunderlin, R.P. 1998 Guide to
the Vascular Plants of Florida. University Press of Florida, Gainesville.
Includes unknown species that occur in more than one sample.
Scientific name Code Family
Bacopa carohniana BAC Scrophulariaceae
Blechnum serrulatum BLS Blechnaceae
Cephalanthus occidentalis CEO Rubiaceae
Chara spp. CHsp Characeae
Cladium jamaicense Alive CLA Cyperaceae
Cladium jamaicense Dead CLD Cyperaceae
Crinum americanum CRA Amaryllidaceae
Cyperus haspan CYH Cyperaceae
Dryopteris ludoviciana DRY Dryopteridaceae
Eleocharis elongata ELG Cyperaceae
Eleocharis spp. Elsp Cyperaceae
Fuirena breviseta FUB Cyperaceae
Hymenocallis sp. HYsp Amaryllidaceae
Leersia hexandra LEH Poaceae
Ludwigia spp. Lusp Onagraceae
Nymphaea odorata NYO Nymphaeaceae
Nymphoides aquatic NMA Menyanthaceae
Osmunda regalis OSR Osmundaceae
Panicum hemitomon PAH Poaceae
Paspalidium geminatum PDG Poaceae
Peltandra virginica PEV Araceae
Polygonum spp. POsp Polygonaceae
Pontederia cordata PNC Pontederiaceae
Potamogeton spp. PTsp Potamogetonaceae
Rhynchospora tracyi RHT Cyperaceae
Sagittaria lancifolia SAL Alismataceae
Salix carolniana SAC Salicaceae
Typha domingensis Dead TYD Typhaceae
Typha domingensis Alive TYA Typhaceae
Unk. Jointed stem UnkJS
Unk. Segmented rush UnkSR
Unk. Triangular stem UnkTS
Unk. Sawgrass-like grass UnkSG
Utricularia spp. UTsp Lentibulariaceae
Vallisneria sp. VAsp Hydrocharitaceae
Vine Unkn VIN
Woodwardia virginica WOV Blechnaceae









June 2003, D -November 2003, and J June 2004. For example, P18E2 refers to the

prairie community in plot 18, transect 2, sampled on event E (June 2003).

Prior to relativizing the data, I deleted samples that were missing count or

biomass data. This amounted to approximately 1% of the total sample. I also removed

samples that occurred adjacent to the ecotone. Locations of ecotones were determined in

the field by noting the samples between which dominant species appear and disappear,

indicating a different physiognomic type. The definitions of the apriori communities

were used to determine physiognomic types. Ecotones in the conservation area are

typically sharp and distinguishable allowing for minimal observer error in designating

ecotone location. This was done to remove samples that may be considered to be

transitional or not a typical representation of that community unit. Approximately 85%

of the samples remained in the analysis and are assumed to be representative of the

community units sampled.

The community data was converted into relative proportions for each community

unit sampled. Counts for each species in every sampled community were expressed as

the relative density of that species. For example, the relative density of Eleocharis in

community unit P18E2 equals the total count of Eleocharis stems in P18E2 divided by

the total count of all species in community unit P18E2. Relative biomass was calculated

in similar fashion equaling the proportion of the total biomass of a species in that

community unit to the total biomass of all species in that unit.

Averaging the relative density and relative biomass results in an importance value

for each species in each community unit. The advantage of using importance values in

ecological community analysis is that they are equally influenced by large biomasses and









large stem densities, so that species that differ in size and density can be compared within

the same sample unit. The disadvantage of importance values is that a species that has

large biomass values and sparse densities can have the same importance value as a

species with small biomass values and high densities (McCune and Grace 2002). Later I

will discuss how I tested the assumption that importance values can distinguish different

community stands, regardless of the vulnerability associated with importance values.

The resulting importance values for each species in each community unit were

transcribed into a data matrix for analysis (see Table 2-2). The data matrix is then ready

to be processed for multivariate analysis including clustering, indicator species analysis,

and ordination.

Table 2-2. An abridged species data matrix with importance values for species in each
community unit.
Units Species
BAC CEO CLA CLD CRA ELG ELsp
C2D1 0.512563 0.00000 11.09662 5.36581 0.00000 0.31718 45.34719
C2D2 3.045374 0.00000 20.70394 15.23677 1.30300 0.00000 0.00000
GOD1 13.694759 0.00000 38.46195 6.61118 0.00000 22.69860 0.25166
GOD2 0.000000 0.00000 70.29885 18.51327 5.63622 0.00000 0.00000
GOD3 1.566028 0.00000 60.90582 17.21278 0.00000 0.00000 0.00000
G10D1 31.561145 0.39661 27.15960 4.80673 3.86429 0.00000 0.25813
G10D2 0.000000 2.26382 79.35584 6.85827 0.00000 0.00000 0.00000
G10D3 0.000000 22.46180 59.52005 13.52616 0.00000 0.00000 0.00000
G11D1 4.807172 4.24228 22.93135 0.91426 6.15084 44.02458 0.35548
G11D2 0.000000 1.99904 76.09261 18.43122 0.00000 0.00000 0.00000
G11D3 17.612373 6.71708 36.94830 4.02826 15.79072 11.58430 0.00000
G12D1 51.287816 0.00000 14.06211 0.00000 0.00000 20.81183 0.60776
G12D2 0.000000 5.57459 60.60979 23.03208 0.00000 0.00000 0.00000
G13D1 3.280178 16.32098 34.92181 11.40519 0.00000 0.00000 0.00000
G13D2 0.223056 0.00000 32.10193 5.04516 0.00000 0.52885 30.94700
G13D3 22.684489 0.00000 57.68911 8.67561 0.73124 0.00000 0.00000
G14D1 0.000000 0.00000 5.43556 0.00000 5.95042 0.00000 0.00000
G14D2 35.224940 0.00000 2.88301 0.00000 0.00000 0.00000 2.29568














CHAPTER 3
CLASSIFICATION OF META-STABLE STATES

In order to determine how the vegetative habitats of the Everglades change in

response to continuously varying environmental conditions, identification of the meta-

stable states in which they are observed is required. Informal observation of plant

communities in Water Conservation Area 3A yields four physiognomic community

types: sawgrass, wet prairie, slough, and shrub/tree island. Ghost islands are also a

distinguishable community as old sawgrass ridges or islands that have experienced a

disturbance such as extreme flooding or fire. Ghost islands generally have some sparse

sawgrass, pickerelweed and buttonbush associated with them. Cattail (Typha spp.)

communities can also be observed, however there were only two community units

sampled that had cattail as a major component.

Subtle differences in the composition within these physiognomic types require the

statistical analysis of hierarchical classification. Classification through hierarchical

cluster analyses is necessary to identify these meta-stable community states and

recognize the subtle differences in community structure between these states. Meta-

stable states will be identified as discernable subunits of physiognomic types. In the

Everglades, these meta-stable states will be represented through the range of

environmental conditions that physiognomic types in WCA 3A exhibit. After the cluster

analyses, environmental conditions at those sites are investigated to produce profiles of

environmental conditions and thresholds.









Hierarchical Agglomerative Cluster Analysis

I used the species matrices referenced in Chapter 2 to apply a hierarchical

agglomerative cluster analysis. Each matrix contains importance values for every species

in each community unit for a particular sampling event. In all, there were four sample

events yielding four matrices with the same community units in each matrix.

Agglomerative clustering methods build groups hierarchically from the bottom up,

forming groups by fusing similar subgroups together (McCune and Grace 2002). The

optimal number of groups is calculated through an indicator species analysis. Cluster

analyses first calculate a matrix of distances between each pair of entities. Groups that

meet the minimum distance criteria are merged and their attributes combined. The

merging process continues until there is only one group. The result is a dendrogram

complete with a distance measure (from the distance matrix). The distance measure is a

function of the information lost at each clustering step (Wishart 1969).

The cluster analysis was performed on the PC-Ord software using a Euclidian

(Pythagorean) distance measure. Ward's linkage method was chosen for its

combinatorial compatibility. Ward's method also conserves the properties of the original

space as group attributes merge, keeping the Euclidian distances consistent throughout

the analysis (Wishart 1969). Community units were color coded by apriori classification

of physiognomic community types based on observation in the field. See Figures 3-1, 3-

2, 3-3, and 3-4 for the resulting dendrograms for each sampling event.

Testing Importance Value Assumptions

As mentioned previously, importance values have one major disadvantage in that a

large, sparse stand has the same value as a small, dense stand. Because the purpose of

this study is to discriminate between vegetative community states by their structure, those














November 2002 Cluster Dendrogram

Distance (Objective Function)
2E400 6.8E404 1 4E+05 2E405 27E4

Information Remaining (%)
100 75 50 25 0


with Eleocharis/Panicum


Blue Slou~g
Pink Wet Prairie



Red Cattail


SGhat IslandJ
S1403
S5G3
Sb6G1

-l Deep Slough


f E. elogata Prairie



"--





-- E. elogata. Slough










Figure 3-1. Cluster dendrogram from November 2002 sampling event. Community units

are listed on the left and color coded with respect to their a priori designation.


7:


05


-- I-.,

















June 2003 Cluster Dendrogram

Distance (Objective Function)
1.8E--00 6.9E--04 1 .4E--05 2 1 E+S05 2.7E--05

Information Rermaining (9')
1O0 75 50 25 O

C2E1
P2E32E2 agaPrairie Legend
P2E3
P1 7E12 -_ Blu. Slou6h
SbI 7E2
P1 E3 Pi.k Wet Prairie
P7E2
P3E1
S1 7E2
S9E2
P1 7E3Pai/Pap
S' 7E3 :Eleochri. Prair Red- Caail
S3E2
P El Eleokri. PPrairi
P2EI1
Pb3EI
C2E2

Sawa CGo.t I.land







Sawra.ia with

P1 E2 Eleoklri./Cria




















E lH eaHv Sawria.








P1 6E2
P1 E E. elonate Prairie
P1 9E2
SOE3
S8E2
Sb9E2
S1 3E2
SSE3
S6E2
S4E1
POE2
P1 3E1
S1 I E1
S4E1
SI 2E1I
S14E1
Sbl 4E2
S9E1
S1 OE2
Sbl OE2
S13E3
Sb4EI
S1 4E2
Sbl OE1
Pbl 9E2
S1 1 E3
SOE1 E. elonajta
SI1 El2
8=Ei
SbII E1E3
S1 8E3
P1 9E1
SOE2
SI OEI
S1 OE3
SSE2
S1 9E3 Dp SI k
S1 2E2 Deep Sloual
S14E3
S4E3
Sb6E1
S1 E2
S5E1
SSEE3
S6E1
S6E3

Figure 3-2. Cluster dendrogram from June 2003 sampling event. Community units are

listed on the left and color coded with respect to their a priori designation.













November 2005 Cluster Dendrogram
Distance (Objective Function)
5 4E-01 7 3E+04 1 5E+05 22E+05 29E4

Information Remaining (%)
100 75 50 25 0


Sawrra-. Prairie


SPrairie


Legeud

Blue Sl0ou
Pink Wet Prairie



Red Cattail


E. elongata Slough


Figure 3-3. Cluster dendrogram from November 2003 sampling event. Community units
are listed on the left and color coded with respect to their a priori designation.


05













June 2004 Cluster Dendrogram

Distance (Objective Function)
34E+00 79E+04 1 6E+05 24E+05 32E4

Information Remaining (%)
100 75 50 25 0


Island


Blue Slou6h
PiLk Wet Prairie


Red Cattail


S15J1


S18J1


Prairie


Figure 3-4. Cluster dendrogram from June 2004 sampling event. Community units are
listed on the left and color coded with respect to their a priori designation.


+05










habitats must be recognized as distinct states in the analysis. To test the assumption that

importance values will discriminate between these community assemblages, I plotted

sawgrass stands from one sample event using the two components of importance values

(relative biomass and relative density) on each axis (Figure 3-5).


All sawgr

100 + -
90- 2
QA 0 A A A


0 *
o
0r
AI


m3
A4
5
X9
,11
+12
-14


.I 1 11
0 10 20 30 40 50 60 70 80 90 100 1
Relative Der

Figure 3-5. A scatterplot of sawgrass communities sampled in November 2002. Axes
correspond to percent relative biomass and percent relative density. Each
point represents one sawgrass unit. Each sawgrass type resulting from the
cluster analysis is coded in the legend.

The points are color coded to match the groups that they were assigned to via the

cluster analysis discussed in the following section. If the importance values were

distinguishing the differences between large/sparse and small/dense stands, then the

points of the same group would be clustered together on the bi-plot. While the

assumption does not hold up perfectly, there is a definite clustering effect. Obviously,

the importance values are not distinguishing the difference between the large/sparse and

small/dense stands due to the previously stated disadvantages involving importance

values, but rather certain associative species may be more prevalent in one or the other. I

conclude that it is safe to assume, for the purpose of this study, that the disadvantage of









using importance values in this study is not relevant due to mitigating variables such as

species associates.

Indicator Species Analysis

Following the generation of cluster dendrograms, an indicator species analysis

provides a subjective determination of the optimal number of groups based on how well

any of the species acts as a significant indicator of a group. Dufrene and Legendre's

(1997) method of calculating species indicator values combines information on the

concentration of species importance values and the faithfulness, or endemism, of a

species to a particular group. Indicator values are tested for statistical significance using

1000 Monte Carlo randomizations.

Each sample event was subjected to an indicator species analysis on the PC-Ord

software 29 times, testing statistical significance of every species from 30 groups to 2

groups. The program provided a table for each species and a p-value, or the proportion of

randomized trials with an indicator value equal to or exceeding the observed indicator

value. Average p-values for each run and number of significant species (p<0.05) were

plotted in a spreadsheet (see Appendix A). Both plots were used to determine the optimal

number of groups to prune the cluster dendrogram. Low average p-values across the

suite of species, and high numbers of significant species determined the number of

groups. Thirteen groups of community units developed from the November 2002, June

2003, and June 2004 sampling data. Fourteen groups of community units developed

from the November 2003 sampling data.

The indicator species analysis also produces a table of indicator values, or the

percent of perfect indication based on combining the values for relative magnitude and









relative frequency of importance values, for each species. These tables were translated

into graphical signatures for each community state (See Appendix B). The resulting

structural signature was an important guide to describing the groups, or meta-stable

community states. Species with high indicator values (>15%) were significant species of

that community. The resulting community states and their descriptions are shown on the

cluster dendrograms in Figures 3-1, 3-2, 3-3, and 3-4.

Matching Similar Community Descriptions Between Sampling Events

The groups resulting from the cluster analyses were translated into meta-stable state

entities, as defined in Chapter 1. In this study, the meta-stable states are groups of similar

clusters that were consolidated across sample events and in some cases within events. As

a result, the community states identified represent the range of states that occurred

throughout the landscape through different seasonal and annual environmental

conditions. Some of these states were persistent through the study period, and some

occur infrequently. The following is a description of the methods used to group clusters

together across and within sample events, and define them as meta-stable community

states.

A general description of each group arising from the cluster and indicator species

analyses was constructed for each sampling event. In order to facilitate matching

community states between events, a multi-response permutation procedure (MRPP) was

utilized. MRPP is a nonparametric procedure for testing the hypothesis of no difference

between entities (Biondini et al. 1985). Each community unit was tested for

heterogeneity over time. In other words, if a community unit was similar over the four

sampling events, or minimal change had occurred over the course of monitoring, then









that unit would receive a low p-value. Units that exhibit little change can be associated

with the same meta-stable community state described in the indicator species analysis.

The MRPP procedure helped with establishing only four of the community state

descriptions and was insufficient in finding similar states between sample events.

Another approach, used as a complementary analysis to the MRPP, was an agglomerative

clustering process applied to all of the groups resulting from the cluster analyses of the

four sample events. Similar community states should cluster together. Most of the

resulting clusters of groups included only one group from each sample event,

corroborating that those states are unique within sample events and indicating that they

are common throughout the study period. These were designated as the community

states of the major physiognomic types. Two clusters included groups from the same

sample event that were similar: the heavy sawgrass group consists of two combined

clusters in June 2003 and November 2003, and the E. elongata slough consists of two

combined clusters in November 2002. Other community groups may have been

represented only two or three times over the four sampling events. This method of

defining groups over multiple sampling events provided an objective approach to

comparing group structural signatures, and was a potential check on the first set of cluster

analyses. If the differences of groups within sample events are less than the differences

among groups then those clusters can be essentially combined. See Table 3-1 for a list of

meta-stable community states and the frequency of community units in each event.

Discussion of these results continues in chapter 5.









Table 3-1. Meta-stable community states and their frequency at each sample event.
Nov. June Nov. June
2002 2003 2003 2004
Eleocharis Prairie 7 4 4 8
Panicum/Paspalidium/Eleocharis Prairie 9 12 5 11
Eleocharis elongata Prairie 6 13 23 13
Deep Slough 9 16 7 14
Sawgrass Prairie 0 5 11 0
Panicum Slough 4 0 0 5
Eleocharis elongata Slough 31 22 15 0
Shallow Slough 0 0 6 14
Sawgrass with Eleocharis/Panicum 8 0 0 7
Sawgrass with Bacopa Ludwigia 11 5 9 11
Sawgrass with Eleocharis 8 10 6 5
elongata Crinum
Sawgrass monoculture 19 25 27 17
Sawgrass Ghost Island 13 8 1 1
Ghost Island 2 4 4 18
Tree Island 0 7 9 3

Distribution of Meta-Stable States Across the Landscape

In an impounded hydroscape like Water Conservation Area 3A, hydrologic

conditions across the landscape can differ from one end of the drainage basin to the other.

Environmental gradients in WCA 3A such as substrate type and peat thickness in

conjunction with varying hydroperiods should distribute plant communities throughout

the landscape accordingly. Therefore, some insight as to the hydrology of the meta-

stable states that were identified in the cluster analysis may be gained by mapping their

locations. Figure 3-6 splits the study area into four quadrants and identifies the

proportion of each meta-stable state by physiognomic type in each quadrant. Each

quadrant roughly represents the high and low ends of the hydrologic and substrate depth

gradients (i.e. the northwest quadrant is short hydroperiod/shallow peats; southwest

quadrant is long hydroperiod/shallow peats; northeast quadrant is short hydroperiod/deep

peats; southeast quadrant is long hydroperiod/deep peats). The rationale for dividing the



























Slough
* Deep slouh
h Shallow slough
O Panicum slough
O E.eloneata sough
Prairie
" E. elonuaLa prairie
* Elecharis spp. prairie
l PA-H PDG/Elsp prairie
l Sawgrass prairie


Sawgrass
* SG with Elsp and PAH
M SG with Elsp and CRA
L Heavy Sawgrass
O SG with BAC and Lusp
Island
N S'aw'rLLs Ghost Island
N Ghost Island
l Tree Island


Figure 3-6. Shows the distribution of meta-stable states by physiognomic type into four
quadrants of the study landscape.


--









landscape into square quadrants is simply for convenience since the extent of

impoundment effects on the hydroscape have not been determined and a detailed survey

of peat depths in the conservation have yet to be produced.

Examination of the distribution of meta-stable states shows several insightful

trends. Sawgrass communities showed little difference in change across the landscape.

There was a slight increase in the share of Panicum and Bacopa associated sawgrass

communities towards the western extents of the impoundment area. Among prairie

communities the Eleocharis cellulosa state was confined strictly to the west whereas the

Eleocharis elongata states increased dramatically further east, which resembles the peat

depth and substrate type gradients. This suggests that the state that a wet prairie might

exhibit has a lot to do with substrate properties. Not surprisingly, deep sloughs

dominated by water lilies were more prevalent in the longer hydroperiod south. Tree

island types of islands dominated by woody vegetation and shade tolerant herbaceous

species were virtually nonexistent in the southeast.














CHAPTER 4
MULTIVARIATE ANALYSIS AND RESULTS

Hydrology

Selecting Hydrologic Variables

The multivariate approach to statistical analysis of data allows the researcher to test

simultaneously for significance among variables. There are many factors that may have a

hand in the makeup of community assemblages, but hydrology is the main environmental

force driving Everglades community structure. In order to include hydrology in the

statistical analysis of data, it was necessary to first determine which metrics can best

represent the hydrology of the Everglades. A set of hydrological variables was selected a

priori and calculated for each sample unit. Table 4-1 lists the hydrologic variables

selected along with the ecological rationale underlying their use.

Hydrology can be described to reflect one or several different aspects of flooding:

depth, duration, time, and the magnitude of extreme events of flooding and drought

(Richter et al. 1997). The fraction of the year that a given site is inundated was chosen

due to the prominence in the literature, especially regarding the Everglades and ease of

calculation (Toner and Keddy 1997). The mean depth of flooding, whether it be above or

below the ground surface was selected to represent the depth of water that species must

adapt to. Range of depths defines the amount of water level fluctuation that occurs

during a year. Flooding and drought event legacies are a function of the length of time

used for hydrological records to calculate hydrology. Therefore, inundation times were

calculated using one, three, five, and ten-year time records. If there is a consistent period









Table 4-1. Hydrological variables with abbreviations.
Abbreviation Variable Units Ecological implications
(T)yr.inun* Number of days per days Reproduction of some species.
year during which Exposure of soils to oxidation
flooding occurred processes.
(inundated)
MeanDepth Mean depth of feet Establishment of aquatic vs.
flooding over a ten emergent species.
year period
max Average 7-day feet Anaerobic stress in plants.
maximum water Spatial extent of extreme
depths over 10 years conditions.
datemax Average Julien date of day of Coordination of hydrologic factors
maximum water levels year with temperature and photic
factors.
highdurat Duration of high water days Anaerobic stress in emergent
levels wetland species.
min Average 7-day feet Indication of potential oxidation of
minimum water depths soils.
over 10 years Reproduction opportunities.
datemin Average Julien date of day of Coordination of hydrologic factors
minimum water levels year with temperature and photic
factors.
lowdurat Duration of low water days Opportunities for emergents to
levels develop and compete against
floating leaf aquatics.
Exposure of soils to oxidation
processes.
variation Average annual range feet Amount of variation in the
of depth environment that must be
___tolerated.
(T) denotes the length of record used to calculate the metric. Periods of time used to
measure inundation time are 1, 3, 5, and 10 years. All of the other metrics are
calculated using a 10-year time span.

of time that hydrology affects into the future then it would be revealed in the

classification tree analysis. Duration of typical high and low water level events were

calculated using the algorithms in the IHA (Indicators of Hydrologic Analysis) software

(Richter et al 1996). Finally, extremes during the average year were calculated as seven-

day highs and lows. Timing of these extremes was considered. These metrics were

chosen due to literature citing the importance of extreme, stochastic events in the









Everglades that occur periodically and play an important role in community dynamics by

offering opportunities for species recruitment, movement, and nutrient availability that

otherwise are unavailable (Kitchens et al. 2002).

Calculating Hydrologic Variables

The hydrologic data collected was processed extensively in order to be applied to

the multivariate analysis as the metrics that relevantly described the hydrology of the

sites for the community sample units. The monitoring wells were set up shortly

following the first sample event. Water data had to be extrapolated up to ten years prior

to November 2002. The well data also needed to be applied to the various sample units

that it was monitoring. The community units required classification of their elevations in

order to get depths that were relative to the well monitoring that plot. The following is a

discussion of the methodology used to calculate those hydrologic profiles.

Hindcasting using neural networks

To calculate the hydrologic variables mentioned previously, precise water data

dating back ten years was needed for the study plots. Prior to the study there were three

permanent gauging stations, established by various state and federal agencies, within the

study area (See Figure 4-1). Two of these three stations, 3-64 and 3-65, had been

established and collecting data longer than ten years prior to 2002. The agency

monitoring stations and their data was not sufficient for producing hydropatterns at each

of the 20 study plots. A network of monitoring stations needed to be established that

could provide accurate hydrologic data at the community scale.

Although the vast majority ofWCA 3A is flooded most of the time, a flat pool of

water cannot be assumed over such an expansive landscape. Since the plot size was a












































Legend


RegIon Bournda'Jy
SFbI alPdYpn
A pa Rmnml CGagr3 3ljdn
f rrmm av Dda Lnmm


Figure 4-1. Green tangles represent the monitoring stations set up by various agencies.
These stations upload real-time data to the web daily. Yellow circles indicate
the temporary stations that were established in December 2002.









kilometer squared, a flat pool was assumed within a kilometer radius. Temporary data

loggers, designed to monitor water depth were installed in December 2002 at each plot

with a couple of exceptions. Plot 7 and plot 4 each were within one kilometer of an

agency monitoring station. Plots 13 and 15 shared a station, as well as plots 10 and 11.

The wells are driven through the peat substrate to the limestone. The peat soils

usually provide enough stabilization to prevent the wells from leaning even in tropical

storm force winds. In the few areas where the substrate is insufficiently thick, the wells

are stabilized by makeshift tripods. The data loggers are attached to wells that measure

surface water depths from its base. The depth from the substrate is simply calculated by

subtracting the amount of the well that is buried in the peat. The data loggers measure

the water depth at their respective stations twice a day. Every month, data is downloaded

from the data logger to a laptop.

Neural networks are a pattern recognition statistical application that search for

patterns over time with the use of multiple model runs. They are especially useful in

situations that have static as well as dynamic properties (Bishop 1995). The landscape

position remains static between the monitoring stations, yet the data are dependent with

time. After a significant amount of data was collected from the new monitoring stations,

they were joined with the agency stations in a neural network to produce a "constructed-

topographical" model of the water surface. Using neural networks, models were applied

to hindcast, or extrapolate, the depth of each of the new stations to produce hydrologic

data for all of the study plots. Table 4-2 contains the results of the neural network

analysis.









Table 4-2. Neural network model statistics for each station hindcasted. PME (percent
model error) = RMSE (root mean-square error) / (range of measured data).
Plot n R R2 Mean RMSE PME (%)
Error (ft) (ft)
0 694 0.995 0.990 0.028 0.072 2.6%
1 352 0.999 0.997 0.001 0.041 1.5%
2 594 0.992 0.985 0.026 0.067 3.0%
3 301 0.996 0.992 0.009 0.066 2.7%
5 563 0.999 0.997 0.005 0.044 1.5%
6 682 0.994 0.988 0.034 0.075 3.0%
7 222 0.997 0.994 0.004 0.038 1.8%
8 690 0.997 0.995 0.011 0.049 1.8%
9 567 0.993 0.986 0.007 0.082 3.1%
11 658 0.996 0.992 0.026 0.070 2.3%
12 603 0.998 0.996 0.025 0.055 1.6%
14 674 0.998 0.996 0.052 0.072 2.2%
15 659 0.997 0.994 0.013 0.064 1.9%
16 377 0.997 0.994 0.004 0.054 1.8%
17 392 0.991 0.982 0.001 0.089 2.8%
18 613 0.988 0.976 0.033 0.111 3.8%
19 426 0.994 0.988 0.020 0.080 2.7%

Extrapolating from well data to sample unit data

The continuous well data that dated back at least 10 years, produced by the neural

network models, corresponded to a point within each plot. Data for each community

sample unit on transects within the plots needed to be calculated. During each sample

event, water depths were taken at each vegetation sample. For every community sample

unit, those depths were averaged to get the average depth of that community unit for that

date. From that date, the water depth at that site was extrapolated from its corresponding

well for ten years into the past by subtracting the difference between the well depth and

the average community unit depth and applying it as a constant differential. As a result, a

10-year historical hydrologic record for each community sample unit was created. It was

assumed that the water within the square kilometer area of a plot is a flat pool in order to









make this extrapolation. The hydrologic variables mentioned previously were calculated

using the resulting records.

Nonmetric Multidimensional Scaling

After the hydrologic variables were calculated for each community unit, the matrix

of potential environmental drivers was complete. Table 4-3 lists the environmental

factors used in the multivariate analysis of the community data. Some of the variables

were relativized to vary within two orders of magnitude (0-10). Transformations such as

this are imperative in producing rational results in multivariate community analysis

(McCune and Grace 2002).

Table 4-3. Environmental variables used in the multivariate analyses and how they were
relativized if a transformation was appropriate.
Environmental variable Abbreviation Range Relativization (if any)
Peat Depth PeatDepth 0.07-5.74 feet no relativization
Mean Depth MeanDepth 0.09-3.49 feet no relativization
Minimum Water Depth min -0.97-2.38 feet no relativization
Maximum Water Depth max 0.82-4.29 feet no relativization
Timing of Minimum datemin day 147-153 Range from 0 (January
Water Depth 1) to 1 (December 31)
Timing of Maximum datemax day 278-311 Range from 0 (January
Water Depth 1) to 1 (December 31)
Duration of High Water highdurat 3.25-89 days Proportion of year
Levels (range from 0-1)
Duration of Low Water lowdurat 2.25-115.8 days Proportion of year
Levels (range from 0-1)
Average Range of Water variation 0.70-3.11 feet no relativization
Depths in a year
Inundation times (T)yr.inun 51%-100% of Proportion of year
_year (range from 0-1)

Ordination of community data organizes the structural composition of the sample

units into a multidimensional space using nonparametric scaling techniques. The output

of sample units as points in space is the result of multiple runs using various numbers of

dimensions to find the best fit of the data on a hypothetical landscape that minimizes the









"stress" of a solution. Stress is a measure of departure from monotonicity in the

relationship between the distance in the community data and the distance in the resulting

multidimensional space. This particular method of ordination is well suited to

nonnormal, arbitrary scales that are commonplace in community ecology due to its

distance-preserving properties (Clarke and Ainsworth 1993).

A Sorenson distance measure was used for the ordinations because community

analyses in ecology require a metric, city-block distance calculation to handle the

intricacies and occasionally long distances that can occur between species. The program

PC-ORD provided the algorithms for the NMS procedures. See Mather (1976) and

Kruskal (1964) for the methods. The program supplied a random starting configuration.

Each analysis was run 15 times with the real data and a Monte Carlo test was performed

30 times for comparison. Appendix C provides the results of the Monte Carlo tests and

the probabilities that a similar final stress could have been obtained by chance. Scree

plots in Appendix C show the stress reduced per dimension added and provide a visual

check on the stability criterion (.0001 st. dev. in stress over last 10 iterations).

The ordination procedure was first performed using all of the community data and

grouped by the physiognomic groups that resulted from the cluster analyses. The whole-

scale community ordination grouped the physiognomic types distinctly in a two-

dimensional space. Figure 4-2 shows the ordination plot and the vectors that represent

significant environmental gradients (r2 > 0.2) that correspond to (in this case) axis 2.

Minimum depth and mean depth were the environmental variables that explain the

distinction between sample units and communities in Water Conservation Area 3A.








41



Test ordination whole


lA 2




A 03J2 "A R a B, E GAES
GoA Glu A A A 3
C2E1 0203 G G3 12E 2
G0J U3J01 GEGD2 lu C

0b 21 P J3 A 0 D 0 29I P A TA .. 912
J1TeJE1 1012J1 P293 Pl D1 G 1 ~G 0 P0D1 P2D3t
P132 P001 G2J1 12S01I 2 A A
A Pb301 1, A02 B
A A P2G1 P23 PTsp A P11G1 P1D3
Pb3D1 A T9E2 A p + GDU A A
T2 +4E GEG 1rD1t
A13 P b+1D3 011
P1543 A A1 P 02 014E J1401
As
A C 1 P + +,


PIAIA a l A FHT2 A
A T o-i I aX .9E2 C, irjj POC62
A P1102 p o
A3 2PI
rlG 'j. *[2 .'l 0001 JG 023M



--ir nm-tl-ra ie (mini 'p ,
Sa p. t







sawgrassphytn oe stype 4f island-pypsi m. There i no c o
e '.1 + the ori__P..n:ai ax.e .,


AL, P sD2

Axis 1

Figure 4-2. A whole scale ordination plot of the community sample units. Triangles
represent individual sample units and crosses represent species. The key to
the legend: 1--prairie physiognomic type; 2-slough physiognomic type; 3-
sawgrass physiognomic type; 4--island physiognomic type. The
environmental gradients (minimum depth and mean depth) are represented
with red vectors closely aligned with axis 2.


The ordination procedure was repeated using each of the physiognomic types


separately to attempt to discover the relationship between states of each physiognomic


type and the environment. Figure 4-3 is an ordination plot of island-type communities.


The three meta-stable states of island-types are best fit into a three-dimensional space.


Shown are ordination plots of axis 1 and axis 2 and a plot of axes 2 and 3, providing a


less than clear distinction between states of island-types. There is no correlation of


environmental variables to the ordination axes.








42





Island-type ordination Island-type ordination
NewCom NewCom
A 1 A 1
2 2
3 3
4% p
+ ++


A A 4








Axis 1 Axis 2
+ +

A A A +





2- t isl ; 3 e i d.











Figure 4-4 shows the ordination plot of slough communities grouped by their

respective meta-stable states. The procedure resulted in a fit within two dimensions, but



did not recognize an overall significant environmental variable that explains the

distinction between states No significant environmental gradients were found to explain





An NMS ordination plot shows the four states of sawgrass-type communities in


multivariate space (Figure 4-5). A three-dimensional plot was determined as the most

stress-reducing solution. The groups are clearly distinct in a plot of axes 2 and 3, though


no environmental variable met the criteria to be established as a significant gradient in







peat depth gradient, which is strongly correlated with longitude, is aligned with axis two
the position of community units in ordination space.



















peat depth gradient, which is strongly correlated with longitude, is aligned with axis two













































Figure 4-4. Slough-type ordination plots. Triangles represent individual sample units and

crosses represent species. The key to the legend: 4-deep slough; 6-
Panicum slough; 7-E. elongata slough; 8-shallow slough.


Sawgrass-type ordination
NewCom
A 9
10
Vfp A 11
.+ A A12


A
A\ A


A A


A UESREI A A A


+ + NMA ZX 8 .
A-1 ^^~rfA A A

NIN A^ + A
A A +T Ako A A

noD ++
+


Axis 2

Figure 4-5. Sawgrass-type ordination plots. Triangles represent individual sample units
and crosses represent species. The key to the legend: 9-sawgrass with
Eleocharis sp./Panicum; 10-sawgrass with Bacopa/Ludwigia; 11-sawgrass
with E. elongata/Crinum; 12-sawgrass monoculture.







44


Prairie-type ordination
NewCom
l 1
e Xteypt 2
3







.J1
c/N
+ +

-A A
A CRA









Axis 1
Figure 4-6. Prairie-type ordination plots. Triangles represent individual sample units and
crosses represent species. The key to the legend: 1--Eleocharis sp. prairie;
2-Panicum/Paspalidium/Eleocharis sp. prairie; 3-E. elongata prairie; 5-
sawgrass prairie. Environmental gradients shown as red vectors closely
aligned with axis 2.

(r2=0.386 along axis two). Eleocharis sp. prairies and sawgrass prairies are clearly

situated in shallow peat substrates. In the middle of the plot along axis two is the

Panicum/Eleocharis/Rhynchospora prairie located intermediately with respect to peat


depths. Eleocharis elongata prairies are characterized by deep peat depths and located at

the top of the plot along axis two. It can be concluded that the species of Eleocharis are

distributed along the peat depth gradient, determining what state of wet prairie will

manifest when the conditions prevail for a wet prairie.

Inspection of the ordination results can give insight into the strength or weakness of

the distinction between meta-stable states determined by the cluster analyses. The results

also provide indications of prevailing driving forces determining the community structure

of such states.
of such states.









Classification Trees and Characterization of Meta-Stable States by Environmental
Variables

The use of classification trees can determine the environmental variables that

distinguish different groups based on community structure. In the case of this study

classification trees define the groups based on the designation they were assigned by the

cluster analyses. The environmental variables are analyzed for each group and branches

are created in the classification tree where appropriate to predict what community state

will result under significant environmental conditions. The result is a dichotomous key

classifying states by the variables that distinguish the groups quantitatively (Urban 2002).

More specifically, environmental thresholds can be determined for each meta-stable state

at the leaf of the classification tree. For the example of this study the dependent variable

is the community state, whereas the predictor variables are the environmental variables

listed in Table 4-3. A complete interpretation of the classification trees for each

community state is included as separate sections. The importance of various

environmental variables is graphed in Appendix D and indicates the significance of a

specific variable in indicating a meta-stable state. A discussion of the classification trees

and their implications is included in Chapter 5.

Figure 4-7 is a classification tree of all of the identified meta-stable states. A

combination of environmental variables distinguishes each community state. This

particular classification tree is meant to be only an overview and should be a general

guide to understanding the relationship between physiognomic types and their

community states. The explanatory power of this tree is minimal (variation explained =

23%) due to the number of states it was meant to classify. Classification by












physiognomic type is a more useful method of determining the factors that drive


community state structure.






ELEO prair
ELG pr-lr'
ELG bu.gh
DAN slaga
SG G .,,
SG paID.
SchEnBACeLLID
SGlkhELEOIPAN
SCl hELGiC RA
S:IIomSL
Tnebbnd


ttalrax-c arlI 1 d alB'naxn' 81A I Pealtepin2.Al^&l PEa10cclhi2.4lEj





B9)6 K


_"____ I .... 1 I
--- -- e ls M

EL -.*. Pi-'t rPD ELE-L r."'- 136)
i' i ;i -', SG mono


ELG prae SG nvr
4 I) (18)
Figure 4-7. Classification tree for the meta-stable states on 10 environmental variables.
The number of sample units in each leaf are shown in parentheses below each
bar graph, which shows the compositions of communities within each leaf.


Slough Physiognomic Type


Figure 4-8 is the classification tree for the slough physiognomic type. The first


node in the classification tree is the timing of the maximum water depth during the year


(before or after October 21st). Variation of mean depth and the depth of maximum water


levels are the second level of branching in the tree. Finally, other hydrologic factors


work in concert to determine slough meta-stable states. The misclassification rate of the


models was 36% and the amount of variation explained was 54% (1-Relative Error).













Deep slough
ELGslough
PAN slough
ShallowSL














max>2 91







Deep slough
(37)


datemax<0.804111 datemax>0.80411


max<2 91



minimum>0.73E minimum<0.735




ELGslough PAN slough
(6) (4)


variation<2.29 variation>2.29


ShallowSL
(13)


ELGslough
(85)


Figure 4-8. Classification tree for 4 slough community states on 10 environmental
variables. This model was pruned from a tree size of 7 leaves to five, based
on a cost complexity pruning curve, selecting the smallest tree within one
standard error of the best. The number of sample units in each leaf is shown
in parentheses below each bar graph, which shows the compositions of
communities within each leaf.

Deep sloughs

Deep sloughs are typically located in the southern extent of the study area, but


found throughout the basin. Associate species include Nymphoides aquatica, Nymphaea


odorata, and Utricularia sp. According to the classification tree analysis deep sloughs


occur in areas where the timing of maximum water depths is earlier than the 293rd Julien


day of the year (October 21st) and maximum depths are at least 2.91 feet. The


misclassification rate of deep sloughs is very low.









Eleocharis elongata sloughs

This particular slough community state occurs everywhere except in the

southeastern portion of the water conservation area, which is typified by deep peats, long

hydroperiods, and floating mats. Associate species of this community include Eleocharis

geniculata, Nymphaea odorata, Utricularia, and Hymenocallis. According to the

classification tree they occur where variation in water depths is less than 2.29 feet during

the year.

Panicum sloughs

Panicum hemitomon sloughs are mostly limited to the northwestern reaches of the

study area. Associate species of this community include Panicum hemitomon,

Paspalidium geminatum, and some of the typical slough species (N. odorata, N. aquatic,

and Utricularia). According to the classification tree this community state occurs in

relatively shallow site, requiring maximum water depths of less than 2.91 feet and

minimum water depths of less than 0.735 feet.

Shallow sloughs

Shallow sloughs are evenly distributed across the landscape. They consist of

mainly Utricularia, but associates may include P. hemitomon, P. geminatum, N. odorata,

and N. aquatica. According to the classification tree they prefer maximum depths late in

the year and occur in sites with greater variation in water depths throughout the year.

Wet Prairie Physiognomic Type

Figure 4-8 is the classification tree for the wet prairie physiognomic type. The first

node and most important explanatory variable distinguishing conditions in wet prairie

communities is peat depth. Duration of high water levels, hydroperiod, timing of

maximum water levels, duration of low water levels, hydroperiod and annual average of











minimum water levels are the hydrologic factors that determine community state of wet


prairies. The misclassification rate of the models was 28% and the amount of variation


explained was 54% (1-Relative Error).


ELEO prairie
=LG prairie
AN.PDGELEO prairie
SG prairie


Lowdurat<0.19863


ELG prairie
(44)


ELG prairie
(6) PANIPDGIELEO prairie
(36)


Sinun

SG prairie
(11)
ELEO pradtfRNPDGIELEO prairie
(16) (5)


Figure 4-9. Classification tree for 4 wet prairie community states on 10 environmental
variables. This model was pruned from a tree size of 11 leaves to eight, based
on a cost complexity pruning curve, selecting the smallest tree within one
standard error of the best.

Eleocharis sp. prairie

This particular state of wet prairie is confined to the western half of the study area,


concentrated mostly in the southwest. The dominant species is Eleocharis cellulosa with


some Eleocharis equisifoides occurring on occasion. According to the classification tree,


this community state flourishes in areas that have short durations of low water levels, and


ELG prairie
(5)


.2136









low peat depths. Other criteria include minimum water levels greater than 0.75 feet and

long hydroperiods.

E. elongata prairie

This community state occurs throughout the study area, however it tends to be

concentrated in the eastern reaches and is the only type of prairie in the northeastern

quadrant. This prairie state is dominated by Eleocharis elongata, but can also include

associate species such as Bacopa caroliniana, Hymenocallis, N. odorata, P. hemitomon,

P. geminatum, and Utricularia. According to the classification tree, E. elongata prairies

occur in areas with a long duration of low water levels and intermediate peat depths.

Panicum/Paspalidium/Eleocharis prairies

These prairies occur mainly in the western reaches of the study area. Dominant

species include P. hemitomon, P. geminatum, and E. cellulosa. Associate species include

B. caroliniana, N. odorata, N. aquatic, Hymenocallis, and Utricularia. According to the

classification tree these prairies tend toward peat depths less than 2.67 feet, shorter

hydroperiod sites than sawgrass prairies, and lower minimum water levels than E.

cellulosa prairies.

Sawgrass prairies

Sawgrass prairies occur almost exclusively in the southeast portion of the study

area. They are similar in structure to Eleocharis sp. prairies except for the inclusion of

sawgrass. According to the classification tree, sawgrass prairies occur in shorter

hydroperiod sites than the other prairie communities and in peat depths less than 1.21

feet.







51



Sawgrass Physiognomic Types

Figure 4-10 is the classification tree for the sawgrass physiognomic type. The first


node in the classification tree is peat depth. Ten-year hydroperiod, timing of extremes


and maximum water levels are also considered in the classification scheme.


Misclassification rates for this community state were low. The misclassification rate of


the models was 28% and the amount of variation explained was 54% (1-Relative Error).



PeatDepth>2.5748 PeatDepth<2.5748
SG mono
'...r.b'. LI.' datemax>0.773973 datemax<0.773973 max>2.875 max<2.875
SGwIthELG/CRA




PeatDepth>4.2476 PeatDepth<4.2476
tenyr.inun0.980868
datemin<0.412329 datemin>0.4123awithBACILUD
(3) PeatDepth>1 2136 PeatDepth SG mono
(25)
SG mono
(34) SO oo
(6)







SG mono SGwithELGICRA
(26) (41)



SGwithBACAUrGwithELEOPAN
(23) (10)



Figure 4-10. A classification tree for 4 sawgrass community states on 10 environmental
variables. This model was pruned from a tree size of 14 leaves to eight, based
on a cost complexity pruning curve, selecting the smallest tree within one
standard error of the best.


Sawgrass monoculture (heavy sawgrass)

This sawgrass community state is evenly distributed throughout WCA 3A. It is


dominated by Cladiumjamaicense, but associate species include Crinum americana and









Ludwigia spp. According to the classification tree, heavy sawgrass communities are well

distributed across all conditions. Heavy sawgrass is generally found in deeper peat.

Sawgrass with Bacopa and Ludwigia

This state of sawgrass occurs throughout the study area, but is concentrated in the

southwest reaches. Typical species include sawgrass and B. caroliniana. Associate

species include Ludwigia spp. and C. Americana. According to the classification tree,

this community state occurs in peat depths ranging from 1.2 ft to 2.6 ft. Maximum water

levels should remain under 2.9 ft.

Sawgrass with Eleocharis sp. and Panicum

The sawgrass with Eleocharis and Panicum state occurs throughout the study area

except for the southeastern quadrant. This community state is dominated by sawgrass

and Eleocharis sp. Associate species include P. hemitomon, P. geminatum, Typha sp.

and Leersia hexandra. According to the classification tree, this state occurs in peat

depths of less than 1.2 ft and maximum water levels that remain under 2.9 ft.

Sawgrass with E. elongata and Crinum

This community occurs throughout the study area, yet is rarely found in the

southeastern quadrant. A diverse community state, dominant species include sawgrass,

E. elongata, and C. americana. Associate species include B. caroliniana, Cephalanthus

occidentalis, Hymenocallis, L. hexandra, Peltandra virginica, Pontederia cordata,

Ludwigia sp., and E. equistifoides. According to the classification tree, this community

state occurs in peat depths that are between 2.6 ft and 4.2 ft, and relatively long

hydroperiods compared to heavy sawgrass. Timing of extreme water levels should occur

later in the year.











Island Physiognomic Types

Figure 4-11 is the classification tree for the island physiognomic type. The term

island is being used loosely, and typically is associated with woody vegetation and can

occur as fringe marshes adjacent to true tree islands. Islands can also exist as strands of

sawgrass that include woody vegetation and typical tree island associates. The first node

in the classification tree is duration of high water levels, followed by a second branch

representing duration of low water levels. Classification of the island-type communities

broke out simply and evenly with only minor instances of misclassification. The

misclassification rate of the models was 54% and the amount of variation explained was

63% (1-Relative Error).


highdurat>0.2068491 highdurat<0 206849
Ghostlsland
SG Ghost
Treelsland








Lowdurat>0 106164 Lowdurat<0.106164


Ghostlsland
(23)









SG Ghost Treelsland
(25) (21)


Figure 4-11. Classification tree for 3 island-type community states on 10 environmental
variables. This model was pruned from a tree size of 10 leaves to eight, based
on a cost complexity pruning curve, selecting the smallest tree within one
standard error of the best.









Ghost islands

The ghost island community state occurs throughout the study area. Dominant

species include P. virginica, P. cordata, and C. occidentalis. Associate species include

Crinum, Ludwigia, Sagittaria lancifolia, and sawgrass. According to the classification

tree, ghost island communities require durations of high water levels for over 20% of the

year.

Sawgrass ghost islands

This community state occurs throughout the study area except in the northwest

quadrant. Sawgrass ghost islands are dominated by sawgrass, L. hexandra, P. virginica,

S. lancifolia, and C. occidentalis. Associate species include P. cordata, Blechnum

serrulatum, and Ludwigia sp. According to the classification tree, these communities

prefer durations of high water less than 20% of the year and durations of low water

greater than 10% of the year.

Tree islands

This community state occurs throughout the study area except in the southeast.

Due to the low number of samples however, this community probably occurs in all

quadrants of the study area. The dominant species are the ferns Osmunda regalis, and B.

serrulatum. Typical associates are Ludwigia, Typha, S. lancifolia, P. cordata, C.

occidentalis, and sawgrass. According to the classification tree, tree island communities

occur in areas where duration of high water levels is less than 20% of the year and

duration of low water levels is less than 10% of the year.














CHAPTER 5
SUMMARY AND CONCLUSIONS

Discussion

This thesis examines community-scale vegetation dynamics across the landscape-

level processes and drivers of Water Conservation Area 3A, of the Everglades.

Hydrologic processes vary seasonally and annually, but leave legacies that last years or

decades. The other processes that were examined operate on the temporal scale of

decades, such as soil accretion and erosion. Some of the processes examined are

anthropogenically influenced as well as driven by weather patterns. Vegetative

communities respond differentially to these drivers, and comprehension of how

communities respond is essential to management and restoration efforts.

Decompartmentalization and reorganization of the hydroscape for restoration purposes

illustrates the need to grasp the intricacies of community dynamics in the Everglades.

Only through the understanding of how the major drivers of the landscape operate, can

decision-makers rationalize a truly bottom-up approach to management of such a unique

and complex system of habitat and wildlife. Models based on a few key environmental

variables can be valuable tools in conservation management of dynamic wetlands (Toner

and Keddy 1997). This study provides the next step in the identification of those

variables and how they influence community structure.

Identification of meta-stable states of communities is essential for examination of

whole-scale community dynamics. Processes may have only subtle effects on

community structure over a short period of time. These subtleties can provide clues as to









the drivers that are causing potential shifts in landscape-level configurations of

vegetation. Chapter 3 identifies the meta-stable states of marsh physiognomic types of

WCA 3A, using a hierarchical clustering technique to analyze the community data

monitored semiannually over two years. Appendix B provides the structural signature of

each of these states. Four slough states were identified: deep slough, shallow slough,

Panicum slough, and Eleocharis elongata slough. Prairie communities manifest as four

different states: Eleocharis sp. prairie, Panicum/Paspalidium Eleocharis prairie, E.

elongata prairie, and sawgrass prairie. Four states of the sawgrass physiognomic type

were identified: sawgrass with Eleocharis/Panicum, sawgrass with Bacopa/Ludwigia,

sawgrass with E. elongata Crinum, and heavy sawgrass. Finally, shrubby marsh

communities, noted in this thesis as island-types, manifest as three different states:

sawgrass ghost island, ghost island, and tree island.

The majority of analyses performed on the data collected were multivariate in

approach due to the complex interactions and combinations of environmental variables

working in concert to create a unique, dense matrix of conditions in the study area.

Community ecology is best studied in a multivariate framework because of the

complexities of ecological interactions. Univariate comparisons, however, can be useful

in discovering basic trends in distributions of communities across a single gradient.

Figure 5-1 plots the mean water depths of the meta-stable states identified by the cluster

analyses. The chart illustrates both how slight the differentiation, as well as the

distinctive differences of hydrologic factors are between physiognomic types and

community states are. Deep slough and E. elongata slough states are the deepest of the

communities, though they also persist in a relatively wide range of hydroperiods. The E.






57


elongata prairie is the next deepest followed by the Panicum prairie, E. cellulosa prairie,

sawgrass prairie, and shallow slough. The sawgrass communities and sawgrass ghost

island are the next shallowest communities, though the heavy sawgrass community is

evenly distributed along a relatively wide hydrologic range. Finally, the ghost island and

tree island communities have the shortest hydroperiods of the community states. The

Panicum slough community did not cluster around a definable range, probably due to the

low number of sample units.


4 Eleocharis prairie
PAN/PDG/Elsp prairie
ELG prairie
S3 A Sawgrass prairie
SA x Deep Slough
S25 I Panicum slough
S* *ELG slough
a 2 I : x x Shallow slough
I I I SG with Elsp/PAN
S1.5 x x SG with BAC/LUD
S** SG with ELG/CRA
S I Heavy sawgrass
0.5 x x SG ghost island
x x x Ghost island
0 x- x Tree island

Figure 5-1. Distribution of sample units of each community state along a hydrologic
variable (mean annual water depth). Communities are grouped by
physiognomic type: squares=prairies, triangles=sloughs, circles=sawgrass,
crosses=islands.

Peat depths were plotted for each sample unit grouped by community state on the

x-axis in Figure 5-2. The soil gradient is distributed among the physiognomic types more

than the hydrologic gradient. Peat depths are high among the island-types, the heavy

sawgrass community, sawgrass with E. elongata Crinum, sawgrass with

Bacopa/Ludwigia, shallow slough, E. elongata slough, deep slough, and E. elongata

prairies. Shallow peat depths are characteristics of sawgrass with E. cellulosa/Panicum,







58


sawgrass prairies, Panicum prairies, and E. cellulosa prairies. Again, the Panicum slough

could not be located on the soil gradient because of the small number of samples.


7 Eleocharis prairie
PAN/PDG/Elsp prairie
6 ELG prairie
Sawgrass prairie
5 x x Deep Slough
S: i Panicum slough
4- : I ELG slough
S. x x Shallow slough
3 i SG with Elsp/PAN
S. x SG with BAC/LUD
2 x
2 x SG with ELG/CRA
S* xS
S. x Heavy sawgrass
1 x SG ghost island
S I : x x x Ghost island
0 x Tree island

Figure 5-2. Distribution of sample units of each community state along a peat depth
gradient. Communities are grouped by physiognomic type: squares=prairies,
triangles=sloughs, circles=sawgrass, crosses=islands.

The amount of overlap between community states along the individual

environmental variable illustrates why multivariate statistics are the preferred method of

analysis in community ecology. Combinations of environmental conditions create unique

conditions that are favorable for specific communities to develop. It is interesting to note

that communities that are not differentiated by soil characteristics, typically are

differentiated by hydrology. For this reason, it can be concluded that these are the two

major driving factors determining community structure.

Table 3-1 shows the frequency of the community states for each of the sampling

events. Many of the community states were present in steady numbers throughout the

study, such as the heavy sawgrass community. However, there were some states that

occur in large numbers at one event, only to virtually disappear in other sample events.

The Panicum/Paspalidium/Eleocharis prairie fluctuates with respect to the wet and dry










season variation. Communities such as the Eleocharis elongata slough are common at

the beginning of the study, only to drop out completely by the end of the study. One

explanation for this phenomenon could be the differences between the water years of

2003 and 2004. Figure 5-3 is a time series graph of water stage during the time of

sampling. The Everglades was subjected to a moderate fluctuation of water levels from

November 2002 to June 2003. From that time forward extreme levels were reached

during the wet and dry seasons, potentially influencing the state of physiognomic types in

the study area. It can be speculated that these hydrologic extremes discouraged the

persistence ofE. elongata causing a shift to the shallow slough state, a less diverse

community. Although actual shifts in state were not documented, the fluctuation in

frequencies of community states can be interpreted as shifting meta-stable states.


SITE 65 IN CONSERVATION AREA 3R NERR COOPERTOWN, FL
Station: 3-65 Freq: DR DataType: STG Stat: MERN Agency: USGS Recorder: ???? opNum: Dbkey: 16538
13.0





,,, 10 .0 '- .--- - - - - - - - - - --.
12.0





-J


9.0


8.0
Nov Dec Jan Feb Mar Apr May Jun Jul Rug Sep Oct Nov Dec Jan Feb Mar Apr May Jun
2002 2003 2004

Figure 5-3. A time-series graph of water stage at a monitoring station within Plot 4.
Note the extreme highs and lows of the second water year compared to the
first water year.

The distribution of community states across the landscape provides evidence of the

environmental thresholds intrinsic to those states. Compartmentalization of the water









conservation areas creates a relatively flat hydroscape to the elevational gradient. The

same physiognomic types are exposed to differential hydroperiods, depending on their

position along a north-south axis. Soil types and characteristics vary along a longitudinal

axis, with shallow peat depths in the west and deep peat depths to the east. Figure 3-6 is

a graphical representation of the distribution of community states across the landscape. It

is apparent through examination of the community distribution figure that species such as

Panicum hemitomon and E. cellulosa are endemic to areas within the region that parallel

environmental requirements for those species. P. hemitomon is almost exclusively

located in the northern reaches of the study area where shorter hydroperiods persist.

Prairies dominated by E. cellulosa are confined to the western reaches where shallow

peats are typical. The scale of the study allows observation of the range of communities

and environmental conditions within WCA 3A.

An investigation of the environmental conditions of each of these community states

offers insight as to how they are related in terms of successional order and environmental

thresholds. When variables that influence the state of a physiognomic type are identified,

they can be quantified using classification trees. Thresholds of hydrologic variation and

extremes are discovered and a picture of where meta-stable states lie on the continuum of

a multidimensional environmental space begins to emerge. Relative positions of

physiognomic types in this space are already common knowledge to ecologists. In order

of short to long hydroperiod communities, the sequence goes: island, sawgrass, prairie,

slough. The evidence presented through this study shows some overlap in hydrology

between these physiognomic types, though the basic relationships remain.









The classification tree analyses in Chapter 4 describe how the community states

identified by the clustering techniques break out in terms of environmental conditions.

Slough communities were determined by hydrologic factors such as timing of maximum

water depths and the magnitude of extreme water levels. Deep slough communities

require early timing of maximum water levels. Early timing of water levels suggests

location further to the north where there is little time lag compared to the southern extent

of the conservation area where water tends to pile up later along Tamiami Trail.

However, high maximum water levels are required to drown out typical slough associates

like Panicum hemitomon. Panicum sloughs favor shorter hydroperiod sites, as revealed

by the classification tree. E. elongata sloughs favor late timing and minimal variation of

water depths, whereas shallow sloughs tolerate a wide range of water depths. The

implications for this could mean that Eleocharis elongata, as a species, may be less

resilient and more sensitive to hydrologic disturbance. Shallow sloughs tend to be less

diverse, so in the event of massive variation between seasons, E. elongata sloughs could

convert into a shallow slough state.

Peat depth was the major determinant for wet prairie meta-stable states. A close

inspection of the classification tree reveals that E. elongata prairies tend to persist in deep

peat depths and relatively long duration of lower water levels. Confined mostly toward

the eastern half of the study area, this prairie state occurs where soil conditions are

conducive for its establishment. Eleocharis sp. prairies, on the other hand, prefer shallow

peats, which are typical of the western reaches of the study area. Sawgrass prairies have

similar requirements as Eleocharis prairies, but tend towards shorter hydroperiod sites.









Finally Panicum/Paspalidium Eleocharis prairies lack a restrictive soil requirement, but

prefer even shorter hydroperiod sites than sawgrass prairies.

In the cluster analyses sawgrass units were grouped mainly by diversity. Heavy

sawgrass states were fairly common. The classification tree analysis reveals that heavy

sawgrass is ubiquitous throughout the study area and occupies the range of environmental

conditions that are characteristic of the sawgrass physiognomic type. Sawgrass with

Bacopa/Ludwigia occur in shorter hydroperiod sites. Also occupying short hydroperiod

sites is the sawgrass with Eleocharis/Panicum state, yet it is confined to areas of shallow

peats, as are most Eleocharis cellulosa associated communities. Sawgrass with E.

elongata Crinum communities prefer areas of long hydroperiods and deep peat depths.

This community state is probably the sparse sawgrass communities of the southeastern

WCA 3A. These tall sawgrass communities are exposed to continuous inundation and

recruitment of sawgrass individuals is nonexistent. The resilience of these communities

is evident, as these conditions have prevailed for years. However, they are slowly being

replaced by some prairie and primarily slough associates in the absence of drawdown

events.

Island-type communities were determined entirely by duration of high and low

water events. The duration of annual extreme water level conditions has implications on

the balance of competitive and stress-tolerant organisms, as well as anaerobic stress in

plants. Ghost islands characterized by arrowheads and woody shrubs prefer high

duration of water levels. This may suggest they are or have been subject to soil

subsidence or oxidation. Long duration of low water levels is required for sawgrass

ghost islands, which may be old tree islands with slightly higher water levels, or sawgrass









ridges that have been colonized by woody shrubs for one reason or another. The tree

island community state is classified as a community that prefers short durations of

extreme water levels in general. This may suggest that tree islands are not resilient to

shifts in hydrologic regime and require minimal annual variation of water levels.

Comparing NMS and Classification Tree Techniques

One interesting observation worth discussion is the fact that the NMS ordination

joint plots for each of the physiognomic types did not reveal relationships to

environmental variables (except for the prairies). The ordination of the physiognomic

types confirmed what we know about the relationships between them. Hydrology is

clearly the factor determining physiognomic type. When the NMS ordination did not

reveal the same relationship between meta-stable states, it was somewhat surprising.

However, the ordination plots did show distinct patterns that distinguish the meta-stable

states from each other in species space.

The classification tree analyses interpret groups using the environmental variables

given. In this case, the meta-stable states are classified by the variables that distinguish

them. Misclassification is common and overfitting the data could be an issue. The

interpretability of the classification trees, however, provide insight into the conditions

that may be distinguishing the community states. The diversity of variables that were

determined to be associated with each physiognomic type outlines the complexity of

environmental variation between the meta-stable states. This may explain why individual

variables were not determined to be explanatory to structural variation in the NMS.

Perhaps the creation of an alternative metric that accounts for the variables determined by

the classification tree would be useful including in the ordination procedure.









Overall, the NMS technique was helpful in visualizing the differences between the

meta-stable states in terms of community structure. The classification trees provided the

environmental context of each physiognomic type, by distinguishing particular

combinations of conditions that determine a meta-stable state. In the case of this study,

neither method was "more correct" than the other. Both analyses helped to provide the

whole explanation behind what the meta-stable states are.

Review of Methodology and Future Tracks of Research

The analytical techniques of clustering and classification trees were successful at

identifying subtle differences in community structure and their corresponding

environmental characteristics. These specific multivariate analyses are geared toward the

nonparametric nature of community ecology. Sampling was sufficient for establishing

environmental criteria for each of the physiognomic types, with probable exception to

island-types, although even these communities were distinct in the classification tree

analysis.

Importance values are a useful tool in summarizing the complex structure of

community data. However, the addition of relative frequency to the computation of

importance values should be considered. This metric would include clustering and

evenness of species within a community to the equation, providing a more descriptive

index of community structure. In the case of sparse sawgrass communities such as the

"sawgrass with E. elongata Crinum" state, the unevenness of sawgrass would be

accounted for, rather than assumed due to the types of associates.

Although the set of environmental variables analyzed to differentiate the meta-

stable states were comprehensive in regard to hydrology, many soil parameters were

lacking. Soil nutrient concentrations, acidity, and bulk density, although shown in









previous studies to be uniform on a landscape scale in WCA 3A (given sufficient distance

from point sources i.e. canals, etc.), could be a factor in "hot spots" of wildlife activity.

Alligator holes and bird roosts tend to alter local soil chemistry characteristics.

Expansion of community dynamic studies in the Everglades should consider landscape

geometry and proximity to seed banks. It was not necessary to include these landscape

metrics in this study, because all of the research was located in one impounded section of

the Everglades that is relatively distant from disturbed upland habitat that would provide

a source of ruderal recruitment.

Future research should include the investigation of the "movement" of individual

sample units over time. NMS is a useful tool for quantifying and visualizing the

trajectory of a sample in species space. Coupled with changes in seasonal environmental

conditions, additional insights could be made into landscape level trends of community

shifts. Inferences into the resilience of community states could be made through NMS or

Multiresponse Permutational Procedures (MRPP) of individual sample units. Long-term

studies should monitor the movement of ecotones relative to hydrologic dynamics.

Reduction of the Everglades ecosystem to half of its size, loss of sheetflow through

the system, and loss of habitat diversity are some of the functional losses due to human

engineering efforts of the mid-20th century (Davis et al. 1994). Monitoring and

predicting landscape-scale vegetation dynamics is essential to the restoration efforts. The

potential for community dynamic research using the concepts of meta-stable states is

considerable. An adaptive management approach to Everglades restoration means that

hydrologic regimes will continue to change along with management strategies. The need

for greater understanding of the implications of future scenarios, will at the same time






66


allow researchers to further examination of the intricacies of Everglades community

dynamics.
















APPENDIX A
INDICATOR SPECIES ANALYSIS TABLES AND FIGURES

November 2002 Indicator Species Analysis Graphs





0.19
0.18
0.17
0.16
0.15
0.14
0.13
0.12
0.11
0.1
0.09
0.08


Number of clusters

Figure A-i. Change in p-value from the randomization tests, averaged across species at
each step in the clustering.





18
16
14
12
10 -
8 -
6-
4-
2
0


Number of clusters

Figure A-2. Number of species with p < 0.05 for each step of clustering.







68


June 2003 Indicator Species Analysis Graphs


0.2 -
0.18 -
0.16 -
0.14 -
0.12 -
0.1 -
0.08 -
0.06 -
0.04
0.02 -
0-


n r^ r1 r '1 rD '- 0 0 N 0 8 0 'o t
Number of clusters

Figure A-3. Change in p-value from the randomization tests, averaged across species at
each step in the clustering.


Number of clusters

Figure A-4. Number of species with p < 0.05 for each step of clustering.










November 2003 Indicator Species Analysis Graphs


0.19
0.18 -
0.17
0.16 -
0.15 -
0.14 -
0.13
0.12 -
0.11
0.1


Figure A-5. Change in p-value from the randomization tests, averaged across species at
each step in the clustering.


Number of clusters


Figure A-6. Number of species with p < 0.05 for each step of clustering.










70


June 2004 Indicator Species Analysis Graphs


0.18 -
0.16-
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0-
1P


Figure A-7. Change in p-value from the randomization tests, averaged across species at
each step in the clustering.


U0 r^ r r r i i i^ i^ is i i O 8 b O 2
Number of clusters

Figure A-8. Number of species with p < 0.05 for each step of clustering.


Number of clusters
Number of clusters














APPENDIX B
COMMUNITY STATES AND THEIR STRUCTURAL SIGNATURES

The groups from the dendrograms in Figures 3-2, 3-3, 3-4, and 3-5 were run

through another cluster analysis to match similar groups between sampling events. Those

groups from the second clustering run were designated as the community states of Water

Conservation Area 3A. Two groups from the same sampling event were occasionally

pooled together due to the similarity between community structures. The following

figures are the structural signatures of the community states resulting from cluster

analyses of community units. There were a total of fifteen community states.

Species codes align the x-axis (see Table 2-1 for species names and codes). The y-

axis corresponds to the percent of perfect indication relating each species to that

particular community state. The percent of perfect indication combines relative

abundance (the average abundance of a given species in a given group of communities

over the average abundance of that species in all communities) and the relative frequency

(percent of community units in given group where given species is present). This value is

how well that species is an indicator of that community state. When all species are

included on the same graph a "signature" develops that is unique to that community state.

The signatures for each community state at each sampling event (sometimes there were

two in a sampling event, sometimes there were none) are aligned on the z-axis.







72



Panicum/Paspalidium/Eleocharis Prairie


40

35
O

0 20 I- .................-.---- ---
' 30

S25


S 15- un-03
S10
Nov-02

4- *.un-04


Figure B-1. Structural signature of the Panicum/Paspalidium/Eleocharis Prairie.


Figure B-2. Structural signature of the Shallow Slough.


Shallow Slough



50

. 40 -----------
.2 -- ~-------------- -- -

S30 -
- - - - -
20- U Nov-03
o10 ---- I Jlun-04
S10

0
,. L _

-> (D: > .

Species z
Z 7)












Ghost Island


Figure B-3. Structural signature of the Ghost Island.


Figure B-4. Structural signature of the Deep Slough.


O Nov-02
OJun-03
* Nov-03
IJun-04


Deep Slough

60


5 40
v -, -------------------------

g 30 --------Nov-02
S20 DJun-03
S 2 0 ------------------
0o. Nov-03
..................... Jun-04
I, -- - -- - -- --.. .. .. . .



































Figure B-5. Structural signature of the Eleocharis elongata Slough.


Figure B-6. Structural signature of the Eleocharis elongata Prairie.


Eleocharis elongata Slough


ONov-02
ONov-02
IJun-03
*Nov-03


Eleocharis elongata Prairie


O Nov-02
CJun-03
* Nov-03
IJun-04
































Figure B-7. Structural signature of the Sawgrass Prairie.


Figure B-8. Structural signature of the Eleocharis Prairie.


Sawgrass Prairie

30
o
2 5 ----------------- ---.
25

" 2 0 . .
.E ......... -- _
t 15 ------------- -
- - - -
-------- .--------------- -----
S10 Ol un-03
-U-- Nov-03
0 5 --
- - -


Eleocharis Prairie


0 Nov-02
0Jun-03
U Nov-03
13un-04










Panicum Slough


4-1


C
0 c


Figure B-9. Structural signature of the Panicum Slough.


Figure B-10. Structural signature of the Tree Island.


O Nov-02
SJun-04


Tree Island

100
. 90
m 80-
S70
.E 60
t 50
40 EJun-03
a 30 --Nov-03
'5 20 I *Jun-04
S 10 ---
A ---------I ------ -- IVI











Sawgrass with Bacopa and Ludwigia


0 Nov-02
DJun-03
* Nov-03
IJun-04


Species
Species 3


Figure B-11. Structural signature of Sawgrass with Bacopa and Ludwigia.




Sawgrass Ghost Island


O Nov-02
DJun-03


Figure B-12. Structural signature of the Sawgrass Ghost Island.












Sawgrass with E. elongata and Crinum


C
0 o


o'a
C


Figure B-13. Structural signature of Sawgrass with E. elongata and Crinum.


Heavy Sawgrass


30
o



0 Nov-02
Slun-03
4-

10-1 11------------~ ~~~---------------------------------- Olun-03

SNov-03
S 5 1_ Nov-03
0 lJun-04
...'-'- -. .



Species D

Figure B-14. Structural signature of Heavy Sawgrass.


O Nov-02
DJun-03
* Nov-03
*Jun-04







79




Sawgrass with Eleocharis and Panicum

25

20


L 15---

Sc O Nov-02
5I J un-04


Species


Sz,
(n U)


Figure B-15. Structural signature of Sawgrass with Eleocharis and Panicum.


















APPENDIX C
RESULTS OF THE NONMETRIC MULTIDIMENSIONAL SCALING ANALYSES

The following are results of the NMS ordination analyses used to plot the


community sample units in multidimensional ordination space. Included are scree plots


used to assess the dimensionality of the data set. The figures plot the final stress vs. the


number of dimensions. Stress is an inverse measure of fit to the data. The randomized


data from a Monte Carlo test are analyzed as a null model for comparison. The


dimension selected is prior to which additional dimensions provided only small


reductions in stress. Also included are tables comparing the solution to the Monte Carlo


result. Finally, the stress and stability of the solution are included. Stress and stability


were listed in the numerical output of the NMS.

Slough-type ordination
60
Real Data Rando-mied Data
Mea




40






20







2 3 4
Dimensions


Figure C-1. A scree plot for the slough-type ordination.







81


Table C-1. Stress in relation to dimensionality for slough NMS. A two-dimensional
solution was chosen.
Stress in real data Stress in randomized test
Axes Minimum Mean Maximum Minimum Mean Maximum p
1 28.676 44.690 57.313 38.592 48.260 57.336 0.0323
2 13.422 16.720 41.513 16.868 21.001 41.495 0.0323
3 8.469 10.355 32.522 11.537 12.579 14.161 0.0323
4 6.675 12.480 26.998 9.128 10.829 26.956 0.0323



Final stress for two-dimensional solution = 15.10441. Final instability = 0.0001.




Prairie-type ordination
60 *"- ,
Real Data Randomized Data
Maximum
Mean
Minimum




40







20 *







0


Dimensions


Figure C-2. A scree plot for the prairie-type ordination.


Table C-2. Stress in relation to dimensionality for prairie NMS. A two-dimensional
solution was chosen.
Stress in real data Stress in randomized test
Axes Minimum Mean Maximum Minimum Mean Maximum p
1 24.387 38.624 57.243 42.741 48.798 57.302 0.0323
2 14.463 15.606 18.013 21.872 25.496 29.070 0.0323
3 9.502 14.279 32.404 14.924 16.221 17.261 0.0323











4 17.149 20.353 26.866 11.636 16.581 26.869 0.0323



Final stress for two-dimensional solution = 14.71455. Final instability = 0.00007.


Sawgrass-type ordination
60 *
Real Data Randomized Data
Maximum
Mean
Minimum




40 *








20


2 3 4
Dimensions
Figure C-3. A scree plot for the sawgrass-type ordination.

Table C-3. Stress in relation to dimensionality for sawgrass NMS. A three-dimensional
solution was chosen.
Stress in real data Stress in randomized test
Axes Minimum Mean Maximum Minimum Mean Maximum p
1 36.107 45.100 57.495 43.616 49.797 57.390 0.0323
2 21.343 22.866 24.880 24.024 25.749 27.922 0.0323
3 13.965 14.658 15.821 15.967 17.322 18.811 0.0323
4 12.419 16.130 23.565 11.609 17.810 27.082 0.2903



Final stress for three-dimensional solution = 16.53583. Final instability = 0.01579.







83


Island-type ordination
Real Data Randomized Data
Maximum
Mean
Minimum


Dimensions
Figure C-4. A scree plot for the island-type ordination.


Table C-4. Stress in relation to dimensionality for island NMS. A three-dimensional
solution was chosen.
Stress in real data Stress in randomized test
Axes Minimum Mean Maximum Minimum Mean Maximum p
1 34.018 46.591 56.898 38.144 48.194 56.891 0.0323
2 21.074 22.272 23.640 22.478 25.339 40.804 0.0323
3 14.394 14.962 17.061 15.748 17.117 18.931 0.0323
4 10.780 13.211 26.054 12.075 13.028 13.898 0.0323



Final stress for three-dimensional solution = 14.70141. Final instability = 0.00283.















APPENDIX D
IMPORTANCE CHARTS OF ENVIRONMENTAL VARIABLES FROM
CLASSIFICATION TREE ANALYSIS

The following charts rank the importance of the environmental variables in

explaining the differences between groups in the classification tree analysis. The groups,

or meta-stable states of communities, were analyzed as physiognomic types. The

importance scale is based on rankings of explanatory power for each physiognomic type.

Interpretations and discussion of these charts can be found in Chapter 5.



Slough physiognomic type predictor variables








a














Figure D-1. Importance rankings of predictor variables for the slough physiognomic
type.









Prairie physiognomic type predictor variables



0

CD

C I

CD-


d 0 llllll,



Figure D-2. Importance rankings of predictor variables for the prairie physiognomic
type.


Sawgrass physiognomic type predictor variables







E o









Figure D-3. Importance rankings of predictor variables for the sawgrass physiognomic
type.







86




Island physiognomic type predictor variables










CO
U
0
















Figure D-4. Importance rankings of predictor variables for the island physiognomic type.