<%BANNER%>

Estimating Transition Probabilities among Everglades Wetland Communities Using Multistate Models

Permanent Link: http://ufdc.ufl.edu/UFE0023989/00001

Material Information

Title: Estimating Transition Probabilities among Everglades Wetland Communities Using Multistate Models
Physical Description: 1 online resource (66 p.)
Language: english
Creator: Hotaling, Althea
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2008

Subjects

Subjects / Keywords: adaptive, community, everglades, multistate, transition
Interdisciplinary Ecology -- Dissertations, Academic -- UF
Genre: Interdisciplinary Ecology thesis, M.S.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: There has been a drastic loss of wetlands throughout the world with an estimated loss of 53% in the United States since the late 1700s. Half of the Everglades, a south Florida wetland, has been lost to agriculture and urban development in the last hundred years but now efforts are being made to restore it. Managers and decisions makers faced with the task of carrying out Everglades restoration are in dire need of robust statistical estimates that relate changes in water levels to changes in plant communities. To address this need I present a comprehensive framework for investigating multiple competing hypotheses about the factors governing transition probabilities among vegetative community states in the Everglades. The first step in this analysis was to use multivariate analyses to classify vegetation communities into states that are particularly relevant to specified management problems. I then applied likelihood based multistate models in order to evaluate multiple competing hypotheses; and to provide estimates of transition probabilities and associated measures of uncertainty. These estimates can then be incorporated into management models. In addition, to being useful for management of the Everglades I believe that our framework can be used to address adaptive management problems in other ecosystems.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility: by Althea Hotaling.
Thesis: Thesis (M.S.)--University of Florida, 2008.
Local: Adviser: Kitchens, Wiley M.
Electronic Access: RESTRICTED TO UF STUDENTS, STAFF, FACULTY, AND ON-CAMPUS USE UNTIL 2009-06-30

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2008
System ID: UFE0023989:00001

Permanent Link: http://ufdc.ufl.edu/UFE0023989/00001

Material Information

Title: Estimating Transition Probabilities among Everglades Wetland Communities Using Multistate Models
Physical Description: 1 online resource (66 p.)
Language: english
Creator: Hotaling, Althea
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2008

Subjects

Subjects / Keywords: adaptive, community, everglades, multistate, transition
Interdisciplinary Ecology -- Dissertations, Academic -- UF
Genre: Interdisciplinary Ecology thesis, M.S.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: There has been a drastic loss of wetlands throughout the world with an estimated loss of 53% in the United States since the late 1700s. Half of the Everglades, a south Florida wetland, has been lost to agriculture and urban development in the last hundred years but now efforts are being made to restore it. Managers and decisions makers faced with the task of carrying out Everglades restoration are in dire need of robust statistical estimates that relate changes in water levels to changes in plant communities. To address this need I present a comprehensive framework for investigating multiple competing hypotheses about the factors governing transition probabilities among vegetative community states in the Everglades. The first step in this analysis was to use multivariate analyses to classify vegetation communities into states that are particularly relevant to specified management problems. I then applied likelihood based multistate models in order to evaluate multiple competing hypotheses; and to provide estimates of transition probabilities and associated measures of uncertainty. These estimates can then be incorporated into management models. In addition, to being useful for management of the Everglades I believe that our framework can be used to address adaptive management problems in other ecosystems.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility: by Althea Hotaling.
Thesis: Thesis (M.S.)--University of Florida, 2008.
Local: Adviser: Kitchens, Wiley M.
Electronic Access: RESTRICTED TO UF STUDENTS, STAFF, FACULTY, AND ON-CAMPUS USE UNTIL 2009-06-30

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2008
System ID: UFE0023989:00001


This item has the following downloads:


Full Text

PAGE 1

1 ESTIMATING TRANSITION PROBABILIT IES AMONG EVERGLADES WETLAND COMMUNITIES USING MULTISTATE MODELS By ALTHEA HOTALING 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 2008

PAGE 2

2 2008 Althea Hotaling

PAGE 3

3 To my family

PAGE 4

4 ACKNOWLEDGMENTS I am grateful to all the st udents and technicians who have worked at the Florida Cooperative Fish and Wildlife Research Unit si nce the fall of 2002 for their help during the biannual vegetation samples in WCA3A with speci al thanks to those whom I have personally worked with in the last 2 years. Thanks go to Simon Fitz-William, Tayl or Tidwell, Lara Drizd and Brandon VanNuys for their help in continuing and expanding the project. Andrea Bowling was an invaluable resource during data analysis. I also appreciate Eric Powers help in setting up the project and Christa Zweigs help with some of the initial analyses. I would also like to thank all of my office mates at the coop for letting me bounce ideas o ff of them and for keeping the mood fun and light at the coop. Thanks go to Andrea Bowling, Lara Drizd, Brian Reichert, Melissa DeSa, Zach Welch, Christa Zweig, Brad Shoger, and Chris Cattau. The Jacksonville District of the US Army Corps of Engineers and the Vero Beach field office of the US Fish and Wildlife Service provided funding for the project. Julien Martin deserves full credit for coming up with the innovative idea of using MARK to track transition probabilities in wetland plant communities. He was also instrumental in helping me to get started on the analysis and manuscr ipt. I am not sure what my thesis would be on or whether I would even be close to done yet without him. Wiley Kitchens is an excellent wetland ecologist with a thorough un derstanding of the ecosystems of the Everglades. He helped me to focus on and understand the ecology of th e system and made sure I gave it primary importance in my analysis. Debbie Miller and Leonard Pearlstine provide d constructive review comments and helped me to improve my thesis with their input.

PAGE 5

5 TABLE OF CONTENTS page ACKNOWLEDGMENTS...............................................................................................................4 LIST OF TABLES................................................................................................................. ..........7 LIST OF FIGURES.........................................................................................................................8 ABSTRACT.....................................................................................................................................9 CHAP TER 1 OVERVIEW...........................................................................................................................10 Introduction................................................................................................................... ..........10 Background.............................................................................................................................11 Wetlands..........................................................................................................................11 The Everglades................................................................................................................12 Flora and Fauna of the Everglades.................................................................................. 13 2 METHODS.............................................................................................................................15 Study System..........................................................................................................................15 Sampling Design.....................................................................................................................15 Sampling..........................................................................................................................16 Hydrology........................................................................................................................17 Data Analysis..........................................................................................................................18 Multivariate Community Classifying Analysis............................................................... 18 Hierarchical Clustering Analysis to Categorize Wet and Dry years ............................... 20 Multistate Modeling........................................................................................................ 20 Model selection........................................................................................................ 23 Effect size.................................................................................................................24 3 ESTIMATING TRANSITION PROBABILITIES AMONG EVERGLADES WETLAND COMMUNITIES USING MULTISTATE MODELS ....................................... 28 Introduction................................................................................................................... ..........28 Hypotheses and Predictions.................................................................................................... 30 Hypothesis 1: Wet and Dry Seasons Infl uence the Conversion of Sloughs and W et Prairies.........................................................................................................................30 Hypothesis 2: Wet and Dry Years Subs tantially Influence the Process of Conversion of Sloughs and W et Prairies..................................................................... 30 Hypothesis 3: Probabilities of Transiti on between Sloughs and W et Prairies are Substantially Influenced by Impoundment.................................................................. 31 Methods..................................................................................................................................31 Study Area and Sampling Methods.................................................................................31

PAGE 6

6 Data Analysis...................................................................................................................32 Multivariate analysis to classify communities......................................................... 32 Hierarchical clustering analysis to categorize wet and dry years .............................33 Multistate modeling.................................................................................................. 33 Model selection........................................................................................................ 34 Effect size.................................................................................................................34 Results.....................................................................................................................................35 Multivariate Analysis to Classify Communities..............................................................35 Hierarchical Clustering Analysis to Categorize Wet and Dry Years .............................. 35 Multistate Modeling........................................................................................................ 36 Discussion...............................................................................................................................37 4 CONCLUSIONS AND NEXT STE PS..................................................................................51 Hypotheses and Predictions.................................................................................................... 51 Hypothesis 1: Transition Probabilities W ill Be Higher in th e North Than in the South............................................................................................................................51 Hypothesis 2: The Hydrologi cal Variables That Drive Transitions in the North W ill Be Different from the Hydrological Variables That Drive Transitions in the South............................................................................................................................52 Hypothesis 3: Transition Probabilities Will Have a Directio nality That is Related to Water Levels............................................................................................................ 52 Preliminar y Results ............................................................................................................ .....52 Annual Transition Probability Estimates......................................................................... 53 Separate Estimates for the North and South.................................................................... 54 Two-community states: wet prairie and slough ....................................................... 54 Three-community states: wet pr airie, transition, and slough ...................................54 Discussion..................................................................................................................... ...55 WORK CITED...............................................................................................................................62 BIOGRAPHICAL SKETCH.........................................................................................................66

PAGE 7

7 LIST OF TABLES Table page 3-1 Multistate models of transition probabilities. .................................................................... 434-1 Results when the cluster including both wet and dry season data was used to make the input file for annual transition estimates. ................................................................... 574-2 When the communities were reclustered using only wet season sampling data, the results, as seen here, were quite different and more ecologically reasonable. ................. 58

PAGE 8

8 LIST OF FIGURES Figure page 2-1. Location of WCA3A and the study plots.............................................................................. 25 2-2. Belt transect sample collection.......................................................................................... ....26 2-3. Water levels at S65 and S64: a new hydrological era began in 1992. ................................... 27 3-1. Hydrograph of water levels in Water Conservation Area 3A from 1992 to 2006.................. 44 3-2. Our study area, southern WCA3A, is shown here with the twenty study plots in black.. ...... 45 3-3. Stage water levels in WCA3A since 1953 from gauge station 3-65, site 3A-28.. .................46 3-4. Transition probabilities shows that sloughs will trans ition to wet prairies with a certain probability or remain as sloughs........................................................................................ 46 3-5. Results from the Agglomerative Cluster analysis used to determ ine the vegetation communities in WCA3A at each sampling event.............................................................. 47 3-6. Cluster analysis used for all dry seasons since water year 1992 to determ ine which dry seasons were dry (dark grey) and whic h were normal/wet (light grey)............................. 48 3-7. Cluster analysis used for all wet seasons since water year 1992 to determ ine which wet seasons were wet (black) and whic h were normal/dry (light grey)................................... 48 3-8. Transition estimates from our most parsim onious model )3( catWDyr (AIC ( w ) =0.685)...............................................................................................................................49 3-9. Transition estimates for wet prairie and slough communities from our second most parsim onious model )2( catWDyr (AIC(w)=0.089).......................................................50 4-1. Most parsimonious model in model se t that attem pted to get annual transition probability estimates.......................................................................................................... 59 4-2. Model averaged results from the model set u sed to get annual transition probability estimates...................................................................................................................... .......59 4-3. Results from 2 model sets, one for the north and one for the south, with 2 comm unity states, slough and wet prairie that were used to determine what the hydrologic drivers are for each area..................................................................................................... 60

PAGE 9

9 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 ESTIMATING TRANSITION PROBABILIT IES AMONG EVERGLADES WETLAND COMMUNITIES USING MULTISTATE MODELS By Althea Hotaling December 2008 Chair: Wiley M. Kitchens Major: Interdisciplinary Ecology There has been a drastic loss of wetlands throughout the worl d with an estimated loss of 53% in the United States since the late 1700s. Half of the Everglades, a south Florida wetland, has been lost to agriculture and urban developm ent in the last hundred years but now efforts are being made to restore it. Mana gers and decisions makers faced with the task of carrying out Everglades restoration are in dire need of robust statistical estimates that relate changes in water levels to changes in plant communities. To address this need I present a comprehensive framework for investigating multiple comp eting hypotheses about the factors governing transition probabilities among vegetative community st ates in the Everglades. The first step in this analysis was to use multivariate analyses to classify vegetation communities into states that are particularly relevant to specified management problems. I then applied likelihood based multistate models in order to evaluate multiple competing hypotheses; and to provide estimates of transition probabilities and asso ciated measures of uncertainty. These estimates can then be incorporated into management models. In ad dition, to being useful for management of the Everglades I believe that our framework can be used to address adaptive management problems in other ecosystems.

PAGE 10

10 CHAPTER 1 OVERVIEW Introduction Wetlands are transition zones between terres trial and true aquatic ecosystem s. By cleansing polluted waters, preventing floods, protecting shorelines and recharging groundwater aquifers, wetlands fill an important ecological role. They function as sources, sinks, and transformers of chemicals, and biological materi al through their extensive trophic chains and rich biodiversity. Their soils are waterlogged for some portion of the year resulting in anoxic conditions which makes them unlike other ecosystems (Mitsch and Gosselink 1993). The no net loss policy, enacted in Janua ry of 1989, set a goal for the Unite d States to stop the decrease in wetland area occurring throughou t the country (Salzman and Ruhl 2006). No net loss, along with increased public know ledge, has lead to the undertaking of the Comprehensive Everglades Restoration Plan (CERP), one of the largest wetla nd restoration efforts ever. As a part of the restoration effort, certain sp ecies have been identified as indicator species of Evergladess health. The snail kite ( Rostrhamus sociabilis ) is one of these species because it is completely dependent on the South Florida wate rshed for its entire life cycle (Sykes et al. 1995). The fate of the snail kite is inextricably tied to the health of th e Everglades. The attempt to restore the Everglades will be aided by incr eased knowledge of the ha bitats necessary for a healthy, proliferating snail kite population. A robust proliferating snail kite population would be a performance measure of restoration success. Th e snail kite is also an endangered species and as outlined in the Endangered Species Act of 1973, Federal agencies are responsible for protecting and conserving the ecosystems that e ndangered species depend on. Federal agencies must also cooperate with State and local agencies to resolve water resource issues in concert with conservation of endangered species.

PAGE 11

11 To help meet these aims, this study will track the loss of wet prairie, the preferred foraging habitat of the snail kite (Karuna ratne et al. 2006). The proposed reason for this loss of critical habitat is a change in the hydrologic regime due to impoundment. Water Conservation Area 3A (WCA3A) in South Florida was historically very important to the snail kite, so efforts to investigate this link have been focused there (Kitchens et al. 2002). The Water Conservation Areas are also perfect candidates for restoration and critical spatially and functionally to the water management of the Everglades. It is hoped that an improved management strategy for WCA3A and the snail kite will be pr oduced as a result of this study. Background Wetlands It is estimated that there are 7 to 10 million km2 of wetlands on Earth which is about 5 to 8 % of the Earths land surface (M itsch and Gosselink 2007). About 50% of endangered species, 80% of the U.S. breeding bird population and 50% of protected migratory birds depend on wetlands for their survival. Wetlands also m oderate the effects of floods and storms, and improve water quality while recharging the aquifer. Over half the wetlands in the U.S. have been lost and of the total wetlands left 95% are freshwater wetlands (Dahl 2000). The U.S. currently has a no-net lo ss policy for wetlands, which means that for all wetlands destroyed, new wetlands have to be created or degraded wetland s must be restored. (Dahl 2000) estimated that about 23,675 hectares of wetlands are lost annually. The most important factor in wetland restoration is restoration of the natural hydrologic conditions. Hydrologic conditions affect abiotic factors like soil type and nutrient availabi lity, which in turn determine the flora and fauna that will inhabit a wetland. Hydrology encomp asses a variety of parameters, such as hydroperiod, seasonal pulses, flow pattern, and re tention times. Hydrope riod is the hydrologic signature of each wetland, which is determined by the seasonal water level pattern of a wetland.

PAGE 12

12 The rise and fall of surface and subsurface water is characteristic for each type of wetland and can be called its hydroperiod (Mitsch and Go sselink 1993). This study will look at the link between hydrology and vegetative communities in the Everglades. The Everglades The Everglades form ed 5000 years ago in a time of rapid rise in sea level that created the current hydrologic conditions th at have been changing ever since(Gunderson and Pritchard 2002). From the regional landscap e level the Everglades appear to have a uniform topological gradient and vegetative type. At the local level, however, it is a changing and dynamic mosaic driven by topological features such as water depth, hydroperiod, and geology (rock, peat depth and type, or sand). Since the early 1900s the South Florida freshwater wetlands have been reduced from 3.5 million hectares to about 1.8 million hectares and have been impounded by about 2000 km of dikes and canals (Kitchens et al. 2002). In recent years there has been a massive effort to restore the Everglades and protect the endangered sp ecies that are native to it. In order to restore the Everglades, however, a comprehensive monitoring and management plan must be developed and our understanding of the dynamic Everglades system must be improved. In order to restore what remains of the Ever glades, the key ecological driving forces need to be determined (Sklar et al. 2002). As a result of impoundment caused by the building of dams, dikes, and canals, the vegetative comm unities and subsequently the fauna of the Everglades are being altered. Th ere is a flattening of the topo graphy created by the slow flow, about 2 cm/km, of the watershed before impoundment (Leach et al. 1972). Over time, the Everglades are becoming a monoculture due to lack of flow and lack of variation in hydrology. An ecosystem needs all of the historical natural diversity of abiotic processes to survive disturbance (Gunderson and Holling 2002). The Ev erglades have lost biotic and process

PAGE 13

13 diversity needed to support the native flora a nd fauna through disturbances like drought, fire, flood, and hurricanes. Flora and Fauna of the Everglades Seven physiographic landscapes comprised th e Everglades before drainage. Three of these landscape types have been lost to devel opment and the largest remnant of the wet prairieslough-tree island-sawgrass mosaic can be f ound in Water Conservation Area 3A (WCA3A) (Davis et al. 1994). Of these ha bitats wet prairie has been identified as the type most used for foraging by the endangered snail kite (Karunaratne et al. 2006). Generally wet prairies are found in areas covered by surface water for much of the year and where the water level is not more than 30.48 cm below ground level except in extrem e drought. They frequently form transition zones between the sawgrass communities and th e slough communities. Some of the indicator species for wet prairies are beak rush ( Rhynchospora spp .), madiencane (Panicum spp .) and spike rush ( Eleocharis spp .). Sloughs are shallow, from a few centimeters to a meter or two deep, natural channels, which typically have wate r most of the year. They are characterized by white water-lily ( Nymphaea odorata ) and bladderwort ( Utricularia). Sawgrass strands are typically drier communities composed primarily of Cladium jamaisences (Loveless 1959). Changes in hydrology (source, timing, durati on, and depth of water) change plant communities (Mitsch and Gosselink 1993). Huma n induced development and management reduces the natural variation a nd extremes in hydroperiod. Hydrol ogy and fire are the dominant influences on the current condition of plant co mmunities. Both hydrology and fire have been altered by past human activities a nd are under management control so they are central parameters in future management schemes. Hydrology in fluences primary production, decomposition, and export of particulate organic ma tter. Nutrient cycling and nu trient availability are both significantly influenced by hydrologic conditions.

PAGE 14

14 WCA3A South has a total area of about 62,000 hectares and is bounded by Tamiami Trail (Hwy 41) to the south, Alligator Alley (I-75) to the north, Big Cypress National Preserve to the west, and Water Conservation Area 3B to the east. WCA3A was the main reproductive unit for the snail kite, which is an enda ngered species whose U.S. populati on is restricted to the South Florida freshwater wetlands. The snail kite feeds mainly on apple snails ( Pomacea paludosa ) and forages most effectively in a wet prairie habi tat where it is easier for them to see the snails when they crawl up the emergent vegetation to breat he (Kitchens et al. 2002). There is a matrix of freshwater habitats in the study area from relatively dry tree islands and sawgrass strands to wet prairies and deep-water sloughs. It is believed that the foragi ng and nesting habitat of the sna il kite is decreasing in extent and quality, and that this is c ontributing to a populati on decline. In recen t years wet prairie communities used for foraging by have been transforming to slough communities because of excessive hyrdoperiods (Kitchens et al. 2002, Zweig and Kitchens 2008a ). Before this trend, the water management in the 70s, 80s, and 90s ke pt WCA3A drier than it historically was. During this time period there was a 25% loss in prairie and slough and a correlated gain in sawgrass (White 1994). Essentially, management has been counter to c onditions necessary for wet prairie communities for the past 40 years by moving from one extreme to another. Recognizing the magnitude of this critical habi tat lost and understandi ng the reason it was lost, will lead to better management of the Everglades. A better management plan will lead to a more successful restoration attempt. There have been a number of descriptive studies describing this trend from wet prairie to slough (Wood and Tanner 1990, David 1996) but this study will quantify it for the first time.

PAGE 15

15 CHAPTER 2 METHODS Study System Water Conservation Area 3A ( WCA3A), the la rgest intact remnant of ridge and slough landscape left in the Everglades, is located in the South Florida watershe d as shown in Figure 21. The Everglades formed just 5,000 years ago( Gleason et al. 1984), but ha ve been drastically changed in the last 100 years by impoundment for development. WCA3A South has a total area of about 62,000 hectares and is bounded by Tamiami Trail (Hwy 41) to the south, Alligator alley (I-75) to the north, Big Cypress National Preser ve to the west, and Water Conservation Area 3B to the east. There is a matrix of freshwater habitats in WCA3A South ranging from relatively dry tree islands and sawgrass strands to wet prairi es and deep-water sloughs. The plots for this study are located in Southern WCA3A. The St udy Area does not extend all the way to Alligator Alley but stops in central WCA3A. This central area of 3A is referred to in this thesis as north as it is the location of the no rthern study plots. Sampling Design This is an ongoing study for which bia nnual sampling has been carried out since Nove mber of 2002. The sampling took place in twenty 1 km2 plots which were established in a stratified random manner over sout hern WCA3A (Figure 2-1). Thr ee gradients were considered when deciding plot placement: snail kite nesting density, impoundme nt effects, and ground elevation. Snail kite nesting act ivity and density was determined from nesting records for the past six years. The impoundment effect relate s to the differing hydrol ogy of central and south WCA3A. Water was not always deeper in sout hern WCA3A, the deeper water is a result of impoundment, associated with the Tamiami Trail. By placing half of our plots in central 3A in relatively intact ridge and slough ha bitats with sheet flow and the other half in southern 3A it

PAGE 16

16 was possible to compare vegetation changes in a relatively unimpounded wetland to those in a mostly impounded system. The last gradient considered was peat depth with little to no peat in the west and gradually deeper pe at to the east (Powers 2005). Sampling In each p lot 2 or 3 belt transects were placed perpendicular to the plant community zonation. Transect varied in length because they started in one community and moved into at least one more if not many more. For example a typical transect might have run from a sawgrass strand through a wet prairie to e nd in a slough. A belt tran sect consisted of th ree pairs of sample transects and extended through tw o or three plant communities. Transects varied in length because they had to start in the middle of a clearly definable reference community and extend until they were in the middle of another clearly definable reference community. There were three 1 meter wide paths/walkways between the sampling transects and the distance between the centers of the walkways/transects was 4 meters. An aerial view of a typical belt transect can be seen in Figure 2-2. Samples, taken every three meters along a single trans ect, consisted of all the rooted vegetation in a 0.25 m2 area. Biannual sampling has ta ken place since 2002, once at the end of the dry season in April or May and once in October or November at the peak of the standing crop or end of the wet season. The tran sect within the belt transect to be sampled was selected randomly by choosing a letter C to N excluding any letters that had already been sampled. For example, samples C, D, E, and F were on the same transect but C and D were on one side of the transect and E and F were on the other side. C and E were offset 1.5 meters from D and F and in this way with just 3 transect lines, 12 different sampling events could occur (Powers 2005).

PAGE 17

17 Hydrology To quantify the independent variables associat ed with hydroperiod, water levels have been monitored since 2002. There is spatial variabilit y in rainfall so data from 17 wells spread throughout south WCA3A at our sa mpling plots was used. Water depths at our sampling sites can be hindcast back using wells in place before 2002 to look at the historical hydrology of the site (Conrads et al. 2006). Th e hydrologic variables thought to be of the most importance in determining which seasons are wet and which are dry are minimum, maximum, and mean seasonal water depths as well as the duration of the wet or dr y event (calculated by taking the percent of time water levels fell in the lower qu artile of water levels for that season and the percent of time wate r levels fell in the upper qu artile of water levels for that season). Lower and upper quartiles for wet and dry seasons were de termined by combining all depths for a season from 1992 to 2007 and dividing them into quartiles. Using quartiles made it possible to obtain an idea of inundation and exposure time, which could not be obtained from a minimum or a maximum, and has a significant effect on vegeta tive communities. The dry season was defined as occurring from November to June and the wet season was from June to November. Water levels back to 1992 were used b ecause it was the beginning of the newest hydrological era in WCA3A (Figure 2-3). Hydrologi c alterations of the na tural system began in South Florida around the turn of the century. A consequence of these alterations is the management or regulation of the natural system for a variety of uses. Water levels in WCA3A were not recorded until 1953 but from that time on changes can be seen in regulation schedules starting with a relativel y dry era in the 1950s and getting pr ogressively wetter as the needs of wildlife began to be considered. The hydrological eras in WCA3A do not correspond to natural droughts or floods lasting 15 to 20 y ears but to changes in manageme nt priorities and consequent changes in regulation schedules.

PAGE 18

18 Data Analysis There were several steps in the analysis to look at the transition probabilities of various Everglade s vegetative communities and the factors dr iving the transitions. The first step was to use a multivariate community classifying technique in program PC-ord to classify the communities in WCA3A at each sampling occasion. H ydrological variables were used to cluster water levels for each season into categories of dry, normal, and wet time periods. This data was then put into a matrix for use in program MARK which estimates the transition probabilities for the vegetative communities in WCA3A. Multivariate Community Classifying Analysis The vegetation sam pled for this study had a wide variety of growth forms from Cladium Jamaicense with a few blades per sample, but considerable biomass per blade, to Eleocharis elongata with many stems per sample, but low biomass per stem. To deal with this problem a relativizing index termed importance values (IV) from McCune and Grace (2002) was used. Calculating IV helps when high density and low biomass species need to be compared to low density high biomass species. It was important to know the relative im portance of each species in each a priori community so that each community with its specific ratio of importance values could be clustered to find which communities were similar to each other. To calculate IV the first step is to determ ine the relative density and biomass for each species present in a community group. Relative density is calculated by taking the density of a species in a particular 0.25 mete r sample and dividing it by the sum of the density of all the species in the plot. Relative biomass is calcul ated in the same way using the biomass of an individual species in an indivi dual sample and dividing it by th e sum of the biomass of all the species in the plot. IV is simp ly relative density plus relati ve biomass divided by 2 and then

PAGE 19

19 multiplied by 100 (McCune and Grace 2002). Each species in each sample has an IV which can be grouped with the IVs of the same species in each a priori designated community. Using the multivariate statistics program, PC-ORD (McCune and Grace 2002), a hierarchical, agglomerative cluster analysis was done using the IV for each species from each of these communities for each sampling occasion to determine if the commun ities remained in the same cluster or moved to a different one at each sampling occasion. Sixty six a priori community groups were used in the cluster anal ysis with about 9 sampling occasions in each a priori community group for a total of 477 communities for PC-ord to cluster. The distance measure used to cluster was Sorensen or Bray-C urtis which is shared abundance divided by total abundance. This is a good distance measure fo r non-negative proportion (IV) data (McCune and Grace 2002). The group linkage method was flexible beta which is combinatorial and flexible with respect to space. The optimal number of clusters was chosen us ing an indicator species analysis which also allowed us to identify the most important spec ies in each cluster. A species which has an indicator value of 100 is a perfect indicator of that group or occurs exclusively and always in that group. Indicator values of 0 mean no indication. Indicator values were also tested for statistical significance using 1000 Monte Carlo runs. If groups are too fine ly divided, their indicator values will be low and if groups are too large, their internal heterogene ity will reduce indicator values. Indicator values were calculated for each species at 20 clusteri ng levels (from 1 to 20 clusters) and the p-values for all species at each clustering level were averaged. The cluster level with the smallest average p-value is the most informativ e and best fitting (McCune and Grace 2002). The clusters were then designated as wet prairie, slough, sawgrass, or tree island using our knowledge of the species compositions of each of these community types.

PAGE 20

20 Hierarchical Clustering Analysis to Categoriz e Wet and Dry years Water levels from the 17 wells placed in WCA3 A near the sampling sites were hindcast to 1992 using artificial neural networks (Conrads et al. 2006). Figure 2-3 explains why water levels from 1992 to present are considered to be in th e same era. The hydrologic variables from the hindcast water levels that were thought to be of the most importance in determining which years were wet or dry were; minimum, maximum, a nd mean seasonal water depths as well as a duration of high and low water proxy which was the percent of time water levels fell in the lower quartile of water levels for that season and the percent of time water levels fell in the upper quartile of water levels for that season. These values were ca lculated for each wet and dry season and run through separate hierarchical clus ter analyses; one for wet seasons and one for dry seasons. The hierarchical cluster analysis was run in program R using a set of dissimilarities for the objects being clustered. When R clustered, it initial ly assigned each object to its own cluster and then used an algorithm to proceed iteratively, at each stage joini ng the two most similar clusters, continuing until there was just a single cluster. At each stage distances between clusters were recomputed by the LanceWilliams dissimilarity update formula according to the complete clustering method. The complete cl ustering method finds similar clus ters. The algorithm used in hclust command in R orders the subtrees so that the tighter cluster is on the left (the last, i.e., most recent, merge of the left subtree is at a lowe r value than the last merge of the right subtree). The results from these cluster analyses allowed us to classify each wet season as either wet or normal and each dry season as either dry or normal. Multistate Modeling Only wet prairie and s lough community states from the multivariate community classifying analysis were used in the input matrix fo r program MARK. For each sampling occasion each a

PAGE 21

21 priori community was labeled either P for wet prai rie or S for slough based on the results of the community cluster. Each a priori community was also grouped as either N or S depending on whether it was located in the north or south of our sampling area. This input matrix was put in MARK where it was possible to classify each samp ling occasion as either wet, normal, or dry by assigning it a parameter. Models were run to test for a north/sout h effect, a seasonal effect, and a wet/dry time period effect. Likelihood based multistate models were used to estimate transition probabilities among vegetative community states. I defined AB as the probability that a community in state A at time t is in state B at time t +1. In our application, there were two states; slough communities denoted S and wet prairies communities denoted P Communities had two options from one sampling occasion to the next, they c ould start as wet prairie ( P ) and persist as wet prairie (P ) or transition to slough ( S ). Accordingly if they started as sloughs ( S) they could pers ist as sloughs ( S) or transition to wet prairies ( P ). I wanted to know what the prob abilities of these transitions were for each sampling occasion and what environmental drivers could be linked to these transitions. Four environmental factors that could influe nce transition probabilities were considered. The effect of wet and dry season on was denoted SEAS, and, by extension, the model that included a seasonal effect on was denoted (SEAS). We also include d wet and dry years as a factor (denoted WDyr). Models that had three groups of years (WDyr3cat), wet, dry, and normal were used, as were models with just two groups (WDyr2cat), wet or dry, to determine the effect of wet and dry years. A covari ate (covar) of percent of time wa ter levels were in the lower quartile of all water levels for that season was al so used to test for the effect of wet and dry years. The effect of the spatia l location of the study site, north versus south, was denoted (NS) and can be considered an indicator of impoundme nt effects. In addition, we considered two

PAGE 22

22 temporal structures: time variation denoted ( t, which assumed that vary over time), and no time variation denoted (. which assumed that remain constant over time). Finally we allowed some of the factors to interact (the interaction between two factors was de noted *, for instance model (WDyr*NS)). In general when running models in MARK the sin link function was used because provides constraints that keep the real parameters in the [0,1] interval, ye t allows the number of parameters to be correctly estimated. The identity matrix was used for all the models except the models containing real covariates (covar) when the design matrix had to be used. When models with covariates were run the logit link function was used. The logit link function is used with design matrices and real covariates because the logit link function is monotonic and the sin link function is not. This means that multiple values of the sin function will produce exactly the same transformation (sin(x), sin(x+2), and sin(x+4 ) all produce the same transformation) which is not true for the logit function (Cooch and White 2008). All link functions are transformations of proba bility such that the transformed probability changes from being either 0 or 1 (P or S) to They make it so that the probability of an event (transition) is a linear f unction of a vector of explanator y variables. Link functions are essentially performing a regression (in the case of the logit link function a logistic regression) and the resulting parameters are the estimates of the slope in the linear model. Logit link functions have problems estimating parameters that are close to 0 or 1. For this reason the sin link function should be used when possible but when the design ma trix and therefore the logit link function must be used and there are proble ms with the transitions from a stratum not summing to one, mlogit should be used to constrai n the parameters to sum to 1. It was not

PAGE 23

23 necessary in my analysis to use mlogit as the transitions from a stratum summed to one because there were only 2 transit ons (Cooch and White 2008). The parameters for survival (S) and detecti on (p) were both fixed to 1 as we were not trying to estimate those quanti ties. In our application surv ival and detec tion of plant communities were set to 1 because communities were always detected and transitioned to one of the two states but never died. The alternat e optimization method used for these models was simulated annealing which made sure that th e global maximum and mini mum and not the local maximum and minimum were found (Cooch and White 2008). Model selection We developed a set of candidate m o dels in order to evaluate our a priori hypotheses. Each model corresponded to a mathematical form ulation of our hypotheses. For example [ps sp](WDyr3cat) means that the probability of transitioning from wet prairie to slough is not equal to the probability of transitioning from slough to wet prairi e and that this shift is a function of wet, normal, and dry time period classificatio ns. We used Akaike Information Criterion (AIC) to select the models that provided the most parsimonious desc ription of the variation in the data (i.e., model with the lowest AIC) (Bur nham and Anderson 2002). Adding parameters to a model increases the fit of the model but reduces the precision of parameter estimates. AIC values account for this by penalizing the better fit or lower deviance of more parameterized models for the reduced precision of the estimates themselves. AIC= -2ln(L)+2K, L is the model likelihood which goes up as fit ge ts better and makes the AIC go down. K is the number of parameters which causes the AIC to increase or get worse. We used AICc weight ( w ) as a measure of relative support for each model. Values of w range from 0 to 1 with 0 indicating no suppor t from the data, and 1 i ndicating maximum support

PAGE 24

24 (Burnham and Anderson 2002). We also presented AICc ( AICc for the i th model was computed as AICci min (AICc), see Burnham and Anderson 2002). Effect size Effect size is used to determ ine whether the size of the difference between two estimates is significant. The difference between two estimates of transition probabilities were computed by calculating the arithmetic difference between these estimates to get an effect size (ES). The difference between the two estimates of transition pr obabilities were considered to be statistically significant when the 95% CI of the ES did not overlap 0 (Cooch and White 2008).

PAGE 25

25 Figure 2-1. Location of WC A3A and the study plots.

PAGE 26

26 Figure 2-2. Aerial schematic of be lt transect sample collection. A be lt transect consists of 3 sets of four sample transects, labeled C th rough N. A 1 m wide walkway is exists between each pair of sample transects. The distance between the center of a walkway and the center of the next walkway is 4 m. The transect to be sampled on a particular date is randomly selected (see Belt Transect field book). All the vegetation rooted in 0.25 m2 plots will be collected every 3 m al ong the transect that is sampled on that date. The second time that transect is sampled the plots are offset 1.5 m from the previous starting point. Poles have been pl aced in the field at the beginning and end of transect G/H/I/J to guide future placemen t of the other transects. Transect E is always to be positioned to the left of the G/H/I/J transect poles from the slough end (start) of the transect. H IJ EF Transect Poles 0.25 m 2 sam p le Walkwa y 3 m G DC KL MN

PAGE 27

27 10.00 12.00 14.00 16.00 18.00 20.00 22.00 24.00 26.00 28.00 30.00 32.001 95 3 1956 1 95 9 1 96 2 1965 1 96 8 1971 1 97 4 1977 1980 1 98 3 1986 1 98 9 1992 1 99 5 1 99 8 2001 2 00 4 2007Water Level (cm) Figure 2-3. Water levels at S65 and S 64: a new hydrological era began in 1992.

PAGE 28

28 CHAPTER 3 ESTIMATING TRANSITION PROBABILIT IE S AMONG EVERGLADES WETLAND COMMUNITIES USING MULTISTATE MODELS Introduction South Florid a freshwater wetlands have been reduced from 3.5 million hectares to 1.8 million hectares in extent and have been im pounded by 2000 km of dikes and canals as a result of agricultural and urban development in the last hundred years (Mitsch and Gosselink 1993, Kitchens et al. 2002). In order to reverse some of the adverse im pacts of impoundment one of the most ambitious ecosystem restor ation projects ever, the Compre hensive Everglades Restoration Project (CERP) has been undertaken (RECOVER 2005). One of the stated goals of this project is to promote conditions that will increase the abundance and diversit y of native species by regulating water in the system (R ECOVER 2005). In order to accomplish this goal, it is critical to develop reliable models of how hydrology aff ects the dynamics of plant communities in the Everglades. Unfortunately, there is very little information in this critical area. To address this issue, our study will present a comprehensive fr amework for investigating multiple competing hypotheses about the factors gove rning transition probabilities among vegetative community states in the Everglades. This framework allows for the calculation of robust estimates of transition probabilities and estim ates of uncertainty (process and sampling variance associated with these estimates). The estimation of various t ypes of uncertainty is pa rticularly important for making informed decisions for cons ervation (Martin et al. 2008c). Our study focused on the transition probabi lities between wet prairie and slough community states because of their impor tance to the endangered snail kite ( Rostrhamus sociabilis ) population, which has been selected as one of the key performance measures of the ongoing restoration activities a ssociated with CERP (RECOV ER 2005, Martin et al. 2007a, Martin et al. 2008c). Wet prairies are defined as areas that are covered in surface water for much

PAGE 29

29 of the year and where the water level does not drop more than a foot below ground level except in extreme drought (Loveless 1959). They fre quently form transition zones between drier sawgrass communities and wetter slough communities. We t prairie habitat is ideal for snail kite foraging because of its sparse emergent vegetati on. Sparse emergent vegetation enables the snail kite to easily see its primary food source, the apple snail, when it emerges to breathe, making wet prairie the habitat in which they forage most ef fectively (Kitchens et al 2002, Karunaratne et al. 2006). Some of the indicator species for wet prairies are beak rush ( Rhynchospora spp. ), maidencane ( Panicum spp. ), and spike rush ( Eleocharis spp. ). Sloughs are shallow, a few inches or feet deep, natural channels, that have water most of the year. They are characterized by white water-lily ( Nymphaea odorata ) and bladderwort Utricularia (Loveless 1959). The primary objective of our study was to pr ovide the first estim ates of transition probabilities between wet prairie and slough communities using multistate models. Although a number of authors (Kolipinski and Higer 1969, McPherson 1973, Dineen 1974, Alexander and Crook 1975, Zaffke 1983, Wood and Tanner 1990, Davi s et al. 1994, David 1996) have proposed verbal or conceptual models of how these tr ansitions may proceed, there are few mechanistic mathematical models that can translate conseque nces of environmental variation or management actions on community dynamics in the Everglades. Here we use likelihood-based multistate models to estimate transition probabilities am ong wet prairie and slough communities. This type of model is now commonly used to estimat e movement probabilities of organisms among discrete geographic units or physiologic states (Blums et al. 2003, Martin et al. 2007b). However, these models have rarely been used to evaluate multiple competing hypotheses about factors governing the dynamics of plant communities.

PAGE 30

30 Hypotheses and Predictions Hypothesis 1: Wet and Dry Seasons Influence the Conversion of Slou ghs and Wet Prairies Precipitation is the main route by which water enters the Everglades ecosystem (Duever et al. 1994) and the dominant source of natural surface freshwater in this part of the Everglades. Rainfall in southern Florida is highly seasonal with 60% occurring from June to September and only 25% occurring between November and April. The result of this rainfall pattern is a hydroperiod that has strong effects on vegetation composition and structure and which exhibits natural periodicity or substantial and predicta ble within year seasonal variation (White 1994) (Figure 3-1). The vegetation of the Everglades is adapted to this seasonal environment in its rhythms of production, decomposition, mortality, and reproduction. Therefore, I predict the transition probabilities from wet prairie communities to sloughs to be greater during wet seasons which occur in the interval from June to N ovember. In contrast, I predict the transition probabilities from slough communities to wet pr airie communities to be greater during dry seasons which occur in the interval from November to June. Hypothesis 2: Wet and Dry Years Substantially Influence the Process of Conversion of Sloughs and Wet Prairies The hydroperiods of m ost wetlands vary significan tly from year to year with 10 to 20 year cycles (Mitsch and Gosselink 2007) (Figure 3-1). In South Florida wetlands, precipitation which has a significant impact on hydroperiod has hi gh interannual variabil ity with documented extremes from 86 cm to 224 cm for the pe riod from 1951 to 1980 (NOA A 1985, Obeysekera et al. 1999). The El Nino Southern Osci llation is responsible for much of the variability in rainfall (Puckridge et al. 2000), but it is difficult to detect a clear in terannual wet dry cycle in South Florida as hurricanes are frequently the cause of wet years. Extreme values of precipitation are encountered in the Everglades on a time period of 3 to 10 years. With this in mind, I predict the

PAGE 31

31 transition probabilities from sl ough communities to wet prairies to be greatest during dry intervals. In contrast, I predict the transition probabilities from wet prairie to slough communities to be greater during wetter intervals. Hypothesis 3: Probabilities of Transiti on betw een Sloughs and Wet Prairies are Substantially Influenced by Impoundment Impoundment has eliminated sheet flow fr om the Everglades and caused excessive ponding in the southern ends of the Water Cons ervation Areas (WCAs) while over-draining the northern ends (Dineen 1972, Light and Dineen 1994). Impounded wetlands have vertical rather than lateral expansions/retractions which cause a loss in intra and inter wetland heterogeneity (Kitchens et al. 2002). This is causing conversio n from wet prairie and sawgrass communities to deeper, more aquatic slough habitats in the so uthern area of the WCAs due to prolonged hydroperiods (Kitchens et al. 2002). In southern sites, I predict there will be more conversion from wet prairies to sloughs and less conversion from sloughs to wet prairies. In the northern sites, I predict less conversion fr om wet prairies to sloughs and more conversion from sloughs to wet prairies. Methods Study Area and Sampling Methods Our study was located in the southern porti on of Water Conservation Area 3A ( WCA3A) in the Everglades of South Florida, USA (Figure 3-2). In the fall of 2002, twenty 1 km2 plots were placed across three landscap e gradients: an east-west peat depth gradient, an artificial north-south water depth gradient, and a Florida snail k ite nesting activity gradient in a random stratified manner. Two or three belt transects were placed in each plot perpendicular to ecotones or moving from one a priori community type (slough, sawgrass, tree/shrub island, Typha, and wet prairie) into another. Samples were collected every 3 meters along belt transects twice a year

PAGE 32

32 at the end of the dry (May/June) and wet (N ovember/December) seasons. A sample is a 0.25 m2 area from which all standing biomass has been cl ipped at peat level including any submerged aquatic plants. The 0.25 m2 samples represent pse udorepeated measures, as destructive samples were taken and I could not resample the exact location. There were eight sampling events from November 2002 to June 2006 during which 33,501 samp les were taken. The samples were sorted by species, counted, dried, and weighed. In a ddition, 17 water level monitoring wells were placed in the plots to take twi ce daily water level readings. Data Analysis Multivariate analysis to classify co mmunities The relative density and bioma ss for each species present in a plot were calculated to determine an importance value (IV) for each species in each a priori community in the plot. Relative density or biomass is calculated by ta king the sum of the density or biomass for each species and dividing it by the sum of the density or biomass of a ll species in the plot. IV is simply relative density plus relative biomass divided by 2 and then multiplied by 100. IV is a relativizing index that helps to account for high density and low biomass species and high biomass low density species (McCune and Grace 2002). A priori community designations were used to group each 0.25 m2 sample into communities for each plot. Using the multivariate statistics program, PC-ORD (McCune and Grace 2002), a hierarchical, agglomerative cluster an alysis was done using IV from each of these communities for each sampling occasion to determin e if the communities remained in the same cluster or moved to a different one. The optimal number of clusters was chosen using an indicator species analysis which also allowed us to identify the most important species in each cluster. The clusters were then designated as wet prairie, slough, sawgra ss, or tree island using our knowledge of the species compositions of each of these community types.

PAGE 33

33 Hierarchical clustering analysis to categoriz e wet and dry years Water levels at our sampling site s were hindcast using artificial neural networks to look at the historical hydrology of the s ite (Conrads et al. 2006). As th e newest hydrological era in WCA3A began in 1992, water levels from the past 16 years have been featured in Figure 3-3 and used in the cluster analysis. The hydrologic variables that were thought to be of the most importance in determining which years were wet or dry were; percent of time water levels fell in the lower quartile of water levels for that s eason, minimum seasonal water level, percent of time water levels fell in the upper quartile of water levels for th at season, maximum seasonal water level, and mean seasonal water depth. These va lues were calculated for each wet and dry season since 1992 and run through separate agglomerative cluster analyses; one for wet seasons and one for dry seasons. This allowed us to classify each wet season as either wet or normal and each dry season as either dry or normal. Multistate modeling Likelihood based m ultistate models were used to estimate transition probabilities among plant community states. I defined AB as the probability that a community in state A at time t is in state B at time t +1. In our application, there were two states; slough communities denoted ( s ) and wet prairies communities denoted ( p) (Figure 3-4). I considered four factors that could influence transition probabilities. Th e effect of wet and dry season on was denoted SEAS, and, by extension, the model that included a seasonal effect on was denoted )( SEAS I also included wet ( W ) and dry (D ) years (yr) as a factor (denoted )( WDyr ). Models that had three categories () 3(][catWDyrspps ) for years, wet, dry, and normal were used, as were models with just two categories ()2(][catWDyrspps ), wet or dry, to determine the effect of wet and dry years. A covariate )(cov ar of percent of time water levels were in the lower quartile of all water

PAGE 34

34 levels for that season was also used to test for th e effect of wet and dry years. The effect of the spatial location of the study site, north versus south, was denoted ()( NS ) and can be considered an indicator of impoundment effect s. In addition, I considered two temporal structures: time variation denoted ( t, which assumed that vary over time), and no time variation denoted (. which assumed that remain constant over time). I allowed some of the factors to interact (the interaction between two factors was denoted *, for instance model )*( NSWDyr ). Finally, for all models I assumed that transition probabilities ps and sp were either identical (denoted ][ spps ); or were different (denoted][ spps ). I used program MARK to develop and analyze multistate models (White and Burnham 1999). Model selection I developed a set of candidate m odels in order to evaluate our a priori hypotheses. Each model corresponded to a m athematical formulation of our hypotheses. I used Akaike Information Criterion (AIC) to select the models that provi ded the most parsimonious description of the variation in the data (i.e., model with the lowest AIC) (Burnham and Anderson 2002). I used AICc weight ( w ) as a measure of relative support for each model. Values of w range from 0 to 1 with 0 indicating no support fr om the data, and 1 indicating maximum support (Burnham and Anderson 2002). I also presented AICc ( AICc for the i th model was computed as AICci min (AICc), see Burnham and Anderson 2002). Effect size Effect size (ES) was calculated by taking the arithmetic difference between the two estimates of transition probabilities from the same model that were being compared. The

PAGE 35

35 difference between the two estimates of transition probabilities was considered to be statistically significant when the 95% CI of the ES did not overlap 0 (Cooch and White 2008). Results Multivariate Analysis to Classify Communities The indicato r species analysis, based on hierarchical, agglomerat ive cluster analysis of the Everglades WCA3A vegetation monitoring da ta, indicated that there were eleven communities/clusters (Figure 3-5). Using our know ledge of the system, I determined that there were 2 slough, 3 wet prairie, 4 sawgrass strand, and 2 tree island communities. Communities that were not initially classified as slough or wet prairie were removed from the data set used in the multistate analysis. There are several reasons why I removed the other communities from the data set. Most importantly, the data available would not have supported m odels with more than two vegetation states. Secondly, slough and wet prairie are the co mmunity types that are most relevant to management of snail kite habitat. Finally, one motivation of our study is to provide models of system behavior for the adaptive ma nagement of Everglades and WCA3A and most decision making tools require simple system models (e.g., Stochastic Dynamic Programming, (Martin et al. 2008c). Indeed, using more paramete rized models (models with more states) would substantially increase the state space and, therefore, would increase the difficulty of solving the decision problem. Hierarchical Clustering Analysis to Categoriz e Wet and Dry Years The agglomerative hierarchical cluster analysis of all dry seasons since 1992 found that the dry seasons of water years 1992, 2000, 2001, 2004, a nd 2006 clustered together and were dry. The dry seasons of water years 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2002, 2003, and 2005 clustered together and could be considered wet or normal (Figure 3-6). The wet seasons of water years 1995, 1996, 1998, 2000, 2004, and 2006 clustered togeth er and were wet. The wet seasons

PAGE 36

36 of water years 1993, 1994, 1997, 1999, 2001, 2002, 2003, 2005, and 2007 clustered together and were normal to dry (Figure 3-7). This lead to the designation of two dry time periods November 2003 to June 2004 and November 2005 to June 2006 and two wet to normal time periods November 2002 to November 2003 and November 2004 to November 2005. This designation could be further broken down to include two wet time periods June 2003 to November 2003 and June 2005 and November 2005 and two normal tim e periods November 2002 to June 2003 and November 2004 to June 2005 for a total of thr ee water categories: we t, dry, and normal. Multistate Modeling The m ost parsimonious model based on AIC weight was model ) 3(][catWDyrspps (AIC ( w ) =0.685) (Table 1). This model is a mathemati cal formulation of the hypothesis that wet and dry years influence transition probabilities be tween slough and wet prai rie communities. Based on this model, I found that estimates of ps were greater during normal years (ps =0.119 (SE=0.050)) than during dry years (ps = 0 (SE=0)) and wet years (ps = 0.042 (SE=0.041)). The difference in ps between normal and dry years was statistically significant (ES=0.119 (95%CI= 0.019 to 0.219)), but it wa s not statistically significant between normal and wet years (ES=0.077 (95%CI= -0.052 to 0.206)), nor betw een wet and dry years (ES=0.042 (95%CI= 0.039 to 0.123)). Also, based on model ) 3(][catWDyrspps I found that that estimates of sp were greater during dry years (sp = 0.181 (SE=0.067)) than during wet years (sp = 0.111 (SE=0.052)) and that there were no transitions from sloughs to we t prairies during normal years (sp =0 (SE=0)). The difference in sp between normal and dry years was statistically significant (ES=0.181 (95%CI=0.047 to 0.316)), as was the difference between normal and wet years (ES= 0.111 (95%CI=0.006 to 0.216)). However, the difference in sp from wet to dry

PAGE 37

37 years was not statistically significant (ES=0.071 (95%CI= -0.099 to 0.241)) (Figure 3-8). Based on AIC weight, all the other models received less support from the data. However, it is interesting to note that based on our second most parsimonious model )2(][catWDyrspps (AIC ( w ) =0.089), which used only wet and dry to categorize year s, estimates of ps were greater during wet years (ps = 0.091 (SE=0.035)) than during dry years (ps = 0 (SE=0)), and this difference was statistically significant (ES=0.091 (95%CI=0.020 to 0.162)). Similarly, based on model )2(][catWDyrspps estimates of sp were greater during dry years (sp = 0.182 (SE= 0.067)) than during wet years (sp = 0.048 (SE=0.023)), but this diffe rence was not statistically significant (ES=0.134 (95%CI= -0.008 to 0.276)) (Figure 3-9). Discussion This study provides the first estim ates of transition probabilities between slough and wet prairie communities in the Everglades ecosystem s from likelihood based multistate models. This approach allowed us to evaluate hypotheses ab out the factors governing the shifts from one community type to another and to relate such shifts to water conditions. Our results provided support for our 2nd hypothesis, that the probability of c onversion from wet prairie to slough is greater during normal and wet years than during dr y years, whereas the probability of transition from slough to wet prairie is greater during dry y ears than normal and wet years. In determining which years were wet, normal, and dry I used mean, minimum, maximum water depths, as well as a duration proxy which was the percent of ti me water levels were in the upper or lower quartile of all water levels for that season. In essence I combined many of the factors found in other studies (Kolipinski and Hi ger 1969, Dineen 1974, Zaffke 1983) to be correlated with plant community conversion to categorize each year for which I had plant community data. It is not therefore, surprising that the model ) 3(][catWDyrspps was the most parsimonious in the model

PAGE 38

38 set. The estimates found here support the con ceptual models posed by (Kolipinski and Higer 1969, Dineen 1974, Zaffke 1983, Zweig and Kitchens 2008b) but are based on empirical data and on statistically robust estimators. In the most parsimonious model ) 3(][catWDyrspps there were a few anomalous transitions from slough to wet prairi e, during wet years, for which there is a logical explanation. Almost all the anomalous transitions occurr ed in northern WCA3A in 2005. Hurricane Wilma passed over the Everglades in October 2005. There wa s not a lot rainfall in the Everglades from this hurricane but there were wind speeds up to 195 km/h. The hurricane force winds caused an interesting phenomenon, which has been noted in other studies as well, to occur (ScienceCoordinationTeam 2003, Larsen et al. 2007). The wind blew the submerged aquatic vegetation, a main indicator of sloughs, out of the sloughs maki ng samples taken in November appear like wet prairie samples in the cluster analysis because they had lost their main slough indicator species. The 3rd most parsimonious model in the model set supported to some extent the 1st hypothesis and included the intera cting affects of wet/dry ye ars and seasons while the 4th most parsimonious model showed some support for the 3rd hypothesis and included the interacting effects between wet/dry years and the north/south impoundment a ffect. The vegetation of the Everglades is adapted to cyclical seasonal changes in water levels and has shown a high resilience to seasonal change as explained in our 1st hypothesis (White 1994). Species of Eleocharis have been shown to recover from comple te die off due to drought in just 9 weeks when the water returns (Edwards et al. 2003). It is not, therefore surp rising that although seasonality is important in community transitions it is not the main driver and that longer time scale cycles of wet and dry years are. It was su rprising to learn that impoundment as outlined in

PAGE 39

39 our 3rd hypothesis did not have more of an effect on community transitions showing up first as an interacting factor in our 4th model but this may be a questi on of time scale. The impoundment effect in WCA3A started when levee 29 was c onstructed in 1962, which means that all of the plots in the southern area of WCA3A had been subjected to the effects of inundation for 40 years before the first samples were taken. Much of th e plant community conversion due to deep water in southern WCA3A had probably already occurred by 2002 as it wa s noted in a number of past studies (Alexander and Crook 1975, Zaffke 198 3, Wood and Tanner 1990, David 1996). Our results show that the over riding driver of plant community conversion in WCA3A at the 4 year time scale are cycles of wet and dry years. The estimates provided in this study from our most parsimonious models are very valuable for Everglades restoration and management. Indeed, our estimates can be incorporated into management models (e.g., Markov chain models) to predict how management actions, like water level regulations, will affect the proportion of habitat occupied by wet prairie or slough communities (Martin et al. 2008c). These models can also be used to predict the effect of global changes on the dynamics of vegetative communities in the Everglades. Dynamics of wet prairie and slough communities can be described by the expression below 1 tt Where ssps sppp is a projection matrix, s p t is a vector with p representing the occupancy of wet prairies (i .e., proportion of habitat oc cupied by wet prairies) and srepresenting the occupancy of slough communities. If the probability of transition among the community states can be assumed to be constant over time then a system governed by the above expressions will attain dynamic equilibrium (C aswell 2001, MacKenzie et al. 2006, Martin et al.

PAGE 40

40 2008b). The equilibrium occupancy for each commun ity state, or the proportion of habitat occupied by each community type, can be computed by calculating the first element of the right eigenvector associated with the dominant eigenvalue of the transition matrix For instance, lets assume a 10 year scenario in which there ar e 4 wet years, 3 normal years, and 3 dry years. One can compute the average pr obabilities for each transition (e.g. 10 )33 4( sp Dry sp Normal sp Wet sp ), which if I used estimates from model )3(][catWDyrspps would lead to an averag e probability of 0.099 for sp and an equilibrium occupancy by wet prairies of 0.65 (i.e., at equilib rium occupancy for this scenario, 65% of the habitat would be occupied by wet prairies and th e remaining 35% by sloughs). This is just one example among many of how our estimates can be used to investigate the dynamics of vegetation communities. Our estimates can also be incorporated into more complex and realistic analyses (e.g., explicit incorporation of envi ronmental stochasticity) (Caswell 2001). For instance several scenarios of how alterations associated with global change would affect the dynamic of vegetative communities in the Everglades could be examined by varying the frequency of dry and wet years, see (IPCC 2007). Perhaps, of ev en greater relevance to management of the Everglades, one could use our appr oach to parameterize management models as part of a process of structured decision making and adaptive management (Martin et al. 2008c). The goal of such structured decision process is to determine decisions that are optim al with respect to management objectives (Williams et al. 2002, Martin et al. 2008c). For instance, managers may be interested in atta ining historical proporti ons of wet prairie in the Everglades without compromising the socioeconomic status of South Florida. This goal would be important to many

PAGE 41

41 native species that use wet prairi es but especially for the snail kite whose population is at great risk of extinction (Martin et al. 2007a, Martin et al. 2008a). Advocates of structured decision making and adaptive management emphasize the importance of considering severa l important sources of uncertain ty: model uncertainty, sampling uncertainty and environmental uncertainty. The a pproach that I have developed to model the dynamics of vegetative communities in the Everglad es, explicitly measures all of these sources of uncertainty. Model uncer tainty can be measured by AICc weight at least as an initial step, but a Bayesian approach is necessary for fu rther updating of the model weights at each implementation of management actions (Williams et al. 2002). Environmental uncertainty can be incorporated into the models by providing estimates for contrasted environmental conditions like wet and dry years. Environmental stochasticity can also be measured by computing the process variance associated with each transition probabili ty. Unfortunately, our monitoring data did not include enough years of record to measure process variance, but I believe that it will be possible to estimate this quantity as more monitoring da ta is collected. Finall y, the sampling variance associated with each estimate of transition probabilities can be incorporated into the management models to account for the uncertainty associated with sampling methods. In conclusion, our approach involved three st eps. First, vegetation communities in areas that have been deemed key to Everglades restoration projects were monitored. Second, multivariate analyses were applied to classify vegetative communities into states that are particularly relevant to management problems (e.g., management of vegetative communities that can affect performance measures such as Snail kites). Third, I applied likelihood based multistate models in order to evaluate multiple compe ting hypotheses about factors governing the dynamics of vegetative communities and to provide estimates of transiti on probabilities and associated

PAGE 42

42 measures of uncertainty which can then be in corporated into management models. Although, the models I developed for this study were fairly si mple, they provide a starting point from which additional levels of complexity can be added (as more data become s available). It is also worth noting that most methods to determine optimal decisions require relatively simple models (Williams et al. 2002, Fonnesbeck 2005). I hope th at ecologists and managers will find our framework useful for investigating the dynami cs of other vegetation communities and for implementing this new knowledge into the adaptive management of other parts of the Everglades and possibly other ecosystems.

PAGE 43

43 Table 3-1. Multistate models of transition probabilities ( ) for wet prairie to slough conversions (ps) and slough to wet prairie conversions ( s p). Model AICc AICC w K DEV )3(][catWDyrspps 107.798 0 0.685 6 52.724 )2(][catWDyrspps 111.863 4.064 0.090 4 61.006 )*(][SEAS WDyrspps 112.056 4.258 0.081 8 52.679 )*(][NSWDyrspps 112.916 5.117 0.053 12 44.671 )*(cov][NSarspps 115.153 7.355 0.017 4 64.297 )(][NSspps 116.098 8.299 0.011 2 69.379 )*(][NSWDyrspps 116.111 8.313 0.011 6 61.036 )2(][catWDyrspps 116.547 8.748 0.009 2 69.828 )*(][NSSEASspps 116.939 9.14 0.007 4 66.082 )(cov][arspps 116.947 9.149 0.007 2 70.228 )(][SEASspps 117.186 9.387 0.006 2 70.466 )(cov][arspps 117.537 9.738 0.005 4 66.680 )*(cov][NSarspps 118.068 10.269 0.004 8 58.690 )3(][catWDyrspps 118.167 10.369 0.004 3 69.390 The effect of wet and dry years were tested (WDyr), as was the effect of seasons (SEAS) and the north south water impoundment effect (NS). AICC is the Akaike Information Criterion, AIC is adjusted for sample size, w is AICC weight, K is the number of parameters, DEV is the deviance given by program MARK.

PAGE 44

44 19 21 23 25 27 29 31JanFebMarAprMayJunJulyAugSeptOctNovDecWater Depth in cm 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 Figure 3-1. Hydrograph of water levels in Wate r Conservation Area 3A from 1992 to 2006. The graph demonstrates the define d seasonal pattern in water levels as well the clear variation from year to year.

PAGE 45

45 Figure 3-2. Our study area, southe rn WCA3A, is shown here w ith the twenty study plots in black. The plots were placed in a strati fied random manner across landscape level gradients like peat depth, water level, a nd snail kite nesting co ncentration. All data used in this analysis came from transects placed in these plots.

PAGE 46

46 10 12 14 16 18 20 22 24 26 28 301953 1956 1959 1962 196 5 196 8 197 1 197 4 197 7 198 0 1983 1986 1989 1992 1995 1998 200 1 200 4 200 7YearWater Level in cm Figure 3-3. Stage water levels in WCA3A si nce 1953 from gauge station 3-65, site 3A-28. Several different water regulat ion schedules can be seen in the graph with the driest schedule in the 50s and 60s. Dashed box i ndicates newest era in water regulation schedule which started in 1992. pp Figure 3-4. Transition probabilities shows that sloughs will transi tion to wet prairies with a certain probability or remain as sloughs. We t prairies behave in the same manner, either transitioning to sloughs with a cer tain probability or continuing on as wet prairies. P S ps ss sp

PAGE 47

47 0% 20% 40% 60% 80% 100% Tree Islan d Tree Islan d Saw Grass St ra nd Saw Grass Str a nd Saw Grass Str a nd Saw Grass St ra n d Wet P r ai ri e Wet P r ai ri e Wet P r ai r ie Slo u g h Slo u g h Utricularia spp. Nymphaea odorata Nymphoides aquatica Eleocharis cellulosa Eleocharis elongata Paspalidium geminatum Panicum hemitomon Bacopa caroliniana Peltandra virginica Typha spp. Cladium jamaicense Sagittaria lancifolia Pontederia cordata Osmunda regalis Blechnum serrulatum Cephalanthus occidentalis Figure 3-5. Results from the Agglomerative Cluste r analysis used to determine the vegetation communities in WCA3A at each sampling event. An indicator species analysis was also done and used in combination with our knowledge of the system to label the communities.

PAGE 48

48 Figure 3-6. Cluster analysis used for all dry s easons since water year 1992 to determine which dry seasons were dry (dark grey) and which were normal/wet (light grey). Figure 3-7. Cluster analysis used for all wet s easons since water year 1992 to determine which wet seasons were wet (black) and whic h were normal/dry (light grey).

PAGE 49

49 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 Wet Prairie to SloughSlough to Wet PrairieTransition probabilities Wet Years Normal Years Dry Years Figure 3-8. Transition estimates fr om our most parsimonious model )3( catWDyr (AIC ( w ) =0.685) for wet prairie and slough communitie s using wet, normal and dry year classifications. For transiti on estimates from wet prairi e to slough the difference between normal and dry years was sta tistically significan t (ES=0.119, 95%CI= 0.019 to 0.219), but the difference in ps between normal and wet years as well as the difference in ps between dry and wet years was not significant. For slough to wet prairie conversions, the difference betw een normal and dry years was significant (ES=0.181; 95%CI=0.047 to 0.316) as was th e difference between normal and wet years (ES= 0.111; 95%CI=0.006 to 0.216) However, the difference in sp between dry and wet years was not significant.

PAGE 50

50 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 Wet Prairie to SloughSlough to Wet PrairieTransition Probabilities Wet/Normal Years Dry Years Figure 3-9. The transition estimates for wet prairie and slough communities from our second most parsimonious model )2( catWDyr (AIC(w)=0.089) which used wet and dry year classifications instead of wet, dry, and normal year classifications. The difference in between wet and dry years is signi ficant for wet prairie to slough estimates (ES=0.091, 95%CI=0.020 to 0.162) bu t not for slough to wet prairie estimates.

PAGE 51

51 CHAPER 4 CONCLUSIONS AND NEXT STEPS It was deter mined in the last Chapter that multistate models could be used to obtain transition probability estimates for vegetative commun ities. The next step is to increase sample size by dividing up the a priori community groups. With more samples it will be possible to evaluate a number of hypotheses about the vege tative community dynamics of WCA3A. It will also be possible, with a larger sample size, to obtain annual transition estimates by using data from just the wet season. The selection of a si ngle season to define tran sitions was prompted by the confusion of within year transitions complicating the year to year transitions across the landscape. The last set of an alyses pointed to hydrology as the main driver of community transitions. In this step th e five different aspects of hydr ology (mean, minimum, maximum, % time in upper quartile, and % time in lower quart ile) that were combined before, will be separated to determine which aspe ct of hydrology is most correlate d with community transition. Hypotheses and Predictions Hypothesis 1: Transition Probabilities Will B e Higher in the North Than in the South The northern study area is more hydrologically dynamic, reflecting sheet flow rather than impoundment influences. In theory natural wetland hydrology is associated with high transition probabilities, meaning that community transitions are occurring at a high rate, and communities are not stagnant, or tending toward monoculture (Kitchens et al. 2002). In effect wetland vegetative communities should change in concert naturally with changing water conditions (Mitsch and Gosselink 1993). It is expected that transition probability estimates will be higher for the northern plots than for the southern plots at all times.

PAGE 52

52 Hypothesis 2: The Hydrological Variables That Drive Tr ansitions in the North Will Be Different from the Hydrological Variable s That Drive Transitions in the South The northern study area is more hydrologically dynamic and drier than the southern study area. This leads to the hypothe sis that the hydrologi cal variables of maximum and % of time in the upper quartile will be most correlated with community trans itions in the north. In other words the north has been kept fairly dry so it is expected that most tran sitions will occur during periods of high water. The opposite is true in th e south which has been kept very wet. This leads to the hypothesis that the hydrologic variables of minimum and % of time in the lower quartile will be most correlated with community transitions in the south. Only during extreme dry down will the southern communities begin transitioning. Hypothesis 3: Transition Probabilities Will Have a Directionality Th at is Related to Water Levels The wetland communities in the slough/wet prairie/sawgrass strand/tree island landscape type occur on a hydrologic gradient. Sloughs are typically the wettest communities followed by wet prairies and then sawgrass strands with tr ee islands as the driest communities in this landscape. As conditions dry down, communitie s should begin transitioning up the hydrological gradient to drier physiognomic types. As the system rewets, the communities should begin transitioning back down the hydrological gradient to wetter physiognom ic types. It is predicted that the transition probabilities from slough to wet prairie communities will be higher in dry time periods. Conversely, the transition probabilities from wet prairie to slough communities will be higher in wet time periods. This trend was seen in the last analys es and is expected to be seen again when the larger data set is used. Preliminary Results All of the preliminary analyses presented below will be rerun using the new larger data set but some very ecologically significant trends were found even with a small data set.

PAGE 53

53 Annual Transition Probability Estimates Obtaining annual transition estim ates from this data set was difficult as using data from only wet seasons left just 4 occasions from which transition probabilities could be estimated. This decreased sample size caused an increase in standard errors and confidence intervals from those found when all 8 occasions were used. Th e candidate model set, including a north-south interaction and different hydrologic variables, was essentially the same as the model set in the last chapter although the effect of seasonality could no longer be te sted. The results can be seen in table 4-2. The most parsimonious model in this model set suggested that there were more transitions in the north than in the south but did not include the effect of time or hydrology (ps<>sp *NS) (AIC( w )=0.43350) (Figure 4-1). In th e north there were significantly more transitions from slough to wet prairie than from wet prairie to slough (ES= 0.434 (95%CI=0.084 to 0.785)). There was also significantly more transition in the no rth than in the south from slough to wet prairie (ES=0.402 (95%CI=0.015 to 0.789)). These findi ngs provided support for hypothesis one which was that transition probabilities w ould be higher in the north than in the south. Communities in the north also transitioned more to drier stat es than wetter states which was predicted in hypothesis three although the effect of time was not included in this model. The model-averaged results are also presented here as model averaging was used because there was some model selection uncertainty (Figure 4-2). Confidence intervals on the estimates are very large and there were no statistically significant effect si zes. Still the model averaged results indicate that transition probabilities were higher in the north than in the south. This is more support for hypothesis one which was related to the fact that the south is severely impounded and stagnant while the north is more hydrologically dynamic, reflecting sheet flow rather than impoundment influences.

PAGE 54

54 Separate Estimates for the North and South Two-community states: wet prairie and slough Two separate analyses were performed with two sm aller sets of data created from the initial larger data set. Essentially the same ca ndidate model set was use d, with the readdition of seasonality this time. Results from model averag ing are presented in Figure 4-3. These results showed that in the south wet pr airies were transitioning to slo ughs with more frequency than sloughs were transitioning to wet prairies in all time periods except the dry seasons of 2004 and 2006. On these occasions sloughs were more likely to transition to wet prairies than wet prairies were to transition to sloughs. This supports hypothesis three as the dry seasons of 2004 and 2006 had long duration drydowns that should have m oved communities toward the drier wet prairie state. These trends are ecol ogically interesting although the e ffect sizes on them were not statistically significant. In the north sloughs were more likely to transition to wet prairies than wet prairies were to transition to sloughs on all occasions except the wet season of 2004 and the dry season of 2005. During these relatively wet time periods, wet prairies were more likely to transition to slough than sloughs we re to transition to wet prairies so during wet time periods most movement in the north was toward a wette r community state. Although all these findings are ecologically significant and s upport hypothesis three, the confidence intervals of the effect sizes overlapped 0 and are therefor e not statistically significant. Three-community states: wet prairie, transition, and slough Wet prairie and slough represen t two physiognomic types within which there are multiple community states. Two community states from the vegetative community cluster analysis that had been classified as physiognomic types slough and wet prairie were reclassified as transition states between slough and wet prairie. The same ca ndidate model set as in the previous analysis was used.

PAGE 55

55 As in the previous analysis, transition probabi lities were generally higher in the north than in the south, with the excepti on of the dry seasons of 2004 a nd 2006. The hydrological variables that were driving community transitions were easier to identify in this set of analysis. In the north the third most parsimonious model included th e effect of the percent of time spent in the upper quartile of water levels for that season. Models that included this interaction had a combined AIC weight of 0.16061. This appeared to be the hydrologic variable that best explained community transition in the northern part of the study site. In the south the most parsimonious model included the effect of the percen t of time spent in the lower quartile of water levels for that season. Models th at included the effect of per cent of time spent in the lower quartile of water levels for the season had a combined AIC weight of 0.33407. These results showed the first support for hypothesis two whic h predicted maximum and % of time in the upper quartile would relate best to community tran sition in the north and that minimum and % of time in the lower quartile would relate be st to community transition in the south. Discussion These analy ses can be seen as the comparison of an intact wetland to a degraded and impounded wetland. The preliminary results presen ted here are the first step to further untangling the vegetative community dynamics of WCA3A. Similar analysis will now be run on a larger data set to determine if the findings are correct and statistica lly significant. With annual transition probabilities and the hydrologic drivers of transitions in hand, planning appropriate hydrologic regimes to maintain de sired vegetative communities in WCA3A, and predicting vegetative community sh ifts under certain management schedules becomes possible. In an era of restoration, particul arly Everglades restoration the goal of management schedules is a return to natural community transition probabilitie s. This analysis will provide a path toward

PAGE 56

56 that goal by identifying the aspects of hydrol ogy that are the main drivers of vegetative community transition.

PAGE 57

57 Table 4-1. Results when the clus ter including both wet and dry season data was used to make the input file for annual transition estimates. Model AICc Delta AICc AICc weights Model likelihood Num. par Deviance {sp=ps} 42.759 0 0.269 1 2 15.367 {sp=ps*NS} 42.866 0.108 0.254 0.948 3 13.334 {sp=ps*WDyr} 44.439 1.681 0.116 0.432 3 14.907 {sp<>ps} 44.456 1.697 0.115 0.428 3 14.923 {sp<>ps*NS} 45.544 2.785 0.067 0.248 5 11.576 {sp=ps*NS*WDyr} 46.336 3.577 0.045 0.167 5 12.368 {sp=ps*t} 46.631 3.872 0.039 0.144 4 14.907 {sp=ps*WDyr3states} 46.631 3.872 0.039 0.144 4 14.907 {sp<>ps*WDyr} 47.268 4.509 0.028 0.105 5 13.300 {sp=ps*t*NS} 49.190 6.431 0.011 0.040 7 10.571 {sp=ps*NS*WDyr3states} 49.190 6.431 0.011 0.040 7 10.571 {sp<>ps*t} 51.914 9.156 0.003 0.010 7 13.295 {sp<>ps*WDyr3states} 51.914 9.156 0.003 0.010 7 13.295 {sp<>ps*NS*WDyr} 52.591 9.832 0.002 0.007 9 9.088 {sp<>ps*t*NS} 60.358 17.599 0.000 0.000 13 6.315 The estimate of the probability of wet prairie transitioning to slough was equal to the probability of slough transitioning to wet prairie in the most parsimonious model and received a combined AIC (w) of 0.783.

PAGE 58

58 Table 4-2. When the communities were reclustered using only wet season sampling data, the results, as seen here, were quite different and more ecologically reasonable. Model AICc Delta AICc AICc weights Model likelihood Num. par Deviance {PS<>SP*NS} 72.029 0 0.434 1 4 28.118 {PS<>SP} 72.784 0.755 0.297 0.686 2 33.194 {PS=SP*NS} 75.82 3.791 0.065 0.150 2 36.230 {PS<>SP*wetdry223} 76.095 4.066 0.057 0.131 4 32.184 {PS=SP} 77.295 5.266 0.031 0.072 1 39.794 {PS<>SP*NS*%25} 78.233 6.205 0.019 0.045 8 25.063 {PS<>SP*NS*min} 78.538 6.509 0.017 0.039 8 25.367 {PS<>SP*NS*dist} 78.786 6.757 0.015 0.034 8 25.615 {PS<>SP*NS*avg} 79.111 7.083 0.013 0.029 8 25.941 {PS<>SP*NS*wetdry223} 79.291 7.262 0.011 0.027 8 26.120 {PS<>SP*%25} 79.407 7.378 0.011 0.025 4 35.496 {PS<>SP*NS*%75} 79.411 7.382 0.011 0.025 8 26.240 {PS<>SP*t} 80.302 8.273 0.007 0.016 6 31.869 {PS=SP*t} 80.765 8.737 0.005 0.013 3 39.039 {PS<>SP*NS*max} 80.945 8.916 0.005 0.012 8 27.774 {PS=SP*t*NS} 83.288 11.259 0.002 0.004 6 34.855 {PS<>SP*t*NS} 86.412 14.383 0.000 0.001 12 23.055 {PS<>SP*NS*min*dist} 86.868 14.839 0.000 0.001 12 23.512 The most parsimonious model in this model set s howed that there were more transitions in the north than in the south but did not include the effect of time or hydrology (ps<>sp *NS) (AIC(w)=0.43350). In the north th ere were significantly more transitions from slough to wet prairie than from wet prairie to slough (E S= 0.434 (95%CI=0.084 to 0.785)). There was also significantly more movement in the north than in the south from slough to wet prairie (ES=0.402 (95%CI=0.015 to 0.789)).

PAGE 59

59 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 Wet Prairie to Slough Slough to Wet PrairieTransition Probability South North Figure 4-1. Most parsimonious model in model set that attempted to get annual transition probability estimates. It showed that there we re more transitions in the north than in the south but did not include the effect of time or hydrology (ps<>sp *NS) (AIC( w )=0.43350). In the north there were sign ificantly more transitions from slough to wet prairie than from wet prairie to slough (ES= 0.434 (95%CI=0.084 to 0.785)). There was also significantly more movement in the north than in the south from slough to wet prairie (ES=0.402 (95%CI=0.015 to 0.789)). 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 200320042005200320042005 Wet Prairie to Slough Slough to Wet PrairieTransition Probabilities South North Figure 4-2. Model averaged results from the mode l set used to get annual transition probability estimates. Confidence interv als on the estimates are very large and there were no statistically significant effect sizes. The results show that the transition probabilities were higher from sloughs to wet prairies th an from wet prairies to sloughs at all times. Transition probabilities were also higher in the north than in the south although the effect sizes for these co mparisons were not significant.

PAGE 60

60 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 J-03N-03J-04N-04J-05N-05J-06Transition Probabilities P to S South S to P South P to S NORTH S to P NORTH Figure 4-3. Results from 2 model sets, one for the north and one for the south, with 2 community states, slough and wet prairie th at were used to determine what the hydrologic drivers are for each area. These results show that in the south wet prairies were transitioning to sloughs with more fr equency than sloughs were transitioning to wet prairies in all time pe riods except the dry seasons of 2004 and 2006. On these occasions sloughs were more likely to transiti on to wet prairies than wet prairies were to transition to sloughs. In the north sloughs were more likely to transition to wet prairies than wet prairies were to transiti on to sloughs on all occasions except the wet season of 2004 and the dry season of 2005. Du ring these relatively wet time periods, wet prairies were more likely to transition to slough than sloughs were to transition to wet prairies so during wet time periods mo st movement in the north was toward a wetter community state. Although all these findings are ecologica lly significant, the confidence intervals of the effect sizes overl apped 0 and are theref ore not statistically significant.

PAGE 61

61 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 Jun03 Nov03 Jun04 Nov04 Jun05 Nov05 Jun06Transition Probabality P to T South T to P South T to S South S to T South P to S South S to P South P to T NORTH T to P NORTH T to S NORTH S to T NORTH P to S NORTH S to P NORTH Figure 4-4. Results from two model sets (north and south) with 3 community states: slough, transition, and wet prairie used to determin e hydrologic drivers for each area. There was no movement from wet prairies directly to sloughs; all movement occurred through the transition state. As in the previous analysis, transition pr obabilities were generally higher in the north than in the s outh, with the exception of th e dry seasons of 2004 and 2006. There was still no clear directionality of movement.

PAGE 62

62 WORK CITED Alexander, T. R. and A. G. Crook. 1975. R ecent and Long Term Vegetation Changes and Patterns in South Florida, Part II, Final Report, South Florida Ecological Study, Appendix G. NTIS PB 264462, University of Miami, Coral Gables, FL. Blums, P., J. D. Nichols, J. E. Hines, M. S. Lindberg, and A. Mednis. 2003. Estimating natal dispersal movement rates of female European ducks with multistate modelling. Journal of Animal Ecology 72:1027-1042. Burnham, K. P. and D. R. Anderson. 2002. Mode l Selection and Multim odel Inference: A Practical Information-Theoretic Approac h. 2nd edition. Springer Science + Business Media, New York. Caswell, H. 2001. Matrix population models: Cons truction, analysis, and in terpretation. Sinauer Associates, Sunderland, Massachusetts. Conrads, P. A., E. Roehls, R. Daamen, and W. M. Kitchens. 2006. Using artificial neural network models to integrate hydrologic and eco logical studies of the snail kite in the Everglades, USA. in The 7th International Conferen ce on Hydroinformatics. Research Publishing Services, Nice, France. Cooch, E. and G. White. 2008. Program MARK "A gentle introducti on". Colorado State University, Fort Collins, Colorado. Dahl, T. E. 2000. Status and Trends of Wetlands in the Conterminous United States 1986 to 1997. U.S. Fish and Wildlife Service. David, P. G. 1996. Changes in plant communities re lative to hydrologic cond itions in the Florida Everglades. Wetlands 16 :15-23. Davis, S. M., L. H. Gunderson, W. A. Pa rk, J. R. Richardson, and J. E. Mattson. 1994. Landscape dimension, composition, and function in a changing Everglades ecosystem. Pages 419-444 in S. M. Davis and J. C. Ogden, edito rs. Everglades: The ecosystem and its restoration. St. Lucie Press, Delray Beach, Florida. Dineen, J. W. 1972. Life in the tenacious Ev erglades. In Depth Report 1(5), Central and Southern Flood Control District. Dineen, J. W. 1974. Examination of water manageme nt alternatives in Water Conservation Area 2A, In Depth Report 2(3), Central and S outhern Florida Floo d Control District. Duever, M. J., J. F. Meeder, L. C. Meeder, and J. M. McCollom. 1994. The climate of South Florida and its role in shaping th e Everglades ecosystem. Pages 225-248 in S. M. Davis and J. C. Ogden, editors. Everglades: The ecosy stem and its restoration. St. Lucie Press, Delray Beach, Florida.

PAGE 63

63 Edwards, A. L., D. W. Lee, and J. H. Richar ds. 2003. Responses to a fluctuating environment: effects of water depth on growth and bioma ss allocation in Eleocharis cellulosa Torr. (Cyperaceae). Canadian Journal of Bo tany-Revue Canadienne De Botanique 81:964-975. Fonnesbeck, C. J. 2005. Solving dynamic wild life resource optimization problems using reinforcement learning. Natural Resource Modeling 18:1-40. Gleason, P. J., A. D. Cohen, W. G. Smith, H. K. Brooks, P. A. St one, R. L. Goodrick, and W. S. Jr. 1984. The environmental significance of Holocene sediments from the Everglades and saline tidal plain. Pages 297-351 in P. J. Gleason, editor. Enviro nments of South Florida: Present and Past. Miami Geol ogical Society, Miami, FL. Gunderson, L. H. and C. S. Holling. 2002. Panarc hy: understanding transformations in human and natural systems. Panarchy: understanding transformations in human and natural systems.:i-xxiv, 1-507. Gunderson, L. H. and L. Pritchard. 2002. Res ilience and the Behavior of Large Scale Ecosystems. Island Press, Washington D.C. IPCC. 2007. The physical science basis contribution of working group I the fourth assessment report of the IPCC., Cambridge University Press, Cambridge. Karunaratne, L. B., P. C. Darby, and R. E. Bennetts. 2006. The effects of wetland habitat structure on Florida apple snail density. Wetlands 26:1143-1150. Kitchens, W. M., R. E. Bennetts, D. L. DeA ngelis, J. W. Porter, and K. G. Porter. 2002. Linkages between the snail kite population and wetland dynamics in a highly fragmented South Florida hydroscape. Pages 183-203 in J. W. Porter and K. G. Porter, editors. The Everglades, Florida Bay, and coral reefs of the Florida Keys: an ecosystem sourcebook. CRC Press, Boca Raton, Florida. Kolipinski, M. C. and A. L. Higer. 1969. Some Aspects of the Effects of Quantity and Quality of Water on Biological Communities in Everglades National Park. Open File Report, U.S. Geological Survey, Tallahassee, FL. Larsen, L. G., J. W. Harvey, and J. P. Cr imaldi. 2007. A delicate balance: Ecohydrological feedbacks governing landscape morphology in a lotic peatland. Ecological Monographs 77:591-614. Leach, S. D., H. Klein, and E. R. Hampton. 1972. Hydrologic effects of water control and management of southeastern Florida. Florid a Bureau of Geology Report of Investigations. Light, S. S. and J. W. Dineen. 1994. Water control in the Everglades: A historical perspective. Pages 47-84 in S. M. Davis and J. C. Ogden, editor s. Everglades: The ecosystem and its restoration. St. Lucie Press, Delray Beach, Florida. Loveless, C. M. 1959. A Study of the Vegeta tion in the Florida Everglades. Ecology 40 :1-9.

PAGE 64

64 MacKenzie, D. I., J. D. Nichols, J. A. Royle, K. H. Pollock, L. L. Bailey, and J. E. Hines. 2006. Occupancy estimation and modeling: inferr ing patterns and dynamics of species occurrence. Elsevier, San Diego, CA. Martin, J., W. M. Kitchens, and J. E. Hine s. 2007a. Importance of well-designed monitoring programs for the conservation of endangere d species: Case study of the snail kite. Conservation Biology 21 :472-481. Martin, J., W. M. Kitchens, and J. E. Hine s. 2007b. Natal location influences movement and survival of a spatially structured population of snail kites. Oecologia 153 :291-301. Martin, J., W. M. Kitchens, M. Oli, and C. E. Cattau. 2008a. Relative importance of natural disturbances and habitat degradation on snail kite population dynamics. Endangered Species Research in press Martin, J., J. D. Nichols, C. L. McIntyre, G. Ferraz, and J. E. Hines. 2008b. Perturbation analysis for patch occupancy models. Ecology in press Martin, J., M. C. Runge, J. D. Nichols, B. C. Lubow, and W. L. Kendall. 2008c. Structured decision making in a conceptual framework to identify thresholds for conservation and management. Ecological Applications in press McCune, B. and J. B. Grace. 2002. Analysis of ecological communities. MjM Software Design, Gleneden Beach, Oregon. McPherson, B. F. 1973. Vegetation in Relation to Water Depth in Conservation Area 3, Florida. Open File Report, U.S. Geological Survey, Tallahassee, FL. Mitsch, W. J. and J. G. Go sselink. 1993. Wetland s. 2nd edition. Mitsch, W. J. and J. G. Gosselink. 2007. Wetla nds. 4th edition. John Wiley & Sons, Inc., Hoboken, New Jersey. NOAA. 1985. Climatography of the United States N o. 20, Climate Summaries for Selected Sites, 1951-1980, Florida., National Climatic Data Center, Asheville, NC. Obeysekera, J., J. Browder, L. Hornung, and M. A. Harwell. 1999. The natural South Florida system I: Climate, geology, and hydrology. Urban Ecosystems 3:223-244. Powers, E. 2005. Meta-stable States of Vegetativ e Habitats in Water Conservation Area 3A, Everglades. University of Florida, Gainesville, FL. Puckridge, J. T., K. F. Walker, and J. F. Costelloe. 2000. Hydrologica l persistence and the ecology of dryland rivers. Regulated Rivers-Research & Management 16:385-402. RECOVER. 2005. The RECOVER t eam's recommendations for interim targets for the Comprehensive Everglades Restoration Project. South Florida Water Management District and U.S. Army Corps, West Palm Beach, FL.

PAGE 65

65 Salzman, J. and J. B. Ruhl. 2006. 'No Net Loss' Instrument Choice in Wetlands Protection. in J. Freeman and C. D. Kolstad, editors. Moving to Markets in Environmental Regulation: Twenty Years of Experience. Oxford University Press, United Kingdom. ScienceCoordinationTeam. 2003. The role of flow in the Everglades ridge and slough landscape. South Florida Ecosystem Restoration Working Group. U.S. Geological Survey, Washington, D.C., USA. Sklar, F., C. McVoy, R. VanZee, D. E. Gawlik, K. Tarboton, D. Rudnick, S. Miao, and T. Armentano. 2002. The effects of altered hydrolog y on the ecology of the Everglades. Pages 39-82 in J. W. Porter and K. G. Porter, editor s. The Everglades, Florida Bay, and coral reefs of the Florida Keys: an ecosyst em sourcebook. CRC Press, Boca Raton, FL. Sykes, P. W., Jr., J. A. Rodgers, Jr., and R. E. Bennetts. 1995. Snail kite. Birds of North America 171:1-32. White, G. C. and K. P. Burnham. 1999. Program MARK: survival estimation from populations of marked animals. Bird Study 46:120-139. White, P. S. 1994. Synthesis: Vegetation pattern and process with the Everglades ecosystem. Pages 445-458 in S. M. Davis and J. D. Ogden, edito rs. Everglades: The ecosystem and its restoration. St. Lucie Press, Delray Beach, Florida. Williams, B. K., J. D. Nichols, and M. J. Conroy. 2002. Analysis and management of animal populations: modeling, estimation, and decisi on making. Academic Press edition, San Diego, California, USA. Wood, J. M. and G. W. Tanner. 1990. Graminoid community composition and Structure within 4 Everglades Management Areas. Wetlands 10:127-149. Zaffke, M. 1983. Plant communities of Water Conservation Area 3A: base-line documentation prior to the operation of S-339 and S-340. T echnical Memorandum, South Florida Water Management District, West Palm Beach, FL. Zweig, C. and W. Kitchens. 2008a. Multi-state su ccession in wetlands: a novel use of state and transition models. Ecology in press Zweig, C. and W. M. Kitchens. 2008b. Effect s of landscape gradients on wetland vegetation communities: Information for la rge-scale restoration. Wetlands in press

PAGE 66

66 BIOGRAPHICAL SKETCH Althea Hotaling was born in Athens, Georgia in 1982. She grew up in the countryside around Athens until h er family moved to Central Flor ida. Her love for nature and being outside developed early in these woods and has not waned. She obtained a BS in marine biology from the University of West Florida in 2003. As an undergraduate she worked on a variety of research projects and decided she would lik e to focus on applied research th at really made a difference. Immediately after graduation she married Donald Hagan and joined the Peace Corps, where she served for over 2 years in Bahia de Caraquez, Ecuador, as a natural reso urces volunteer. During her Peace Corps service, she grew to understand that conservation and re storation of natural ecosystems must be something everyone in the comm unity is invested in or it will surely fail and that it is a delicate balance of n eeds. She entered the graduate program at the School for Natural Resources and the Environment in 2006.