Modeling Species Distributions of Three Endemic Florida Panhandle

Home , Anthophora abrupta, Conradina, Macbridea, Macbridea alba, Megachile astragali, Stachydeoma

Florida State University Libraries

Electronic Theses, Treatises and Dissertations The Graduate School

2009 Modeling Species Distributions of Three Endemic Florida Panhandle Mints under Climate Change: Comparing Plant and Pollinator Distribution Shifts under Future Conditions Amanda J. Kubes

Follow this and additional works at the FSU Digital Library. For more information, please contact [email protected] FLORIDA STATE UNIVERSITY

COLLEGE OF ARTS AND SCIENCES

MODELING SPECIES DISTRIBUTIONS OF THREE ENDEMIC FLORIDA PANHANDLE

MINTS UNDER CLIMATE CHANGE: COMPARING PLANT AND POLLINATOR

DISTRIBUTION SHIFTS UNDER FUTURE CONDITIONS

AMANDA J. KUBES

A Thesis submitted to the Department of Biological Science in partial fulfillment of the requirements for the degree of Master of Science

Degree Awarded: Summer Semester, 2009

The members of the committee approve the thesis of Amanda J. Kubes defended on June 23, 2009.

Austin Mast Professor Directing Thesis

Alice A. Winn Committee Member

Brian Inouye Committee Member

Approved:

P. Bryant Chase, Chair, Department of Biological Science

The Graduate School has verified and approved the above-named committee members. ii

ACKNOWLEDGEMENTS ―I have no special talents. I am only passionately curious.‖

-Albert Einstein

This has been one of the most rewarding (and challenging) decisions in my life, and I made it through this everything in no small part thanks to the support of everyone who has taken steps with me along the way. Thank you so much-first, to my family: my loving husband, who doesn‘t know exactly what I do but is behind me every step of the way, my parents who are proud of me no matter what I do, and all of my aunts, uncles, cousins, and grandparents who have given me their encouragement and support. Thanks to all of my closest friends, both old and those I have met here along the way for being there and offering support and suggestions. Thank you to my advisor, Dr. Austin Mast, who helped me along with never-ending patience, guidance, and strong support even when I thought I might not be able to push myself. I want to thank the members of my graduate committee, Drs. Alice Winn and Brian Inouye, who were always ready and willing to give me their time, help, encouragement, and suggestions whenever I needed it. Thank you to Gil Nelson and Loran Anderson for all of the help with plant identification, localities, and ready willingness to help me at all times. Thank you to everyone in the herbarium (past and present), for all of your technical and moral support; especially to Sarah Braun, who gave me ideas and wonderful support to get me through my first year. Funding for my fieldwork came from the Robert K. Godfrey Endowment Award from the Florida State University, for which I am very grateful. Last but not least, I want to thank the following people and organizations for their moral and technical support: Amy Jenkins, Lindsay Horton, and the entire staff at Florida Natural Areas Inventory, Dean Jue, Ken Womble, Louise Kirn, Theresa Pitts-Singer, and Vivian Ortez-Negron.

iii

TABLE OF CONTENTS

List of Tables ...... v List of Figures ...... vi Abstract ...... x INTRODUCTION ...... 1 1. POLLINATOR OBSERVATIONS, EXCLUSION MANIPULATIONS, AND SEED SET OF THREE ENDEMIC FLORIDA PANHANDLE MINTS ...... 3 INTRODUCTION ...... 3 METHODS ...... 14 RESULTS ...... 19 DISCUSSION ...... 31

2. MODELING THE DISTRIBUTIONS OF THREE ENDEMIC FLORIDA PANHANDLE MINTS AND THEIR POLLINATORS UNDER CURRENT AND FUTURE CLIMATE CONDITIONS ...... 36 INTRODUCTION ...... 36 METHODS ...... 43 RESULTS ...... 49 DISCUSSION ...... 77

CONCLUSION ...... 82 REFERENCES ...... 83 BIOGRAPHICAL SKETCH ...... 94

LIST OF TABLES

Table 1.1 Flowering times for three mint species as reported by FNAI and PanFlora...... 14

Table 1.2 Number of potential pollinators observed over 13 hours total for populations of Conradina glabra (2008 and 2009), Physostegia godfreyi, and Stachydeoma graveolens (2008 only for these species) ...... 19

Table 1.3 Composition of the three most observed potential pollinators for Conradina glabra. Shows the total number of each insect type observed, the percent of the total insect observations for Conradina glabra, and the rate of visitation for each insect type shown ...... 23

Table 1.4 Composition of the three most observed potential pollinators for Stachydeoma graveolens. Shows the total number of each insect type observed, the percent of the total insect observations for Stachydeoma graveolens, and the rate of visitation for each insect type shown ...... 27

Table 1.5 Composition of the three most observed potential pollinators for Physostegia godfreyi. Shows the total number of each insect type observed, the percent of the total insect observations for Physostegia godfreyi, and the rate of visitation for each insect type shown ...... 27

Table 1.6 Total number of seeds produced in both open-pollinated and pollinator-excluded flowers for plants in both populations of Conradina glabra (2008 and 2009) and Physostegia godfreyi (2009) ...... 29

Table 1.7 Final generalized linear model (GLM) for Conradina glabra. The variable treatment is responsible for the majority of the deviance in seed set ...... 30

Table 1.8 Final generalized linear model (GLM) for Physostegia godfreyi. The variable treatment is responsible for the majority of the deviance in seed set, although population appears to be slightly significant as well ...... 30

Table 2.1 Number of occurrence records used for each taxon in modeling ...... 44

Table 2.2 Environmental layers used in Maxent modeling ...... 49

LIST OF FIGURES

Figure 1.1 Known distribution and occurrences of Conradina glabra, Physostegia godfreyi, and Stachydeoma graveolens ...... 6

Figure 1.2 Images of Conradina glabra habitat and morphology ...... 9

Figure 1.3 Images of Stachydeoma graveolens habitat and morphology ...... 11

Figure 1.4 Images of Physostegia godfreyi habitat and morphology ...... 13

Figure 1.5 Time of day observations of potential pollinators for Conradina glabra for all four days over two populations in 2008 and 2009 ...... 21

Figure 1.6 Time of day observations of potential pollinators for Stachydeoma graveolens for all four days over two populations in 2008 ...... 22

Figure 1.7 Time of day observations of potential pollinators for Physostegia godfreyi for all four days over two populations in 2008 ...... 22

Figure 1.8 Images of carpenter bees (Xylocopa micans) observed in Conradina glabra population in 2009, robbing nectar from flowers ...... 24

Figure 1.9 Images of Bombylius major (beefly) observed in Conradina glabra populations in 2009 ...... 25

Figure 1.10 Digger bee (Anthophora spp.) observed and captured in Conradina glabra population in 2009 ...... 25

Figure 1.11 Images of digger wasps (Scoliidae), brown skipper moth (Hesperidae), and honey bee (Apis mellifera) observed in 2009 ...... 26

Figure 2.1 Known distributions and occurrences of Conradina glabra, Physostegia godfreyi, and Stachydeoma graveolens ...... 45

Figure 2.2 Occurrence records for the three potential pollinator taxa selected. Megachile spp. (red) are well-represented in the extent of interest. All three taxa have widespread distributions compared to those of the mint species ...... 46

Figure 2.3 Receiver operating curve (ROC) for Conradina glabra. Sensitivity is the fraction of all presences that are correctly identified as such, while the specificity is the fraction of all absences that are correctly identified as such ...... 50

Figure 2.4 Predicted suitability for Conradina glabra modeled with Maxent under current environmental conditions at the default regularization setting of 1.0. The map indicates areas suitable for the species to be present, showing a probability of occurrence from 0-1. Brighter colors such as red and orange indicate the most suitable conditions. Violet dots indicate locations used for testing the model, and white points are the locations used for training the model. All following figures are described by the same coloration ...... 52

Figure 2.5 Predicted suitability for Conradina glabra modeled under future environmental conditions at the default regularization setting of 1.0 ...... 52

Figure 2.6 Predicted suitability for Conradina glabra modeled under current environmental conditions at a regularization setting of 3.0 ...... 53

Figure 2.7 Predicted suitability for Conradina glabra modeled under future environmental conditions at a regularization setting of 3.0 ...... 53

Figure 2.8 Predicted suitability for Conradina glabra modeled under current environmental conditions at a regularization setting of 5.0 ...... 54

Figure 2.9 Predicted suitability for Conradina glabra modeled under future environmental conditions at a regularization setting of 5.0 ...... 54

Figure 2.10 Results of the jackknife performed with training data for Conradina glabra modeled with the default regularization parameter of 1.0, showing the individual contributions of environmental variables to the model when included in isolation and excluded ...... 56

Figure 2.11 Receiver operating curve (ROC) for Stachydeoma graveolens modeled with the default regularization setting of 1.0 ...... 57

Figure 2.12 Predicted suitability for Stachydeoma graveolens modeled under current environmental conditions at the default regularization setting of 1.0 ...... 58

Figure 2.13 Predicted suitability for Stachydeoma graveolens modeled under future environmental conditions at the default regularization setting of 1.0 ...... 58

Figure 2.14 Predicted suitability for Stachydeoma graveolens modeled under current environmental conditions at a regularization setting of 3.0 ...... 59

Figure 2.15 Predicted suitability for Stachydeoma graveolens modeled under future environmental conditions at a regularization setting of 3.0 ...... 59

Figure 2.16 Predicted suitability for Stachydeoma graveolens modeled under current environmental conditions at a regularization setting of 5.0 ...... 60

vii

Figure 2.17 Predicted suitability for Stachydeoma graveolens modeled under future environmental conditions at a regularization setting of 5.0 ...... 60

Figure 2.18 Results of the jackknife performed with training data for Stachydeoma graveolens modeled with the default regularization setting of 1.0 ...... 62

Figure 2.19 Receiver operating curve (ROC) for Physostegia godfreyi modeled with the default regularization setting of 1.0 ...... 63

Figure 2.20 Predicted suitability for Physostegia godfreyi modeled under current environmental conditions at the default regularization setting of 1.0 ...... 64

Figure 2.21 Predicted suitability for Physostegia godfreyi modeled under future environmental conditions at the default regularization setting of 1.0 ...... 64

Figure 2.22 Predicted suitability for Physostegia godfreyi modeled under current environmental conditions at a regularization setting of 3.0 ...... 65

Figure 2.23 Predicted suitability for Physostegia godfreyi modeled under future environmental conditions at a regularization setting of 3.0 ...... 65

Figure 2.24 Predicted suitability for Physostegia godfreyi modeled under current environmental conditions at a regularization setting of 5.0 ...... 66

Figure 2.25 Predicted suitability for Physostegia godfreyi modeled under future environmental conditions at a regularization setting of 5.0 ...... 66

Figure 2.26 Results of the jackknife performed with training data for Physostegia godfreyi modeled with the default regularization parameter of 1.0 ...... 68

Figure 2.27 Receiver operating curve (ROC) for Xylocopa virginica at the default regularization setting of 1.0...... 69

Figure 2.28 Predicted suitability for Xylocopa virginica modeled under current environmental conditions at the default regularization setting of 1.0 ...... 70

Figure 2.29 Predicted suitability for Xylocopa virginica modeled under future environmental conditions at the default regularization setting of 1.0 ...... 70

Figure 2.30 Results of the jackknife performed with training data for Xylocopa virginica modeled with the default regularization parameter of 1.0 ...... 71

Figure 2.31 Receiver operating curve (ROC) for Anthophora abrupta modeled with the default regularization setting of 1.0 ...... 72

viii

Figure 2.32 Predicted suitability for Anthophora abrupta modeled under current environmental conditions at the default regularization setting of 1.0 ...... 73

Figure 2.33 Predicted suitability for Anthophora abrupta modeled under future environmental conditions at the default regularization setting of 1.0 ...... 73

Figure 2.34 Results of the jackknife performed with training data for Anthophora abrupta modeled with the default regularization setting of 1.0 ...... 74

Figure 2.35 Receiver operating curve (ROC) for Megachile spp. modeled with the default regularization setting of 1.0 ...... 75

Figure 2.36 Predicted suitability for Megachile species modeled under current environmental conditions at the default regularization setting of 1.0 ...... 76

Figure 2.37 Predicted suitability for Megachile species modeled under future environmental conditions at the default regularization setting of 1.0 ...... 76

Figure 2.38 Results of the jackknife performed with training data for Megachile modeled with the default regularization setting of 1.0 ...... 77

ABSTRACT

Climate change is expected to drive some species to extinction by restricting the range of their suitable habitat or by causing local extinctions of associated organisms necessary to the species‘ survival and reproduction. However, it is not yet widely appreciated that climate change might directly disrupt or eliminate beneficial ecological interactions between species even before extinctions occur. There is growing interest in utilizing species distribution models as management tools and for predicting changes under future conditions such as climate change (Peterson et al. 2000; Guisan and Thuiller 2005; Whittaker et al. 2005). It is important to develop knowledge about the ecological interactions and reproductive biology of species of conservation concern to inform long-term strategies for conservation under changing climatic conditions, particularly for rare and narrowly distributed species. The Florida Panhandle is home to many endemic plants, and many of these are rare and threatened within their range. Unfortunately, for many of these Panhandle endemics, little is known about their ecological interactions and reproductive biology. Here, I address this need with a two-part research project. The objectives of my research were to (1) observe and identify the potential pollinators of three mint species (Lamiaceae) endemic to the Florida Panhandle, (2) determine if these plants are self-compatible, and (3) predict the current and future areas of suitability for these mint species and their pollinators under a climate change scenario. I address three questions related to the last objective: (1) where are the plants and their pollinators expected to occur given current conditions? (2) where are the plants and their pollinators predicted to occur under future climate change conditions? and (3) do these future predicted suitable areas overlap? The three mint species I examined were Apalachicola false rosemary (Conradina glabra), Godfrey‘s false dragonhead (Physostegia godfreyi), and mock pennyroyal (Stachydeoma graveolens). I performed field observations of these endemic mints to identify the potential pollinators of the species. I also performed pollinator exclusion manipulations in two populations of two of the species to determine if they were self-compatible in the absence of pollinators. I found that all three species are visited by multiple potential pollinators, although the numbers and identities of potential pollinators differed between species (and to a smaller extent between populations). Most of the potential pollinators observed were native bees, but all were from genera

that have fairly widespread distributions compared to those of the plants in this study. I determined that Conradina glabra and Physostegia godfreyi are self-compatible. A smaller percentage of flowers set seeds when pollinators were excluded from these species, but, when I hand-pollinated Conradina glabra, a percentage of flowers similar to the open pollinated treatment set seeds. I did not test the self-compatibility of Stachydeoma graveolens. I conclude from these results that the two examined species are self-compatible but likely rely on pollinators for maximum seed set. I selected three of the most frequent visitors to the three plant species, Xylocopa virginica, Anthophora abrupta, and Megachile species, to create species distribution models. I modeled the areas of suitability for these three mint species and the selected potential pollinators with current conditions and with future conditions in a doubled CO2 climate scenario. I used the Maxent modeling method (maximum entropy modeling) to create the models. The suitability maps were examined to see if separations in predicted ranges between plants and their potential pollinators might occur. Two of the three plant species, Stachydeoma graveolens and Physostegia godfreyi, had expanded future areas of suitability, whereas the area of suitability for Conradina glabra shrank and shifted more than 200 km. The models generated for all three mint species were accurate as determined using test data sampled from the presence records. The potential pollinators‘ areas of suitability were not modeled as well as those of the plants. However, the results suggest that these plant species will not be separated from their pollinators under this climate change scenario. This research represents a novel approach to conservation planning that considers both plants and their pollinators.

INTRODUCTION

The Florida Panhandle is considered to be one of the hotspots of biodiversity and endemism within the United States. It contains a large amount of plant biodiversity, with more than 125 plant taxa that are endemic only to this area (Blaustein 2008). Many of these species are found within the Apalachicola National Forest and in the Apalachicola River region. The reasons for these high levels of endemism and diversity can be attributed to a combination of factors that include Florida‘s geological history, peninsular effects, and humid climate (Myers and Ewell 1990; Estill and Cruzan 2001; Blaustein 2008). Many of these Panhandle endemic species are of conservation concern yet still have little information available on their life histories, distributional patterns, and ecological interactions. Increasing threats from human activities such as climate change make it more urgent to create long-term management and conservation strategies that will enable these species to survive under changing conditions (Hernandez et al. 2006). Plants in particular may be at risk due to their immobility. Studies of rare plants in some habitats have studied the responses of plants to management practices (Hessl and Spackman 1995; Brewer 1999; Lesica 1999), but few studies have examined the effects of these practices on their pollinators or interactions with them. One major reason is that the effective pollinators for many species are unknown. Understanding the reproductive biology of rare and endemic plants is often a major concern for their conservation (Kruckeberg and Rabinowitz 1985; Kaye 1999; Spira 2001). This is of even more concern in plants that have short life spans where the next generation is entirely dependent on seed production. In species that are dependent on seeds for regeneration and are short-lived, conservation might depend on studying and understanding factors that might limit seed production, such as pollination biology (Pavlik et al. 1993). Factors that limit seed production may also play some role in mating system regulation, such as self-compatibility. Mating systems and patterns affect how much genetic variation is found within plant populations (Hamrick et al., 1991; Ellstrand and Elam, 1993), and these can vary even between populations of the same species. In rare and geographically limited species, genetic variation can potentially be important for population stability. It is important to research and document the factors that influence and shape reproductive biology in these rare plant species, including pollinators and their interactions. 1

Climate change has been predicted to separate plants and pollinators by potentially causing distribution and phenology shifts (Kearns et al. 1998; Peñuelas et al. 2004). These shifts could potentially alter other interactions such as competition for pollinators among plants. There are studies that focus on predicting future effects on these interactions (Memmott et al. 2004; Devoto et al. 2007), but none have used species distribution models to predict future range shifts of both rare plants and their pollinators. Modeling plant species‘ distributions in parallel with their pollinators‘ distributions under predicted climate change can help us envision the consequences range shifts may have on plant-pollinator interactions. Species distribution models are now commonly used to predict the suitable habitats and distribution patterns of endemic species, but few attempts have been made to model other species that interact with the selected species.

CHAPTER ONE

POLLINATOR OBSERVATIONS, EXCLUSION MANIPULATIONS, AND SEED SET OF THREE ENDEMIC FLORIDA PANHANDLE MINTS

INTRODUCTION

Florida is home to more than 3800 plant species and ranks third in seed plant species richness out of all the states in the United States (Wunderlin and Hansen 2000). The high levels of biodiversity present in Florida are a result of a combination of historical, geographical, and climatic factors. The terrestrial Florida that we see today covers a large range of landscape ages, from areas that have likely been occupied by plants and animals for more than 20 million years to regions that have only been above sea level for a few thousand years (Myers and Ewell 1990). These patterns lead to diversity that is composed of relict species and species that have more recent origins as colonizers (or invasives) or products of speciation. Florida is unique in the United States because it encompasses both temperate and subtropical zones. Biodiversity tends to increase from temperate to tropical regions (Stein et al. 2000), so this may also explain to some extent the amount of diversity present within the state. Most of the land that shares Florida‘s latitude is desert, but Florida itself has a humid climate because it is a peninsula and surrounded on three sides by oceans. This humidity is also in part responsible for some of the dense vegetation in parts of the state (Myers and Ewell 1990). In the Florida Panhandle, some of these factors have led to the environmental diversity that allows this region to support a high level of biodiversity (Myers and Ewell 1990; Stein et al. 2000). At least 500 Florida plant species are listed as threatened or endangered at the state level (Stein et al. 2000), and more than 50 of these species are on the Federal Endangered Species List. Significant numbers of these endemic species are geographically limited to relatively small areas within the state, such as the Lake Wales Ridge or the Panhandle and these are potentially more vulnerable to habitat fragmentation and human-mediated climate change than more widespread species. Despite the large number of taxa of potential conservation concern, little is known about the ecological interactions between many of these endemic species and co-occurring species that might play important roles as pollinators or competitors.

Pollinators often have a critical role in flowering plant reproduction. They serve multiple functions such as facilitating the transfer of pollen, regulating mating systems, and maintaining genetic diversity within populations (Ellstrand and Elam 1993; Pitts-Singer et al. 2002). The pollinators themselves benefit from, and may depend on, floral resources such as nectar and pollen for their survival and reproduction (Westerkamp and Claben-Bockhoff 2007). Both plant and pollinator species are at increased risks of extinction (locally and globally) from human activities such as habitat loss, introduction of invasive species, and climate change (Kearns et al. 1998; Biesmeijer et al. 2006). Some of these activities threaten the interactions between plants and pollinators, with climate change rising as one of the biggest threats to these interactions (Memmott et al. 2007). Based on the importance of pollinators to native ecosystems, it is logical to assume that they may play a role in the persistence (ability to occupy an area within a range) and distribution of endemic plant species (Mustajarvi et al. 2001; Sargent and Ackerly 2007) that are dependent on pollinators for seed set. In small and fragmented endemic plant populations, genetic diversity can be very low, making a species more vulnerable to local extirpation. A study done on white-birds-in-a-nest (Macbridea alba), a Panhandle endemic mint, has shown this to be the case in this species (Godt et al. 2004). Plants may self-pollinate more often if flowering occurs when pollinators are scarce or unreliable (Memmott et al. 2007), which can ensure seed set but also reduces genetic diversity in small populations. This can render populations of a rare species vulnerable to changes in environmental conditions or unable to persist in, or shift to, potential suitable habitat in the future (Memmott et al. 2004). Pollinator limitation under changing conditions may make this scenario more likely. Reproduction is an important aspect of biology, and plant mating systems are of particular interest because of the wide variety of strategies that plants employ to promote outcrossing, selfing, or in mixed mating, both of these strategies. The persistence of plants in changing habitats depends in large amount on their reproductive biology. It is logical that determining the pollination biology and mating systems of rare and geographically limited species is important for planning future management and conservation programs (Hamrick et al. 1991; DeMauro 1993; Weller 1994; Spira 2001). This information may also aid in determining how vulnerable a species may be to extinction under changing conditions such as the future climate (Bowlin et al. 1993; Carlsen et al. 2002). In this chapter, I introduce the three Panhandle endemic mint species studied in this thesis and present observations and identifications of their potential pollinators obtained from field 4

observations and manipulations within two populations of each species. I also examine if these mints are autogamously self-compatible or if pollinators are necessary for facilitating seed set in these species.

Natural History of Study Organisms I selected three mint species (Lamiaceae) endemic to the Florida Panhandle for my research. These three species are Apalachicola false rosemary (Conradina glabra), mock pennyroyal (Stachydeoma graveolens), and Godfrey‘s false dragonhead (Physostegia godfreyi). These three species were selected for several reasons, not least of which is that the potential pollinators and mode of reproduction in these species remains relatively unstudied and results of this study may directly influence and supplement conservation plans for these species. Another reason is that all three of these species are of conservation concern on at least the state level, with distributions restricted to a few counties within the Panhandle region. They also have distinctive morphologies which allow for identification of individuals in the field even when flowers are not present. Conradina glabra is listed as endangered at the federal and state levels. It is confined to a single county (Liberty Country) within the Florida Panhandle (Figure 1.1). The remaining two species, Physostegia godfreyi and Stachydeoma graveolens, are both listed at the state level as threatened, but not at the federal level. These species are also restricted to the Florida Panhandle, occurring mainly within the Apalachicola National Forest across seven counties each, respectively (Figure 1.1).

Figure 1.1. Known distribution and occurrences of Conradina glabra, Physostegia godfreyi, and Stachydeoma graveolens.

The genus Conradina (Lamiaceae) consists of six species, all of which are endemic to the southeastern United States. All of the species within this genus are aromatic shrubs that have needle-like leaves, and they share unique floral characteristics such as a sharply bent flower tube that set them apart from other closely related mints, such as Calamintha (Gray 1965). Most of the species in Conradina, including C. glabra, are only present in the endangered sand pine scrub or in sandhill habitats of Florida (Edwards et al. 2006). These habitats often contain large numbers of endemic species because they persisted as Pleistocene refugia (Christman and Judd 1990; Myers and Ewell 1990); they are also among the oldest habitats in the state. These habitats are in decline as a result of increasing human impacts such as development and degradation. As a result of habitat loss or destruction, four of the six species in Conradina are federally listed as endangered or threatened. Phylogenetic studies show that Conradina is most closely related to other mint genera endemic to the southeastern United States that have similar morphologies and habitat types (Edwards et al. 2003; Edwards et al. 2006), including another Panhandle endemic, Stachydeoma graveolens. The species of Conradina fall into two clades, a Northern Panhandle group and a Southern Peninsular group (Edwards et al. 2008). It should be noted that although six species of Conradina are recognized, species do cross with each other easily in experimentation (Gray 1965). It seems likely, and has been hypothesized, that Conradina species are reproductively isolated only because they are geographically isolated (Gray 1965). It is possible that when ranges are close to each other hybridization between species occurs. Possible hybrid populations occur in Santa Rosa County in the Panhandle that have intermediate characteristics between C. glabra and C. canescens, another Panhandle endemic (Edwards et al. 2006; Edwards et al. 2008). Conradina glabra was not determined to be taxonomically distinct until 1962 (Shinners 1962). Within Liberty County it occurs in flat sandy areas that have an overstory composed of longleaf pine and turkey oak, and that often are adjacent to steephead ravines (Gray 1965). There is no available information on its natural habitat prior to the 1950s when conversion of pine plantations occurred. It has also been observed to occur along the edges of pine plantations, highways, and utility right-of-ways (Clewell 1985; Wunderlin 1998; FNAI 2000). Its current known distribution is composed of eight populations on Nature Conservancy and Torreya State Park lands adjacent to the bluffs and ravines of the Apalachicola River basin (Figure 1.2). For the purpose of this study I did not consider the putatively hybrid Santa Rosa County populations 7

in this distribution. Also, the populations located within the Nature Conservancy‘s Bluffs and Ravines Preserve are made up of transplanted individuals for research and conservation purposes, and these populations were not examined either. For the purpose of this study, I considered populations to be separate if they are at least 1 km apart (from FNAI classification of a population; FNAI 2000), and the eight populations fit this criterion. Although some pollen transfer research has been conducted on this species (Isom 2000), its ecological relationships remain largely unknown. The mating system in this species is still poorly understood. Conradina glabra is a perennial shrub that grows up to 0.8 m tall and is densely branched. It has evergreen, strongly aromatic green leaves that are arranged opposite to each other and are carried in clusters. The leaves are smooth and hairless on their upper surfaces, and are densely hairy on their lower surfaces, although these hairs can be nearly invisible to the naked eye. This species is known to flower from March through May or June, and very occasionally until December (FNAI 2000; Clewell 1985) (Table 1.1). The flowers are borne in the leaf axils in groups of two to three, and are from 1.3‒1.9 cm in length. It is distinguished from other species in Conradina by its glabrous calyx. The flowers have a lower lip that has three lobes, and they vary in color from white to pale purple or pink, with a band of darker purple spots on the throat (Figure 1.2). Each flower produces a maximum of four nutlets, which is characteristic of members of the Lamiaceae. The exact mode of seed dispersal is not known, although related mint taxa (Dicerandra) in the southeastern scrub mint clade disperse seeds locally, simply with gravity. The seeds do not fall far from the parent plant, with offspring found at a maximum of two meters away (Evans et al. 2007). It seems likely that the mode of dispersal of seeds is very similar in the mints studied here because they are morphologically similar and produce seeds of comparable size. It is not known if the species relies on a seed bank.

Figure 1.2. Images of Conradina glabra habitat (top center) and morphology. Photo credit: Amanda Kubes; Illustration courtesy of FNAI.

The life history of Stachydeoma graveolens, as with Conradina glabra, is also not well documented. The genus Stachydeoma is considered to be monotypic based on genetic analyses (Edwards et al. 2006), so S. graveolens is the sole species within the genus. This species is also

referred to as Hedeoma graveolens and included within the genus Hedeoma by some authors (e.g., Wunderlin 1998), but more recent phylogenetic analyses (Prather et al. 2002; Edwards et al. 2006) indicate that this species is more closely related to the clade of southeastern scrub mints that includes Conradina species but does not include Hedeoma species (Edwards et al. 2006). Based on the recent phylogenetic results of this species, I use the name Stachydeoma graveolens throughout. All members in this Southeastern scrub mint clade share a very similar morphology, including Conradina glabra and Stachydeoma graveolens. Stachydeoma graveolens plants have short woody stems and are densely branched. The branches are upright and can be densely hairy and are small in stature, growing up to 50 cm tall. It has small, opposite oval-shaped leaves that are aromatic and also hairy, sometimes with scattered glandular hairs as well (FNAI 2000). Its flowers are also small (1.3 cm long), and are produced in leaf axils near the top of the stem. In coloration they are bright pink or purple with darker spots lining the throat (Clewell 1985; Wunderlin 1998), and they have three lobes on the lower lip (Figure 1.3). Flowering occurs from May to July (Table 1.1) with observations extending to September (FNAI 2000). This species also produces four nutlets per flower. As with Conradina glabra, the exact mode of seed dispersal in this species is not known, although as mentioned, seem likely to be locally dispersed by gravity. It is not known if this species depends on a seed bank either. Stachydeoma graveolens occurs within sandhills or dry areas in pine-palmetto-wiregrass flatwoods (FNAI 2000). Within suitable sites, plants may be locally abundant in groups that become more scattered at edges of populations.

Figure 1.3: Images of Stachydeoma graveolens habitat (top center) and morphology. Photo credits: Amanda Kubes. Illustration courtesy of FNAI.

Physostegia godfreyi also has little information available on its ecological relationships and reproductive biology. The genus Physostegia consists of 12 species (Cantino 1982). It is restricted to North America, and species in this genus range from being common and widespread across the country (P. virginiana), to rare and of conservation concern, such as P. godfreyi. Physostegia is not related to members of the Southeastern scrub mint clade. Members of this genus are herbaceous, and phylogenetic studies place it as closely related to Macbridea, another genus with a rare species restricted to the Florida Panhandle (Scheen et al. 2004; Scheen et al. 2008). Physostegia godfreyi plants stand fairly tall, growing up to 0.6‒0.9 m. Plants have upright, slender, stems that have two or three branches. It can be confused with other Physostegia species that overlap the range of P. godfreyi. Often the leaves are used to distinguish the species. P. godfryei has narrow leaves that are opposite and can be up to 6.4 cm long. The leaves are shorter at the top of the stems than at the bottom, and are narrow throughout their length. The flowers develop acropetally, and grow along a tall spike at the top of the stem and face in different directions. The flowers are small (1‒1.3 cm long) and tubular with three lobes on the lower lip (Clewell 1985; Wunderlin 1998; FNAI 2000); they are normally pale pink to purple with purple spots or streaks within the throats (Figure 1.3). Flowering occurs from May through July (FNAI 2000) (Table 1.1). This species also produces four nutlets per flower. As with the previous species, the exact mode of seed dispersal is not known. Based on floral morphology and seed shape, gravity dispersal seems likely for this species as well. It is not know if this species maintains a seed bank. This species occupies moister habitats than the previous two species, such as wet flatwoods, prairies, and pitcher plant bogs (FNAI 2000). Within populations it tends to be present in scattered clumps even when abundant, with fewer plants observed at population fringes.

Figure 1.4. Images of Physostegia godfreyi habitat (top center) and morphology. Photo credits: Amanda Kubes.

Table 1.1. Flowering times for three mint species. Highlighted months are those reported as flowering times from FNAI (2000). Months without highlighting are those reported by Gil Nelson‘s PanFlora database alone (http://www.gilnelson.com/PanFlora/). Shows overlap and differences in observed phenologies for the three species in different sources. Physostegia godfreyi flowering information was not available from FNAI, so there are no highlighted entries for this species. Species Name Flowering Phenology Conradina glabra Feb March April May June July Physostegia godfreyi May June July August Sept. Oct. Nov. Stachydeoma graveolens April May June July August Sept.

All three of these species are fire-responsive, and can be induced to flower again shortly following fire (FNAI 2000; personal correspondence with Louise Kirn). Flowering has been observed as late as October and November following late summer fires.

METHODS

Potential Pollinator Observations I selected two populations of each of these three mint species (see Figure 1.1) within which to observe insect visitors and perform pollinator exclusions. Both of the populations of Conradina glabra were located within Torreya State Park lands close to state highway 12, relatively close to each other (<3 km apart; at geographic coordinates -84.929, 30.505 and - 84.899, 30.504 respectively). The populations selected for the other two species were generally much further apart from one another. They also were selected to have variation in habitat between populations selected. The populations of Stachydeoma graveolens were located off of two different state roads, on forest roads within the Apalachicola National Forest. The first population was located off of forest road 173 from state road 12 in the Apalachicola National Forest at Cotton Landing (-85.006, 29.991). The second population was located on forest road 103 off of state road 65 (-84.564, 30.382). The Physostegia godfreyi populations were also both located within the Apalachicola National Forest. The first population was located off of state road 12 on forest road 105 (-84.985, 30.060). The second population was at the same location as the first population of Stachydeoma graveolens, off of state road 12 at forest road 173 (-85.006, 29.991). Due to the drought conditions from the past few years, my goal was to perform

fieldwork for two growing seasons, March‒July 2008 and March‒June 2009 to ensure that observations made in a single field season were not a result of drought conditions or changes in controlled burn scheduling within the Apalachicola National Forest. Within each population, I selected and marked four plots (1m × 1m). Each plot had at least two flowering plants present to observe for potential pollinators. I attempted to select plots that represented habitat variety where variety was present. I made sure there was a minimum distance of 10 meters between each plot. Observations were made within a randomly selected plot for 15 minutes each hour, rotating between different plots every hour throughout the day and recording plant visitors and potential pollinators and their behavior. I observed each population from dawn to dusk for two non-consecutive days in each field season from to determine the identities of visitors and which times of the day visitation rates were highest. I based my hourly observation methods on a similar pollinator observation study on Panhandle endemics performed by Theresa Pitts-Singer (2002). During observation periods, I used a stopwatch with a timer to record the length of time each insect visitor spent contacting a single flower, and I recorded if the same insect moved to different flowers on the same plant or moved to a different plant within the plot being observed. I only followed one insect at a time, observing it until it moved out of the current plot. However, I noted that it was uncommon to find numerous pollinators on the same plant at the same time. When multiple insects were observed on a single plant, I recorded that information as a side note but did not follow and record multiple insects at the same time. I defined ―contact‖ as landing on the lip of the corolla, and ―pollinator behavior‖ as contacting the stamens or probing the corolla throat for nectar and, as a result, potentially contacting the anthers and/or stigma. Insect visitors that did not exhibit potential pollinator behavior were also observed and noted. Non-pollinator behavior was not as common as pollinator behavior, and was exhibited mostly by various beetles, flies, crab spiders, and ants. Observations of non-pollinators were noted, but were not timed during the pollinator observation periods. In all three mint species, the reproductive structures of the flowers were visible by naked eye, and the behaviors of both types of visitors were noticeably distinct from each other. Representative insects that were observed to exhibit potential pollinator behavior were collected in a kill jar during observations at each population in both field seasons to aid in identification. Images of insects were taken during observation times at each population in 2009 15

using a handheld Canon PowerShot SD87015 digital camera to aid in identification of visitors. Identification of insect visitors and potential pollinators was performed by 1) taking them to the University of Florida‘s insect collection for comparison and professional identification and 2) searching images online and 3) using field guides such as the Field Guide to Insects and Spiders of North America (Evans 2007).

Pollinator Exclusion Manipulations I performed pollinator exclusion manipulations within each of the plant populations that I visited for pollinator observations. The purpose of these exclusion manipulations was to determine if these species set seed in the absence of pollinators, or if they require pollinators to produce seeds. One of the three mint species, Stachydeoma graveolens, did not receive this treatment in 2008 because the two selected populations underwent controlled burns before it could be implemented. Due to the drought conditions that have existed during the past few years (http://www.dep.state.fl.us/drought), maintaining a normal controlled burn schedule was difficult for the forest service because of the risk of larger, uncontrolled wildfires. The year 2008 was still dry, although wetter than the previous years had been, and controlled burns were performed in several areas within the Apalachicola National Forest to take advantage of the improved conditions. The 2009 field season was planned around the burn schedule, and pollinator exclusion manipulations of all three species were planned. At the time of this writing, only Conradina glabra had flowered in 2009. Physostegia godfreyi and Stachydeoma graveolens were not yet flowering as of the beginning of June 2009, so results from the second field season for these species are not included in this document but I intend to include a second field season when this work is published. I completed the pollinator observations before implementing exclusion manipulations in both field seasons to ensure that my observations would not alter the behavior of insect visitors due to the presence of exclusion bags. Within each of the populations of each species, I randomly selected 10‒25 plants and for each plant placed fine white bridal mesh bags over single branches with at least two unopened flowers, but not more than 10 flowers. No open flowers were bagged, and only branches with unopened flowers were selected to exclude pollinators from. The bags were attached to the branches with white twist ties and secured at least 2 cm

below the selected flowers. This allowed me to observe seed set in open-pollinated and bagged flowers on the same plant. In the 2008 field season, I did not distinguish between possible pollen transfer between flowers that were in the same bag (geitnogamy) and pollen transfer to the stigma within a single flower (autogamy). Since multiple flowers were present in each bag, the possibility of pollen transfer from different flowers existed. All three mint species have anthers that curve along with the stigma along the hood under normal conditions, and pollen transfer could potentially have occurred as the corollas were shed or with direct anther-stigma contact. Wind could potentially aid in transferring pollen between flowers that were excluded from pollinators as well, although these are not considered to be wind-pollinated species. Different mechanisms for self-pollination in bagged flowers should be considered when trying to make informative decisions about the mating system and selfing ability in species of conservation interest. For many pollination studies, it is enough to know whether a plant can set seeds in the absence of pollinators, but there is some room for misinterpretation if proper experimental treatments are not included (Kearns and Inouye 1993). In 2009 I again selected 10‒25 plants within two populations of each mint species to perform pollinator exclusion manipulations on, and these were not necessarily the same plants as selected in the previous year. I also added a hand-pollination treatment in 2009 to determine the extent of seed set with mechanical application of self pollen. At the time of this writing, this treatment was only applied to Conradina glabra as the other two species were not yet flowering. For the hand-pollination treatment, I transferred self-pollen to the stigmas of 10 flowers (representing 10 plants) in each population. I excluded these flowers from pollinators by bagging them prior to floral opening, removing the bags for hand-pollination, and then replacing the bags until flowering was completed. The hand-pollinations were performed by using a small paintbrush to apply self-pollen from a flower‘s anthers to its stigma. All three mint species produced flowers that persisted for about three to four weeks, and once a week during that time bags were checked to make sure no insect visitors had penetrated the mesh and gained access to flowers. Bags that had been opened and flowers that had been damaged were not used in my results. After flowers dehisced, I removed the bags and used Elmer‘s glue to seal the calyces of the excluded flowers to prevent loss of seeds and replaced the bags once the glue dried, on the same day. This treatment was repeated for the hand-pollinated 17

bagged flowers as well. This glue treatment was recommended to me by Dr. Alice Winn as a way to prevent loss of seeds without affecting seed development. I also marked at least five (and up to 10) open-pollinated flowers during flowering with white tie-tags on the same plants selected for exclusion manipulations, and I repeated the Elmer‘s glue treatment on the marked open-pollinated calyces on each of these selected plants to compare seed set between treatments. I attempted to ensure equal numbers of open-pollinated and bagged flowers could be compared for each species. The calyces from the bagged, open-pollinated, and hand-pollinated treatments were collected two weeks after the end of flowering, and seeds were counted for each flower in each treatment. As mentioned previously, all three of these species produce four nutlets per flower, making it relatively easy to count seeds and compare the numbers of seeds produced between bagged and open-pollinated flowers with their maximum number.

Generalized linear models I used Generalized Linear Models (GLMs) to determine if seed set varied significantly between treatments, populations, and in the case of Conradina glabra, years. The seed set data were proportions—I knew how many of the maximum possible number of seeds were produced in each treatment (open-pollinated vs. bagged). Because of the data type, I used GLM with binomial distributions to look at seed set results for both Conradina glabra and Physostegia godfreyi for each population and year. The GLMs were run using the statistical software R (v. 2.7.0, R development Core Team 2008). I created initial models that looked at the individual variables, and the combinations of interactions for all of the variables. For Conradina glabra, models were created that used the variables treatment type, population, year (since C. glabra was observed over two years), and the interactions of these variables. Physostegia godfreyi was only observed in 2008. Initial models were produced with all of the variables and interactions, and then models were re-fitted with any non-significant interaction terms removed until the simplest possible models remained. These GLMs were used to determine if any of the variables or their interactions had significant effects on seed set.

Models were also checked for overdispersion, or the possibility that the data contained greater variance then expected for a binomial distribution. If this was the case, then models were re-fit using a quasibinomial function in R to account for this overdispersion.

RESULTS

Insect Abundance All three mint species were each observed for four days in 2008: two days for each population of a species observed. In 2008, I observed each species for 13 hours total (15 minutes/hour over four days); in 2009 I observed Conradina glabra for a total of 14 hours (slightly longer days). Observations could not be made for two species (Physostegia godfreyi and Stachydeoma graveolens) in 2009 because they were not yet flowering when this document was written. Insects were observed visiting all three plant species, and pollinator behavior was observed from several of these taxa. Some of these potential pollinator species were observed visiting all three mint species, but there were differences in the abundance and composition of potential pollinators that visited each species. In all three species there was also some variation in potential pollinator observations between populations of the same species, mostly in abundance but also in composition in the case of Physostegia godfreyi. Relatively small numbers of insect visitors were observed in the 2008 field season for all three mint species, although Conradina glabra had the most insects observed with nearly twice the number of potential pollinator visits recorded as the other two (Table 1.2). Stachydeoma graveolens and Physostegia godfreyi had comparable numbers of potential pollinators observed. These species are more patchily distributed within populations than Conradina glabra, which could indicate why fewer potential pollinators were documented during observation periods. In 2009, there were almost twice as many observations of potential pollinators as compared the 2008 field season for Conradina glabra (Table 1.2).

Table 1.2. Number of potential pollinators observed for populations of Conradina glabra (in 2008 and 2009), Physostegia godfreyi (in 2008), and Stachydeoma graveolens (in 2008). Populations of Conradina glabra Number of Insect Visitors Observed 2008

Table 1.2 continued. Number of potential pollinators observed for populations of Conradina glabra (in 2008 and 2009), Physostegia godfreyi (in 2008), and Stachydeoma graveolens (in 2008).

Populations of Conradina glabra Number of Insect Visitors Observed 1 46 2 60 Total 106 2009 1 108 2 77 Total 185 Populations of Stachydeoma graveolens Number of Insect Visitors Observed 1 29 2 28 Total 57 Populations of Physostegia godfreyi Number of Insect Visitors Observed 1 28 2 16 Total 44

Time of Day Observations of Pollinators Pollinator visitations in the 2008 field season usually peaked twice a day in all populations observed (Figures 1.5‒1.7) for all three species. In 2008, the peak times of visitation overall for all three mints occurred between 9 a.m. and 11 a.m., and also between 1 p.m. and 3 p.m. There was some variation observed between species and between populations. In 2009 a similar pattern was again observed for Conradina glabra. The composition of potential pollinators did not appear to change throughout the day for all three plant species, although the low numbers of recorded visits might biase this observation. The number of observations declined closer to dusk within all three species. No potential pollinators visited the flowers during thunderstorms, rain, or other wet conditions, and fewer were observed when it was overcast.

Figure 1.5. Time of day observations of potential pollinators for Conradina glabra for all four days over two populations in 2009.

Figure 1.6. Time of day observations of potential pollinators for Physostegia godfreyi for all four days over two populations in 2008.

Figure 1.7. Time of day observations of potential pollinators for Stachydeoma graveolens for all four days over two populations in 2008.

Potential Pollinators Most of the potential pollinators observed at all three species were bees and insects native to North America and were all from genera that are widespread within the state (and outside of it) than those of the mint species observed. Within the observed populations of Conradina glabra the composition of plant visitors was very similar. The potential pollinators observed in 2008 for C. glabra consisted primarily of beeflies (Bombylius major), carpenter bees (Xylocopa micans), and to a smaller extent digger bees (Anthophora abrupta) (Table 1.3, Figures 1.8‒1.11). There were other potential pollinators observed, but these were less abundant than these three taxa. The identifications of some of these other pollinators are tentatively given to family, and include sweat bees (Halictidae), digger wasps (Scoliidae), mining bees (Andrenidae), and a brown skipper moth (Hesperidae) (see Table 1.3, Figure 1.10). In 2008 10 sweat bees, two brown skipper moths, two scoliid wasps, and two mining bees were observed visiting Conradina glabra flowers. In the 2009 field season, there were almost twice as many potential pollinators observed in these populations than in 2008 (Table 1.2). The potential pollinators observed the most often in 2009 again included Xylocopa micans and Bombylius major, but many more individuals of Anthophora abrupta were observed in both populations. The latter made up the largest percentage of observations for C. glabra in 2009. Representatives of Halictidae, Scoliidae, and Hesperidae were observed again, but new solitary bees were also observed in 2009. These included honeybees (Apis mellifera, Figure 1.11) and leaf cutting bees (Megachile spp., Figure 1.11). Twenty‒two various mining bees, 14 brown skipper moths, two sweat bees, two scoliid wasps, and a honeybee were observed in total, far more bees than in 2008.

Table 1.3. Composition of the three most observed potential pollinators for Conradina glabra. Shows the total number of each insect type observed, the percent of the total insect observations for Conradina glabra, and the rate of visitation for each insect type shown. Conradina glabra Bombylius Xylocopa Anthophora 2008 major virginica abrupta Total Observed 65 21 9 Percent of Total Observations 61.30% 19.80% 8.50% Rate of Visitation 5.00/hour 1.62/hour 0.69/hour 23

Table 1.3 continued. Composition of the three most observed potential pollinators for Conradina glabra. Shows the total number of each insect type observed, the percent of the total insect observations for Conradina glabra, and the rate of visitation for each insect type shown. Conradina glabra Bombylius Xylocopa Anthophora 2009 major virginica abrupta Total Observed 50 36 63 Percent of Total Observations 27.00% 19.50% 34.00% Rate of Visitation 3.57/hour 2.57/hour 4.50/hour

Figure 1.8. Images of carpenter bees (Xylocopa micans) observed in Conradina glabra population in 2009, robbing nectar from flowers. Photo credits: Amanda Kubes.

Figure 1.9. Images of Bombylius major (beefly) observed in Conradina glabra populations in 2009. Photo credits: Amanda Kubes.

Figure 1.10. Digger bee (Anthophora spp.) observed and captured in Conradina glabra population in 2009. Photo credits: Amanda Kubes.

Figure 1.11. Less frequently observed visitors to Conradina glabra. From top left clockwise: digger wasp (Scoliidae) observed in 2008 and 2009, brown skipper moth (Hesperidae) observed in 2008 and 2009), and honey bee (Apis mellifera) observed in 2009. Other potential pollinators include sweat bees (Halictidae), mining bees (Andrenidae), and leaf cutter bees (Megachile spp.). Photo credits: Amanda Kubes.

Stachydeoma graveolens pollinator observations were only conducted in 2008. There are no results for 2009 yet. There were far fewer potential pollinator visits recorded than for Conradina glabra. This species was visited primarily by leaf cutter bees (Megachile spp.) at both populations (Table 1.4). Other potential pollinators were observed and many were the same taxa observed visiting Conradina glabra populations. Anthophora abrupta, scoliid wasps, halictid bees, and brown hesperiid moths were the other potential pollinators observed visiting S. graveolens populations, with five sweat bees and four brown skipper moths. The composition and abundance of pollinators was similar within both populations, although one site had more Anthophora abrupta individuals recorded visiting than the other.

Table 1.4. Composition of the three most observed potential pollinators for Stachydeoma graveolens. Shows the total number of each insect type observed, the percent of the total insect observations for Stachydeoma graveolens, and the rate of visitation for each insect type shown. Stachydeoma graveolens Anthophora 2008 Megachile spp. abrupta Scoliidae Total Observed 35 9 4 Percent of Total Observations 61.40% 15.80% 7.01% Rate of Visitation 2.7/hour 0.69/hour 0.31/hour

Physostegia godfreyi populations were only observed in 2008. There are no results yet for 2009. The two populations of P. godfreyi populations differed in potential pollinator composition, possibly due to differences in the habitat each population occupied. One population was located in an open wiregrass pitcher-plant savanna (Population 1), and the other population was in a wiregrass-saw palmetto savanna with a pine overstory (Population 2). Although there were differences in the composition of insects between populations, the taxa observed were primarily the same as observed for Conradina glabra and Stachydeoma graveolens. The main visitors to Population 1 were Xylocopa virginica and Anthophora abrupta. This population also had fewer individual visitors observed than Population 2 (see Table 1.2). The main visitors to Population 2 were Megachile spp. Halictid bees and brown hesperiid moths were a observed in both populations. In terms of the potential pollinators observed the most often for Physostegia godfreyi, Megachile spp. made up the largest percent of visitors observed, followed by halictid bees and Anthophora abrupta (Table 1.5). The other visitors observed included four carpenter bees, three brown skipper moths, and a singled gossamer-winged butterfly. Of the three mints observed, Physostegia godfreyi had the most variability in potential pollinator visitors, perhaps due to differences in habitat type between the two populations.

Table 1.5. Composition of the three most observed potential pollinators for Physostegia godfreyi. Shows the total number of each insect type observed, the percent of the total insect observations for Physostegia godfreyi, and the rate of visitation for each insect type shown. Physostegia godfreyi Anthophora 2008 Megachile spp. Halictidae abrupta

Table 1.5 continued. Composition of the three most observed potential pollinators for Physostegia godfreyi. Shows the total number of each insect type observed, the percent of the total insect observations for Physostegia godfreyi, and the rate of visitation for each insect type shown. Physostegia godfreyi Anthophora 2008 Megachile spp. Halictidae abrupta Total Observed 19 9 8 Percent of Total Observations 43.10% 20.50% 18.20% Rate of Visitation 1.46/hour 0.70/hour 0.62/hour

All three mint species are visited by multiple potential pollinator taxa (Tables 1.3, 1.4, 1.5), and none appear to be limited to few or specialist species. Stachydeoma graveolens was visited by the smallest number of different taxa and was visited the most often by a single potential pollinator (Megachile spp.) compared to the other two species. Conradina glabra had the most pollinator observations observed in 2008 and nearly twice that amount in 2009, and also had the greatest number of different taxa observed. Physostegia godfreyi had the smallest numbers of observed potential pollinators, but had more diversity in composition than S. graveolens.

Pollinator Exclusions, Mating Systems, and Seed Set The two plant species for which I performed exclusion manipulations are self-compatible. Conradina glabra and Physostegia godfreyi both produced seeds in the absence of pollinators (Table 1.6), although the specific method of autogamy and/or geitnogamy is uncertain. In general, 2‒4 times the number of open-pollinated flowers produced seeds as did pollinator- excluded flowers for these species (Table 1.6). Between 76 and 92% of open-pollinated flowers set seed, whereas between 18 and 39% of pollinator-excluded flowers set seed. A total of 20 flowers, 10 from each population of Conradina glabra in 2009, were hand- pollinated and excluded from pollinator visitation. 13 (65%) of the 20 hand-pollinated, pollinator-excluded flowers produced seeds. This is approaching the percent of flowers that produced seed in the open-pollinated treatments and suggests that pollen transfer is inefficient in the absence of pollinators. Generalized linear models were performed in R (v. 2.7.0, R development Core Team 2008) for both Conradina glabra and Physostegia godfreyi to examine the effects of year,

population, pollination treatment, and the interactions of these variables on seed set. Year, population, and their interactions did not appear to have significant differences in seed set in Conradina glabra, but the variable treatment alone showed a highly significant difference in the number of seeds produced (Table 1.7). In Physostegia godfreyi, the interaction between treatment and populations was not significant. Population on its own appears to be significant, but the treatment variable again, as with Conradina glabra appears to be highly significant with regards to differences in the number of seeds produced (Table 1.8).

Table 1.6. Total number of seeds produced in both open-pollinated and pollinator- excluded flowers for plants in both populations of Conradina glabra (2008 and 2009) and Physostegia godfreyi (2009). Conradina glabra-Population 1 2008 Open Pollinated Bagged Hand Pollinated Number of flowers (20 plants) 173 173 na Number of flowers with seeds 137 47 na Number of seeds 404 127 na % of flowers that set seeds 79% 27% na Conradina glabra-Population 2 2008 Open-Pollinated Bagged Hand Pollinated Number of flowers (15 plants) 143 143 na Number of flowers with seeds 109 56 na Number of seeds 286 139 na % of flowers that set seeds 76% 39% na Conradina glabra-Population 1 2009 Open Pollinated Bagged Hand Pollinated Number of flowers (17 plants) 98 98 20 Number of flowers with seeds 88 27 13 Number of seeds 271 70 40 % of flowers that set seeds 90% 28% 65% Conradina glabra-Population 2 2009 Open Pollinated Bagged Hand Pollinated Number of flowers (18 plants) 100 100 na Number of flowers with seeds 82 29 na Number of seeds 236 56 na % of flowers that set seeds 82% 29% na Physostegia godfreyi-Population 1

Table 1.6 continued. Total number of seeds produced in both open-pollinated and pollinator-excluded flowers for plants in both populations of Conradina glabra (2008 and 2009) and Physostegia godfreyi (2009). Physostegia godfreyi-Population 1 2008 Open-Pollinated Bagged Hand Pollinated Number of Flowers (10 plants) 97 97 na Number of Flowers with Seeds 77 17 na Number of Seeds 263 41 na % of flowers that set seeds 79% 18% na Physostegia godfreyi-Population 2 2008 Open-Pollinated Excluded Hand Pollinated Number of flowers (10 plants) 94 94 na Number of flowers with seeds 86 34 na Number of seeds 288 71 na % of flowers that set seeds 92% 36% na

Table 1.7. Final generalized linear model (GLM) for Conradina glabra. The variable treatment is responsible for the majority of the deviance in seed set. Variables Deviance Residual Deviance F Pr(>F) Population 3.6 3571 1.4521 0.228478 Year 1.2 3569.8 0.4746 0.491039 Treatment 679.2 2890.6 275.2345 <2.2e-16*** Population:Treatment 10.3 2880.3 4.1679 0.041454* Year: Treatment 25.1 2855.2 10.1523 0.001485** Significance codes: 0 ‗***‘ 0.001 ‗**‘ 0.01 ‗*‘ 0.05 ‗.‘ 0.1 ‗ ‘ 1

Table 1.8. Final generalized linear model (GLM) for Physostegia godfreyi. The variable treatment is responsible for the majority of the deviance in seed set, although population appears to be slightly significant as well. Variables Deviance Residual Deviance F Pr(>F) Population 14.16 1598.11 5.2407 0.02262* Treatment 562.75 1035.36 208.2485 <2e-16*** Significance codes: 0 ‗***‘ 0.001 ‗**‘ 0.01 ‗*‘ 0.05 ‗.‘ 0.1 ‗ ‘ 1

DISCUSSION

A growing concern in conservation is that disruption of pollination systems and declines of certain types of pollinators have been reported on every continent except Antarctica (Kearns et al. 1998). Although large areas of different countries and continents have not been studied, we can assume that disruption is widespread because the causes are associated with human activities that are widespread, such as habitat fragmentation, invasive species spread, and climate change. There is concern about the preservation of existing interactions, particularly those of rare and geographically limited plant species. For plants that are self-incompatible, or even in facultatively selfing species, if seed production is dependent on pollinators then it could become limited under conditions where pollinator abundance declines or pollinator behavior (Evans et al. 2003). This pollinator limitation or failure could cause seed production in populations to fall below what will be needed to maintain these populations. This provided the motivation behind this research. I discuss the results of my pollinator observations and pollinator exclusion manipulations in the context of floral morphology and similar research and similar studies on other rare Florida endemic mint species.

Pollinator observation results I found that insects visited all three mint species, and many of these visitors exhibited pollinator behavior. Conradina glabra had the greatest number of insect visitors observed as well as the most diversity in composition, and Stachydeoma graveolens had the least on both counts (Tables 1.2, 1.4). Based on the diversity and numbers of potential pollinators observed for all three species, it does not seem likely that these mint species are in immediate danger of pollinator limitation were one or two species to be become unavailable. All of the insect taxa observed were generalists and are known to visit other flowering plants as well. The majority of the potential pollinators observed for all three species were native bees, including Eastern carpenter bees (Xylocopa virginica), leaf cutting bees (Megachile spp.), digger bees (Anthophora abrupta) and sweat bees (Halictidae). Other insects were also observed that exhibited pollinator behavior such as bee flies (Bombylius major), digger wasps (Scoliidae), and brown skipper moths (Hesperidae). Although there were some differences in the composition of insect visitors

(in diversity and abundance) observed within each mint species, representatives of virtually all of these groups were present in all three species observed. Surveys of insect visitors to plants in the Florida Panhandle could be performed to make species lists of the insect visitors present within sites of interest. None currently exist for the Apalachicola National Forest, although similar pollinator observation studies have been performed on other rare mint species (e.g., Theresa Pitts-Singer 2002; Evans et al. 2004). A species list would be useful to catalogue other potential pollinators for these species and others of interest in the area (e.g., as in Bartholomew et al. 2006). In this research, the most frequently observed insect taxa are all representatives of genera that have large numbers of species, many which have widespread distributions throughout the state or parts of the country. Most of these taxa are known to be generalist pollinators that are not restricted to a single species of flower. Although most of these taxa are considered to be useful pollinators, it is noted that some of these also act in ways that do not appear to be beneficial for pollination. For example, Xylocopa species are known to be flower robbers, and research has been performed on members of the genus to determine their effectiveness as pollinators in different systems (Castro et al. 2008; Zhang et al. 2007). Members of this genus are known to be nectar robbers, and in mint species (and other flowers) will make a slit at the base of the corolla to obtain nectar. Robbing flowers can be detrimental to reproduction in some casesbecause it might prevent pollen deposition or cause tissue damage that results in no seed set. On the other hand, flower robbers that make slits in corollas to obtain nectar have been found to positively affect pollination in some plant species such as rabbiteye blueberry (Vaccinium ashei) (Maloof and Inouye 2000; Irwin et al. 2001; Sampson et al. 2004). Flowers that have slits at the corolla base can attract short-tongued bees such as honeybees (as secondary robbers) to visit flowers more often which can lead to increases in pollen transfer (Sampson et al. 2004). Based on the known research, here I consider Xylocopa virginica to be a potentially important pollinator of these mint species. Although the pollination and reproductive biology of the three mint species was relatively unknown, it was not surprising to observe that each species seems to be pollinated by multiple insect taxa. There are several reasons why. Most plant species are visited by multiple floral visitors (Memmott 1999; Memmott and Waser 2002; Vazquez and Simberloff 2002; Jordano et al. 2003). Similar studies on other threatened Florida endemic mints such as Dicerandra, 32

Macbridea alba, and Scutellaria floridana (Pitts-Singer 2002; Evans et al. 2007) indicate that similar pollinators also visit these species and appear to be important to their reproductive biology in terms of seed set and genetic diversity. The structure of Lamiaceae flowers itself suggests an adaptive response to pollinators (Westerkamp and Claben-Bockhoff 2007). Although in this research we consider bees only in the context of their contributions to pollination, bees also take pollen to store as food for their larva (Westerkamp and Claben- Bockhoff 2007), in some cases leaving very little pollen left over for pollination of other flowers. Most bees use pollen-carrying structures such as brushes or hairs on the abdomen and/or the hindlegs, which helps the bee retain pollen but effectively prevents pollen from being used in pollination by flowers. The best way for a flower to transfer pollen to a pollinator but avoid losing large amounts of pollen is for it to be deposited on the back (dorsal side) of a pollinator. Bilabiate (two-lipped) flowers accomplish this as an evolved defense against pollen-collecting bees and a means to ensure more efficient pollination. Bilabiate flowers are a hallmark of mint species, and they have also arisen in parallel in numerous other angiosperm groups, meaning there may be strong selection pressure towards this floral morphology (Westerkamp and Claben- Bockhoff 2007) to protect pollen and improve pollination efficiency. It should be noted that although I included the insects that were observed the most as the most important potential pollinators, this may not reflect the true relationships between these mints and their effective pollinators. The most frequent visitors to a plant may not be the most effective or important pollinators (Schemske and Horvitz 1984). This has been shown to be true in studies that have examined pollen transfer by insects, and in some cases large bees such as Xylocopa spp. are less efficient as pollinators than smaller and less abundant taxa (Sampson et al. 2004). These results should be considered in the context of additional research to determine the actual relationships of specific pollinator taxa with these mint species. Although these plants do not appear to have specialist relationships with pollinators, it is still useful to know who the effective and important pollinators are. Changing conditions could disrupt these interactions, and although they may be generalist in preference, if potential pollinators are scarce or ineffective these species may be at more risk of extirpation.

Seed set

Differences in floral structure can be adaptations to certain pollinators or groups of pollinators, but can also affect mating systems. For example, the placement of stigmas and style in self-fertile taxa can control the degree of geitnogamy or xenogamy. It is known that different mechanisms in the Lamiaceae are used to promote out-crossing (Rodriguez-Riano and Dafni 2007), but self-compatibility is also known. It has been observed in rare and fragmented populations of some species as well (e.g., Macbridea alba; Walker and Madsden 1997). Therefore, it seems likely that a mixed mating system is present in these mints as well. I predicted that pollinators would be important to seed set in the mint species observed in this research, based on previous research (e.g., for Dicerandra by Evans et al. 2004). After performing pollinator exclusion manipulations on flowers and analyzing the resulting seed set, it was determined that pollinators do significantly affect seed set in the two species studied. Significantly fewer seeds were produced in Conradina glabra and Physostegia godfreyi flowers that had been bagged to prevent pollinator visitation. However, the hand-pollination treatment of Conradina glabra in 2009 indicates that pollinator-assisted selfing yields seed set comparable to open-pollinated flowers. These results were not surprising based on the results of similar research on another southeastern endemic mint, Dicerandra. The study of Evans et al. (2004) focused on two rare endemic species of Dicerandra (D. frutescens and D. christmanii) and found that pollinators are important to seed set in the populations studied. And similar to my results, Dicerandra flowers that had been hand- pollinated and bagged produced more seeds than flowers that were simply excluded from visitation (up to 80% of the amount open-pollinated flowers produced). This indicates that Dicerandra depends on pollinators for maximum seed set. All of the species in this study showed similar patterns in seed set. Dicerandra is included within the southeastern scrub mint clade that contains Conradina spp. and Stachydeoma graveolens, and this clade is characterized by similar morphology and habitat preference. It seems logical that they would also have similar pollination requirements. Further studies to determine if pollinator composition changes over time or if the ability to autonomously self-pollinate changes under different conditions (drought, pollinator limitation) should be considered. Although these results indicate that pollinators might be important for the production of maximum seed set, it does not necessarily mean that larger numbers of seeds are beneficial to the survival of a species. In general, when large numbers of seeds are produced by relatively long- 34

lived species, seeds are less important to population growth and maintenance than the survival of established plants (Silvertown et al. 1993). This view is less often considered in conservation biology, where seed set is often related to ecological interactions and is viewed as a proxy for survivial and success of a species (Dewenter and Tscharntke 1999). However, even if large numbers of seeds are not important to the persistence of a species, other problems could arise if the majority of seeds produced (even if numbers are small) are from self-pollen, such as inbreeding depression. It has been observed that this can lead to low genetic diversity in small populations, which can increase the risk of a rare species to extirpation or an inability to recover under changing conditions (Godt et al. 2004). Evans et al. (2004) make the point that the reproductive biology of rare species is not always predictable, and to improve this more consideration should be made of the life history of a species, ecological relationships, and phylogenetic history. This underlines the point that the more information we have about a rare species, the better able we will be to predict how it will respond to changing conditions over time.

CHAPTER 2 MODELING THE CURRENT AND FUTURE DISTRIBUTIONS OF THREE ENDEMIC FLORIDA PANHANDLE MINTS AND THEIR POLLINATORS

INTRODUCTION

Knowledge of species‘ distributions and ecological interactions are important for the conservation and maintenance of threatened species (Soberon and Peterson 2005; Ferrier 2002, Funk and Richardson 2002, Rushton et al. 2004). Species distribution models are increasingly used to help make conservation and management-based decisions (Peterson et al. 2000; Guisan and Thuiller 2005; Whittaker et al. 2005). They have been used to find previously undiscovered populations of rare and endangered species (Engler et al. 2004) and have been able to identify areas that have high potential for colonization of species of interest (Hirzel et al. 2002), including invasive species (Peterson 2003; Peterson et al. 2003). These models are also being used to examine the potential impacts of climate change (Busby 1991; Guisan and Theurillat 2000; Dirnböck et al. 2003) on species‘ distributions. Species restricted to small areas or specific habitat types that may be threatened by human activity (Myers et al. 2000) are an important focus for conservation, particularly in light of the threat of climate change. Although there have been numerous studies that predict species distributions under future climate conditions (Thuiller et al. 2005; Bakkenes et al. 2002; Viney et al. 2007), few have attempted to examine the potential effects of climate change on ecological interactions. It is recognized that climate change may affect interactions between species, such as between plants and their pollinators (Memmott et al. 2007), but in the case of many species of conservation concern these interactions remain relatively unknown. This was illustrated in Chapter 1, where the potential pollinators of three Florida Panhandle endemic mint species were observed and documented as a first step towards determining the potential effects of climate change on these species. Modeling potential shifts between plant and pollinator distributions can be used to inform us if these species will be at increased risk in the future if their distributions no longer overlap.

In this chapter, I use the distribution modeling method of maximum entropy modeling (Phillips et al. 2006) to create potential distribution maps for each of the three Florida Panhandle endemic mint species and for three of the potential pollinators observed in Chapter 1. I create species distribution models for these taxa under current conditions and under a future climate change scenario to look at potential range shifts between the mints and their pollinators. I examine the sensitivity of the results against changes in a parameter (the regularization multiplier parameter) that affects the closeness of fit of the model to the training to explore the effect on the predicted distributions for the three mint species. I present the modeling results for the mints and pollinators and discuss potential future areas of improvement. This chapter answers the following questions: (1) where are the plants and their pollinators expected to be today, (2) where are the plants and their pollinators expected to be in the future?, and (3) do these future predicted ranges overlap?

Species distribution modeling Species distribution models are most often used to create predictions of potential distributions by correlating the current presence or abundance of a species to environmental conditions that are considered to be relevant to the distribution of the species of interest. Most of the modeling approaches that have been created for predicting species distributions have their roots in efforts to quantify the relationships between species and their environments. These models are spatially explicit representations of these relationships (Austin 2002; Guisan and Thuiller 2005; Elith et al. 2006), meaning the output map gives a snapshot of the predicted distribution at a specific point in time. These models have a long history of use in ecology and have their origins in the concept of the ecological niche (Grinnell 1924; Hutchinson 1957). The ecological niche of a species is often considered in terms of its fundamental niche and its realized niche. The fundamental niche of a species is defined by the total range of environmental conditions that are suitable for existence in the absence of limiting factors such as competition from other species or predation (Silvertown 2004; Hutchinson 1957). The realized niche is the amount of environmental space within the fundamental niche that the species actually occupies, and so species distribution model outputs are usually considered to be representing the realized niche of a species (Guisan and Zimmerman 2000; Pearson and Dawson 2003; Araujo and Guisan

2006). I consider this research in the context of modeling the realized niche of the species of interest. Most of the research on developing distribution modeling methods has focused on creating models that use presence and absence or abundance data that have been systematically collected (Hirzel and Guisan 2002, Cawsey et al. 2002). However, the distribution data that exist for most species have been recorded without a systematic sampling method, and most of these data are presence only records from museum or herbarium collections that are accessible electronically (Graham et al. 2004, Huettmann 2005, Soberon and Peterson 2005, Stockwell and Peterson 2002). For species of conservation concern, occurrence records are often scarce and reflect presence-only observations (Hernandez et al. 2008), which make selecting a modeling method that will provide reliable predictions of distributions from presence-only data important. Although there is a large number of different species distribution modeling methods available including statistical methods (generalized linear and additive models, regression-based methods), climatic envelope methods, and machine learning methods (Guisan and Thuiller 2005; Elith et al. 2006), there are few comparative reviews to aid a researcher in model selection (Elith et al. 2006). Comparing the performance of multiple modeling methods across an array of datasets can provide information on which methods consistently perform well in predictive ability. The improvement of presence-only methods and the means of evaluating how well they make predictions are important for the creation of models that accurately reflect a species‘ relationship to its environment. Early attempts to model distributions with presence-only data used methods that calculated the climatic envelope of a species or used simple distance-based measures taken from the known distribution (Silverman 1986; Busby 1991; Walker and Cocks 1991; Carpenter et al. 1993) to make predictions of the potential distribution. Species distribution modeling then turned to using presence-absence data (those data that can be modeled by a binomial response). Using presence and absence data can improve predictions of a species‘ distribution by giving a researcher the locations and environmental conditions where it is present and where it is not. In reality it is difficult to obtain accurate absence data, particularly for rare species or species in hard to navigate areas. Some of the more recent modeling methods (reviewed in Elith et al. 2006) can utilize presence and absence data to test how well models that predict distributions from presence-only data perform. This is done by withholding samples of presence and absence data and examining how well the models predict the withheld data (Stockwell and Peters 1999; 38

Boyce et al. 2002; Ferrier et al. 2002; Zaniewski et al. 2002; Keating and Cherry 2004). The latest modeling methods have their foundations in ecological and statistical research, and appear to perform well in predicting distributions with presence-only records, particularly when modeling species with small numbers of occurrence records (Hernandez et al. 2008; Elith et al. 2006). Based on the comparative review of modeling methods by Elith et al. (2006), I selected Maxent (Phillips et al. 2006) as the modeling method for my study. It creates pseudo-absences to use in the model by taking random samples of locations in the known range and environmental conditions given where the organisms have not been shown to be present (Stockwell and Peters 1999; Elith et al. 2006). It generally outperformed other methods in the study, including such commonly used modeling methods such as Bioclim and GARP (Elith et al. 2006). It has performed well in multiple studies that include future predictions of species‘ distributions under climate change (Phillips et al. 2006; Hijmans et al. 2006; Heikkinen et al. 2006; Peterson et al. 2007; Phillips and Dudik 2008).

Evaluation of species distribution models Selecting a species distribution modeling method needs to take into account the ecological question being asked and the properties of the data being used, since different methods require different input data and also have differences in how they model the distribution. There are different ways to evaluate the performance of a modeling method. Most commonly the numbers of prediction errors (omission or commission errors, Fielding and Bell 1997) made by its models are used to evaluate a method. Accuracy and the ability of its models to be generalized for use with different study systems are also considered when determining an appropriate method (Fielding and Bell 1997). A method is of more use for conservation management if it can be used across multiple species that may have different distributions, environmental conditions, and variable amounts of occurrence data available. There are different statistics that are used to evaluate the predictive performance (accuracy) of modeling methods, such as the AUC-ROC (area under the receiver operating curve) and Kappa statistics. The AUC-ROC measures a model‘s ability to distinguish between where a species is present and where it is absent. It can also be useful if you are trying to rank contributions to an output such as environmental variables (Fielding and Bell 1997; Elith et al. 2006). This statistic is used in this research to assess the predictive performance of the models.

Maximum entropy I used the Maxent software (v. 3.2.19, http://www.cs.princeton.edu/~schapire/maxent/) to create species distribution models. Maxent follows the principle of maximum entropy, which states that the best approach is to make sure our prediction satisfies any constraints on the unknown distribution that we have set but that it otherwise has maximum entropy (Jaynes 1957; Phillips et al. 2006). Entropy is considered as a measure of how much choice is involved in the model selection process—the more entropy the distribution has, the less constrained it is. It uses environmental predictors (grids of environmental variables) at a species‘ known presence locations as well as 10,000 randomly selected locations (pseudo-absences) that are not known locations of the organisms from the spatial extent to create probability distributions of a species‘ occurrence for each of the grid cells (Phillips et al. 2004; Phillips and Dudik 2008). The modeling algorithm used by the software undergoes an iterative process to build the model, which begins by assuming a uniform probability distribution for species occurrences. It then excludes all possibilities that are outside of the target distribution that are specified by the constraints of maximum entropy to create predicted distributions based on the relationship between the environmental variables and species occurrences. Maxent performs internal validation tests to improve the model prediction by dividing these occurrence data into training and testing datasets (Phillips et al. 2006; Ortega-Huerta and Peterson 2008). Although it is among the newer methods in use in ecology, it has already been used in numerous studies on different organisms, from predicting invasive species‘ spread (Ward 2007) to climate change modeling (Phillips and Dudik 2008).

Study System Species distribution models for current and future climate conditions were created for the three Florida Panhandle endemic mint species observed in Chapter 1: Apalachicola rosemary (Conradina glabra), Mock pennyroyal (Stachydeoma graveolens), and Godfrey‘s false dragonhead (Physostegia godfreyi). Three of the most ubiquitous potential pollinators observed from Chapter 1 were also modeled: Carpenter bees (Xylocopa virginica), digger bees (Anthophora abrupta), and leaf-cutter bees (Megachile spp.). The mint species selected were chosen because they were all of conservation concern at the state, and in one case federal, levels. The three pollinator species were chosen because they 40

were observed the most frequently and deemed likely to be important potential pollinators of the mint three species. They also had available occurrence records, which is not the case for many pollinator species (Sabrowsky 1962). Lack of available occurrence records for North America prevented the inclusion of bee flies (Bombylidae) in the modeling. Plant species. Conradina glabra is a species that is listed as endangered at the federal and state levels. It is a perennial shrub that grows up to 0.8 m tall and is densely branched. This species is known to flower from March through May or June, and very occasionally until December (FNAI 2000; Clewell 1985) (see Chapter 1, Table 1.1). Flowers are borne in the leaf axils in groups of two to three, and are from 1.3–1.9 cm in length. It is distinguished from other species of Conradina by its glabrous calyx. The flowers have a lower lip that has three lobes, and they vary in color from white to pale purple or pink, with a band of darker purple spots on the throat. It is narrowly distributed within a single county in the Florida Panhandle (Liberty County) and occupies dry, sandhill or scrub habitat that often borders steephead ravines (Gray 1965; Clewell 1985; FNAI 2000). There are 8 recorded natural populations (Table 2.1), occurring mainly within Torreya State Park and near the Park boundaries. It reproduces by seeds, and each flower produces up to four nutlets, which is characteristic of members of the Lamiaceae. Stachydeoma graveolens is not a federally listed species, but is listed as threatened within the state due to its limited distribution. It is a perennial shrub that has a short woody stem and that is densely branched. Its flowers are small (1.3 cm long), and are produced in leaf axils near the top of the stem. They are bright pink or purple with darker spots lining the throat (Clewell 1985; Wunderlin 1998), and they have three lobes on the lower lip. Flowering occurs from May to July (see Chapter 1, Table 1.1) with observations extending to September (FNAI 2000). Stachydeoma graveolens is found only within seven counties in the Florida Panhandle, with most populations occurring within the Apalachicola National Forest. This species occurs within sandhills or dry areas in pine-palmetto-wiregrass flatwoods (FNAI 2000). Stachydeoma graveolens reproduces by seeds, producing up to four nutlets per flower. Physostegia godfreyi is not a federally listed species, but is also listed as threatened within the state due to its limited distribution. It is an herbaceous mint that grows up to a height of 0.6–0.9 m. The flowers develop acropetally, and grow along a tall spike at the top of the stem and face in different directions. The flowers are small (1–1.3 cm long) and tubular with three 41

lobes on the lower lip (Clewell 1985; Wunderlin 1998; FNAI 2000); they are normally pale pink to purple with purple spots or streaks within the throats. Flowering occurs from May through July (FNAI 2000), although it is a fire-responsive species and may be induced to flower later if fire treatment is applied (see Chapter 1, Table 1.1). This species occurs within seven counties of the Florida Panhandle, with most population occurrences documented in the Apalachicola National Forest (FNAI 2000). This species occupies moister habitats than the previous two species, preferring wet flatwoods, prairies, and pitcher plant bogs (FNAI 2000). This species also produces four nutlets per flower. Insect taxa. Xylocopa virginica is a member of the Apidae, a family of solitary and social bees that includes bumblebees and honeybees. Xylocopa virginica is not threatened or endangered federally or within the state of Florida. It closely resembles a bumblebee in size and coloring, except that it lacks hairs on the abdomen (Gerling 1976). The male and female of this species can easily be distinguished by the presence of a yellow or white patch on the face of males. Males are also unable to sting. They make their nests within sound dead wood and occupy diverse habitats, including suburban areas. Females tunnel into wood and create one to several partitions or cells which they use to lay eggs in. After the eggs hatch and larva mature into adult bees, usually in the late summer, they hibernate over the winter and emerge again in the spring to look for mates (Balduf 1962; Gerling 1976). They are being increasingly valued as pollinators, although they are known to rob nectar from flowers by boring a hole at the base (Sampson et al. 2004; Zhang et al. 2007; Castro et al. 2008). Females collect pollen as well to provide for brood cells. Pollen is likely to be deposited on the backs of these bees when they visit flowers due to the floral structures of the mint flowers (Westerkamp and Bockhoff 2007). Xylocopa virginica has a more widespread distribution than any of the three mint species, with occurrences documented from New England and nearby Canada south to Florida and Arkansas and west to Nebraska, Kansas, Oklahoma, and east Texas (Gerling 1976; Barthell and Baird 2004). Anthophora abrupta is also within the family Apidae. This species is not listed as threatened or endangered. Individuals are medium-sized bees that are yellow and black and resemble bumblebees. They have less hair on the abdomen and are smaller than bumblebees (Norden 1984), and they do not sting. Males can be distinguished by the presence of a ‗mustache‘ of black hairs on the face that are used for collecting plant fragrances to mark 42

territories (Norden 1984; Norden 1985). They are solitary bees, and females build nests in hard bare soil, sand, or clay. As with Xylocopa virginica, this species collects pollen as a food source for larvae and is known to serve as a pollinator for numerous plant species including the three mint species (Chapter 1) and others such as Asclepias, Azalea, Iris, Rosa, and Delphinium (Norden 1984). Anthophora abrupta is distributed across most of the eastern United States, from Florida through Texas and north as far as Ohio. They are solitary bees, and females build nests in hard bare soil, sand, or clay. Megachile is a very large genus of bees within the Megachilidae, with more than 500 species and 50 subgenera known worldwide. None of the species that occur in Florida are listed as threatened or endangered. They are medium-sized bees that often have gray or yellow stripes on the abdomen. They are also distinguished by the presence of a pollen basket on the underside of the abdomen in females, compared to most bees which have specialized hairs on their legs to serve as pollen baskets (Evans 2007). Megachile bees dig burrows in sandy loose soils, stems, or rotten wood and create brood cells which females stock with pollen and nectar for developing larva (Evans 2007). They are often observed in spring visiting flowers in mainly open habitats (Evans 2007). There are more than 10 Megachile species that occur in Florida, and all of these species are distributed throughout the eastern United States, from Texas eastward and north as far as Vermont (Whitfield and Richards 1992; Evans 2007).

METHODS

Plant Occurrence Data Florida Natural Areas Inventory (FNAI) tracks species of interest, with regular surveys of these species to document population occurrences [personal correspondence with Louise Kirn and Amy Jenkins]. I obtained population occurrence records for Conradina glabra, Stachydeoma graveolens, and Physostegia godfreyi from FNAI in an ArcGIS shapefile. A shapefile is used in mapping software such as a GIS (geographical information system); it is usually viewed as a map that shows features such as roads, water bodies, and points of interest. In this case, the shapefile used was a map of Florida divided into counties that showed all of the known occurrence records for all three mint species as points on the map. I used this shapefile in ArcMap 9.3 to create individual shapefiles for each of the mint species‘ population occurrence

records (Figure 2.1). Conradina glabra has only eight recorded population occurrences, but the remaining two species had more than 40 population occurrence records each (Table 2.1).

Potential Pollinator Distributions Shapefiles for the distributions of the three potential pollinators were created in ArcMap 9.3 from occurrence records obtained from the Global Biodiversity Information Facility (http://www.gbif.org/, accessed 2009-06-10; Table 2.1). These occurrence records were given in geographic coordinates and projected onto a map of state boundaries to visualize the current distributions of these species (Figure 2.2).

Table 2.1. Number of occurrence records used for each taxon in modeling. Taxon Name Total Number of Occurrence Records Used Conradina glabra 8 Physostegia godfreyi 46 Stachydeoma graveolens 54

Xylocopa virginica 15 Anthophora abrupta 9 Megachile spp. 367

Figure 2.1. Known distributions and occurrences of Conradina glabra, Physostegia godfreyi, and Stachydeoma graveolens

Figure 2.2. Occurrence records for the three potential pollinator taxa selected. Megachile species (red) are well-represented in the extent of interest. All three taxa have widespread distributions compared to those of the mint species.

Modeling

The Maxent software has a graphical user interface that is relatively easy to use. To run Maxent, the user needs to supply a file containing presence localities, a directory where the desired environmental variables are stored, and an output directory. For my research I retained most of the default settings. I changed the display options for the predicted suitability maps by selecting the options to (1) create response curves, (2) make pictures of the predictions, and (3) perform jackknifes to measure the contribution of each variable to the model. Maxent offers different options for the model output formats, and I used the default logistic output. The map gives the probability of a species‘ presence from 0–1. The output map is colored with warmer colors (red and orange) representing high probabilities and cooler colors (green and blue) representing lower probabilities. The software can also be directed to split an existing dataset into training and testing data under model settings in the setting ―random test percentage‖, although the default setting does not split the data. For my models, I set 25% of the occurrence records for the mint and pollinator species as testing data. This reduces the number of records available for model training, but the test data are used to run statistical analyses that can be useful in evaluating model performance. Maxent‘s ―regularization multiplier‖ parameter affects how fitted the output distribution is to the training data. This parameter can be adjusted to help avoid model over-fitting. The default setting is 1.0, and lowering the value of this parameter can lead to model over-fitting by increasing model complexity (Phillips et al. 2006). Since this parameter affects the predicted output distribution spread (from very fitted to generalized), I created three different models for each of the mint species under current conditions and future conditions, respectively, setting each model to a different regularization parameter. One model was run at the default regularization parameter setting of 1.0, one at a regularization parameter setting of 3.0, and one of a regularization parameter setting of 5.0. The pollinators all had more widespread distributions than those of the mint species, with all taxa being distributed across multiple states (Figure 2.2). The default regularization parameter setting was deemed appropriate to modeling predicted suitable areas for these insects.

Data for Modeling All of the environmental data layers used were obtained from Worldclim v. 1.4 (Hijmans et al. 2004). The environmental variables I selected included altitude and 19 bioclimatic variables (Table 2.2) that are derived from current averages of temperature and precipitation 47

based on suggestions for modeling with Maxent (Phillips et al. 2006). I also based the selection of these variables on the many studies that have utilized these same bioclimatic variables for predictive modeling using Maxent. All of the bioclimatic variable and altitude grids were originally downloaded at a resolution of 30 arc-seconds (or ~1 km) in the ESRI grid format, which is specifically for use in ArcGIS and other ESRI products. I considered a resolution of 1 km as sufficient for modeling shifts in plant species‘ distributions based on other studies modeling plant distributions (Pearson et al. 2004; Hernandez et al. 2006; Ibanez et al. 2009). I used the altitude grid in ArcGIS 9.3 to create two additional grids for slope and aspect which were also included. I clipped all of the grids to a geographic extent that included 9 states (Florida, Georgia, South Carolina, North Carolina, Tennessee, Alabama, Mississippi, Louisiana, and Arkansas), and then converted them from grids to ASCII files in ArcGIS 9.3. These files are the required format for use in the Maxent software. I also obtained future environmental data layers from Worldclim v. 1.4. These were the same 19 bioclimatic variables created from current averages of temperature and precipitation and projected to future conditions based on a doubled CO2 scenario (Govindasamy et al. 2003). These layers were also downloaded at a resolution of 30 arc-seconds (~1 km) and clipped and converted for use in Maxent. The 19 future environmental layers have the same file names and definitions as given for the current layers (refer to Table 2.2). The future environmental variables were only available in a generic grid format, which were not compatible with ArcGIS software. I formatted them in the DIVA-GIS software (v. 5.4, Hijmans et al. 2001) an open- source geographic information system. To ensure uniformity in data layers for current and future conditions, I obtained the 19 bioclimatic variables and altitude generic grids from Worldclim v. 1.4 for the current conditions as well and clipped and converted these files to my extent in DIVA-GIS. The altitude grid from the current conditions was used in DIVA-GIS to create grid files for slope and aspect. These layers were also used with the future data under the assumption that elevation, slope, and aspect likely would not change noticeably between current and future conditions. I did not use the original current environmental variables that were obtained in ESRI format. For each of the plant and insect species distribution models created in Maxent, all 22 of the environmental layers were used.

Table 2.2. Environmental layers used in Maxent modeling. Abbreviation Description Source Abbreviation Description Source Annual mean precipitation of bio 1 temperature Worldclim bio 14 driest month Worldclim mean diurnal range (mean of monthly precipitation bio 2 max temp-min temp) Worldclim bio 15 seasonality Worldclim precipitation of bio 3 Isothermality Worldclim bio 16 wettest quarter Worldclim Temperature precipitation of bio 4 seasonality Worldclim bio 17 driest quarter Worldclim Max temperature of precipitation of bio 5 warmest month Worldclim bio 18 warmest quarter Worldclim Min temperature of precipitation of bio 6 coldest month Worldclim bio 19 coolest quarter Worldclim Temperature of bio 7 annual range Worldclim altitude Altitude (DEM) Worldclim Derived Mean temperature of from bio 8 wettest quarter Worldclim aspect aspect altitude Derived Mean temperature of from bio 9 driest quarter Worldclim slope slope altitude Mean temperature of bio 10 warmest quarter Worldclim Mean temperature of bio 11 coldest quarter Worldclim bio 12 Annual precipitation Worldclim precipitation of bio 13 wettest month Worldclim

*The future environmental variables are the same 19 bioclimatic variables projected to 2090 under the CCM3- 2×CO2 scenario. The altitude, aspect, and slope files were also included in the future environmental variables dataset.

RESULTS

Plant Modeling Conradina glabra. The receiver operating curves (ROC) for Conradina glabra, calculated for both the training and test data, are shown in Figure 2.3. This plot also gives the area under the ROC (the AUC). Sensitivity is the fraction of all presences that are correctly

identified as such, while the specificity is the fraction of all absences that are correctly identified as such. The light blue line shows what you would expect if your model performed no better than random. If the dark blue falls below this line, it indicates that the model performs worse than random in correctly predicting occurrences. In the case of Conradina glabra, the model appears to perform well, predicting all of the withheld (test) data correctly. For all three models run for Conradina glabra, the response curves were the same and had the same values as Figure 2.3.

Figure 2.3. Receiver operating curve (ROC) for Conradina glabra. Sensitivity is the fraction of all presences that are correctly identified as such, while the specificity is the fraction of all absences that are correctly identified as such.

Figure 2.4 shows the suitable area predicted for Conradina glabra under current environmental conditions with the default regularization parameter settings of 1.0. The suitable area includes the Florida Panhandle, where the Conradina glabra populations are currently found, encompassing all of the known occurrences. There are also a few smaller predicted 50

locations elsewhere in southeast Florida, Louisiana, and the edge of North Carolina, but they are very small and widely separated from the actual known occurrences. When the model with default regularization settings for Conradina glabra was projected to the future environmental conditions, the predicted area of suitability shrank (Figure 2.4), with the current known locations no longer considered to be suitable for the species. The only area with high predicted suitability for species occurrences in this scenario is a single small area in the western Panhandle. The current and future predicted suitable areas for Conradina glabra (Figures 2.5 and 2.6) with a regularization parameter of 3.0 show very similar results to the predicted areas at the default settings. The predicted area is slightly more diffuse for the current environmental conditions at the parameter of 3.0, but the future area remains limited to the small area in the western Panhandle. The current and future predicted suitable areas for Conradina glabra with a regularization parameter of 5.0 show similar predictions that are still more diffuse. The same suitable areas within the current distribution are predicted, with a larger area around the current known range, as well as slightly larger predicted areas in Louisiana, South Florida, and North Carolina. All three models under current environmental conditions predicted as suitable the area where the known presence records occurred. All three future models predicted a westward shift in suitable area as well as a reduction in the size of the area, with only small area in the western Panhandle showing suitability for Conradina glabra.

Figure 2.4. Predicted suitability for Conradina glabra modeled with Maxent under current environmental conditions at the default regularization setting of 1.0. The map indicates areas suitable for the species to be present, showing a probability of occurrence from 0-1. Brighter colors such as red and orange indicate the most suitable conditions. Violet dots indicate locations used for testing the model, and white points are the locations used for training the model. All following figures are described by the same coloration.

Figure 2.5. Predicted suitability for Conradina glabra modeled under future environmental conditions at the default regularization setting of 1.0.

Figure 2.6. Predicted suitability for Conradina glabra modeled under current environmental conditions at a regularization setting of 3.0.

Figure 2.7. Predicted suitability for Conradina glabra modeled under future environmental conditions at a regularization setting of 3.0.

Figure 2.8. Predicted suitability for Conradina glabra modeled under current environmental conditions at a regularization setting of 5.0.

Figure 2.9. Predicted suitability for Conradina glabra modeled under future environmental conditions at a regularization setting of 5.0.

Variable contributions to Conradina glabra

A jackknife test of the importance of variables was performed to evaluate the contribution of each variable to the predicted model construction. Each of the variables is excluded from a model in turn. Then, a model is created for each variable in isolation. Finally, a model is created using all of the variables together. Figure 2.10 shows the results of the jackknife test for the Conradina glabra training data run with the default regularization parameter of 1.0. The jackknife figures are shown in a bar chart that lists the gain of the modeling performance with the inclusion and exclusion of each variable with the model. The gain is closely related to deviance, which is a measure of the goodness of fit of a model. The environmental variable with the highest gain when used alone is bioclimatic variable 8 (bio 8, mean temperature of wettest quarter), which indicates that it has the most useful information to the model. The environmental variable that decreases model performance the most when it is left out is bioclimatic variable 15 (bio15, precipitation seasonality), which means it appears to have the most information that is not present in the other variables. The same results are observed for the jackknife on the test data. The jackknife results for environmental variable contributions to the model construction for Conradina glabra were the same for all three of the models run with different regularization parameter values.

Figure 2.10. Results of the jackknife performed with training data for Conradina glabra modeled with the default regularization parameter of 1.0, showing the individual contributions of environmental variables to the model when included in isolation and excluded.

Stachydeoma graveolens Figure 2.11 shows the ROC curve produced for Stachydeoma graveolens with the current environmental variables at the default regularization setting. The model fit is the same as that produced for Conradina glabra. The model fit was the same for all three models run for Stachydeoma graveolens with different regularization parameters.

Figure 2.11. Receiver operating curve (ROC) for Stachydeoma graveolens modeled with the default regularization setting of 1.0.

Model outputs for Stachydeoma graveolens The predicted suitable area for the current conditions using the default regularization parameter (Figure 2.12) encompasses the current known species occurrences, and also extends east and west of the occurrences in the Florida Panhandle with a few smaller, less suitable areas predicted going south along the coast. A small area in Louisiana also shows an area of potential suitability for this species, although areas with high suitability are restricted to the current known distribution. The model created with the future environmental conditions for Stachydeoma graveolens under the default regularization parameter predicts a large increase in the areas of suitability (Figure 2.13). This predicted area is much larger in the Panhandle, and there are new areas within the Carolinas. The current and future predicted areas of suitability under the regularization parameters of 3.0 and 5.0 for Stachydeoma graveolens (Figures 2.14–2.17) are very similar in appearance to the respective areas with the default setting.

Figure 2.12. Predicted suitability for Stachydeoma graveolens modeled under current environmental conditions at the default regularization setting of 1.0.

Figure 2.13. Predicted suitability for Stachydeoma graveolens modeled under future environmental conditions at the default regularization setting of 1.0.

Figure 2.14. Predicted suitability for Stachydeoma graveolens modeled under current environmental conditions at a regularization parameter of 3.0.

Figure 2.15. Predicted suitability for Stachydeoma graveolens modeled under future environmental conditions at a regularization parameter of 3.0.

Figure 2.16. Predicted suitability for Stachydeoma graveolens modeled under current environmental conditions at a regularization parameter of 5.0.

Figure 2.17. Predicted suitability for Stachydeoma graveolens modeled under future environmental conditions at a regularization parameter of 5.0.

Variable contributions to Stachydeoma graveolens

Figure 2.18 shows the results of the jackknife test for Stachydeoma graveolens run with the default regularization parameter of 1.0. This figure shows the results for the training data. The environmental variable with the highest gain when used alone is bioclimatic variable 13 (bio 13, precipitation of wettest month), which indicates that it has the most useful information by itself. The environmental variable that decreases model performance the most when it is left out is bioclimatic variable 19 (bio19, precipitation of coldest quarter), which means it appears to have the most information that is not present in the other variables. The same results are observed for the jackknife results with the test data. The jackknife results were the same for all three models run for Stachydeoma graveolens with the different regularization parameters, with the exception of the model run with a regularization parameter of 5.0. In this model, bio13 contained the most useful information by itself for the model and it decreased the gain the most when it was left out.

Figure 2.18. Results of the jackknife performed with training data for Stachydeoma graveolens modeled with the default regularization setting of 1.0.

Physostegia godfreyi Figure 2.19 shows the ROC curve produced for Physostegia godfreyi with the current environmental variables at the default regularization setting. The models performed just as well as those produced for the previous two species. The ROC-AUC curve was the same for all three models run for Physostegia godfreyi with the different regularization parameters.

Figure 2.19. Receiver operating curve (ROC) for Physostegia godfreyi.

Model outputs for Physostegia godfreyi Figure 2.20 shows the area of suitability produced for Physostegia godfreyi at the default regularization parameter under current environmental conditions. The current areas of suitability are very similar to those predicted for Stachydeoma graveolens; they encompass all of the known occurrences and extend slightly west and east of the current range. There is also an area of predicted suitability in Louisiana. The future predicted areas of suitability with the default regularization parameter shows an increase in the predicted distribution of Physostegia godfreyi that extends from the Panhandle eastward up along the entire Carolina coast (Figure 2.21), with a gap of potentially unsuitable area separating the predicted area in north central Florida. The current and future predicted areas of suitability for Physostegia godfreyi at the regularization parameter of 3.0 (Figures 2.22 and 2.23) show larger areas than those predicted from the default setting. The current predicted distribution shows an increase in suitable range from the default parameter model, increasing to a region in Louisiana and a few smaller locations within central- south Florida. 63

Figure 2.20. Predicted suitability for Physostegia godfreyi modeled under current environmental conditions at the default regularization setting of 1.0.

Figure 2.21. Predicted suitability for Physostegia godfreyi modeled under future environmental conditions at the default regularization setting of 1.0.

Figure 2.22. Predicted suitability for Physostegia godfreyi modeled under current environmental conditions at a regularization setting of 3.0.

Figure 2.23. Predicted suitability for Physostegia godfreyi modeled under future environmental conditions at a regularization setting of 3.0.

Figure 2.24. Predicted suitability for Physostegia godfreyi modeled under current environmental conditions at a regularization setting of 5.0.

Figure 2.25. Predicted suitability for Physostegia godfreyi modeled under future environmental conditions at a regularization setting of 5.0.

Variable contributions to Physostegia godfreyi

Figure 2.26 shows the results of the jackknife test for Physostegia godfreyi run with the default regularization parameter of 1.0. This figure shows the results for the training data. The environmental variable with the highest gain when used alone is bioclimatic variable 16 (bio 16, precipitation of wettest quarter). The environmental variable that decreases the gain the most when it is left out is bioclimatic variable 19 (bio19, precipitation of coldest quarter). The same results are observed for the jackknife on the test data. The jackknife results are the same for all three models run for Physostegia godfreyi with different regularization parameters, with the exception of the model run with the default regularization parameter. In this model, bioclimatic variable 16 was the most informative on its own. For the remaining two models, bioclimatic variable 13 (bio 13, precipitation of wettest month) was the most informative variable by itself to the model. The environmental variable that decreased the gain the most when left out was bioclimatic variable 19 for all three models.

Figure 2.26. Results of the jackknife performed with training data for Physostegia godfreyi modeled with the default regularization parameter of 1.0.

Insect Modeling Xylocopa virginica. Figure 2.27 shows the ROC curve produced for Xylocopa virginica with the current environmental variables and the default regularization. The potential pollinators were all modeled using the default regularization setting (1.0) only. The ROC curve for the training and test data from Xylocopa virginica are quite different; the test data is poorly predicted by the model.

Figure 2.27. Receiver operating characteristic (ROC) curve for Xylocopa virginica modeled with the default regularization parameter of 1.0.

Most of the occurrence records for Xylocopa virginica fall within the predicted area of suitability (Figure 2.28), which extends through north Florida and the Florida Panhandle, although in no places is it predicted to be highly suitable to the species‘ presence. Figure 2.29 shows the predicted future areas of suitability are expanded in the Florida Panhandle and the coast of the Carolinas.

Figure 2.28. Predicted suitability for Xylocopa virginica modeled under current environmental conditions at the default regularization setting of 1.0.

Figure 2.29. Predicted suitability for Xylocopa virginica modeled under future environmental conditions at the default regularization setting of 1.0.

Figure 2.30 shows the results of the jackknife test for the models for Xylocopa virginica. This figure shows the results for the training data. The environmental variable with the highest gain when used alone is bioclimatic variable 8 (bio 8, mean temperature of wettest quarter). The environmental variable that decreases the gain the most when it is left out is also bio 8. However, unlike the previous models run for the mint species, different results were obtained for test data, with most of the variables including bio 8 indicating negative gains.

Figure 2.30. Results of the jackknife performed with training data for Xylocopa virginica modeled with the default regularization parameter of 1.0.

Anthophora abrupta. Figure 2.31 shows the ROC curve produced for Anthophora abrupta with the current environmental variables at the default regularization setting. Anthophora abrupta has similar AUC results for both test and training data that indicated that the model performed well.

Figure 2.31. Receiver operating characteristic (ROC) curve for Anthophora abrupta modeled with the default regularization setting of 1.0.

Figure 2.32 shows the predicted current areas of suitability for Anthophora abrupta under the default regularization setting of 1.0. The suitability of areas with the current climate generally increases from the northwest to the southeast (Figure 2.32). The future predicted areas of suitability for Anthophora abrupta (Figure 2.33) shrink but the areas that show the greatest suitability remain in the Florida Peninsula.

Figure 2.32. Predicted suitability for Anthophora abrupta modeled under current environmental conditions at the default regularization setting of 1.0.

Figure 2.33. Predicted suitability for Anthophora abrupta modeled under future environmental conditions at the default regularization setting of 1.0.

Figure 2.34 shows the results of the jackknife test for Anthophora abrupta run with the default regularization parameter of 1.0. This figure shows the results for the training data. The environmental variable with the highest gain when used alone is bioclimatic variable 18 (bio 18, precipitation of warmest quarter). The environmental variable that decreases the model performance the most when it is left out is also bio 18.

Figure 2.34. Results of the jackknife performed with training data for Anthophora abrupta modeled with the default regularization setting of 1.0.

Megachile spp. Figure 2.35 shows the ROC curve produced for Megachile with the current environmental variables and the default regularization. Megachile has similar AUC

results for both test and training data, with both above 90%. This indicates that the model performed well for this genus.

Figure 2.35. Receiver operating characteristic (ROC) curve for Megachile spp. modeled with the default regularization setting of 1.0.

Figure 2.36 represents the predicted current areas of suitability for Megachile under the default regularization parameter 1.0. The predicted areas are for the genus; multiple species are represented in the data points (see Table 2.1). The areas of suitability are smaller for the genus than they are for the other two pollinators (Figure 2.36). The future predicted areas of suitability for Megachile spp. (Figure 2.37) show a general shift and expansion towards central Florida and the east coast and away from the Panhandle, with expansions in the Carolinas as well.

Figure 2.36. Predicted suitability for Megachile species modeled under current environmental conditions at the default regularization setting of 1.0.

Figure 2.37. Predicted suitability for Megachile species modeled under future environmental conditions at the default regularization setting of 1.0.

Figure 2.38 shows the results of the jackknife test for Megachile run with the default regularization parameter of 1.0. The environmental variable with the highest gain when used alone is bioclimatic variable 1 (bio 1, precipitation annual mean temperature). The environmental variable that decreases model performance the most when it is left out is bioclimatic variable 18.

Figure 2.38. Results of the jackknife performed with training data for Megachile modeled with the default regularization setting of 1.0.

DISCUSSION

It has been recently suggested that we are undergoing a global pollination crisis (Steffan- Dewenter et al. 2005; Biesmeijer et al. 2006). Although we know that the interactions between 77

plants and pollinators are of conservation concern, recent reviews show that research on the effects of climate change on these mutualistic interactions is still limited (Walther et al. 2002; Visser & Both 2005). Disruption of these interactions is a growing concern, particularly their effects on rare plants (Schemske et al. 1994). The interactions between plants and their pollinators can be disrupted through at least two ways: by temporal (phenological) and spatial (distributional) disjunctions (Visser and Both 2005; Heglend et al. 2009). Most of the research on these interactions focuses on potential temporal shifts under climate change conditions, with very little work on predicting separations in species distributions. Studies such as Memmott et al. (2007) have observed that predicted changes in phenology can decrease the resources available to both plants and pollinators, even for generalist relationships. To my knowledge, there are no other studies that demonstrate future separations in plant and pollinator interactions. However, one study (Devoto et al. 2007) created simulations of future climate distribution shifts of plants and their pollinators along a rainfall gradient and determined that few species of either plants or pollinators would be in danger of extirpation. Their results indicate that some pollination systems seem to be resistant to climate change. Separations in plants and their pollinators are still rarely explored and the consequences of climate change on these systems still remain relatively unknown. A need for more data to inform the conservation of important interactions in rare species provided the motivation behind this research. I attempted to create a preliminary study that accomplished the twin goals of providing more ecological information for species of conservation concern and creating predicted distribution models of suitable areas under climate change to determine if separations of important interactions may occur. Here I discuss the results of species distribution modeling for rare plants and their pollinators in the context of current research and conservation planning.

Plant species distribution modeling Three different models were run for each of the three mint species. The models were identical except for changes made to the regularization parameter, which affects how closely the model fits the data. Within each species, the different models for both current and future predicted distributions were similar. As expected, altering the regularization parameter made the predictions differ only in how closely the model was fitted to the data given. The higher values 78

for the regularization parameter resulted in slightly larger predicted areas of suitability, but all models for each species had close agreement in the predicted areas of suitability under current and future conditions. Current predicted suitable areas for all three mint species closely fit those of the known occurrences, with a few additional areas predicted to be suitable as well for each species. The future predicted areas of suitability for all three mint species indicated potential shifts in species distributions would occur under the doubled CO2 climate scenario. Physostegia godfreyi and Stachydeoma graveolens are predicted to have larger areas of suitable habitat than those at present, and so potential range expansions seem likely under this scenario (Figures 2.14, 2.16, 2.18, 2.22, 2.24, and 2.26). Conradina glabra is predicted to have a much smaller area predicted suitable under future conditions (Figures 2.6, 2.8, 2.10). The current range this species occupies is predicted to become unsuitable. Of additional concern is that the future predicted area suitable for this species is smaller than the current distribution and is not located on or near conservation lands. These models were evaluated with test data drawn from the occurrence records for each species. All three mint species appeared to be modeled very well, with AUC results above 90% for each. Maxent is known to produce models that perform well, particularly for rare species like these mints that have small numbers of occurrence records and narrow geographical and environmental ranges (Phillips et al. 2006; Phillips and Dudik 2008). It should be noted that in some cases sampling bias, such as only arises when occurrences that are convenient to record are sampled, can lead to similar results that may not accurately reflect the species‘ potential distribution or the model‘s performance (Phillips et al. 2008). It was not surprising that Maxent would perform well in modeling the current distributions of these species. This method was selected because it is a flexible method that works well with large to very small numbers of occurrence records (Phillips et al. 2006; Pearson et al. 2007). A few studies suggest that other approaches might outperform Maxent for small numbers of occurrences (Hernandez et al. 2006), but in general this method is considered to work well for modeling rare species (Elith et al. 2006; Phillips et al. 2006; Pearson et al. 2007). I assume that, if the modeling method has high accuracy in predicting the current distribution, it will perform as well for future predictions by extension. However, it is not always possible to know how a species will respond to changing conditions without detailed historical knowledge or experimentation. It is also worthwhile to

look at the sensitivity of results against a wider set of probable future climate conditions than done here.

Insect species distribution modeling Three of the potential pollinators from the results in Chapter 1 were selected based on the availability of datasets: Xylocopa virginica, Anthophora abrupta, and Megachile spp. These insects were also selected because they were observed to visit all three mint species at high frequencies. A single model was created for each of the pollinator taxa, using the default regularization parameter. All three of the potential pollinators have observed and predicted current areas of suitability that are close to those of the mint species and appear to be overlapping. Unlike the predicted mint distributions, the suitability of cells to the pollinator taxa was never as sharply delineated. One of the problems with creating equivalent predictive distribution models for plant species and their pollinators is that the distributions of most pollinator taxa are poorly characterized and sampled. This is a problem present with many different groups of organisms, but it is especially prevalent for invertebrate taxa (Wilson 1992). Xylocopa virginica in particular was poorly modeled by Maxent, with an AUC for the test data of 0.341. This could be due to the fact that the pollinator occurrence records were much more spread out over geographical and environmental space and made response modeling difficult. In the case of Xylocopa virginica, there were just 15 records available in the extent of the modeling area. As with the mint species, all three pollinator taxa are predicted to undergo some shifts in suitable predicted area under future conditions. Anthophora abrupta is predicted to experience a decrease in the predicted suitable area down into South Florida (2.34). The results are interesting in that they also predict range expansions for two of the three taxa, just as the results for the mint species did. This leads to the question of why some of these organisms are predicted to succeed under this climate change scenario, and why some are not.

Improvements and Future Directions Not all of the environmental variables selected for use in the modeling proved useful, and future modeling work could perhaps exclude them. For example, 14 of the variables used to model Anthophora abrupta (Figure 2.36) did not add to the training gain when used alone or resulted in a decreased training gain (decreased model performance) when removed. At an 80

extreme, only four of the variables used to model Xylocopa virginica (Figure 2.32) resulted in a positive training gain when used alone and only one of them resulted in a decreased training gain when removed. However, a new assessment of the value of these variables would be warranted if significantly more occurrences were included in the new analysis. I suspect that the poor performance of the models for the pollinators stemmed from the small number of occurrences relative to their widespread distributions. The generality of the plant-pollinator relationships in these species are likely to make continuation in pollen delivery for these species more robust with climate change (Devoto et al. 2007; Heglend et al. 2009) than in cases (e.g., orchids) where pollination is more specialized. Although pollinator limitation seems unlikely under to occur in these species, Conradina glabra may be in jeopardy if its future area of suitability is smaller and disjunct with today‘s area. What can be done? Methods such as transplantation of individuals and even populations to new locations could be feasible. It has been performed on Conradina glabra with some success already (Isom 2002). Conservation of a species is more likely to be successful within the context of a system that is managed to maintain natural functions and processes (Saunders et al. 1991; Gordon 1996), and this kind of research aims to plan ahead for the conservation of interactions instead of on a species by species basis.

CONCLUSION

It is generally known that plant and pollinator species are at risk of local and global extinction (Cane and Tepedino 2001; Memmott et al. 2007) in the face of climate change, invasive species introduction, and land conversion. In this study, I documented potential pollinators of three Panhandle endemic mint species and their importance to seed set in those taxa. Then, I used distribution data to model areas of suitability today and in the future for the mints and a selection of their potential pollinators. I discovered that multiple potential pollinators visit all three mint species, although one of the species (Stachydeoma graveolens) was visited by fewer taxa than the other two species during periods of observation. All of the potential pollinators observed are from species that are more widespread in their distributions than the mint species. The mints do not appear to be involved in specialist plant-pollinator relationships. Pollination by a vector (insect or human) seems to be necessary to ensure that maximum seed set occurs for Conradina glabra and Physostegia godfreyi. It does not appear that any of these three species are in danger of pollinator limitation at present given the diversity of pollinator taxa that co-occur. I created species distribution models with Maxent for the mint species and three of their observed pollinators with their current distributions and determined suitability for the species in today‘s climate and the likely climate in 90 years. I also examined the sensitivity of the results to changes in the regularization parameter that determines the model‘s fit to the training data. The plant distributions were modeled more accurately than those of the potential pollinators. The models predicted changes in the suitability of areas for all six species, with striking reductions in the area of suitability for Conradina glabra. Based on the modeling results, it does not appear that the plants will be endangered by potential separations from their pollinators. I suggest that new modeling efforts for these species examine the sensitivity of the results to the removal of apparently uninformative variables from the models. I also suggest checking the sensitivity of the results to the modeling method used and the future climate that is assumed. This research represents a novel approach to understanding the potential for future disruptions between plants and their pollinators.

REFERENCES

Araújo, M. and Guisan, A. 2006. Five (or so) challenges for species distribution modeling. Journal of Biogeography 33: 1677–1688.

Austin, M.P. 2002. Spatial prediction of species distribution: an interface between ecological theory and statistical modelling. Ecological Modelling 157(2-3): 101–118.

Bakkenes, M., Alkemade, J.R.M., Ihle, F., Leemans, R., and Latour, J.B. 2002. Assessing effects of forecasted climate change on the diversity and distribution of European higher plants for 2050. Global Change Biology 8: 390–407.

Balduf, W.V. 1962. Life of the carpenter bee, Xylocopa virginica (Linn.) (Xylocopidae, Hymenoptera). Annals of the Entomological Society of America 55: 263–271.

Barthell, J.F. and Baird, T.A. 2004. Size variation and Aggression among Male Xylocopa virginica (L.) (Hymenoptera: Apidae) at a Nesting Site in Central Oklahoma. Journal of the Kansas Entomological Society 77: 10–20.

Bartholomew, C.S., Prowell, D., and Griswold, T. 2006. An Annotated Checklist of Bees (Hymenoptera: Apoidea) in Longleaf Pine Savannas of Southern Louisiana and Mississippi. Journal of the Kansas Entomological Society 79(2): 184–198.

Biesmeijer, J.C., Roberts, S.P.M., Reemer, M., Ohlemuller, R., Edwards, M., Peeters, T., Schaffers, A.P., Potts, S.G., Kleukers, R., Thomas, C.D., Settele, J., and Kunin, W.E. 2006. Parallel declines in pollinators and insect-pollinated plants in Britain and the Netherlands. Science 313: 351–354.

Blaustein, R.J. 2008. Biodiversity hotspot: The Florida Panhandle. Bioscience 58(9): 784– 790.

Bowlin, W.R., Tepedino,V. J., Griswold, T. L. 1993. The reproductive biology of Eriogonum pelinophilum (Polygonaceae). Pp. 296–302 in R. Sivinski and K. Lightfoot [eds.], Southwestern rare and endangered plants. Miscellaneous publication No. 2. New Mexico Forestry and Resources Conservation Division, Santa Fe, New Mexico, USA.

Boyce, M. S., Vernier, P.R., Nielson, S.E., and Schmiegelow, F.K.A. 2002. Evaluating resource selection functions. Ecological Modelling 157: 281–300.

Brewer, J. S. 1999. Effects of competition, litter, and disturbance on an annual carnivorous plant (Utricularia juncea). Plant Ecology 140: 159–165.

Busby, J. R. 1991. BIOCLIM: a bioclimate analysis and prediction system. Pp. 64–68 in Margules, C. R. and Austin, M. P. (eds), Nature conservation: cost effective biological surveys and data analysis. CSIRO.

Cane, J.H. and Tepedino, V.J. 2001. Causes and Extent of Declines among Native North American Invertebrate Pollinators: Detection, Evidence, and Consequences. Ecology and Society 5: 1. [online] URL: http://www.consecol.org/vol5/iss1/art1. CSA.

Cantino, P.D. 1982. A monograph of the genus Physostegia (Labiateae). Contributions of the. Gray Herbarium 211: 1–105.

Carlsen, T. M., Espeland, E. K., and Pavlik, B. M. 2002. Reproductive ecology and the persistence of an endangered plant. Biodiversity and Conservation 11: 1247–1268.

Carpenter, G., Gillison, A. N. and Winter, J. 1993. DOMAIN: a flexible modelling procedure for mapping potential distributions of plants and animals. Biodiversity Conservation 2: 667–680.

Castro, S., Silveira, P., and Navarro, L. 2008. Consequences of nectar robbing for the fitness of a threatened plant species. Plant Ecology 199: 201–208.

Cawsey, E.M., Austin, M.P., and Baker, B.L. 2002. Regional vegetation mapping in Australia: a case study in the practical use of statistical modeling. Biodiversity and Conservation 11(12): 2239–2274.

Christman, S.P. and Judd, W.S. 1990. Notes on plants endemic to the Florida scrub. Florida Scientist 53: 52–73.

Clewell, A.F. 1985. Guide to Vascular Plants of the Florida Panhandle. Tallahassee, Florida: Florida State University Press. 605 pp.

DeMauro, M.M. 1993. Relationship of breeding system to rarity in the lakeside daisy (Hymenoxys acaulis var. glabra). Conservation Biology 7: 542–550.

Devoto, M., Zimmerman, M., and Medan, D. 2007. Robustness of plant-flower visitor webs to simulated climate change. Ecologia Australia 17: 37–50.

Dewenter, S. and Tscharntke, T. 1999. Effects of habitat isolation on pollinator communities and seed set. Oecologica 121(3): 432–440.

Dirnböck, T. , Dullinger, S., and Grabherr, G. 2003. A regional impact assessment of climate and land use change on alpine vegetation. Journal of Biogeography 30: 401–418.

Edwards, C.E., Soltis, D.E, and Soltis, P.S. 2003. Molecular phylogeny and conservation genetics of Conradina (Lamiaceae). Gainesville, Florida: Interim report for Florida Division of Forestry.

Edwards, C.E., Soltis, D.E., and Soltis, P.S. 2006. Molecular phylogeny of Conradina and other scrub mints (Lamiaceae) from the Southeastern USA: evidence for hybridization in Pleistocene refugia? Systematic Botany 31(1): 193–207.

Edwards, C.E., Soltis, D.E., and Soltis, P.S. 2008. Isolation, characterization and cross-species amplifications of microsatellite loci from Conradina (Lamiaceae). Molecular Ecology Resources 8(2): 363–366. Elith, J., Graham, C.H., Anderson, R.P., Dudık, M., Ferrier, S., Guisan, A., Hijmans, R.J., Huettmann, F., Leathwick, J.R., Lehmann, A., Li, J., Lohmann, L.G., Loiselle, B.A., Manion, G., Moritz, C., Nakamura, M., Nakazawa, Y., Overton, J. McC., Peterson, A.T., Phillips, S.J., Richardson, K., Scachetti-Pereira, R., Schapire, R.E., Soberon, J., Williams, S., Wisz, M.S., and Zimmermann, N.E. 2006. Novel methods improve prediction of species‘ distributions from occurrence data. Ecography 29: 129–151.

Ellstrand, N.C., and Elam, D.R. 1993. Population genetic consequences of small population size: implications for plant conservation. Annual Review of Ecology and Systematics 24: 217–242.

Engler, R., Guisan, A., and Rechsteiner, L. 2004. An improved approach for predicting the distribution of rare and endangered species from occurrence and pseudo-absence data. Journal of Applied Ecology 41: 263–274.

Estill, J.C., and Cruzan, M.B. 2001. Phytogeography of rare plant species endemic to the Southeastern United States. Castanea 66(1-2): 3–23.

Evans, A.V. and Tufts, C. 2007. National Wildlife Federation Field Guide to Insects and Spiders & Related Species of North America. Sterling Publishing Company, New York, New York. 496 pp.

Evans, M.E.K., Menges, E.S., and Gordon, D.R. 2003. Reproductive biology of three sympatric endangered plants endemic to Florida scrub. Biological Conservation 111: 235–246.

Evans, M.E.K., Menges, E.S., and Gordon, D.R. 2004. Mating systems and limits to seed production in two Dicerandra mints endemic to Florida scrub. Biodiversity and Conservation 13: 1819–1832.

Evans, M.E.K., Holsinger, K.E., and Menges, E.S. 2007. Modeling the effect of fire on the demography of Dicerandra frutescens ssp. frutescens (Lamiaceae), an endangered plant endemic to Florida scrub. Population Ecology 50(1): 53–62.

Ferrier, S., Watson, G., Pierce, J., and Drielsma, M. 2002. Extended statistical approaches to modelling spatial pattern in biodiversity: the north-east New South Wales experience. I. Species-level modelling. Biodiversity Conservation 11: 2275–2307.

Fielding, A.H. and Bell, J.F. 1997. A review of methods for the assessment of prediction errors in conservation presence/absence models. Environmental Conservation 24: 38–49.

FNAI. 2000. Field Guide to the Rare Plants and Animals of Florida online. Florida Natural Areas Inventory. Accessed 10/15/08.

Funk, V.A. and K.S. Richardson. 2002. Systematic data in biodiversity studies: use it or lose it. Systematic Biology 51(2): 303–316.

GBIF Data Portal (www.gbif.net). 2009-06-10. Ohio State University, 15 records; USDA-ARS, 9 records, USDA-ARS, 367 records.

Gerling, D. and Hermann, H.R. 1976. Biology and Mating Behavior of Xylocopa virginica L. (Hymenoptera, Anthrophoridae). Behavioral Ecology and Sociobiology 3: 99–111.

Godt, M.J.W., Walker, J. and Hamrick, J.L. 2004. Allozyme diversity in Macbridea alba (Lamiaceae), an endemic Florida mint. Journal of Heredity 95(3): 244–249.

Gordon, D.R. 1996. Experimental translocation of the endangered shrub Apalachicola rosemary (Conradina glabra) to the Apalachicola Bluffs and Ravines Preserve, Florida. Biological Conservation 77: 19-26.

Govindasamy, B., Duffy, P. B., and Coquard, J. 2003. High-resolution simulations of global climate, part 2: effects of increased greenhouse cases. Climate Dynamics 21: 391–404.

Graham, C. H., Ferrier, S., Huettman, F., Moritz, C., and Peterson, A.T. 2004. New developments in museum-based informatics and applications in biodiversity analysis. Trends in Ecology and Evolution 19: 497–503.

Gray, T.C. 1965. A monograph of the genus Conradina A. Gray (Labiatae). Ph.D. thesis: Vanderbilt University. Nashville, TN.

Grinnell, J. 1924. Geography and evolution. Ecology 5: 225–229.

Guisan, A. and Theurillat, J.P. 2000. Equilibrium modeling of alpine plant distribution and climate change: how far can we go? Phytocoenologia 30(3-4): 353–384.

Guisan, A. and Zimmermann, N.E. 2000. Predictive habitat distribution models in ecology. Ecological Modelling 135(2-3): 147–186.

Guisan, A. and Thuiller, W. 2005. Predicting species distribution: offering more than simple habitat models. Ecology Letters 8: 993–1009.

Hamrick, J. L., Godt, M. J.W., Murawski, D. A., and Loveless, M. D. 1991. Correlations between species traits and allozyme diversity: implications for conservation biology. In:

D. A. Falk and K. E. Holsinger [eds.], Genetics and Conservation of Rare Plants. Oxford University Press, New York, New York, USA. 75-86.

Heglend, S.J., Nielson, A., Lazaro, A., Bjerknes, A.L., and Totland, O. 2009. How does climate warming affect plant-pollinator interactions? Ecology Letters 12: 184–195.

Heikkinen, R. K., Luoto, M., Araujo, M.B., Virkkala, R., Thuiller, W., and Sykes, M.T. 2006. Methods and uncertainties in bioclimatic envelope modelling under climate change. Progress in Physical Geography 6: 751–777.

Hernandez, P.A., Graham, C.H., Master, L.L., and Albert, D.L. 2006. The effect of sample size and species characteristics on performance of different species distribution modeling methods. Ecography 29: 773–785.

Hernandez, P.A., Franke, I., Herzog, S.K., Pacheco, V., Paniagua, L., Quintana, H.L., Soto, A., Swenson, J.J., Tovar, C., Valqui, T.H., Vargas, J., and Young, B.E. 2008. Predicting species distributions in poorly-studied landscapes. Biodiversity Conservation 17: 1353–1366.

Hessl, A. and Spackman, S. 1995. Effects of Fire on Threatened and Endangered Plants: An Annotated Bibliography. Information and Technology Report 2. U.S. Department of the Interior, Washington, DC. 55 pp.

Hijmans R. J., Cruz, M., Rojas, E., and Guarino, L. 2001. DIVA-GIS, version 5.4. A geographic information system for the management and analysis of genetic resources data. Manual. International Potato Center and International Plant Genetic Resources Institute, Lima, Peru. URL: http://www.diva-gis.org/.

Hijmans, R.J., Cameron, S., and Parra, J. 2004. WorldClim, version 1.2. A square kilometer resolution database of global terrestrial surface climate. http://www.worldclim.org/ accessed 10/20/08.

Hijmans, R.J. and Graham, C.H. 2006. The ability of climate envelope models to predict the effect of climate change on species distributions. Global Change Biology 12: 2272– 2281.

Hirzel, A., and Guisan, A. 2002. Which is the optimal sampling strategy for habitat suitability modeling? Ecological Modelling 157(2-3): 331–341.

Huettmann, F. 2005. Databases and science-based management in the context of wildlife and habitat: towards a certified ISO standard for objective decision-making for the global community by using the internet. Journal of Wildlife Management 69: 466–472.

Hutchinson, G.E. 1957. Concluding remarks. Cold Spring Harbor Symposia on Quantitative Biology 22 (2): 415–427.

Ibanez, I., Silander, J.A. Jr., Wilson, A.M., LaFleur, N., Tanaka, and N., Tsuyama, I. 2009. Multivariate forecasts of potential distributions of invasive plant species. Ecological Applications 19(2): 359–375.

Irwin, R.E., Brody, A.K., and Waster, N.M. 2001. The impact of floral larceny on individuals, populations, and communities. Oecologica 129: 161–168.

Isom, P.S. 2000. Pollen transfer between three translocated populations of the endangered mint, Apalachicola rosemary (Conradina glabra), at the Apalachicola Bluffs and Ravines Preserve. In: Gordon, D.R.; Slapcinsky, J.L. (eds), Annual Research Report: A Compilation of Research Conducted or Supported by The Nature Conservancy in Florida. Florida Science and Stewardship Programs and The Nature Conservancy.

Jaynes, E.T. 1957. Information theory and statistical mechanics. Physical Review 106: 620– 630.

Jordano P., Bascompte, J., and Olesen, J.M. 2003. Invariant properties in coevolutionary networks of plant–animal interactions. Ecology Letters 6: 69–81

Kaye, T.N. 1999. From flowering to dispersal: reproductive ecology of an endemic plant, Astragalus australis var. olympicus (Fabaceae). American Journal of Botany 86: 1248– 1256.

Kearns, C.K., and Inouye, D.W. 1993. Techniques for Pollination Biologists. University Press of Colorado, Niwot, CO. 583 pp.

Kearns, C.K., Inouye, D.W., and Waser, N.M. 1998. Endangered mutualisms: the conservation of plant-pollinator interactions. Annual Review of Ecological Systems 29: 83–112.

Keating, K. A. and Cherry, S. 2004. Use and interpretation of logistic regression in habitat selection studies. Journal of Wildlife Management 68: 774–789.

Kruckeberg, A.R. and Rabinowitz, D. 1985. Biological aspects of endemism in higher plants. Annual Review of Ecology and Systematics 16: 447–479.

Lesica, P. 1999. Effects of fire on the demography of the endangered, geophytic herb Silene spaldingii (Caryophyllaceae). American Journal of Botany 86(7): 996–1002.

Maloof, J.E. and Inouye, D.W. 2000. Are nectar robbers cheaters or mutualists? Ecology 81: 2651–2661.

Memmott, J. 1999. The structure of a plant-pollinator food web. Ecology Letters 2: 276–280.

Memmott, J. and Waser, N.M. 2002. Integration of alien plants into a native flower-pollinator visitation web. Proceedings of the Royal Society of London Series B 269: 2395–2399.

Memmott, J., Waser, N.M., and Price, M.V. 2004. Tolerance of pollination networks to species extinctions. Proceedings of the Royal Society of London Series B 271: 2605–2611.

Memmott, J., Craze, P.G., Waser, N.M., and Price, M.V. 2007. Global warming and the disruption of plant-pollinator interactions. Ecology Letters 10: 710–717.

Mustajarvi, K., Siikamaki, P., Rytkonen, S., and Lammi, A. 2001. Consequences of plant population size and density for plant-pollinator interactions and plant performance. Journal of Ecology 89: 80–87.

Myers, R.L. and Ewell, J.J. (eds). 1990. Ecosystems of Florida. University of Central Florida Press, Orlando, Florida. pp 3-100.

Myers, N., Mittermeier, R.A., Mittermeier, C.G., de Fonseca, G.A.B., and Kent, J. 2000. Biodiversity and Hotspots for conservation priorities. Nature 403: 853–858.

Norden, B.B. 1984. Nesting Biology of Anthophora abrupta. Journal of the Kansas Entomological Society 57(2): 243–262.

Norden, B.B. 1985. Male bees sport black mustaches for picking up parsnip perfume. Proceedings of the Entomological Society of Washington 87(2): 317–322.

Ortega-Huerta, M. A. and A. T. Peterson. 2008. Modeling ecological niches and predicting geographic distributions: A test of six presence-only methods. Revista Mexicana de la Biodiversidad 79: 205–216.

Pavlik, B.M., Ferguson, N. and Nelson, M. 1993. Assessing limitations on the growth of endangered plant populations, II. Seed production and seed bank dynamics of Erysimum capitatum ssp. angustatum and Oenothera deltiodes ssp. howellii. Biological Conservation 65: 267–278.

Pearson, R.G. and Dawson, T.P. 2003. Predicting the impacts of climate change on the distribution of species: Are bioclimate envelope models useful? Global Ecology and Biogeography 12: 361–371.

Pearson, R.G. Dawson, T.P., and Liu, C. 2004. Modelling species distributions in Britain: a hierarchical integration of climate and land-cover data. Ecography 27(3): 285–298. Pearson, R.G., Raxworthy, C.J., Nakamura, M., and Peterson, A.T. 2007. Predicting species distributions from small numbers of occurrence records: a test case using cryptic geckos in Madagascar. Journal of Biogeography 34: 102–117.

Penuelas, J., Fillela, I., Zhang, X., Llorens, L., Ogaya, R., Lloret, F., Comas, P., Estiarte, M., and Terradas, J. 2004. Complex spatiotemporal phenological shifts as a response to rainfall changes. New Phytologist 161(3): 837–846.

Peterson, A. T. , Egbert, S. L., Sánchez-Cordero , V., and Price, K. P. 2000. Geographic analysis of conservation priorities using distributional modelling and complementarity: endemic birds and mammals in Veracruz, Mexico. Biological Conservation 93: 85–94. Peterson, A.T. 2003. Predicting the geography of species' invasions via ecological niche modeling. Quarterly Review of Biology 78: 419–433.

Peterson, A. T., Papes, M., and Kluza, D. A. 2003. Predicting the potential invasive distributions of four alien plant species in North America. Weed Science 51: 863–868.

Peterson, A. T., Benz, B.W., and Papes, M. 2007. Highly pathogenic H5N1 avian influenza: Entry pathways into North America via bird migration. PLoS ONE 2(2): e261.

Phillips, S. J., Dudik, M., and Schapire, R.E. 2004. A maximum entropy approach to species distribution modeling. Pp. 472–486 in Proceedings of the Twenty-First International Conference on Machine Learning. ACM Press, New York.

Phillips, S.J., Anderson, R.P., and Schapire, R.E. 2006. Maximum entropy modeling of species geographic distributions. Ecological Modelling 190: 231–259.

Phillips, S.J., and Dudik, M. 2008. Modeling of species distributions with Maxent: new extensions and a comprehensive evaluation. Ecography 31: 161–175.

Pitts-Singer, T.L., Hanula, J.L., and Walker, J.L. 2002. Insect pollinators of three rare plants in a Florida longleaf pine forest. Florida Entomologist 85(2): 308–316.

Prather, L.A.., Monfils, A.K., Posto, A.L., and Williams, R.A. 2002. Monophyly and phylogeny of Monarda: evidence from the internal transcribed spacer (ITS) region of nuclear ribosomal DNA. Systematic Botany 27: 127–137.

R Development Core Team. 2008. R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. ISBN 3-900051-07-0, URL: http://www.R-project.org.

Rodriguez-Riano, T. and Dafni, A. 2007. Pollen–stigma interference in two gynodioecious species of Lamiaceae with intermediate individuals. Annals of Botany 100: 423–431.

Rushton, S.P., Ormerod, S.J., and Kerby, G. 2004. New paradigms for modelling species distributions? Journal of Applied Ecology 41(2): 193–200.

Sabrosky, C.W. 1962. Mating in Xylocopa virginica. Proceedings of the Entomology Society of Washington 64: 184

Sampson, B.J., Danka, R.G., and Stringer, S.J. 2004. Nectar robbery by bees Xylocopa virginica and Apis mellifera contributes to the pollination of rabbiteye blueberry. Journal of Economical Entomology 97(3): 735–740.

Sargent, R.D., and Ackerly, D.D. 2007. Plant-pollinator interactions and the assembly of plant communities. Trends in Ecology and Evolution 23(3): 123–130.

Saunders, D. A., Hobbs, R. J., and Margules, C. R. 1991. Biological conservation of ecosystem fragmentation: a review. Conservation Biology 5: 18-32.

Scheen, A.C., Brochmann, C., Brysting, A.K., Elven, R., Morris, A., Soltis, D.E., Soltis, P.S., and Albert, V.A. 2004. Northern hemisphere biogeography of Cerastium (Caryophyllaceae): insights from phylogenetic analysis of noncoding plastid nucleotide sequences. American Journal of Botany 91: 943–952.

Scheen, A.C., Lindqvist, C., Fossdal, C.G., and Albert, V.A. 2008. Molecular phylogenetics of tribe Synandreae, a North American lineage of lamioid mints (Lamiaceae). Cladistics 24: 299–314.

Schemske, D.W. and Horvitz, C.C. 1984. Variation among floral visitors in pollination ability: A precondition for mutualism specialization. Science 225: 519–521.

Schemske, D.W., Husband, B.C., Ruckelshaus, M.H., Goodwillie, C., Parker, I.M., and Bishop, J.G. 1994. Evaluating approaches to the conservation of rare and endangered plants. Ecology. 75:584-606.

Shinners, L.H. 1962. Synopsis of Conradina (Labiatae). Sida 1,2: 84–88

Silverman, B.W. 1986. Density estimation for statistics and data analysis. In: Monographs on Statistics and Applied Probability. Chapman and Hall, London.

Silvertown, J., Franco, M., Pisanty, I., and Mendoza, A. 1993. Comparative Plant Demography- -Relative Importance of Life-Cycle Components to the Finite Rate of Increase in Woody and Herbaceous Perennials. Journal of Ecology 81(3): 465–476.

Silvertown, J. 2004. Plant coexistence and the niche. Trends in Ecology and Evolution 19(11): 605–611.

Soberon, J., and Peterson, A.T. 2005. Interpretation of models of fundamental ecological niches and species‘ distributional areas. Biodiversity Informatics 2: 1–10.

Spira, T. P. 2001. Plant–pollinator interactions: a threatened mutualism with implications for the ecology and management of rare plants. Natural Areas Journal 21: 78–88.

Steffan-Dewenter, I., Potts, S.G., and Packer, L. 2005. Pollinator diversity and crop pollination services are at risk. Trends in Ecology and Evolution 20: 651–652.

Stein, B.A., Kutner, L.S., and Adams, J.S. [eds]. 2000. Precious Heritage: The Status of Biodiversity in the United States. Oxford University Press, USA. 399 pp.

Stockwell, D. and Peters, D. 1999. The GARP modeling system: problems and solutions to automated spatial prediction. Integrated Journal of Geographic Information Science 13: 143–158.

Stockwell, D. R. B. and Peterson, A. T. 2002. Effects of sample size on accuracy of species distribution models. Ecological Modelling 148: 1–13.

Thuiller, W., Richardson, D.M., Pysek, P., Midgley, G.F., Hughes, G.O., and Rouget, M. 2005. Niche-based modelling as a tool for predicting the risk of alien plant invasions at a global scale. Global Change Biology 11: 2234–2250.

Vázquez, D. P., and Simberloff, D. 2002. Ecological specialization and susceptibility to disturbance: conjectures and refutations. American Naturalist 159: 606–623.

Viney, N.R., Bates, B.C., Charles, S.P., Webster, I.T., and Bormans, M. 2007. Modelling adaptive management strategies for coping with the impacts of climate variability and change on riverine algal blooms. Global Change Biology 13(11): 2453–2465.

Visser, M.E. and Both, C. 2005. Shifts in phenology due to global climate change: the need for a yardstick. Proceedings of the Royal Society B 272: 2561–2569.

Walker, P. A. and Cocks, K. D. 1991. HABITAT: a procedure for modelling a disjoint environmental envelope for a plant or animal species. Global Ecological Biogeography Letters 1: 108–118.

Walker, J.L. and Madsen, D. 1997. The breeding system of Macbridea alba, a federally threatened mint. Associated SE Biological Bulletin 44: 127–128.

Walther, G.R., Post, E., Convey, P., Menzel, A., Parmesan, C., Beebee, T.J.C., Fromentin, J.M., Hoegh-Guldburg, O., and Bairlein, F. 2002. Ecological responses to recent climate change. Nature 416: 389–395.

Ward, D.F. 2007. Modelling the potential geographic distribution of invasive ant species in New Zealand. Biological Invasions 9 (6): 1573–1474.

Weller, S. G. 1994. The relationship of rarity to plant reproductive biology. Pp. 90–117 in M. L. Bowles and C. J. Whelan [eds.] Restoration of endangered species. Cambridge University Press, Cambridge, UK.

Westerkamp, C., and Claben-Bockhoff, R. 2007. Bilabiate flowers: the ultimate response to bees? Annals of Botany 100: 361–374.

Whitaker, R. J. and Banfield, J. F. 2005. Population dynamics through the lens of extreme environments. Reviews in Mineralogy and Geochemistry 59: 259–277.

Whitfield, G. H. and K. W. Richards . Temperature-dependent development and survival of immature stages of the alfalfa leafcutter bee, Megachile rotundata (Hymenoptera: Megachilidae). Apidologie 11: 23.

Wilson, E. O. 1992. The Diversity of Life. W. W. Norton and Company, Inc., New York, New York. 424 pp.

Wunderlin, R.P. 1998. Guide to the vascular plants of Florida. University Press of Florida, Gainesville, Florida. 806 pp.

Wunderlin, R.P. and Hansen, B.F. 2000. Flora of Florida: Volume I Pteridophytes and Gymnosperms. University Press of Florida, Gainesville, Florida. 384 pp. Zaniewski, A. E., Lehmann, A. and Overton, J. M. 2002. Predicting species distribution using presence-only data: a case study of native New Zealand ferns. Ecological Modelling 157: 261–280.

Zhang, Y.W., Robert, G.W., Wang, Y., and Guo, Y.H. 2007. Nectar robbing of a carpenter bee and its effects on the reproductive fitness of Glechoma longituba (Lamiaceae). Plant Ecology 193: 1–13.

BIOGRAPHICAL SKETCH

Amanda J. Kubes was born in 1982 in Winter Haven, Florida, and grew up and pursued schooling in Jacksonville, Florida. With a strong interest in Biology, she attended the University of North Florida in Jacksonville and transferred to Florida State University to complete her B.S. degree. She received her B.S. in Biology with a minor in Chemistry from Florida State University in 2006. During this time, she worked in Dr. Austin Mast‘s laboratory as a Lab Technician and also worked with an environmental consulting laboratory as a Field and Lab Technician and Laboratory Manager for three years prior to entering the Graduate program at Florida State University. She has also worked as the Curator for the Robert K. Godfrey Herbarium at Florida State University for two years.Her interests and work experience have led her to develop a strong interest in conservation biology and botany, and the integration of biology and geographical information systems (GIS) for use in conservation planning. Future plans include pursuing her Ph.D. and continuing research in her fields of interest.