xml version 1.0 encoding UTF-8
REPORT xmlns http:www.fcla.edudlsmddaitss xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.fcla.edudlsmddaitssdaitssReport.xsd
INGEST IEID EUB710FUF_G5260Q INGEST_TIME 2014-05-13T21:04:01Z PACKAGE UFE0046408_00001
AGREEMENT_INFO ACCOUNT UF PROJECT UFDC
1 DISTRIBUTION AND HABITAT USE OF FLORIDA ENDEMIC SOLITARY BEE HESPERAPIS ORARIA AND HOST PLANT BALDUINA ANGUSTIFOLIA By HANNAH K HUNSBERGER A THESIS PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR TH E DEGREE OF MASTER OF SCIENCE UNIVERSITY OF FLORIDA 2013
2 2013 Hannah K Hunsberger
3 To EPD and ISH
4 ACKNOWLEDGMENTS I would like to express my very great apprecia tion to my advisors Dr. Debbie Miller and Dr. Mack Thetford for their expertise and guidance. I would also like to offer a special thanks to my committee member Dr. Jaret Daniels for his advice and support throughout this project. I wish to acknowledge E mily Leary for her invaluable help with my statistical analyses. I am particularly grateful for the support and funding from Gulf Islands National Seashore. I would also like to thank the following people for their contributions: Tova Spector and John Be nte with Florida State Parks for their enthusiasm and assistance with the collection of my data; the resource management staff at Eglin Air Force Base for field work support; Dr. James Cane for providing me with valuable insight into H. oraria ecology. Fi nally, I wish to thank my family for their support and encouragement throughout my study. The output and data analysis for this paper was generated using SAS software, Ve rsion 9.2 of the SAS System for Windows. Copyright 2008 SAS Institute Inc. SAS and all other SAS Institute Inc. product or service names are registered trademarks or trademarks of SAS Institute Inc., Cary, NC, USA. The analysis and maps for this paper was generated using ESRI 201 1. ArcGIS Desktop: Release 10. Redlands, CA: Environmental Systems Research Institute
5 TABLE OF CONTENTS page ACKNOWLEDGMENTS ................................ ................................ ................................ .. 4 LIST OF TABLES ................................ ................................ ................................ ............ 6 LIST OF FIGURES ................................ ................................ ................................ .......... 7 ABSTRACT ................................ ................................ ................................ ..................... 8 CHAPTER 1 INTRODUCTION ................................ ................................ ................................ .... 10 Threats to Specialized Solitary Bees ................................ ................................ ...... 10 Hesperapis oraria and Balduina angustifolia ................................ ........................... 1 7 2 HABITAT SUITABILITY OF BALDUINA ANGUSTIFOLIA AND HESPERAPIS ORARIA AT TWO SPATIAL SC ALES ................................ ................................ .... 19 Background ................................ ................................ ................................ ............. 19 Study Area ................................ ................................ ................................ .............. 21 Methods ................................ ................................ ................................ .................. 23 Data C ollection ................................ ................................ ................................ 25 Statistical A nalyses ................................ ................................ .......................... 27 H. Oraria Habitat U se ................................ ................................ ....................... 28 B. Angustifolia Habitat U se ................................ ................................ ............... 31 Results ................................ ................................ ................................ .................... 32 H. Oraria Habitat U se ................................ ................................ ....................... 33 B. Angustifolia Habitat U se ................................ ................................ ............... 35 Discussion ................................ ................................ ................................ .............. 37 3 CHANGES IN BALDUINA ANGUSTIFOLIA AND HE SPERAPIS ORARIA DENSITY OVER TIME ................................ ................................ ............................ 64 Background ................................ ................................ ................................ ............. 64 Methods ................................ ................................ ................................ .................. 66 Data C ollection ................................ ................................ ................................ 66 Statistical A nalyses ................................ ................................ .......................... 67 Results ................................ ................................ ................................ .................... 68 Discussion ................................ ................................ ................................ .............. 69 4 FUTURE DIRECTIONS AND MANAGEMENT IMPLICATIONS ............................. 79 LIST OF REFERENCES ................................ ................................ ............................... 84 BIOGRAPHICAL SKETCH ................................ ................................ ............................ 90
6 LIST OF TABLES Table page 2 1 Indicator species of the 4 optimal clusters from the cluster analysis performed based on environ mental variables. ................................ .................... 42 2 2 Total sampling effort for H.oraria in coastal and non coastal mainland areas with and without host plant B.angustifolia. ................................ .......................... 43 2 3 Summary of H. oraria count data and B. angustifolia charactersti cs by location, unit and site. ................................ ................................ ........................ 44 2 4 Prediction success (of the test data) for the classification tree i n predicting site habitat use by H.oraria at a regional scale ................................ .................. 46 2 5 Variables important for predicting the site habitat use of H.oraria at a regional scale. ................................ ................................ ................................ ................. 47 2 6 Prediction success (of the test data) for the classification tree in predicting habitat patch use by H.oraria at a landscape scale ................................ .......... 48 2 7 Variables import ant for predicting the habitat patch use of H.oraria at a landscape scale. ................................ ................................ ............................... 49 2 8 Prediction success (of the test data) for the classification tree in predicting site habitat use by B. angust ifolia at a regional scale ................................ ........ 50 2 9 Variables important for predicting the site habitat use of B.angustifolia at a regional scale. ................................ ................................ ................................ .... 51 2 10 Prediction success (of the test data) for the classification tree in predicting B.angustifolia habitat patch use at a landscape scale ................................ ....... 52 2 11 Variables important for predicting the hab itat patch use of B.angustifolia at a landscape scale. ................................ ................................ ................................ 53 2 12 Site surveyed for H.oraria, including the 1993 1994 field seasons (Cane et al. 1996) and 2011 2012 field seasons. ................................ ................................ 54 3 1 Summary statistics of B.angustifolia patch characteristics that showed significant changes between 2011 and 2012. ................................ ..................... 74 3 2 Wilcoxon rank sum test results showing H.oraria density and B.angustifolia patch characteristics that changed significantly between 2011 and 2012. ......... 75 3 3 Parameter estimates for the full regression model compar ing characteristics of foraging habitat ( B.angustifolia patches) against H.oraria density. ................. 76
7 L IST OF FIGURES Figure page 2 1 The geographical range o f Hesperapis oraria documented by in 1993 1994 field seasons ( C ane et al. 1996 ) . ................................ ................................ ....... 56 2 2 Sites visited, fall 2012, where B.angustifolia was present and H.oraria was either present or absent. ................................ ................................ ..................... 57 2 3 Sites sampled (with transects), fall 2012, where B.angustifolia and H.oraria were present. ................................ ................................ ................................ ...... 58 2 4 Sites grouped into dominant community types via hierarchi cal clustering using Sorensen (Bray Curtis) for the distance measure and Flexible Beta for the linkage method in PC ORD. ................................ ................................ ......... 59 2 5 Classification tree relating probability of site use by H.oraria to landscape features over a regional scale. ................................ ................................ ........... 60 2 6 Classification tree relating probability of patch use by H.oraria to landscape features over a landscape scale. ................................ ................................ ........ 61 2 7 Classification tree relating probability of occupancy of B.angustifolia to landscape features over a large scale. ................................ ............................... 62 2 8 Classification tree relating p robability of habitat patch occupancy of B.angustifolia to landscape features. ................................ ................................ .. 63 3 1 Gulf Islands National Seashore units Perdido Key, Fort Pickens, and Santa Rosa. ................................ ................................ ................................ .................. 77 3 2 Total monthly precipitation for months preceding and for the duration of H.oraria foraging and B.angustifolia flowering. ................................ ................... 78
8 Abstract of Thesis Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Master of Science DISTRIBUTION AND HABITAT USE OF FLORIDA ENDEMIC SOLITARY BEE HESPERAPIS ORARIA AND HOST PLANT BALDUINA ANGUSTIFOLIA By Hannah K Hun sberger December 2013 Chair: Mack Thetford Cochair: Debbie Miller Major: Interdisciplinary Ecology The solitary ground nesting bee Hesperapis oraria is endemic to the barrie r islands of the Gulf of Mexico Significant not only for its restricted geogr aphic distribution but also for its specialization to a sole floral host, Balduina angustifolia this rare bee warrants further investigation into its habitat needs. H. oraria density decreased overall from 2011 to 2012 and was lower on barrier island sit es when compared with coastal mainland sites. No H. oraria were found on non coastal mainland sites. This study also quantifies the habitat use of H. oraria and B. angustifolia at two spatial scales using a classification tree species distribution modelin g approach. Analys e s determined the regional distribution of H. oraria was driven by the number of B. angustifolia patches while at the landscape scale H. oraria was predicted most strongly by B. angustifolia patch area and density as well as proximity t o ephemeral wetlands. T he regional distribution of B. angustifolia was influenced by the proximi ty to the bay and within a landscape B. angustifolia distribution wa s strongly influenced by the stability of the landscape feature in which it is found. Resu lts from Wilcoxon rank sum tests indicate that B. angustifolia patch size (m) and flower density (no/m 2 ) decreased from 2011 to 2012. Establishing a long term monitoring program to
9 determine the population status of this rare bee is vital to it conserva tion. It is also important to protect B. angustifolia habitat by maintaining the limited stability of these coastal ecosystems.
10 CHAPTER 1 INTRODUCTION Threats to Specialized Solitary Bees Biological conservation traditionally focuses on conservin g charismatic vertebr ates, plants, and communities and, t here fore, there has been a distinct lack of effort towards the conservation of insect species (Gullan and Cranston 2010). The conservation of rare insects is often threatened simply because insects are frequently overlooke d due to low detection (Brust et al 2008 ; Much of the effort in insect conservation has been focused on charismatic insects which can act as flagship species. However, e ven the plight of flagship species can prove hard to detect. For example, a threat to the narrow food source ( Myrmica ant societies) of the globally threatened Maculinea arion or Large Blue Butterfly went unnoticed until the butterfly came close to extinction (Thomas et al 2009). T he con servation of insects is important for many reasons. Insects can act as keystone species, meaning their loss would collapse the wider eco system of which they are a part, or indicator species, species whose population health can be used to gauge the health of the ecosystem at large. Knowledge of the importance of insects has grown and there are numerous examples of successful insect conservation endeavors. For example, the El Segundo blue, Euphilotes battoides spp allyni lives primarily in sand dunes nea r the Los Angeles airport. Urban sprawl threatened this habitat but with education and negotiations 80 hectares of their prime habitat was set aside as a reserve and surroundin g golf courses agreed to preserve thei r larval food plant Erigonum parvifolium. This reserve not only protects the El Segundo blue but also an endangered habitat and oth er threatened insects (Gullan and Cranston 2010). General s olutions t o mitigate threats to rare and
11 threatened insects include increasing knowledge about the existe nce of these species, determining where they are found, and quantifying their habitat needs. In recognition of the importance of even the smallest members of an ecosystem, there is a recent increase in research efforts to gather precise information about requirements and determine habitats that do and could support endangered insects (Hill Horak 2011). About 87% of angiosperms are pollinated by animals (Ollerton et al. 2011) and i nsects Insect plant interactions in the form of pollination is a mutualistic relationship, the insect gathers resources from the plant a nd in turn aids that plant in sexual reproduction Threats to pollinators include loss and fragmentation of habitat, climate change and the introduction of exotic species (Rathcke and Jules 1993; Allen Wardell et al. 1998; Debinksy and Holt 2000). U nderstanding the plant pollinator community structure is vital to c onservation efforts focusing on pollinator and hos t plant populations (Aizen 2007; Pemberton 2010; Potts et al. 2003). Bees are considered the most important group of pollinator insects because and are economically vital for their pollination services to crops (Klein et al 2007; Potts et al 2010). There has been a recent rise in co ncern over increasing reports on the loss of bee diversity worldwide (Brown and Paxton 2009; Brown 2011; National Research Council of the National Academies 2006). Despite a general consensus of their importance and imperiled status, bee population dynamics and conservation knowledge is grossly inadequate; especially when compared with similar taxa (Biesmeijer et al 2006, Kremen et al. 2007, Potts et al 2010). Therefore, more research on bee ecology, population genetics,
12 population dynamics and behavior is critical to make informed decisions and effective conservation efforts (Cerna et al 2013; Franzen and Nilso n 2013). Insects display a variety of social behavior ranging from e to solitary. Eusocial insects co operate for the common good of their colony ; they are defined by (1) a division of l abor with a rigid caste system which inc ludes non reproductive individuals, (2) co oper ation in caring for the young and (3) an overlap in generations where all work simultaneously for a common goal of the persistence and success of the colony Solitary insect s exhibit none of these co operativ e behaviors Solitary insects care for their brood on their own, in fact, nearly all solitary insects are parasitoids where adults immobilize arthropod prey for their offspring to feed on. Non social insects in the Hymenoptera order are restric ted to a fe w aculeate (ovipositor is modified as a sting) in the Apocrita suborder within the superfamilies Chrys idoidea, Vespoidea, and Apoidea (Gullan and Cranston 2010). Solitary bees (Apoidea) use pollen to provision the ir nest and feed their larvae instead of arthropod prey Every solitary female bee is fertile and inhabits the nes t she built alone. Eusocial bees have some advantages over solitary bees since foraging, feeding the queen, caring for offspring, and nest maintenance are all performed simultaneous ly by different castes where solitary bees must perform all these tasks in sequence. As a result individual errors for eusocial bees have little or no consequence compared with solitary bees where one mistake could be fatal to an individual female and the refore all her potential offspring. Furthermore, solitary bees lack the ability to defend against more formidable predators that eusocial bees could cooperatively thwart. Despite their disadv antages
13 so litary bees make up a significant portion of the wor roles in the functioning of their ecosystem. Solitary bees provide important pollination services These bees can be especially important for outcrossing plant species, even when they are not the only pollina tors visiti ng the plant (Bischoff et al 2013). A rare solitary bee Hylaeus matamoko was found to be responsible for 90% and 95% of the pollination of two outcross ing plant species in the New Zealand alpine which were visited j ust as often by a fly speci es (Bischoff et al 2013). Zych and Stpiczynska (2012) studied the pollination effectiveness of various pollinators on Fritillaria meleagris (Liliac eae) and concluded that while Bo mbus queens are the most frequent visitors, the small bodied solitary bees (genera Andrena and Lasioglossum) carried almost three times larger pollen loads. The researchers concluded that small solitary bees could therefore be more effective pollinators in terms of quality of pollination (Zych and Stipiczyska 2012). These studi es also point to the importance of understanding the effectiveness of each visit by a pollinator in addition to the rate of visitation to identify the importance of each pollinator species for a plant or community (Bischoff et al 2013; Zych and Stipiczysk a 2012). Solitary bees might not be the most frequent visitors but possibly because they are doing all the work by themselves they can still be more effective and are therefore more important than casual observation might indicate. In most pollinati on systems insects broadly pollinate all or most flowering plants without discrimination. However some insects are faithful pollinators toward one or a few select plants. For example, fig wasp species are known to be specific pollinators of specific fig species and some moth proboscis lengths match exactly to the flower depth
14 of only certain species (ie Angaecum sesquipedale and Xanthopaan morgana paraedicta ). Specialist bees use olfactory receptors on their antennae to detect flower odorants and can dis tinguish plant species as well as differentiate between rewarding and non rewarding flowers within species (Dobson a nd Bergstroim 2000; Raguso 2008 ). Examples of specialist bees include Heriades truncorum which are restricted to heaths (Praz et al. 2008) Hoplitis adunca which solely collects pollen from plants in the Echium genus (Boraginaceae) (Burger et al. 2010 ) and Andrena vaga (Andrenidae) which specializes on the genus Salix (willow) as a host plant (Cerna et al. 2013) Specialist bees may be rest ricted to their host by physiological adaptations for pollen digestion and/or neurological adaptations to identify floral features (Sedivy et al. 2008) but they find their host plant using taxon specific olfactory cues (Dotterl and Vereecken 2010). Older bees can also use learned features of host plants to locate them (Dobson and Bergstrom 2 000; Dotterl and Vereecken 2010 ; Milet Pin heiro et al. 2012). While m any solitary bees are oligolectic, or specialist pollinators to a few related plant species True monolectic bees, bees which only pollinate one species of flower, are extremely rare. Bee populations worldwide are threatened and the ecology of specialist bees makes them particularly imperiled. Habitat degradation and fragmentation from landscape c hanges are major threats to bee populations (e.g. Klein et al. 2007; Kremen e t al. 2007). Rare and endangered species are often threatened because of their specialized food and habitat requirements; this is especially true of many rare insects (Brust et a Thomas et al. 2009 ; Snep et al. 2011; Horak 2011). Specialist bees seem to be especially sensitive to negative effects of habitat
15 fragmentation compared with generalists species (Biesmeijer et al. 2006; Cane et al. 2006). For bees that pollinate more than one flower species there is opportunity to optimize the best balance of water and sugar by varying the foraging patterns appropriately. However, monolectic bees that rely on a single floral host usually specialize in the most abundant and reliable host species, often one with simple flowers that can be pollinated by many species (Dotterl and Vreecken 2010). The availability of one reward over another can be limiting to the bee species that specialize in this way (Pemberto n 20 1 0, Willmer 1986). Pemberton (2010) documented a correlation between the reductions in the abundance of a specialist bee to the decline of the only oil reward plant in an area. Monolectic bees also do not have the luxury of choosing their foraging t ime to optimize water sugar ratios in flower nectar because this would susceptibility to nest robbing and parasitism (Willmer 1986). Often, the only remaining option for sp ecialist bees to optimize foraging is the choice between flo ral patches of the same species. R esource inputs are controlled by the number of flowers visited in a given trip (Willmer 1986). Because of the lack of control over energy water ratios, monolect ic solitary bees are thought to forage primarily to fulfill their own physiological water needs and, assuming the monolectic relationship occurs because of co evolution, it is likely that sugar needs are met by default (Willmer 1986). For example, a study of a monolectic solitary bee, Chalicodoma sicula concludes that it chooses its foraging strategy in relation to optimal water, rather than caloric rewards in arid environments (Willmer 1986). Pollen specialization can also affect the reproductive succes s of the host plant as both the quality and quantity of pollination available to them can be limited
16 (Aizen 2007). Since specialists rely on specific habitats and narrow food sources, anthropogenic disturbance of either is a significant threat (Brust et al 2008; Hill and Snep et al. 2011). Pollinator specialist are thus vulnerable to food loss due to habitat fragmentation, and monolecty further increases this sensitivity as the reliance on one host plant increases the cha nces that landscape change can likely affect the species entire food source (Bommarco et al. 2010). Physiological factors influence the distance an individual bee can travel from nest to foraging patch. Therefore, the density of small bodied specialized s pecies is more likely to decrease over time in small habitat patches than larger habitat patches (Bommarco et al. 2010) Bo mmarco et al. (2010) concludes that a high proportion of small specialist bees are extinct because of habitat fragmentation ; demonst rated by low species richness found in small habitat patches (Bommarco et al. 2010). Solitary bees are also highly vulnerable to habitat change. Solitary bees rely on flowers for both energy and water needs as well as brood provisions; this is unlike man y social bees that often take breaks from pollen foraging to meet their water needs from a different source (Pemberton 2010 ; Willmer 1986). Solitary bees must return to their nests multiple times throughout a day to supply their brood cells with pollen an d nectar (Zurbuchen et al. 2010b ). Therefore, habitat fragmentation and landscape barriers can limit the resources available to a population (Zurbuchen et al 2010c). Furthermore, solitary bees are susceptible to negative impacts from landscape fragmenta tion or any change that might increase the distance between their nests and floral host patches because they have a higher frequency of trips from nesting sites to foraging sites th an social bees (Zurbuchen et al. 2010b ). A study by Peterson and Roitberg (2006)
17 demonstrated that the reproduction success of some solitary bee species suffered when the foraging distance was increased. Understanding the distance a bee is capable of covering between the nest and host plant patch is therefore critical in the pre servation of endangered bee speci es populations (Zurbuchen et al. 2010 b ). However, the maximum foraging distance is not always representative of the actual distances females will travel, as some species are shown to often stop short of their ma ximum (Zurbu chen et al. 2010 a ; Zurbuchen et al. 2010 c ). Overall, solitary bees have less flexibility and are therefore more vulnerable to landscape change and habitat fragmentation than their social counterparts. The vulnerability of a host plant species to habitat fragmentation depends on the extent of their mutualism with a pollinator (Aizen et al. 2002b). Research in pollinator and host plant interaction indicates that habitat fragmentation reduces the abundance of pollinators and is therefore responsible for a decline in pollination rates that results in a decrease in fruit and/or seed production of host plants (Stef fan Dewenter and Tscharntk 1999; Cunningham 2000). For example, Cunningham (2000) found flowers of A cacia brachybotrya and Eremophila glabra decrea sed fruit production when growing in fragmented habitats because they are pollen limited. Hesperapis oraria and Balduina angustifolia Hesperapis oraria is a ground nesting solitary bee endemic to coastal areas on northern Gulf of Mexico. Recently describ ed, this bee is a species of interest for its limited geographic range and its dependence on a single floral host plant, Balduina angustifolia, for all its nectar and pollen needs (Cane et al. 1996). There is limited information on the habitat needs of th e bee and much of what is known is inferred from its closest relative, Hesperapis carinata, found in mid and north western United States.
18 Preliminary studies indicate that H. oraria is rare and likely endangered as a result of habitat loss and fragmentati on from both natural and anthropogenic disturbances (Cane et al. 1996). In order to protect this rare species it is important to determine and characterize its geographic distribution. Their apparent reliance on a single host plant species makes the char acteristics of B. angustifolia habitat important as well. The host plant is dependent on pollination for seed production and therefore, H. oraria may be important for its reproductive success in these coastal habitats. However, B. angustifolia can be fo und in locations where H. oraria is not present indicating that the environmental factors and is pollinated by other insects.
19 CHAPTER 2 HABITAT SUITABILITY OF BALDU INA ANGUSTIFOLIA AND HESPERAPIS ORARIA AT TWO SPATIAL SCALES Background Described in 1996 Hesperapis oraria an endemic species with a range from eastern Mississippi barrier islands to the eastern coast of the Florida panhandle, is the only known species of the Dasypodaidae family to occur east of the Mississippi River (Cane et al. 1996). In 1993 and 1994 entomologists from Auburn University and the U.S. Department of Agriculture determined endemism to select coastal locations on the northern Gulf of Mexico. They also presented it as a rare monolectic species, meaning the female bees gather all their pollen for brood provision from only one species of plant, Balduina angustifolia (Cane et al. 1996). Monolecty is extremely rare; in fact, many s pecies initially documented as monolectic have since been documented collecting pollen on a substitute host when their preferred host plant is unavailable. (Dotterl and Vreecken 2010 ; Muller and Kuhlmann 2008; Westrich 2008). The nests of H. oraria have n ot been located but information on their lifecycle can be inferred from foraging behavior and general knowledge of other ground nesting solitary species. H. oraria emerges for about three weeks from the end of September to the middle of October to reprodu ce, females likely mate shortly after emergence and spend the rest of their time provisioning their nests (Cane 1994) H. oraria is therefore a species of interest because of its limited geographic range, the threatened state of its limited habitat, and i ts monolecty. Beyond the study conducted in 1993 and 1994, there is no information on the distribution or habitat needs for H. oraria The genus Balduina is composed of three species found in southeastern United eceptacular bractlets connected into
20 honeycomb like structures surr Parker 1975). Balduina angustifolia or coastal plain honeycomb head is a self incompatible annual. It grows in dry soils of deep sand ridges along rivers and on shallow dunes of the Atlantic and Gulf Coast beaches and can be found throughout Florida and into southern Georgia and Alabama (Parker 1975). The plant is the sole known host plant for H. oraria, however, it is regularly visited by a broad variety of pollinators. In coastal areas, coastal plain honeycomb head has a patchy distribution and only flowers for a short period coinciding with foraging activity of H. oraria The coastal areas of the Gulf of Mexico are under the pressure of constant change from wind, waves, sedimentations, sea level rise punctuated by storms and hurricanes that cause sporadic changes in geomorphic features (Lucas and Carter 2010). The limited flowering period coupled with the dynamic landscape of the gulf coast habitats co ntributes to the patchy distribution of the plant in these environments. Human development and habitat degradation threaten many species that live entirely in these dynamic habitats making these species especially sensitive to further disruption from alte rations of ve getation (Lucas and Carter 2010; Pries et al. 2009). Therefore, it is important to understand what drives the distribution of this host plant if we aim to understand the distribution of H. oraria. In order to understand the agents driving t he distribution of H. oraria and B. angustifolia it is necessary to examine their interactions with each other and abiotic and biotic factors at different scales (Urban et al. 1987). The complexity of the interactions between these factors and how spatial scale affects them can be overwhelming. Therefore, this study attempts to simplify the system into basic components operating at different scales. By designing analyses at two different scales (Simon 1962) it is
21 possible to predict how spatial scale inf luences the interactions between H. oraria, B. angustifolia and their environment. Study Area Barrier islands, coastal mainland areas, and non coastal mainland areas were surveyed for H. oraria and B angustifolia throughout most of the known range of H oraria in Florida (Figure 2 1). Sites were located on lands managed by state and federal agencies and included: Gulf Islands National Seashore (GINS), Eglin Air Force Base (EAFB), Florida Park Service land, and Florida State Forest land (Figure 2 2). Thr ee units of GINS were included: Perdido Key, Fort Pickens, and Santa Rosa Island. Both non coastal mainland and Santa Rosa Island sites were sampled on EAFB. Florida State lands included Big Lagoon State Park, Henderson Beach State Park, Topsail Hill Sta te Park, Grayton Beach State Park, Deer Lake State Park, St. Andrews State Park, and Point Washi ngton State Forest. An area owned by the University of West Florida on Santa Rosa Island was also included. Gulf of Mexico barrier islands are narrow strips of sand running parallel to the mainland. Barrier islands included in the study were Perdido Key and Santa Rosa Island, Florida. Together they stretch from approximately two miles west of the Santa Rosa Island is about 60 km long and 0.5 km wide The island has foredunes near the Gulf of Mexi co that run intermittently the length of the island parallel to the shoreline. Secondary dunes are found behind foredunes closer to t he bay. The area between these dunes consists of sand flats and rolling grassland s with ephemeral wetlands, called swales, found scattered throughout. Dominant species of the foredunes include sea oats ( Uniola paniculata) cakile ( Cakile spp. ), beach morn ing glory ( Ipomoea
22 imperati ), and seashore elder ( Iva imbricata ). The secondary dunes are dominated by woody species such as false rosemary ( Ceratiola erico i des ), woody goldenrod ( Chrysoma pauciflosculosa ), sand live oak ( Quercus geminata ), and sand pine ( Pinus clausa ) (Pries et al. 2008 ). Perdido Key is about 24 km long and 0.5 km wide, has fewer well developed dunes and generally lacks the woody secondary dunes compared to Santa Rosa Island (Looney et al. 1993). However, the species composition of the islands is comparable. Perdido Key also features a wider section of the island where the secondary dunes develop into a scrubby flatwood forest with a slash pine ( Pinus elliotii) overstory (FNAI 2010). Coastal mainland sites included all the Florida State Parks (Fig 2 2) and were defined as the first 0.5 km of mainland closest to the Gulf of Mexico These sites are composed of foredunes close to the gulf, behind which are secondary woody dunes similar in composition to those found on the barrier islands. However, behind the woody scrub dunes is denser woody vegetation composed of evergreen shrubs and, in some cases, a pine canopy (scrub and scrubby f latwoods) Dominant species include Florida rosemary ( Ceratiola ericoides ), sand pine ( Pinus clausa ), myrtl e oak ( Quercus myrtifolia ), sand live oak ( Quercus geminata Quercus chapmanii ) and saw palmetto (Serenoa repens ). Interspersed through the woody species are opening s consisting of bare ground and sparse herbaceous species (FNAI 2010). Th e non coastal mainland sites included in this study consist of Mesic and Scrubby Flatwoods and Sandhill co mmunities located further than 0.5 km from the Gulf of Mexico. Mesic and Scrubby flatwoods have an open canopy of longleaf pine ( Pinus palustris) and slash pine ( Pinus elliottii ). These communities consist of low shrubs
23 including saw palmetto ( Serenoa repens ), inkberry ( Ilex glabra ), and fetterbush ( Lyonia lucida ). Dominant understory species are dwarf live oak ( Quercus minima ), runner oak ( Q. elliot tii ), shiny blueberry ( Vaccinium myrsinites ), Darrow's blueberry ( V. darrowii ), dwarf huckleberry ( Gaylussacia dumosa ), wiregrass ( Aristida stricta var beyrichiana ), pineywoods dropseed ( Sporobolus junceus ), panicgrasses ( Dichanthelium spp.), and broomsed ges ( Andropogon spp.). Sand hill communities are composed of a longleaf pine ( Pinus palustris ) overstory and a midstory dominated by turkey oak ( Quercus laevis ), bluejack oak ( Q. incana ), sand live oak ( Q. geminata ), sand post oak ( Q. margaretta ), saw palm etto ( Serenoa repens ), and sparkleberry ( Vaccinium arboreum ). The understory is composed of a diversity of grasses and forbs, the dominant of these are wiregrass ( Aristida stricta var. beyrichiana) dwarf huckleberry ( Gaylussacia dumosa ), pricklypear ( Opu ntia humifusa ), gopher apple ( Licania michauxii), pineywoods dropseed ( Sporobolus junceus ), lopsided indiangrass ( Sorghastrum secundum ), several species of bluestems ( Andropogon spp.), and little bluestem ( Schizachyrium scoparium ) (FNAI 2010) Methods To confirm specialization on B. angustifolia and determine H. oraria and B. angustifolia distribution in coastal and non coastal areas, locations visited in previous years were re surveyed. In the fall of 1993 and 1994 entomologists from Auburn University searched for H. oraria on barrier islands, coastal mainland areas, and non coastal mainland areas from Mississippi to the eastern Florida panhandle (Cane et al. 1996). The researchers searched for the bee in areas with abundant B. angustifolia and areas with little or no B. angustifolia, collecting voucher specimen s for species verification. They did not have GPS/GIS technology to accurately document the
24 location of their search areas, s o this study relied on resulting publications and reports along with correspondences bet ween Cane and land managers to locate sampling areas Ten of the same locations initially sampled i n 199 3 19 94 were resampled within H. oraria To determine the spatial distribution of H. ora ria and B. angustifolia 24 large scale grids were established at each site. Each site was sampled using belt transects orientated to fit within the boundaries of the undeveloped sections of the state and federal lands. (Figure 2 3). Using 2006 Digital Or thophoto Quarter Quads ( DOQQs ) from the United State Geological Survey (USGS), transects were placed through a variety of landscape feature s and possible habitat types. Length and configuration of transects was confined by the Gulf of Mexico, bays (Chocta whatchee Bay, Santa Rosa Sound, Pensacola Bay, and Big Lagoon), highways, parking lots, housing developments and other anthropogenic features. As a result, transect length varied in two ways. At 18 sites four two meter wide and 400 meters long belt tran sects were established, the transects were all 50 meters apart and ran perpendicular to the Gulf of Mexico within undeveloped areas. Every 100 meters along the transects a circular point count area of 28 m eters was established. At six sites, 10 two meter wide and 200 meters long belt transects were established. These transects were all 50 meters apart and ran perpendicular to the Gulf of Mexico also exclusively within undeveloped areas. There was a point count covering 28 meters every 100 meters along th e length of these transects. Overall, approximately 4000m 2 were sampled at each site which encompasses more than the estimated maximum foraging distance of H. oraria based on behavior and body size ( Gathmann and Tscharntke 2002, Zurbu chen et al. 2010 c )
25 D ata C ollection To determine habitat use in 2011 and 2012 observations occurred within large scale sampling site transects. However, for descriptive purposes only, obs ervations outside sampling transects were included to denote the presence of the bee when it was not always detected during formal sampling. Furthermore, Camp Helen State Park and Perdido Key State Park were visited wi thout establishing sampling transects These locations were informally searched for two hours or until the bee w Sampling for H. oraria took place from October 4 th through October 22 nd 2012. To eliminate bias, two samplers were randomly assigned a site, sampling time, starting location and the role of recorder or surveyor. At the beginning of the survey, a Safe Easy digital thermometer. At all point count locations, a hand held Kestral wind meter was used to de termine average wind speed (mph), and maximum wind gusts (mph). H. oraria and other pollinators were searched for at each point count by both samplers who stood back to back for two minutes and recorded all pollinators spotted within a three meter radius from the center of the point count. Four pictures were also taken at each point count, in the four cardinal directions. Between the point count locations transects were surveyed in a consistent manner. T ransects were sampled at a steady pace and all H. oraria and other pollinators up to a meter on either side of the transect line were counted. When an H. oraria bee was found, the location was marked with a flag and geographic coordinates and the landscape featu re in which the bee was found were recorded. A photo of the location was taken where any H. oraria bee was found
26 In order to quantify the landscape in which H. oraria and B. angustifolia were found, landscape features and dominate vegetation were surveyed and quantified at both a regional and lan dscape scale. Nine landscape feature s were identified: w ashover sand flat (an area eroded by storm surge or wind, very flat, sparse or no recovering vegetation), grassy dune field (expanses of dunes dominated by U.paniculata, P.amarum, S.patens, or S.mari timum ), grassy dune (well developed dune dominated by grass cover of primarly U.paniculata, P.amarum, S.patens, or S. maritimum ), mixed grassy/woody dune (well developed dunes with a mix of grass and woody species), woody scrub dune (well developed dunes dominated by woody species), herbaceous interdune (areas between dunes vegetated by herbaceous species), bowl without swale (mostly low lying dunes with irregular and undulated margins that snake along edges of non vegetated swales), swale ridge (mostly lo w lying dunes with irregular and undulating margins that snake along the edges of swales), and swale (densely vegetated ephemeral wetlands) ( Branch et al. 201 1) The length of each landscape feature along the transect was recorded along with the dominant plant species and tota l percent cover. Also, the location and length along the transect of each patch of B. angustifolia was measured The width of the entire patch and distance to the next patch including patches outside the transect was determined to the nearest meter using paces and a measuring tape. B. angustifolia plants and flower density in each patch was determin ed using meter square quadrat s placed randomly within each patch; the number of quadrats used was in proportion to the area covered by the patch. Using a d igital elevation model from USGS Lidar data the average elevation of a site and each B. angustifolia patch was calculated. The presence or absence of
27 foredunes and the average and maximum foredune height (m) at a site were also determined. ArcMap was use d to calculate the maximum and minimum distances of each site to the Gulf of Mexico and the bay. Also, the distance from the center of each B. angustifolia patch to the Gulf of Mexico and the bay, if present, was calculated. Using historic hurricane trac ks from 1990 to 2012 from the National Oceanic and Atmospheric Administration (NOAA) the distance of each site and B. angustifolia patch to the nearest storm landfall location, category of storm, and the dista nce to the nearest category three hurricane lan dfall locations were recorded. Statistical A nalyses The area sampled was calculated separately for all coastal sites (barrier islands and coastal mainland) that had B. angustifolia present and all coastal sites where B. angustifolia was absent. The are a sampled for non coastal mainland was also calculated. Additionally, t he ecological density of H. oraria was calculated using the number of bees counted per area of B. angustifolia sampled. The means of each of the following B. angustifolia characteriza tion variables was calculated: nearest patch distance, flower density, plant density, patch area, and number of patches per site. These values were calculated per site, per location (barrier island, coastal mainland, and non coastal mainland), and o verall across all study sites To determine dominant community types the sampling sites were grouped by similar characteristics via hierarchical clustering using Sorensen (Bray Curtis) for the distance measure and Flexible Beta for the linkage method in PC ORD ( Figure 2 4). This combination was recommended by McCune and Grace (2002) for ecological data. Sorensen (Bray Curtis) was used because it retains sensitivity in more heterogeneous data sets and gives less weight to outliers ( McCune and Grace 2002 ). In dicator species
28 analysis was performed to decide where to prune the dendrogram or how many clusters are optimal. Four community types were selected as the best level of classification. Each group was formed by environmental variables and can be identifie d by species composition (Table 2 1) These groups were then used to represent dominant community types in regional scale species distribution models. A classification tree is a non parametric species distribution model that is often used to handle compl 2000). Classification trees split the data into two groups that are mutually exclusive in terms of each predictive variable (as measured by the Gini impurity index). The goal is a set of 2000). Often this goal is unrealistic and a tree is built until the data cannot be divided any furt her. Four analyses were run in Salford S (Salford RT software, SPM Modeler 7.0; 2013). Two analyses were run using the presence or absence of H. oraria as the dependent variable, the first analysis determined habitat use at a regional scale and the second tree quantified habitat patch use on a landscape scale. Similarly two analyses were run using the presence or absence of B. angustifolia as the target variable on a regional and landscape scale. H. Oraria Habitat U se Regional scale: To develop a model for predicting H. oraria habitat use on a regiona software, SPM Modeler 7.0; 2013) with H. oraria presence or absence at a sampling site as the dep endent variable. Sixteen sites determined to be habitat (i.e. coastal sites whic h had B. angustifolia present ) were included. The 30 predictor variables used
29 were: location of the site (mainland coastal, or barrier island), the presence/absence of a foredune for at least 5% of the coastline of site (a foredune is defined as any dune feature at least two meters tall situated closer than 100 meters from the gulf), the category of the storm with the closest landfall since the 1990s (Tropical depression:SD, Tropical storm: TS, Category 1 hurriane:H1, Category 2 hurricane:H2, Category 3 hu rricane:H3), density of all other pollinators (insects/ m 2 ) excluding H. oraria ), average density of B. angustifolia (plants/m 2 ) average density of B. angustifolia flowers (flowers/ m 2 ), the total number of B. angustifolia patches in a site, average size o f B. angustifolia patches ( m 2 ), total area covered by all B. angustifolia patches ( m 2 ), average distance between B. angustifolia patches (m), average percent cover of B. angustifolia in a patch, minimum and maximum distance to the bay and the gulf (m), tem perature during sampling (C), average height of foredunes (m), maximum foredune height (m), average elevation (m), maximum elevation (m), minimum elevation (m), wind speed during sampling (mph), average gusts during sampling (mph), total cover of washover features (% cover), distance from the center of the site to the nearest named storm event since 1990 (km), distance to the nearest category 3 hurricane landfall since 1990, percent cover of group 1 community type, percent cover of group 2 community type, percent cover of group 3 community type, and percent cover of group 4 community type (Table 2 1). A V fold cross validation method was used on this model because the sample size was too small to withdraw an independent test sample. Using 100% of the data a maximal, exploratory tree was grown and no pruning was performed. Then (by setting V=10) 10 cross validation sub trees were grown on a random 90% of the data, and tested against the remaining 10%. The error rates of the sub trees were combined
30 and map ped to the exploratory tree. The maximal tree was then pruned using the error estimates from each node. The percent of sites correctly classified was used as an overall measure of predictive success. The receiver operating characteristic (ROC) curve was also recorded to measure the accuracy of the model, or how well they separate the data into groups with and without H. oraria or B. angustifolia. An ROC score of 1 indicates perfect predictive accuracy where a score of 0.5 is the predictive accuracy of ra ndom guess. Landscape scale: The second analysis was run to quantify H. oraria foraging habitat use on a landscape scale. The differences between B. angustifolia patches where the bee was present and the patches wher e bee was absent were analyzed to de termine foraging habitat. For this analysis there were eight predictor variables: the landscape feature in which the patch was found, percent cover of B. angustifolia distance to the nearest patch (m), distance to the gulf (m), density of B. angustifolia (plants/ m 2 ), density of B. angustifolia flowers (flowers/ m 2 ), elevation ( m ), and the presence/absence of each feature within 10 meters of the patch (features = washover, grassy dune field, grassy dune, scrub dune, swale, scrub, flatwoods, road, mixed gras sy/woody dune, herbaceous interdune, swale ridge, opening in scrub, and bowl). A series of classification models were run on 2/3 of the original data (learn data) and cross validated with the remaining 1/3 (test data) to determine the misclassification ra te for each tree. The trees were pruned by minimizing misclassification rate. The percent correctly classified was used as an overall measure of predictive success. The receiver operating characteristic (ROC) curve was also recorded to measure the accu racy of the model.
31 B. Angustifolia Habitat U se Regional scale: To develop a model for predicting B. angustifolia habitat use on a regional scale a classification tree analysis was run in CART with B. angustifolia presence or absence at a sampling site as the dependent variable. The predictor variables for the landscape scale analyses are the same as for site suitability of H. oraria excluding the B. angustifolia plant and patch characteristics leaving 23 explanatory variables: location of the site (mainlan d inland, mainland coastal, barrier island), the presence/absence of a foredune for at least 5% of the coastline of site (a foredune is defined as any dune feature at least 2 meters tall situated closer than 100 meters from the gulf), the category of the s torm with the closest landfall since the 1990s (SD, TS, H1, H2, H3), density of all pollinators (insects/m 2 excluding H. oraria ), minimum and maximum distance to the bay and the gulf (m) (or the distance to each body of water to the closest and furthest ed ge of each site), temperature during sampling (C), average height of foredunes (m), maximum foredune height (m), average elevation (m), maximum elevation (m), minimum elevation (m), wind speed during sampling (mph), average gusts during sampling (mph), to tal cover of washover features ( % ), distance from the center of the site to the nearest named storm event since 1990 (km), distance to the nearest category 3 hurricane landfall since 1990, total cover of group 1( % ), total cover of group 2( % ), total cover o f group 3 ( % ), and total cover of group 4( % ). A gain, a V fold cross validation method was used on this model with 10 cross validated sub trees. The percent of patches correctly classified was used as an overall measure of predictive success. The receiver operating characteristic (ROC) curve was also recorded
32 Landscape scale: The second analysis was run to quantify B. angustifolia habitat use on a landscape scale Patches of B. angustifolia documented along each transect were used as habitat patches and using ArcMap, a total of 450 random points were generated across all transects excluding known B. angustifolia patches to represent non habitat units. For this analysis there were five predictor variables: the landscape feature in which the patch was foun d, distance to the gulf (m), distance to the bay (m), elevation (m), and the presence/absence of each feature within 10 meters of the patch (features = washover, grassy dune field, grassy dune, scrub dune, swale, scrub, flatwoods, road, mixed grassy/woody dune, herbaceous interdune, swale ridge, opening in scrub, and bowl). A classification tree was run to examine habitat preference of Balduina angustifolia Again, the model was run on 2/3 o f the original data and cross validated with the remaining 1/3 to determine the misclassification rate for each tree. The percent correctly classified was used as an overall measure of predictive success. The receiver operating characteristic (ROC) curve was noted to determine model accuracy Results Comparing sites searched in the 1993 1994 field seasons to sites searched in the 2012 field season resulted in no differences in the range or overall regional distribution of H. oraria In both studies H. oraria was only found in coastal areas where B. angustifolia was p resent. In the 2012 field season H. oraria was absent from Camp Helen State Park but was found during the survey of Perdido Key State Park. H. oraria was also absent at all point count surveys. B. angustifolia was present on 0.16 ha or 10% of the non coa stal mainland sites where H. oraria was absent Also, H. oraria was absent o n the 6.57 ha or 82% of barrier island and coastal mainland (coastal) sites
33 surveyed without B. angustifolia. (Table 2 2) Although present outside of sampled transects, H. orari a were not found within transects established on either barrier island within GINS and UWF sties, and only three bees were found in EA FB sites on Santa Rosa barrier i sland. In contrast, a total of 44 H. oraria were counted both within transects on coastal mainland site s, and only the site on Big Lagoon State Park was absent of the bee during the surveys (Table 2 3). Coastal mainland sites had more B. angustifolia patches, the patches covered more area, and the host plant and flower density were higher th an on barrier island sites. T he estimated density of H. oraria in sampled B. angustifolia patches situated in coastal sites was 39 bees/ha contrasting with an average of 2 bees/ha in sampled B. angustifolia patches located o n barrier islands (Table 2 3). H. Oraria Habitat U se Regional scale: The analysis of H. oraria habitat use on a regional scale yielded a tree with two branches (Figure 2 5) which fits the data well while minimizing the misclassification error rate. (Table 2 4). The number of B. an gustifolia patches per site was identified as the most important variable predicting H. oraria presence. The distance between patches of B. angustifolia (m), total area covered by B. angustifolia patches (m 2 ), maximum distance to the bay (m), plant commun ity group 3 (% cover), and the minimum distance to the bay (m) were also identified as important variables in that order (Table 2 5). These variables are important but do not show up as a splitter in the classification tree, indicating they are surrogate splitters, or back up splitters that could be used when the primary splitter is missing. The classification tree predicts that H. oraria will be present when there are at least 12 patches of B. angustifolia at a site, which was correct 85% of the time. T he tree also predicts that if there are less than 12
34 patches at a site then the area is not used by H. oraria, this prediction was correct 67% of the time (Table 2 4). No other variables offered further explanation of H. oraria presence or absence at a si te. The entire tree correctly classified 75% of cases. The relative cost, which measures the performance of the test sample scaled with 1.00 being random and 0 a perfect scenario, is 0.49 which is more than would be expected by chance. However, the ROC is only 0.5 which indicates a predictive performance no better than chance. This discrepancy occurs because classification rates are dependent on the distribution of observation s in the classification tree. Therefore, the uneven number of sites with H. or aria absent versus sites with H. oraria present can cause some inaccuracies in measuring the predictive success of the model. An ROC curve calculates the sensitivity and specificity for each cutoff value and can therefore account for inequalities in observ ations. In conclusion, this is not a strongly predictive model but the relationships laid out in this tree are notable and warrant further investigation. Landscape scale: The analysis of H. oraria habitat use of B. angustifolia patches at a landscape sc ale yielded nine classification trees. The tree with five branches was (Figure 2 6) optimal because it fit the data well with the lowest misclassification rate (Table 2 6). This tree identifies B. angustifolia patch size (m 2 ), B. angustifolia density with in the patch (plants/m 2 ), the landscape in which the B. angustifolia patch was located, the presence of a swale in the surrounding environment, distance to the Gulf of Mexico (m), and the density of B. angustifolia flowers (flowers/m 2 ) as important variabl es in that order. The landscape feature and B. angustifolia flower density were important but were not splitter s in the classification tree (Table 2 7). The classification tree predicts that B. angustifolia patches less than or equal to 111.5m 2
35 were not used as foraging habitat by H. oraria, this prediction was correct 100% o f the time. If the patch size was greater than 111.5 m 2 and there was a swale found within 10 meters of the patch, the patch was unused by foraging H. oraria ; again, this prediction was correct 100% of the time. The model predicts H. oraria used foraging habitat if the patch was greater than 111.5m 2 had no swale within 10 meters, wa s no further than 475.37 meters from the Gulf of Mexico, and had a density of more than 2.5 B. angusti folia plants per m 2 This prediction was correct 74% of the time. Overall, the tree correctly classified 75.5% of the cases at the relative cost of 0.447, which is more than would be expected by chance (Table 2 6) The ROC value is 0.78 which indicates a good predictive accuracy. B. Angustifolia Habitat U se Regional scale: The analysis of habitat use by B. angustifolia at a regional scale yielded two possible classification trees. A tree with two branches (Figure 2 7) is optimal because it minimized the miscl assification error rate and fit the data well (Table 2 8). This tree identified minimum distance of the site to the bay as the most important variable. The maximum elevation (m), cover of group 4 communities (%) the maximum distance to the bay (m), the area covered by washover sand flats, and the cover of group 2 communities were also identified as important variables in that order (Table 2 9). The optimal classification tree predicts B. angustifolia was present when the site was greater than 25 me ters from the bay at its closest point and absent if the site was 25 meters or nearer to the bay, these predictions are correct 100% of the time. No other variable offered further explanation of H. oraria presence or absence at a site. The entire tree co rrectly classified 75% of cases. (Table 2 8) The relative cost
36 measuring the model performance was 0.49 which is more than would be expected by chance. The ROC of 0.92 provides further evidence to accurate model performance. Landscape scale: The analysi s of habitat use by B. angustifolia at a landscape scale yielded 28 possible classification trees. A tree with six branches was chos en (Figure 2 8). This tree did not have the lowest misclassification error rate (Table 2 10) but still fit the data well a nd had a low misclassifi cation rate; the optimal tree was overly complicated for practical predictiv e purposes. Furthermore, it contained end nodes created from a limited number of cases leading to uncertainty of its predictions. The six branch tree chos en for this model identified landscape feature, washover sand flat p resence in the surrounding 10 m B. angustifolia patch size (m 2 ), distance of the patch from the gulf (m), presence of a swale in the surround ing 10 m presence of a herbaceous interdunal area and presence of a mixed grassy/woody d une in the surrounding 10 m as important variables in that order. The chosen tree did not include patch size, presence of a herbaceous interdunal area and presence of a mixed grassy/woody d une in the surrounding 10 m but all could serve as surrogate splitters in the place of primary splitters (Table 2 11). The cho sen classification tree predicted a habitat patch situated within an herbaceous interdunal area, swale or washover sand flat would not be used by B. a ngustifolia and this prediction was correct 88 % of time. The model indicated that a habitat patch located in a bowl, flatwoods, grassy dune, grassy dune field, mixed grassy/woody du ne, scrub dune, scrub opening, or swale ridge, without a washover sand fla t in the surrounding 10 meters was used by B. angustifolia. If the patch was further than 110 m inland from t he Gulf of Mexico the habitat was suitable but if the patch is located less than or equal to 110 m inland from
37 the gulf then it was predicted to be unsuitable for B. angustifolia. The first prediction was correct 70% of the time and the second prediction was correct 65% of the time. Branch five predicted the habitat patch located in a bowl, grassy dune, grassy dune field, scrub dune, or scrub ope ning situated further than 10 meters from the nearest washover sand flat with no swale present is unsuitable for B. angustifolia This prediction was correct 83% of the time. L astly, the model predicted a patch found in a bowl, flatwoods, mixed grassy/wo ody dune, or swale ridge can still be suitable for B. angustifolia even with a washover sand flat in the surrounding area (69% of the cases in this branch classified correctly). Overall, the tree correctly classified 70.9% of the cases at the relative cos t of 0.528 which is more th an would be expected by chance (Table 2 10) The ROC indicates an accurate model performance with a value of 0.77. Discussion The original study of the bee, its western range was det ermin ed to spread from Horn Island, Mississ ippi east to Saint Andrews State Park in Panama City, Florida (Cane et al. 1996) Entomologists searched for H. oraria both inland and along the immediate coast along the northern Gulf Coast (Cane et al. 1996). The bee was not detected at any non coastal sites even when B. angustifolia was abundant, nor was it observed anywhere without B. angustifolia. A comparison of sites searched in the 1993 1994 field seasons to sites searched in the 2012 field season found no differences in the range or overall regional distribution of the bee. H. oraria density estimates suggest the bees are influenced by variability in B. angustifolia characteristics among coastal sites and B. angustifolia characteristics vary with occurrence on or off barrier islands and with the dominant plant com munities on site. Coastal mainland habitat support ed higher bee densities than barrier island
38 foraging habitat and had more B. angustifolia patches, the patches covered more area and the host plant and flower density were higher than in barrier island sit es. The average number of patches on a barrier island site was 10 where the average number of patches on a coastal mainland site was 36 patches per 120,000 m 2 There was also variability in plant and bee quantity among coastal mainland sites in response to the increase of habitats unsuitable to B. angustifolia such as dense scrub and flatwoods with dense midstory of shrubs. These scrub habitats are densely wooded except for small openings between shrubs and B. angustifolia only grows in the small openings in the woody vegetation. Big Lagoon and St. Andrews State Parks both ha d high numbers of B. angustifolia patches but low patch areas when compared with other state park s sampl ed because these parks consisted primarily of rosemary scrub and scrub dunes. Coastal sites in other state park s consist ed of more washover sand flats and grassy dunes and less scrub than Big Lagoon and St. Andrews State Parks. However, B. angustifolia plant and flower density was comparable at these sites with smaller total patch areas to sites with greater total patch area. Furthermore, t he total area of B. angustifolia within a state park was not always the highest where abundance of H. oraria was the greatest Topsail Hill ha d the most H. oraria but the smallest total area of the host plant. This result ed in a significantly higher ecological density (bees/ B. angustifolia area) of H. oraria than at the other similar sites. In summary, the B. angustifolia and the density of host plants was more descriptive of a site supporting a higher abundance of H. oraria than total patch area Observations in 2012 support o riginal conclusions that H. oraria is only found in coastal habitats where B. angustifolia is abundant and it appears the bee does not
39 occupy non coastal mainl a n d areas even if its host plant is plentiful (Cane et al. 1996) A recent study on pollinator distribution reported about 11% of bee species found in open sites and 2% of forest bee species were specialists and concluded open habitat bee species have narrower diets than f orest bee species (Grundel et al. 2010). H. oraria appear s to uphold these findings and does not occupy forested habitats. Further work would be needed to determine if non coastal mainland sites are suitable but there are barriers stopping dispersal inla nd fro m coastal sites Most coastal sites are separated from inland open pine forest communities supporting B. angustifolia by dense scrubby vegetation and/or roads and buildings. The surrounding plant communities have a history of fire; however, with in creased development and fragmentation, fire has long been excluded from these areas allowing a thick mid story to develop (Jokela et al 2006). F ire is also a positive predictor for specialist bee abundance ( Grundel et al. 2010) Recent acquisit ions by Florida State land agencies land that includes coastal to forested habitats uninterrupted by major development offers hope that better land management in the future will increase connect ivity between coastal and non coastal sites. For now, it can be concluded that H. oraria as a specialist bee, is restricted to foraging in open coastal and barrier island habitats where B. angustifolia is abundant with no major changes to their range in Florida occurring in the last 18 years. Disturbance has b een shown to greatly influence the spatial distribution of populations of many species of bee (Potts et al. 2003a, Williams and Kremen 2007, Russell et al. 2005, Winfree et al. 2007) and H. oraria distribution and habitat use appears to be driven by a mode rate amount of disturbance. H. oraria show ed a preference for sites with more B. angustifolia patches and patches of a substantial size
40 indicating the bees are attracted to a site by pollen and nectar rewards (Potts et al. 2004, Larsson and Franzen 2007). It is therefore valuable to identify the habitat needs for this plant in order to understand comprehensive conservation needs for H. oraria B ee species that specialize are more likely to use foraging habitat that consists of open areas rather than forests ( Grundel et al. 2010) T his association is linked to the presence of disturbance dependent plants, often of the family Asteraceae, preferred by specialists and found in disturbed sandy soils (Swink and Wilhelm 1994). H. orari host plant B. angustifolia soils, not subjected to prolonged flooding and sand movement For example, the bees and the plant were predicted to be absent from certain landscape feature s s uch as swales, which are subject to periodic flooding, and any feature with sparse vegetation, such as interdunal ar eas, or washover sand flats, which experience constant sand movement as windy conditions dominate and inadequate vegetative roots provide li ttle to hold the sand particles in place Both environments may be inhospitable for annual plants. However, B. angustifolia was predicted to use the more stable fea tures of the coast when there was a washover sand flat nearby indicating limited disturban ce is preferable to no disturbance. Hurricanes may cause the most drastic disturbance throughout the range of the H. oraria but this analysis did not indicate that the proximity to and strength of recent storms directly impacted H. oraria or B. angustifo lia distribution. However, the undeniable impact of these storms on the vegetation and geomorphology of these landscapes, especially with respect to dune development, points out an indirect relationship between storm activity and H. oraria and B. angustif olia (Pries et al. 2008 ). For example, bee density on barrier islands was much
41 reduced when compared to coastal mainland, this may be a result of increased storm activity and/or susceptibility of the barrier islands within this study. In conclusion, some disturbance such a low intensity fires and moderate openings in vegetation, create useable habitat for B. angustifolia and H. oraria. Distance to the Gulf of Mexic o and to the bay (if present) was an important predictor for habitat suitability of both s pecies at both regional and landscape scales. For b arrier islands, this distance was important as it indicates the overall stability of the landscape. Generally, a wider portion of a barrier island is less susceptible to storm overwash and contain larger dunes and more established vegetation. For coastal sites attached to the mainland, the distance to the gulf was an important indicator of increased woody vegetation from scrub dunes transitioning into scrub or flatwood s forest communities. It would be i nteresting to run this analysis separately for barrier island and mainland sites to create more site specific distribution predictions. Within a landscape H. oraria was predicted to use foraging patches less than 475 meters from the gulf and B. angustifol ia used habitat greater than 110 meters from the gulf. These two models can be combined to predict that land between 110 475 meters from the Gulf of Mexico is the interception of suitable habitat for H. oraria as well as B. angustifolia making this area critical foraging habitat for H. oraria
42 Table 2 1. Indicator species of the 4 optimal clusters from the cluster analysis p erformed based on environmental variables. Cluster Species* Group1 HYPSPP, ILEGLA, POLGRA, POLPOL, SCHMAR Group2 CERERI, CHR PAU, CHRSPP, LICMIC, MAGGRA, QUEGEM, QUEMYR Group3 ILEVOM, MAGVIR, PINELL, PINPAL, QUEINC, QUELAE, TAXASC Group4 IVAIMB, PANAME, UNIPAN *CERERI = Ceratiola ericoides CHRPAU = Chrysoma pauciflosculosa CHRSPP = Chrysopsis species HYPSPP = Hypericum spec ies ILEGLA = Ilex glabra ILEVOM = Ilex vomitoria IVAIMB = Iva imbricata LICMIC = Licania michauxii MAGGRA = Magnolia grandiflora MAGVIR = Magnolia virginiana PANAME = Panicum amarum PINELL = Pinus Elliotii PINPAL = Pinus palustris POLGRA = Polygonella gra cilis POLPOL = Polygonella polygama QUEGEM = Quercus geminata QUEINC = Quercus incana QUELAE = Quercus laevis QUEMYR = Quercus myrtifolia SCHMAR = Schizachyrium maritimum TAXASC = Taxodium ascendens UNIPAN = Uniola paniculata
43 Table 2 2 Total sam pling effort for H.oraria in coastal and non coastal mainland areas with and without host plant B.angustifolia. Coastal Area Sampled in 2012 Field Season (ha) Non coastal Mainland Area Sampled in 2012 Field Season (ha) B. angustifolia present B. angus tifolia absent B. angustifolia present B. angustifolia absent Hectares Sampled 1.432 6.567 0.1586 1.441 Percent of Total Area (%) 18 82 10 90 H. oraria Count (no.) 47 Not Found Not Found Not Found Density (no./ha) 7.88 Not Found Not Found Not Found
44 Table 2 3. Summary of H. oraria count data and B. angustifolia characterstics by location, unit and site. Gulf of Mexico barrier islands are narrow strips of sand running parallel to the mainland. Coastal non island areas consist of the fi rst 500 meters of mainland closest to the Gulf of Mexico. The non coastal mainland sites included in this study consist of Mesic and Scrubby Flatwoods and Sandhill communities located further than 0.5 km from the Gulf of Mexico. Park units include Gulf I slands National Seashore (GINS), University of West Florida (UWF), Eglin Air Force Base (EAFB), Florida State Parks, and Florida State Forests. Location Park Unit Site (120,000m 2 ) H. oraria count (no.) B. angustifolia patch count (no.) H. oraria ecologica l density (no./ha) Total B. angustifolia patch size (m 2 ) B. angustifolia density (no./m 2 ) Flower density (no./m 2 ) Average distance between B.angustifolia patches (m) Barrier Island Perdido Key (GINS) 1 0 17 0 69 14 33 23 2 0 0 0 0 0 0 0 3 0 7 0 929 2 28 52 Fort Pickens (GINS) 1 0 10 0 496 14 45 28 2 0 15 0 2281 10 39 26 3 0 0 0 0 0 0 0 Santa Rosa Island (UWF) 1 0 30 0 908 5 34 21 Santa Rosa Island (GINS) 1 0 10 0 1144 9 51 23 2 0 0 0 0 0 0 0 Santa Rosa Island (EAFB) 1 0 5 0 485 10 54 20 2 1 24 16 602 8 36 22 3 2 20 14 1219 6 24 27 4 0 5 0 398 11 24 29 5 0 0 0 0 0 0 0 Barrier Island Total Sum 3 143 30 8530 89 366 270 Average 0 10 2 609 6 26 19
45 Table 2 3. cont. Location Park Unit Site (120,000m 2 ) H. oraria count ( no.) B. angustifolia patch count (no.) H. oraria ecological density (no./ha) Total B. angustifolia patch size (m 2 ) B. angustifolia density (no./m 2 ) Flower density (no./m 2 ) Average distance between B.angustifolia patches (m) Coastal Mai nland Big Lagoon State Park 1 0 50 0 185 6 28 21 Henderson Beach State Park 1 14 13 42 6448 9 18 18 Topsail Hill State Park 1 15 35 114 529 9 32 12 Grayton Beach State Park 1 12 15 43 3825 13 39 20 Deer Lake State Park 1 1 23 10 2840 13 48 19 St. Andrews State Park 1 2 81 25 189 8 37 6 Coastal Mainland Total Sum 44 217 234 14016 59 202 96 Average 7 36 39 2336 10 34 16 Non coastal Mainland Eglin Air Force Base 6 0 2 0 1806 10 23 15 Grayton Beach State Park 2 0 10 0 34 3 14 27 Deer Lake Stat e Park 2 0 4 0 13 2 13 7 Point Washington State Forest 1 0 29 0 1806 8 42 21 Non coastal Mainland Sum 0 45 0 3659 23 91 70 Average 0 11 0 915 6 23 17 Total Sum 47 405 264 26205 171 660 437 Average 2 17 11 1092 7 27 18
46 Table 2 4. Prediction succes s (of the test data) for the classification tree in predicting site habitat use by H.oraria at a regional scale Class 0 indicates sites where H.oraria was not detected, class 1 indicates sites where H.oraria was present. The error (%) is the percent of s ites misclassified, and the cost of each misclassification is of equal weight. Actual Class Sites within each class Sites classified correctly (%) Sites misclassified (%) Class 0 (n=4) Class 1 (n=12) 0 3 66.67 33.33 2 1 1 13 84.62 15.38 2 11 Average* 75.64 Overall Correct** 81.25 *Average = average model prediction success for the test sample, or the average percent of sites classified correctly. **Overall correct = overall model prediction success for the test sample, or the total percent of sites classified correctly regardless of class.
47 Table 2 5 Variables important for predicting the site habitat use of H.oraria at a regional scale. The relative importance values are specific for the given model from 0 100. Var iable Relative Importance Score Total number of B.angustifolia patches 100 Average distance (m) between B.angustifolia patches 72 Total area (m 2 ) covered by B.angustifolia patches. 72 Maximum distance (m) of the site to the bay 68 Community group 3 ( % cover) 60 Minimum distance (m) of the site to the bay 56 Community group 3: Ilex vomitoria Magnolia virginiana Pinus Elliotii Pinus palustris Quercus incana Quercus laevis, Taxodium ascendens
48 Table 2 6. Prediction success (of the tes t data) for the classi fication tree in predicting habitat patch use by H.oraria at a landscape scale Class 0 indicates patches where H.oraria was not detected, class 1 indicates patches where H.oraria was present. The error (%) is the percent of patches m isclassified, and the cost of each m is sclassification is of equal weight. Actual Class Patches within each class (n) Patches classified correctly (%) Patches misclassified (%) Class 0 (n=71) Class 1 (n=27) 0 93 75.27% 23.73 70 23 1 5 80.00% 20.00 1 4 Av erage* 77.63% Overall correct* 75.51% *Average = average model prediction success for the test sample, or the average percent of patches classified c orrectly **Overall correct = overall model prediction success for the test sample, o r the total percent of sites classified correctly regardless of class.
49 Table 2 7 Variables important for predicting the habitat patch use of H.oraria at a landscape scale. The relative importan ce scores are specific for the given model from 0 100. Variable Relative Importance Score B.angustifolia patch size (m 2 ) 100 B.angustifolia density (no./m 2 ) 63 Landscape feature 46 Presence of a swale in the surrounding 10 meters 39 Distance of the patch to the Gulf of Mexico (m) 36 Density of B. angustifolia flowers (no./m 2 ) 34 Cover of B.angustifolia (%) 8
50 Table 2 8. Prediction success (of the test data) for the classification tree in predicting site habitat use by B. angustifolia at a regional scale Class 0 indicates sites wher e B. angustifolia was not detected, class 1 indicates sites w here B. angustifolia was present. The error (%) is the percent of sites misclassified, and the cost of each misclassification is of equal weight. Actual Class Sites within each class (n) Sites cla ssified correctly (%) Sites misclassified (%) Class 0 (n=71) Class 1 (n=27) 0 4 100.00 0.00 4 0 1 20 95.00 5.00 1 19 Average* 97.50 Overall correct** 95.83 *Average = average model prediction success for the test sample, or the average percent of sites classified correctly. **Overall correct = overall model prediction success for the test sample, or the total percent of sites classified correctly regardless of class.
51 Table 2 9 Variables important for predicting the si te habitat use of B.angustifolia at a regional scale. The relative importance s cores are specific for the given model from 0 100. Variable Z Relative Importance Score Minimum distance to the bay (m) 100 Maximum elevation (m) 90 Community Group 4 82 Ma ximum distance to the bay 79 Presence of a washover sand flat in the surrounding 10 m. 74 Community Group 2 67 Z Variable Community Group 4: Iva imbricata, Panicum am a rum Uniola paniculata Community Group 2: Ceratiola ericoides, Chrysoma paucifloscul osa, Chrysopsis spp., Licania michauxii, Magnolia grandiflora, Quercus geminata, Quercus myrtifolia
52 T able 2 10. Prediction success (of the test data) for the classification tre e in predicting B.angustifolia habitat patch use at a landscape scale Class 0 indicates patches where B. angustifolia was not detected, class 1 indicates patches w here B. angustifolia was present. The error (%) is the percent of patches misclassified, and the cost of each misclassification is of equal weigh t. Actual Class Patches within each class Patches classified correctly (%) Patches misclassified (%) Class 0 (n=71) Class 1 (n=27) 0 160 62.50 37.50 100 60 1 98 84.69 15.31 15 83 Average* 73.60 Overall correct* 70.93 *Average = average model prediction success for the test sample, or the average percent of patches classified correctly. **Overall correct = overall model prediction success for the test sample, or the total percent of patches classified correctly regardless of class.
53 Table 2 11 Variables important for predicting the habitat patch use of B.angustifolia at a landscape scal e. The relative important scores are specific for the given model from 0 100. Variable Relative Importance Score Presence of a washover s and flat in the surrounding 10 m. 100 B.angustifolia patch size (m 2 ) 84 Distance of patch to the Gulf of Mexico (m) 26 Presence of a swale in the surrounding 10 m. 8 Presence of a herbaceous interdune in the surrounding 10 m. 4
54 Table 2 12. Site surveyed for H.oraria, including the 1993 1994 fi eld seasons (Cane et al. 1996) and 2011 2012 field season s State County Location H.oraria present in 1993 1994? B.angustifolia present 1993 1994? H.oraria present in 2011 2012? B.angustifolia present 2011 2012? Mississippi Jackson Petit Bois Island, Gulf Island National Seashore No Yes Area Not Searched Area Not Searched Mississippi Jackson West & central Horn Island, Gulf Island National Seashore Yes Yes Area Not Searched Area Not Searched Alabama Mobile Dauphin Island No No Area Not Searched Area Not Searched Alabama Baldwin Gulf Shores State Park No Yes Area Not Searched Area Not Searched Alabama Baldwin Ft. Morgan National Monument Yes Yes Area Not Searched Area Not Searched Alabama Ba ldwin Bon Secour National Wildlife Ref., west shore Long Lagoon, south Gator Lake Yes Yes Area Not Searched Area Not Searched Alabama Baldwin Romar Beach (Ala. 182, mile 11) Yes Yes Area Not Searched Area Not Searched Florida Escambia Perdido Key State P ark Area Not Searched Area Not Searched Yes Yes Florida Escambia Gulf Island National Seashore, Perdido Key Unit Yes Yes Yes Yes Florida Escambia Big Lagoon Sate Rec. Area, entrance Yes Yes Yes Yes Florida Escambia Gulf Island National Seashore, Ft. Pic kens, Battery 224, 234 and Cooper on Santa Rosa Island Yes Yes Yes Yes Florida Escambia Santa Rosa Island, 1 mi east of Pensacola Beach Yes Yes Yes Yes Florida Okaloosa Santa Rosa Island, east gate Eglin Air Force Base and site A 7 Yes Yes Yes Yes Flo rida Okaloosa Eglin Air Force Base approx. 6 miles inland Area Not Searched Area Not Searched No Yes Florida Walton Henderson Beach State Rec. Area Yes Yes Yes Yes Florida Walton Blue Mountain Beach Yes Yes Area Not Searched Area Not Searched Florida W alton Grayton Beach State Park Yes Yes Yes Yes Florida Walton Topsail Hill, near coast Yes Yes Yes Yes
55 Table 2 12 continued State County Location H.oraria present in 1993 1994? B.angustifolia present 1993 1994? H.oraria present in 2011 2012? B.angustif olia present 2011 2012? Florida Walton 1.3 mi inland from Grayton Beach State Park No Yes Yes Yes Florida Walton Point Washington State Forest (approx 1.6 miles inland) Area Not Searched Area Not Searched No Yes Florida Walton Deer Lake State Park coast al Area Not Search Area Not Searched Yes Yes Florida Walton Deer Lake State Park (approx. 1 mile inland) Area Not Searched Area Not Searched No Yes Florida Walton West Inlet Beach nr. Seaside No Yes Area Not Searched Area not Searched Florida Walton Cam p Helen State Park Area Not Searched Area Not Searched No Yes Florida Walton North Freeport No Yes Area Not Searched Area Not Searched Florida Bay West Laguna Beach No Yes Area Not Searched Area Not Searched Florida Bay North Fla. 395 (2 mi from coast) No Yes Area Not Searched Area Not Searched Florida Bay Tyndall Air Force Base, east & west ends of Crooked Island, Shell Island No No Area Not Searched Area Not Searched Florida Bay St. Andrews State Park, m from entrance Yes Yes Yes Yes Florida Gulf Mexico Beach No No Area Not Searched Area Not Searched Florida Franklin St. Joseph Peninsula State Park and Cape San Blas No No Area Not Searched Area Not Searched Florida Calhoun St. George Island State Park No No Area Not Searched Area Not Searched Fl orida Walton North Freeport (north Choctawhatchee Bay) No Yes Area Not Searched Area Not Searched Florida Calhoun Clarksville No Yes Area Not Searched Area Not Searched
56 FIGURES F igure 2 1. The geographical range of Hesperapis oraria documented in 1993 1994 field seasons ( C ane et al. 1996 )
57 Figure 2 2. Sites visited, fall 2012, where B.angustifolia was present and H.oraria was either present or absent.
58 Figure 2 3. Sites sampled (with transects), fall 2012, where B.angustifolia and H.ora ria were present.
59 Figure 2 4. Sites grouped into dominant community types via hierarchical clustering using Sorensen (Bray Curtis) for the distance measure and Flexible Beta for the linkage method in PC ORD The optimal number of clust ers was determined to be four after an indicator species analysis yielded the smallest average p value (p = 0.334) and the most number of species with significant p values (13 species).
60 Figure 2 5. Classificati on tree relating probability of site use by H.oraria to landscape features over a regional scale. Categories or numbers on top of each branch indicate the value of the explanatory variable that, when used to split the response variable, leads to maximum h omogeneity of resulting groups. The end of each branch is labeled according to whether H.oraria was present or absent. Numbers at the end of each branch indicate the proportion of observations in that group correctly classified. The total number of obse rvations in each group is listed within parentheses. Total number of B.angustifolia patches < 12 Total number of B.angustifolia 0.67 (4) H.oraria absent 0.85 (12) H.oraria present
61 Figure 2 6. Classification tree relating probability of patch use by H.oraria to landscape features over a landscape scale. Categories or numbers on top of each branch indicat e the value of the explanatory variable that, when used to split the response variable, leads to maximum homogeneity of resulting groups. The end of each branch is labeled according to whether H.oraria was present or absent. Numbers at the end of each br anch indicate the proportion of observations in that group correctly classified. The total number of observations in each group is listed within parentheses.
62 Figure 2 7. Classification tree relating probabil ity of occupancy of B.angustifolia to landscape features over a large scale. Categories or numbers on top of each branch indicate the value of the explanatory variable that, when used to split the response variable, leads to maximum homogeneity of resulti ng groups. The end of each branch is labeled according to whether B.angustifolia was present or absent. Numbers at the end of each branch indicate the proportion of observations in that group correctly classified. The total number of observations in eac h group is listed within parentheses.
63 Figure 2 8. Classification tree relating probability of habitat patch occupancy of B.angustifolia to landscape features. Categories or numbers on top of each branch indicate the value of th e explanatory variable that, when used to split the response variable, leads to maximum homogeneity of resulting groups. The end of each branch is labeled according to whether B.angustifolia was present or absent. Numbers at the end of each branch indica te the proportion of observations in that group correctly classified. The total number of observations in each group is listed within parentheses. Stable Landscape Features: bowl, flatwoods, grassy dune, grassy dune field, mixed grassy/woody dune, scrub dune, scrub opening, swale ridge. Unstabel Landscape Features: herbaceous interdunal area, swale, washover. Landscape Features A: bowl, flatwoods, mixed grassy/woody dune, swale ridge. Landscape Features B: grassy dune, grassy dune field, scrub dune, scrub open.
64 CHAPTER 3 CHANGES IN BALDUINA ANGUSTIFOLIA AND HESPERAPIS ORARIA DENSITY OVER TIME Background In the practice of conservation it is important to identify habitat critical to target species and to understand how changes to habitat impact tar get species survival. Understanding this relationship can be especially important for species occupying dynamic landscapes (Pries et al. 2009) as the availability and quality of habitat can change quickly with stochastic events (Carlsson and K indval 2001; van Horne et al. 1997). Therefore, it is important to examine not only key features that determine spatial distribution or population density, but also how they might change and how those changes might affect the species of interest (Pries et al. 2009). Sufficient foraging habitat is critical for the survival of any species. H. oraria foraging habitat is defined as any area where its preferred host plant B. angustifolia can be found. Threats to H. oraria foraging habitat such as inadvertent damage fro m visitor use and overwash associated with hurricanes could affect the density of B. angustifolia and, therefore, H. oraria By monitoring B. angustifolia patch characteristics and density of both species over time an understanding of the relationships be tween host and bee can be gained to protect and restore critical habitat to better inform population management. The objectives of this chapter are (1) to quantify the changes in B. angustifolia patch size, cover of all vegetation and B. angustifolia pla nt density, flower density, plant height, plant width number of seedheads per plant, and seedhead width from 2011 to 2012 and (2) to quantify changes in H. oraria density in foraging habitat from 2011 to 2012 and (3) to ultimately determine characteristic s of foraging habitat which influence H. oraria density.
65 Density of H. oraria in foraging habitat was determined for three units of the Gulf Islands National Seashore (GINS): Santa Rosa, For t Pickens and Perdido Key (Figure 3 1). The Perdido Ke y u nit is located on Perdido Key, a barrier island located just southwest of Pensacola Florida. Santa Rosa and Fort Pickens u nits are l ocated on Santa Rosa Island southeast of Pensacola Florida. Sampling was conducted in relatively undeveloped areas but occasiona lly in close proximity to roads and, in the Fort Pickens u nit, on manmade earthworks or embattlements. Perdido Key, is about 24 km long and 0.5 km wide. The GINS Perdido Key u nit is located at the eastern most end of Perdido Key (Figure 3 1). Perdido Key has a few well developed foredunes but generally lacks substantial woody secondary dunes. Sampling was conducted close to the park entrance where Perdido Key is wider and where the secondary scrub dunes transition into a scrubby flatwoods with a slash pi ne ( Pinus elliotii) overstory. Sampling was also conducted at the eastern end of the island where grassy dune fields surround the remnant military fortifications of Fort McRee Areas sampled here were located on the sound side of the island behind wet fl atwoods communities. The Fort Pickens u nit, on the western end of Santa Rosa Island (Figure 3 1), had the most manmade structures of the units included in this study. Surrounded by remnants of these structures the host plants are located in grassy dun e fields and swales (ephemeral wetlands) on the gulf side of the island and in flatter, patchy rosemary scrub on the bay side of the island. The Santa Rosa u nit (Figure 3 1) is located about 1.5 miles east of Pensacola Beach. This portion of the island has fragmented foredunes near the Gulf of Mexico
66 and remnant scrub dunes on the sound side of the island but is dominated by undulating grassy dune fields with sandy and herbaceous interd unal areas. On the Santa Rosa u nit the host plants are located on an d around the back dunes as well as in some interdunal areas. The dominant species on these study sites are sea oats ( Uniola paniculata ), bitter panicgrass ( Panicum amarum ), gulf bluestem ( Schizachyrium maritimum ), beach elder ( Iva imbricata ), goldenaster species ( Chrysopsis spp. ), c oastal honeycomb head ( Balduina angustifolia ), and woody goldenrod ( Chrysoma pauciflosculosa) Transects varying in length from 3 84 m, located in 51 B. angustifolia patches were sampled for H. oraria The patches sampled were randomly selected from a near complete survey of all B. angustifolia patches in the GINS units. The number of transects sampled within each unit is proportionate to the number and size of host plant patches within each unit. Within each patch the width, or the baseline, was measured to determine the number of transects needed to sample approximately 20% of the area covered by the patch. This resulted in a total of 58 transects throughout the entire study area Methods Data C ollection The population densi ty of H. oraria was estimated using the perpendicular distance transect method. Data was collected from the 7 th to the 15 th of October in 2011 and 3 rd to the 20 th of October in 2012. This sampling technique required samplers to walk transects at a cons tant pace and mark the location of each bee to the transect. The perpendicular distance (m) of each observed individual bee from the line was measured. The number of B. angustifolia, was recorded in one m 2 quadrats placed
67 systematically (on alternating s ides of the line) every three meters along each transect. Additionally, the total number of flowers in the bottom right quarter of the quadrat was recorded. In December, B. angustifolia seed heads were collected from each patch. A 0.5 m 2 quadrat was pla ced every five meters along the transect within which seeds were collected from randomly selected plants. For the 2012 collection, t he number of seeds per seedhead was counted after the plants were dry. Plant height (cm), width (cm), number of seed heads per plant and width of each seed head collected (cm) were recorded. Lastly, the percent cover of all vegetation in these quadrats was recorded. (NAS) National Oceanic Atmospheric Administration (NOAA) weather station. (Retrieved from wunderground.com) Statistical A nalyses The significance of the change in value of individual predictor variables and bee density between 2011 and 2012 was evaluated with non parametric Wil coxon rank sum tests ( SAS software, Version 9.2; 2008). Because there were no be es found on the Santa Rosa u nit in either 2011 or 2012, analysis was restricted to results fro m Perdido Key and Fort Pickens u nits. Regression analysis (PROC REG) was used t o determine the relationship between the response variable ( H. oraria /ha) and predictor variables: transect length (m) as a proxy for patch size cover of all vegetation (%), B. angustifolia plant density (plants/m 2 ) flower density (flowers/m 2 ) plant he ight (cm) plant width (cm), number of seedheads per plant, and seedhead width (cm). procedure was performed to determine correla tion among predictor variables. The
68 numb er of seeds in each seedhead counted for the 2012 season was us ed to calculate the mean number of seeds per seedhead overall and per GINS unit. P recipitation throughout germination and growing period (March October) was calculated using average daily precipitation, maximum daily precipitation, and t otal average precipitation. The 30 year mean for March through October (1983 2012) for daily precipitation, maximum daily average precipitation, and total average precipitation was calculated. Results B. angustifolia patch size, flower density, and H. ora ria density decreased from 2011 to 2012. Seed head width (cm) increased from 2011 to 2012. The remaining explanatory variables did not chan ge significantly between years (Table 3 1). Median patch size decreased significantly from 2011 (13.5m) to 2012 (3 .0m) (Z= 2.11, 1 d.f., P < 0.0326) (Table 3 2). Median B. angustifolia flower density decreased three fold from 2011 (12/m 2 ) to 2012 (3/m 2 ) (Z= 2.11, 1 d.f., P <0.0336) (Table 3 2). Seed head width (cm) significantly increased from 2011 (0.9 cm) to 2012 (1 .0 cm) (Z=2.82, 1 d.f., P < 0.0038 )(Table 3 2). Mean H. oraria density decreased from 2011 (99/ha) to 2012 (3/ha). Change in H. oraria density cannot be further analyzed because bees were found in 5 of the 51 patches surveyed in 2011 and 2 of the 51 patch es surveyed in 201 2 in the three units of GINS. This resulted in too many zeros in the dataset to determine any pattern through time. The regression analysis determined characteristics of foraging habitat which influence d H. oraria density. However, none of the charact eristics were shown to predict H. oraria density with an acceptable probability in this model (Table 3 3). The overall power of this model was too low to make any predictions on bee density F value
69 procedure ruled out any multicollinearity that could have confounded these results. Data collected from across the range of H. oraria in 2012 better predicted the foraging habitat characteristics for the bee. Models using presence/absence data predict ed B. angustifolia density, and patch size positively correlate d with H.oraria use of the habitat (Chapter 2). The mean number of seeds per B. angustifolia seedhead was 32. On average there were 29 seeds per seedhead at both Santa Rosa Island and Perdido K ey units and 38 seed s per seedhead at Fort Pickens unit. The amount of precipitation between the two years was markedly different (Figure 3 4). Precipitation in 2011 was below average and precipitation in 2012 was above average. Most of the rain in 2012 occurred in events suc h as tropical storm Debby in June, hurricane Issac in August, and an unnamed storm in June. In 2011, the average daily precipitation from March through October (during germination and growth of B. angustifolia ) was 0.33 cm, the maxi mum precipitation experienced in one day was 11.30 cm (July 17th) and the total precipitation was 71.30 cm In 2012, the average daily precipitation during germination and growth from March throug h October doubled to 0.61 cm and the maxim um daily precipi tation was 33.35 cm occurring on June 9 th and the total pre cipitation was 141.55 cm The 30 year mean for Ma rch through October (1983 2012) for daily precipitation was 0.46 cm maximum daily average was 12.60 cm and the total averag e precipitation was 112.29 cm Discussion Specialist bee species have been shown to vary in their population size from year to year correlating positively with variation in their host plant (Minckley et al. 1994 ). The results from this study show patches of B. angustifo lia decreased in size and flower
70 density while H. oraria density also decreased from 2011 to 2012. Therefore, the most general conclusion to be drawn is that smaller patches and fewer flowers of B. angustifolia correlate with fewer H.oraria The first year of sampling was drier than average followed by a wetter than average year Therefore, a year with above average precipitation correlates with smaller B. angustifolia p atches and lower flower density as well as fewer bees. T his relationship indicate s that the heavy rainfall events of 2012 flooded some landscape features containing B. angustifolia These findings are supplemented by regional scale analyses of chapter 2. In the coastal landscapes where H. oraria was found, the distribution and reliab ility of B. angustifolia varied with landscape features. B. angustifolia was found in bowls, flatwoods, grassy dunes, grassy dune fields, mixed grassy and woody dunes, scrub dunes, opening s in scrub, and on the periphery of swales (Chapter 2). Swales flo od periodically and were observed to have flooded significantly and often in 2012. B. angustifolia was observed to have established early in the season in swales but died where flooded before reaching maturity and producing flowers (personal observation). Models from chapter 2 show B. angustifolia rarely grew in swales but was often found on the periphery of swales. The models from chapter 2 were run on data collected during bee foraging and therefore, model B. angustifolia distribution after reaching ma turity and in full flower Furthermore, these models show that although B. angustifolia was commonly found on the periphery of swales H. oraria was rarely found within 10 meters of a swale. All these finding indicate that swales were unreliable habitat f or B. angustifolia growth and H. oraria foraging.
71 H. oraria might use patches of B. angustifolia on the periphery of swales as a foraging resource in a dry year but not in a wet year. Lack of interannual reliability indicates that long term trends could show H. oraria favoring drier landscape features. Studies into the spatial distribution of bees have found their populations to be patchily distributed in the landscape even when resources are broadly d istributed (Minckley et al. 1999 ). This has been att ributed to the interannual reliability of flowers, as bees prefer to forage in areas with dependable resources from ye ar to year (Minckley et al. 1999 ). Previous studies indicate that pollinators occur more frequently at sites where host plants flower mor e predictably from year to year (Herrera 1998; Minckley et al. 1999 ). By tracking B. angustifolia patches over time some patches might be determined as consistently better for H. oraria foraging than others. Changes in patch characteristics indicate the reliability of flowering of B. angustifolia varies between years. The decrease in patch size point s to the uncertainty of a patch persisting over time. Classical metapopulation theory explains occupancy as a function of patch area and connectivit y (Hanski 1999, Franzen and Nilson 2013). Therefore, Bee populations can also change over time for reasons unrelated to variable weather patterns. A study on population dynamics of orchid bees ( Euglossinni) found great variability in population size over two decades; bee abundance varied up to 14 fold with no clear trend or ecological explanation (Roubik 2001 ). Another study by Franzen and Ni llson (2013) examined the population dynamics of solitary bee Andrena humilis and concluded that pollen resources are critical for population persistence and that weather played no role in the population fluctuations. They observed such drama t ic population fluctuations that
72 they could not attribut e the change to any direct or indirect effect of either resources or we ather. They also ruled out the possibility that the fluctuations were a result of the An earlier study by Franzen and Nilsson (2010) drew similar conclusions looking at the solitary bee Andrena hattorfiana at a landscape sca le Again they found local population persistence was related to pollen availability and the local population size of a habitat patch but not to connectivity or weather The variability in H. oraria population fr om 2011 to 2012 might be a part of an ove rall longer temporal trend. Understanding the population dynamics of solitary bees is important because if large population fluctuations exist a long term data set is required for any conclusion to be drawn about the population status of the bee as well a s the pollination services provided by the bee (Franzen and Nillson 2013). Without these long term studies, the population size could be consistently under or overestim ated (Franzen and Nilson 2013). In conclusion, a long term metapopulation study lookin g at colonization/ extinction dynamics as well as reproduction could shed some light on the population dynamics of H. oraria. Another possible explanation for patchy distribution and selectivity of patches could be the bee ecology. Flower m orphology is the result of natural selection enacted by pollinator s (Nattero et al. 2010). Pollinators select flowers based on their perceive d pollen amount and quality. For example, pollinators have been shown to visit large flowers over small ones as f lower size is usually correlated with nectar and pollen reward (Campbell 1991; Vaughton and R a msey 1998; Galen 2000; Fenster et al. 2006). Many studies have also shown floral shape, specifically corolla shape, to be a factor in flower selection as well (H errera 2001; Galen and Cuba 2001 ;
73 Nattero et al 2010 ). Because H. oraria is restricted to B. angustifolia it is likely a main driver in the local scale evolution of the plant. The bee is likely selecting for certain floral traits and floral cues among a nd between patches. A study looking at the visitation rate of individual flowers along with more information on the floral cues of B. angustifolia is necessary to determine what traits H. oraria is selecting for.
74 Table 3 1 Summary s tatistics of B.angustifolia patch characteristics that showed signi ficant changes between 2011 and 2012. Variable n Median Minimum Maximum 2011 2012 2011 2012 2011 2012 2011 2012 B.angustifolia seed head width (cm) 16 14 0.9 1.0 0.6 0.9 1.1 1.1 Length of transects (m) 16 14 13.5 3.0 3.0 3.0 64.0 63.0 B.angustifolia flower density (no./m 2 ) 16 14 12 4 0.4 0.4 36 16
75 Table 3 2. Wilcoxon rank sum test results showing H.oraria density and B.angustifolia pa tch characteristics that cha nged significantly between 2011 and 2012. Variable Units n Sum of Scores Expected Under H0 St. Dev Under H0 Mean Score Two sided Pr >= |S Mean| z statistic 2011 2012 2011 2012 2011 2012 2011 2012 2011 2012 B.angustifolia density no./m 2 16 14 264 201 248 217 23.46 23.46 16.5 14.4 0.5075 0.66 B.angustifolia flower density no./m 2 16 14 298 167 248 217 23.5 23.5 18.6 11.9 0.0336 2.11 Length of transects m 16 14 297 167 248 217 23.21 23.21 18.6 12.0 0.0326 2.11 B.angustifolia plant height cm 16 14 247 218 248 217 24.05 24.05 15.4 15.6 0.9755 0.02 B.angustifolia plant width cm 16 14 236 229 248 217 24.04 24.04 14.8 16.4 0.6300 0.48 B.angustifolia seedheads no./plant 16 14 264 200 248 217 23.96 23.96 16.5 14.3 0.5033 0.67 B.angustifolia seedhea d width cm 16 14 185 279 248 217 22.02 22.02 11.6 20.0 0.0038 2.82 Cover of vegetation % 16 14 228 237 248 217 23.86 23.86 14.3 16.9 0.4254 0.80 H.oraria density no./ha 16 14 270 195 248 217 15.62 16.62 16.9 13.9 0.1928 1.38
76 Table 3 3. Par ameter estimates for the full regression model comparing characteristics of foraging habitat ( B.angustifolia patches) against H.oraria density Variable Units DF Estimate Error t value Pr > |t| Intercept 1 231.698 543.9175 0.43 0.6747 Year 1 152.3 47 154.5879 0.99 0.3362 B.angustifolia plant density no./m 2 1 2.82441 16.03312 0.18 0.8619 B.angustifolia flowers no./m 2 1 13.10636 45.83814 0.29 0.7779 Length of transects m 1 1.77355 3.87008 0.46 0.6517 B.angustifolia plant height cm 1 10.2455 14.19522 0.72 0.4788 B.angustifolia plant width cm 1 10.8039 25.83907 0.42 0.6803 B. angustifolia seedheads no./plant 1 0.56133 12.70219 0.04 0.9652 B.angustifolia seed head width cm 1 495.5865 587.2568 0.84 0.4087 Cover of vegetation % 1 3.84571 5 .30935 0.72 0.4773
77 Figure 3 1. Gulf Islands National Seashore units Perdido Key, Fort Pickens, and Santa Rosa.
78 Figure 3 2. Total monthly precipitation for months preceding and for the duration of H.oraria foraging and B.angustif olia flowering. Data compares the average monthly precipitation from 1983 2012 with 2011 and 2012 sampling season precipitation. 0 5 10 15 20 25 30 35 40 45 50 March April May June July August September October Precipitation (cm) 30 year average 2011 2012
79 CHAPTER 4 FUTURE DIRECTIONS AND MANAGEMENT IMPLICATIONS The results of this study produced useful conclusions about the habi tat needs of Hesperapis oraria and Balduina angustifolia The dependency of H. oraria on B. angustifolia for foraging is clear. At a regional scale H. oraria is only present if the host plant is present and within a landscape the bee primarily uses large r patches with a higher density of the host plant. The habitat needs of B. angustifolia are therefore B. angustifolia requires moderate disturbance too create openings in vegetation where the aster can grow an d precipitation is important for plant recruitment and survival. However, too much disturbance shapes features such as washover sand flats and too much precipitation flooded swales; both of which are unsuitable landscape features for B. angustifolia to fl ower with the reliability from year to year necessary for H. oraria foraging. The questions addressed in this thesis hav e potential for becoming a long term project ; only with multiple year s of data can an accurate population estimate of H. oraria be esta blished. Furthermore, a long term project could create a more complete picture of distribution and project its future survival accounting for changing climatic conditions. Species distribution models have become popular as useful tools to understand species occurrence patterns and forecast effects of environmental change at different spatial scales (Bakkenes et al. 2002; Peterson et al. 2002 ). S pecies distribution models have been used to predict global species extinction risk, exposing the potentially substantial impact of climate change on extinction (Thomas et al. 2004) The species distribution models in this study could be combined into a habitat suitability map of the distribution of H. oraria and B. angustifolia The entire documented range of
80 H. oraria is under threat of both climatic and environmental change. As a dynamic environment coastal landscape s are under constant pressures of change from wind and water punctuated by significant change from storm events. Additionally, climate chang e models predict future change including an increase in storm frequency and intensity as well as significant sea level rise. All of these current and future changes should be included in plans to successfully protect these ecosystems. Therefore, combining the current habitat suitability map with climate change models would produce a powerful predictive tool for extrapolating future distribution of H. oraria and potential threats to the survival of the species and its habitat. In order to determine charact eristics of f oraging habitat which influence H. oraria density, the plots sampled in 2011 and 2012 should continue to be visited for several more years. Resulting time series data could be used to quantify changes in H. oraria density per foraging patch a s well as characteristics of dynamic of patch size and plant density. Ultimately, this data could point to important relationships between precipitation and B. angustifolia density as well as floral reliability and H. oraria population dynamics For inst ance, it would be interesting to determine patches of B. angustifolia that offer a reliable food source for the bee and how the presence of a reliable food source influences H. oraria patch selection. During this project H. oraria was observed resting on and gathering resources from Chrysopsis spp. There are three species of Chrysopsis (goldenaster) found in these habitats. Previously, Cane 1994 documented males on a yellow flowered composite that was not B. angustifolia but later identifies all species of Chrysopsis as coastal plants not visited by Hesperapis bees (Cane, 1994). Therefore, this is the first
81 known documentation of H. oraria visiting this plant. There are a few reasons a monolectic bee might visit another host plant. Bees use olfactory and visual cues to locate their host plants (Filella et al. 2011) and pollination is guided by both innate and learned behaviors (Riffell et al. 2008; Praz et al. 2008). For specialist pollinators precise host recognition is key (Filella et al. 2011). The refore, this could indicate that the bee is not monolectic as previously thought, but is oligolectic (Cane 1994), meaning they specialize in few select species of host plants. Cane mentions this possibility in a research report (1994) but named the bee as a monolectic species in publications. The timing of bee emergence coinciding exactly with B. angustifolia flower indicates it as the strongly preferred host plant at the very least. Possibly the individuals observed on Chrysopsis were stretched for re so urces (Aizen and Harder 2007; Potts et al. 2003). The area where this behavior was observed contained B. angustifolia in great number but the competition among and between species was also high. Lastly, perhaps the bees observed on Chrysposis were only c ollecting nectar and not pollen. Most bee species collect nectar from a wide variety of plants even if they are selective in choosing their pollen source (Filella et al. 2011). All of these reasons are speculative and it is not clear if Chrysopsis is in any way important to H. oraria. More research into this relationship could be key for management and restoration of the bee. An idea first proposed by Cane (1994) suggests an experimental introduction of H. oraria at a new location as a possible manageme nt technique to protect the species from extinction. The populations of H. oraria are so isolated that any major disturbance in one area could wipe out a local population with no hope of re colonization. This experiment could be taken a step further with introduction of the bee to non coastal
82 areas where B. angustifolia is present. If the bee can survive in well managed relict dune communities away from the coast, the threat of sea level rise could be vastly mitigated. Pesticides and invasive fire ants might also threaten the survival of H. oraria (Cane, 1994) All counties in Florida where the bee is found regularly spray pesticides for mosquito and other pest control. Most applications are to ponds and standing water and targeted at mosquito larvae. However, there is a considerable effort toward pest control in the form of aerial and road side aerosol applications. The bee populations on protected land should be shielded from these chemicals however many of the parks are so small that they do not o ffer much of a buffer from public roads where applications occur. There are also many doubts as to the precision of aerial applications. Fire ants ( Solenopsis spp ) are also a threat to many ground dwelling animals and prefer deep sandy soils characterist ic of coastal locations where H. oraria may nest. An invasion of fire ants to these coastal habitats may be devastating to nest survival; therefore, this situation should be monitored. Lastly, the most important next step in conservation efforts for H. o raria is to determine the nesting habitat needs for the bee. S ampler s in 1993 1994 and 2011 2012 were not able to locate the nests even afte r a concentrated effort (Cane et al. 1996) F emales with full pollen loads wer e commonly observed leaving foraging areas, ind icatin g they do not nest among the host plants. Furthermore, males were observed seeking females at host flowers and not searching the ground under host plants as most solitary bee species do to find emerging virgin females (Cane 1994 ) This behavior was also observed in the course of this study which suppor ts the hypothesis that the nesting
83 habitat is not the same as foraging habitat can at best provide an incomplete picture of the habitat requirements of H. oraria It is important for nests to be located and nesting habitat quantified to develop comprehensive conservation plans for this rare, endemic species.
84 LIST OF REFERENCES Aizen MA, Harder LD (2007) Expanding the limits of the pollen limitation concept: ef fects of pollen quantity and quality. The Ecol Soc Am 88 : 271 281 A izen MA, Ashworth L, Galetto L (2002) Reproductive Success in Fragmented Habitats: do compatibility systems and pollination specialization matter? J Veg Sci 13: 885 892 Allen Wardell G Be rnhardt P, Bitner R et al (1998) The potential consequences of po llinator declines on the conser vation of biodiversity and stability of food crop yields. Conserv. Biol. 12:8 17 Bakkenes M, Alkemade JRM (2002) Assessing effects of forecasted climate chang e on the diversity and distribution of Europ ean higher plants for 2050. Glob Change Biol 8 : 390 407 Biesmeijer JC, Roberts SP, Reemer M et al (2006) Parallel declines in pollinators and insect pollinated plants in Britain and the Netherlands. Science 313 :351 354 Bischoff M, Campbell DR, Lord JM et al. (2013) The relative importance of solitary bees and syrphid flies as pollinators of two outcrossing plant species in the New Zealand apline. Austral Ecol 38:169 176 Bom marco R, Biesmeijer JC, Meyer B et al (2010) Dispersal capacity and diet breadth modify the response of wild bees to habitat loss. P Roy Soc B Biol Sci 277: 207 5 2082 Branch LC, Miller DM, Wilkinson EB (2011) Habitat restoration for beach mice: landscape level population studies and dune restor ation, Final Report to the U.S. Fish and Wildlife Service, Panama City Brown MJF, Paxton RJ (2009) The conservation of bees: a global perspective. Apidologie 40:410 416 Brown MJF (2011) Conserva tion: the trouble with bumblebees. Nature 469:169 170 Brust ML, Hoback WW, Wright RJ (2008) A synopsis of Nebrask a Grasshopper Distributions. J Kansas Entomol Soc 81:208 255 Burger H, Ayasse M, Haberlein CM et al cues for host plant recognition by the oligolectic bee Hoplitis adunca. S Afr J Bot 76:788 795
85 Burger H, Ayasse M, Dotterl S et al. (2013) Perception of floral volatiles involved in host plant finding behavior: c omparison of a bee specialists and generalist. J Comp Physiol A 199:751 761 Cane JH (1994) Range, taxonomy and ecology of a newly discovered Gulf Coast solitary bee, Hesperapis oraria, belonging to a Southwestern desert genus. Dept. of Entomology, Auburn University, Auburn AL C ane JH, Snelling RR, Kervin LJ (1996) A New Monolectic Coastal Bee, Hesperapis oraria Snelling and Stage (Hymenoptera: Melittidae), with a Review of Desert and Neotropical Disjunctives in the Southeastern US J Kansas Entomol Soc 6 9:238 247 Cane JH, Minckley RL, Kervin LJ, et al (2006) Complex responses within a desert bee guild (Hymenoptera: Apiformes) to urban habitat fragmentation. Ecol Appl 16:632 644 Cane JH, Sampson BJ, Miller SA (2011) Pollination Value of Male Bees: The s pecialist bee Peponapis pruinosa (Apidae) at Summer Squash ( Cucurbita pepo ). Environ Entomol 40:614 620 Carlsson A Kindvall O (2001) Spatial dynamics in a metapopulation network: recovery of a rare grasshopper ( Stauroderus scalaris ) from population refug es. Ecography 24:452 460 Cerna K, Straka J, Munclinger P (2013) Population structure of pioneer specialist solitary bee Andrena vaga (Hymenoptera: Andrenidae) in central Europe: the effect of habitat fragmentation or evolutionary history? Conserv Genet 1 4:875 883 Cunningham S (2000) Depressed Pollination in Habitat Fragments Causes Low Fruit Set. P Roy Soc B Biol 267:1149 1152 ( 2000 ) Classification and regression trees: A powerful yet simple technique for ecological data analysis Ecology 81: 3178 3192. Debinsky DM, Holt R D (2000) A survey and overview of habitat fragmentation experime nts. Conserv. Biol 14:342 55 Dobson HEM, Bergstrom G (2000) The ecology and evolution of pollen odors. Plant Syst Evol 222:63 87 Dotterl S, Veree cken NJ (2010) The chemical ecology and evolution of bee flower interactions: a review and perspectives. Ca J Zool 88:668 697
86 F ilella I, Bosch J, Llusia J et al ( 2011 ) Chemical cues involved in the attraction of the oligolectic bee Hoplitis adunca to its host plant Echium vulgare. Biochem Syst Ecol 39: 498 508 Florida Natural Areas Inventory (FNAI) (2010) Guid e to the natural communities of Florida: 2010 edition. Florida Natural Areas Inventory, Tallahassee, FL. (FNAI 2010). Franzen M, Larsson M (2 007) Pollen harvesting and reproductive rates in specialized solitary bees. Ann Zool Fenn 44 : 405 414 Franzen M, Nilsson SG (2013) High population variability and source sink dynamics in a solitary bee species. Ecology 94:1400 1408 Gathmann A, Tscharntke T ( 2002 ) Foraging ranges of Solitary bees. Journal of Animal Ecology 71 : 757 764 Gru ndel R, Jean RP, Frohnapple KJ (2010) Floral and nesting resources, habitat structure, and fire influence bee distribution across an open forest gradient. Ecol A ppl 20: 16 78 1692 Gullan P, Cranston P (2010) The insects: An Outline of Entomology (4 ed.). Chichester, West Sussex: Wiley Blackwell A john Wiley & Sons, Ltd. Herrera C (1998) Population level estimates of interannual variability in seed production: what do they actually tell us? Oikos 82 : 612 616 (2010) A picky palate? The host plant selection of an endangered June beetle. J Insect Conserv 14:277 287 Horak J, Chobot K, Gabris R, et al (2011) Uphill distributional shift of an endangered hab itat specialist. J Insect Conserv 15:743 746 J ose S, Jokela EJ, and Miller DL (2006) The Longleaf Pine Ecosystem: Ecology, Silviculture, a nd Restoration. Springer Science+Business Media, LLC Klein AM, Vaissiere BE, Cane JH et al (2007) Importance of p ollinators in changing landscapes for world crops. P Roy Soc B Biol Sci 274:303 313 Kremen C, Williams NM, Aizen MA et al (2007) Pollination and other ecosystem services produced by mobile organisms: a conceptual framework for the effects of land use cha nge. Ecol Lett 10:299 314 Lucas KL, Carter GA (2010) Decadal Changes in Habitat Type Coverage on Horn Island, Mississippi, U.S.A. J Coastal Res 26: 1142 1148
87 McCune B, Grace JB (2002) Analysis of Ecological Communities. MjM Software, Gleneden Beach, Oreg on, USA ( www.pcord.com ) 304 pages. With a contribution by Dean L. Urban McCune B, Mefford MJ (2005) PC ORD, Multivariate analysis of ecological data, Version 5 for Windows edition. MjM Software Design, Gleneden Beach, Oregon USA McCune B, Mefford MJ (2011) PC ORD. Multivariate Analysis of Ecological Data. Version 6. MjM Software Gleneden Beach, Oregon, U.S.A Milet Pinheiro P, Ayasse M, Schlindwein C et al (2012) Host location by visual and Behav Ecol 23:531 538 Minckley RL, Cane JH, Kervin L, Roulston TH ( 1999 ) Spatial predictability and resource specialization of bees (Hymenoptera: Apoidea) at a superabundant, widespread resource. Biol J Linn Soc 67:119 147 Minckley RL, Wcislo WT, Yanega D. et al ( 1994 ) Behavior and phenology of a specialist bee (dieunomia) and sunflower (helianthus) po llen availability. Ecology 75:1406 1419 Natio nal Research Council of the National Academies (2006) Status of pollinators in North America. National Academy Press, Wa shington, DC Ollerton J, Winfree T, Tarra nt S llinated by animals? Oikos 120: 321 26 Parker ES, Jones SB (1975) Systematic study of genus balduina (Compositae, Heliantheae). Brittonia 27: 355 361 Pe mberton RW (2010) Biotic Resource Needs of Specialist Orchid Po llinators. Bot Rev 76:275 292 Peterson AT, Ortega Huerta MA, Bartley J et al ( 2002 ) Future projections for Mexican faunas under global climate chan ge scenarios. Nature 416 : 626 629 Peterson JH, Roitberg BD (2006) Impact of resource level sex ratio and resource allocation in Solitary bee, Megachile rotundata. Environ Entomol 35 : 1404 1410 Potts SG, Biesmeijer JC, Kremen C, et al. (2010) Global pollinator declines: trends, impacts and drivers. Trend s Ecol Evol 25:345 353
88 Potts SG, Vulliamy B, Dafni A et al ( 2003 ) (a) Response of plant pollinator communities to fire: changes in diversity, abundance and floral reward s tructure. Oikos 101:103 112 Potts SG, Vulliamy B Dafni A et al. ( 2003 ) (b) Linking bees and flowers: how do floral communities structure pollinator communities? Ecology 84: 2628 2642 Potts SG, Vulliamy B, Roberts S (2004) Nectar resource diversity organi zes flower visitor community structure. Entomol Ex p Appl 113 :103 107 Praz CJ, Mueller A, Dorn S (2008) Host recognition in a pollen specialist bee: evidence for a g enetic basis. Apidologie 39:547 557 Pries, A. (2006). Hurricane impacts on coastal dunes and spatial distribution of santa rosa beach mice (peromyscus polionotus leucocephalus) in dune habitats (Master's thesi s)Retrieved from http://uf.catalog.fcla.edu/ P r i es AJ Miller D L, Branch LC (2008) Identification of structural and spatial features that influence storm related dune erosion along a barrier island ecosystem in the Gulf of Mexico. J Coastal Res 24: 168 175 Pries AJ, Branch LC, Miller DL (2009) Impact of Hurricanes on Habitat Occupancy and Spatial Distribution of Beach Mice. J Mamm al 90:841 850 Rathcke BJ, Jules E S (1993) Habitat fragmentation and plant pollinator inte ractions. Curr Sci 65:273 277 Riffel l JA, Alarcon R, Abrell L et al (2008) Behavioral consequences of innate preferences and olfactory learning in hawkmoth flower interactions. Proceedings of the national academy of sciences of the United States of America 105 : 3404 3409 Raguso RA (2008 ) St art making scents: the challenge of integrating chemistry into pollination ecology. Entomol Exp Appl 128:196 207 Russell KN, Ikerd H, Doege S (2005) The potential conservation value of unmowed powerline strips for native bees. Biol conserv 124:133 148 Simon HA (1962) The architecture of complexity. Proc Am Philos Soc 106:467 482 Small JK (1933) Manual of the Southeastern flora. New York. Snep RPH, WallisDeVries MF, Opdam P (2011) Conservation where people work: A role for business districts and indu strial areas in enhancing endangered butterfly populations? Landscape Urban Plan 103:94 101
89 SPM Salford Predictive Modeler Software Suite 1984 2013 Salford Systems, CART 1984 2013 California Statistical Software, Inc. Steffan Dewenter I, Tscharntk e T ( 1999 ) Effects of habitat isolation on pollinator communities and seed set. Oecologia 121:432 440 Swink F Wilhem G (1994) Plants of the Chicago region Fourth Edition. Indiana Academy of Sciences, Indianapolis, Indiana, USA. Thomas CD, Cameron A, Gre en RE et al ( 2004 ) Extinction risk from climate change. Nature 427 :145 148. T homas JA, Simcox DJ, Clarke RT (2009) Successful Conservation of a Threatened Maculinea Butterfly. Science 325 :80 83 Urban DL, Oneill RV, Shugart HH (1987) Landscape Ecology. B ioscience 37 :119 127 Van Horne B, Olson G.S, Schooley RL, Corn JG, Burnham KP (1997) Effects of drought habitats. E col Monogr 67:295 315 Williams NM, Kr emen C (2007) Resource d istributions among habitats determine solitary bee offspring production in a mosaic landscape. Ecol Appl 17 :910 921 Willmer PG (1986) Foraging Patterns and Water Balance: Problems of Optimization for a Xerophilic Bee, Chalicodoma sicula J Anim Ecol 55:94 1 962 Winfree R, Griswold T Kremen C (2007) Effect of human disturbance on bee communities in a forested ecosy stem. Conserv biol 21: 213 223 Zurb uchen A, Bachofen C, Mueller A (2010) (a) Are landscape structures insurmountable barriers for foraging bees? A mark recapture study with two solitary pollen specialist species. Apidolog ie 41 :497 508 Zurbuchen A, Cheesman S, Klai ber J e t al. ( 2010 ) (b) Long foraging distances impose high costs on offspring production in solitary bees. J Anim Ecol 79:674 681 Zur buchen A, Landert L, Klaiber J et al. ( 2010 ) (c) Maximum foraging ranges in solitary bees: only few individuals have capability to cover long foraging distances. Biol Conserv 143: 669 676 Zych M, Stpiczyn ska M (2012) N either protogynous nor obligatory out crossed: pollination biology and breeding system of the European Red List Fritillaria meleagris L. (Liliaceae). Plant Biol 14:285 294.
90 BIOGRAPHICAL SKETCH Hannah was born in Butler, Pennsylvania but grew up in Fuqua y Varina, N orth C arolina. She earned a B.S. in e nvironmental s cience from T h e University of North C arolina at Chapel Hill in August 2008. F or over a y ear, she worked for the Studen t Conserva tion Association gaining a broad set of experiences in the natural resource management field. After a season working for the National Pa rk Service in Texas, she w a s then hired by The Joseph W. Jones Ecological Research Center to c ollect data for a long term study of longleaf pine ecosystem restoration and management at Eglin Air Force Base in Florida She started her graduate studies in August 2011 majoring in interdisciplinary ecology. She concentrated on wildlife ecology and conservation and