Environmental and Climatic Predictors of Presence of Adult Amblyomma americanum in Florida

Material Information

Environmental and Climatic Predictors of Presence of Adult Amblyomma americanum in Florida
Kessler, William H
Place of Publication:
[Gainesville, Fla.]
University of Florida
Publication Date:
Physical Description:
1 online resource (87 p.)

Thesis/Dissertation Information

Master's ( M.S.)
Degree Grantor:
University of Florida
Degree Disciplines:
Committee Chair:
Committee Co-Chair:
Committee Members:


Subjects / Keywords:
geography -- sdm -- ticks
Geography -- Dissertations, Academic -- UF
bibliography ( marcgt )
theses ( marcgt )
government publication (state, provincial, terriorial, dependent) ( marcgt )
born-digital ( sobekcm )
Electronic Thesis or Dissertation
Geography thesis, M.S.


The lone star tick, Amblyomma americanum, is the most commonly reported tick in the southern United States and the primary vector of several human and livestock pathogens of public health concern. Although this species is endemic to Florida, little is known about the ecological preferences and spatial distribution of the species in the state. Using occurrence records of adult A. americanum collected between September 2015 and September 2016, a logistic regression model was developed to estimate environmental and climatic associations, as well as predict the fine-scale distribution, of the tick in Florida. Occurrence of adult lone star ticks was found to be associated with a combination of habitat and bioclimatic variables, namely forested land cover and seasonality of precipitation. The estimated spatial distribution, at a resolution of 1 hectare, indicated that probable occurrence decreases from North to South with very little area deemed suitable in the far southern reaches of the state. This trend reaffirms findings in the literature that prevalence of A. americanum on wildlife increases with latitude in Florida. The 1-hectare resolution of the estimated distribution is a significant improvement over distributions currently published in the literature and will better inform the public and state or federal agencies of potential risk of exposure to A. americanum. ( en )
General Note:
In the series University of Florida Digital Collections.
General Note:
Includes vita.
Includes bibliographical references.
Source of Description:
Description based on online resource; title from PDF title page.
Source of Description:
This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Thesis (M.S.)--University of Florida, 2017.
Electronic Access:
Statement of Responsibility:
by William H Kessler.

Record Information

Source Institution:
Rights Management:
Applicable rights reserved.
Embargo Date:
LD1780 2017 ( lcc )


This item has the following downloads:

Full Text




2017 William H. Kessler


To my mother and fath er for their support, wisdom, and encouragement


4 ACKNOWLEDGMENTS I would like to thank my advisor Dr Gregory Glass for his astute observations, subtle guidance, and shared suffering during endless hours of fieldwork in the Florida sun. I would also like to thank Dr. Jason Blackburn for his assistance with the analytical methodologies used in this study and for playing the role of reviewer #2. Thanks also goes to Dr. Katherine Sayler for her expertise in a ll things ticks and for assistance in identifying specimens. I would also like to acknowledge the contributions of my geography cohort, especially Joe Andreoli, Morgan Ridler, Brittany Hodik, and James Richardson in keeping this endeavor in perspective an d providing the motivation to finish. Lastly, I would like to thank my family for their support and encouragement.


5 TABLE OF CONTENTS page ACKNOWLEDGMENTS ................................ ................................ ................................ ............... 4 LIST OF TABLES ................................ ................................ ................................ ........................... 6 LIST OF FIGURES ................................ ................................ ................................ ......................... 7 LIST OF ABBREVIATIONS ................................ ................................ ................................ .......... 8 ABSTRACT ................................ ................................ ................................ ................................ ..... 9 CHAPTER 1 INTRODUCTION ................................ ................................ ................................ .................. 10 2 CHARACT ERIZATION OF ENVIRONMENTAL AND CLIMATIC PREDICTORS OF ADULT AMBLYOMMA AMERICANUM IN FLORIDA ................................ ................ 16 2.1 Background ................................ ................................ ................................ ....................... 16 2.2 Amblyomma americanum : Species of Public Health Concern ................................ ......... 17 2.3 Species Distribution Models ................................ ................................ ............................. 20 2.3.1 Generalized Linear Model Framework for SDMs ................................ .................. 21 2.3.2 Estimating Geographic Distributions of Ixodid Ticks ................................ ............ 22 2.4 Materials and Methods ................................ ................................ ................................ ..... 26 2.4.1 Study Area ................................ ................................ ................................ .............. 26 2.4.2 Tick Collection and Environmental Data Sources ................................ ................. 27 2.4.3 Methodology to Account for Spatial Autocorrelation ................................ ............ 33 2.4.4 Multivariate Logistic Regression ................................ ................................ ........... 35 2.4.5 Model Eva luation ................................ ................................ ................................ ... 37 2.4.6 Spatial Predictions and Gradients ................................ ................................ ........... 42 2.5 Results ................................ ................................ ................................ ............................... 44 2.6 Discussion ................................ ................................ ................................ ......................... 60 3 CONCLUSIONS ................................ ................................ ................................ .................... 67 APPENDIX A LOGISTIC REGR ESSION IN R ................................ ................................ ............................ 70 B CONSIDERATION OF COVARIATE ANOMOLIES ON SPATIAL PREDICTIONS ...... 73 LIST OF REFERENCES ................................ ................................ ................................ ............... 77 BIOGRAPHICAL SKETCH ................................ ................................ ................................ ......... 87


6 LIST OF TABLES Table page 1 1 Yearly incidence of reportab le tick borne diseases in Flo rida from 2010 2016. .............. 13 2 1 Characteristics of variables ................................ ................................ ................................ 29 2 2 Reclassification scheme for land cover variables ................................ .............................. 31 2 3 Specimen totals by site. ................................ ................................ ................................ ...... 45 2 4 Environmental and climatic variables used to model the spatial distribution of Amblyomma americanum in Florida. ................................ ................................ ................. 48 2 5 Environmental and climatic variables with low collinearity determined by Var iance Inflation Factor ................................ ................................ ................................ .................. 49 2 6 Coefficients and variables of final model ................................ ................................ ......... 50 2 7 Model validation metrics ................................ ................................ ................................ .. 51 2 8 Presence absence discrimination for testing data ................................ ............................. 52


7 LIST OF FIGURES Figure page 1 1 Posted safety alert in San Felasco state park warning of exposure risk to ticks ............... 14 2 1 Distri bution of collection sites ................................ ................................ ........................... 46 2 2 Tick collections by month ................................ ................................ ................................ 47 2 3 Spatial distribution of model residuals.. ................................ ................................ ............ 53 2 4 Probabilities of occurrence of Amblyomma americanum in Florida.. ............................... 56 2 5 Estimated distribution of Amblyomma americanum in Florida ................................ ......... 57 2 6 Latitudinal trend in percentage of suitable area ................................ ................................ 58 2 7 Latitudinal trend in high suitability areas ................................ ................................ .......... 59 B 1 Examples of potential interpolation error in WorldClim bioclimatic variables. ............... 75 B 2 Impact of interpolation error in bioclimatic variable Bio13 on the spatial prediction produced by the logistic regression model.. ................................ ................................ ....... 76


8 LIST OF ABBREVIATIONS A. americanum Amblyomma americanum the lone star tick D. variabilis Dermacentor variabilis the American dog tick ENM Ecological Niche Model I. scapularis Ixodes scapularis the black legged tick NDVI Normalized Difference Vegetation Index SDM Species Distribution Model


9 Abstract of Thesis Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requ irements for the Degree of Master of Science ENVIRONMENTAL AND CLIMATIC PREDICTORS OF PRESENCE OF ADULT AMBLYOMMA AMERICANUM IN FLORIDA By William H. Kessler December 2017 Chair: Gregory Glass Major: Geography The lone star tick, Amblyomma americanum, is the most commonly reported tick in the southern United States and the primary vector of several human and livestock pathogens of public health concern. Although this species is endemic to Florida, little is known about the ecological prefere nces and spatial distribution of the species in the state. Using occurrence records of adult A. americanum collected between September 2015 and September 2016 a logistic regression model was developed to estimate environmental and climatic associations as well as predict the fine scale distribution of the tick in Flo rida. Occurrence of adult lone star ticks was found to be associated with a combination of habitat and bioclimatic variables, namely forested land cover and seasonality of precipitation. Th e estimated spatial distribution, at a resolution of 1 hectare, indicated that probable occurrence decreases from North to South with very little area deemed suitable in the far southern reaches of the state. This trend reaffirms findings in the literature that prevalence of A. americanum on wildlife increase s with latitude in Florida. The 1 hectare resolution of the estimated distribution is a sig nificant improvement over distributions currently published in the literature and will better inform the public and state or federal agencies of potential risk of exposure to A. americanum


10 CHAPTER 1 INTRODUCTION In the United States, tick borne diseases are the most commonly reported vector borne disease s outpac ing reports of mosquito borne illness by a n average factor of 9:1 from 2004 2013 (USGCRP 2016) The species of ticks that can transmit these pathogens are not homogeneously distributed which result s in disparate areas of exposure risk and disease incidence. In Florida, five tick species are commonly reported ; the black legged tick ( Ixodes scapularis ), the American dog tick ( Dermacentor variabilis ), the lone star tick ( Amblyomma americanum ), the gulf coast tick ( Amblyomma maculatum ), and the brown dog tick ( Rhipicephalus sanguineus ). Each of these species are considered endemic in the state (Bishopp and Trembley 1945, Yabsley, et al. 2009) In addition, many counties in Florida have reported other tick sp ecies including Ixodes affinis and Ixodes minor (Clark 2004, Forrester, et al. 1996, Wehinger, Roelke and Greiner 1995, Durden, Klompen and Keirans 1993) Collectively t hese species are considered we ll established across much of Florida. Furthermore, reports indicate successful establishment of the exotic species Amblyomma auricularium and A. rotundatum in southern Florida (Oliver Jr, Hayes, et al. 1993, Mertins, Vigil and Corn 2017) Coincidently, the five most commonly reported tick species in Florida are known to transmit a number of pathogens capable of causing debilitating disease in humans and livestock. Ixodes scapularis is most commonly associated with tick borne illnesses as it is the primary North American vector for Lyme disease the most commonly diagnosed vector borne disease in the United States (Armstrong, et al. 2001, Gray 1998) However, it is also known to transmit the pathogens causing anaplasmosis and several other diseases (Adelson, et al. 2004) Dermacentor variabilis and Rhipicephalus sanguineus are both known to vector the causative agent o f Rocky Mountain spotted fever (RMSF); D. variabilis also vectors Francisella tularensis while R.


11 sanguineus additionally vectors canine ehrlichiosis (Bell 1944, Green 1931, Demma, et al. 2005, Smith, et a l. 1976) The gulf coast tick, A. maculatum transmits Rickettsia parkeri a spotted fever rickettsiosis (Sumner, et al. 2007) Amblyomma americanum is the most commonly encountered tick in Florida and other places in the southeastern United States and is capable of vectoring a number of disease causing pathogens (Armstrong, et al. 2001, Childs and Paddock 200 3, Felz, Durden and Oliver Jr. 1996) Whi le the lone star tick does not vector the Lyme disease pathogen, it can transmit Ehrlichia chaffeensis and Ehrlichia ewingii (which cause human ehrlichiosis), Francisella tularensis and the presently unknown agent of Southern Tick Associated Rash Illness known as STARI (Hopla 1955, Hopla 1953, Masters, Grigery and Masters 2008, Goddard and Varela Stokes 2009, Anderson, Greene, et al. 1 992, Anderson, Sims, et al. 1993) In places where A. americanum and I. scapularis co occur perception of Lyme disease may be exaggerated while A. americanum vectored diseases (namely STARI) go unreported; this could be due to growing public awareness of Lyme disease and the association of the disease with ticks in general and the nuisance biting habits of A. americanum (Armstrong, et al. 2001) I. scapularis will f urther confuse diagnoses of Lyme disease and STARI as the presentation of symptoms are similar (Sayler, Boyce and Wisely 2016) The multitude of tick species occurring in the state suggest that climatic and en vironmental conditions in Florida are broadly suitable for the persistence of many tick species and the pathogens they carry. Temperatures and humidity levels, which contribute greatly to tick activity and life cycle, rarely stray beyond what is considered highly suitable for many species (Heath 1981, Yoder and Tank 2006, Yoder and Benoit 2003) High temperatures coupled with host seeking


12 activity and ultimately, desiccation. The monthly average relative humidity for much of Florida ranges from greater than 68% to nearly 90% while average monthly temperatures range from 10 29C (Henry, Portier and Coyne 1994 ) A study by Yoder and Tank (2006) indicated that five common North American species of veterinary importance, including the dry adapted Rhipicephalus sanguineus and the hydrophilic species Amblyomma americanum share similar upper temperature limits (between 30 35C) before rapid moisture loss occurs. These findings American and exotic tick species. Understanding the distribution of arthropod vectors can prov ide insights into at risk ar eas for vector borne pathogens although there are some caveats (Ostfeld, Glass and Keesing 2005) For example, although high predicted exposure risk (based on habitat suitability) to the arthropo d vector ( Ixodes pacificus ) has been positively associated with disease incidence as is the case of Lyme incidence in California, concordance between Lyme disease incidence and habitat suitability is not always high (Eisen, et al. 2006, Atkinson, et al. 2014) Furthermore, c ases of Lyme borreliosis disproportionately cluster around northwestern Wisconsin and on the northeastern seaboard around New Jersey and east central Pennsylvania even though the primary vector for t he disease in the eastern United States is found in 37 states, primarily east of the Mississippi river (Kugeler, et al. 2015, Eisen, Eisen and Beard 2016) Still, s tudies have shown that presence of the primary disease vector and environmental factors associated with high populations of the species increase the risk of some vector borne diseases, for example tick borne Lyme disease (G lass, et al. 1995, Eisen, et al. 2006) Based on this assertion, models that estimate the distribution of a vector species can provide a broad estimate of disease risk (Ostfeld, Glass and Keesing 2005, Eisen, et al. 2006)


13 T icks are common pest s in Florida as indicated by signs and postings in state parks and other public lands warning of the risk of tick exposure (Figure 1 1) Several dozen cases of locally acquired tick borne illness are reported every year in Florida (Table 1 1) however little is known about the statewide distributions of the vector species Table 1 1. Yearly incidence of suspected and confirmed locally acquired reportable tick borne dise ases in Florida from 2010 2016. (Reportable Diseases Frequency Report 2017) Diseases Year 2010 2011 2012 2013 2014 2015 2016 Total Counts Counts Counts Counts Counts Counts Counts 433 41 50 80 53 79 63 67 Anaplasmosis 11 3 6 1 0 0 0 1 Ehrlichiosis 120 8 13 20 17 23 16 23 Lyme Disease 210 22 22 39 21 35 35 36 Spotted Fever Rickettsiosis (RMSF) 92 8 9 20 15 21 12 7 T here have been few studies characterizing the geographic distributions and ecological preferences of ticks in the state at sub county scales Estrada Pena (1998) estimated the probable distribution of Ixodes scapularis across North America at a resolution of 8km using a cokriging interpolation. The consideration of temperature and vegetation conditions improved model fit ( sensitivity and specificity) and found that much of Florida was likely habitat for the species Similarly, J ames, et al. (2015) modeled the distribution of Dermacentor variabilis in the United States us ing biologically relevant climatic variables primarily summer vegetation greenness, at a resolution of 1km and found most of Florida to have high habitat suitability for this species as well. Another important species for public health Amblyomma americanum is the most comm only encountered tick in t he southeastern United States and as noted previously a species of capable of vectoring multiple pathogens including Ehrlichia chaffeensis Ehrlichia ewingii the Panola Mountain Ehrlichia among others (Estrada Pena and Jongejan 1999, Childs and Paddock 2003, Reeves, et al. 2008) A 2015 analysis by Springer and colleagues of the


14 distribution of this species in the United States used a series of ecological niche modeling methods at the county level, and indicated a potential meridional gradient of decreasing suitability across Florida. A national study of I scapularis nymphs also reported no collections of A. americanum from study sites south of roughly 28N in Florida (Diuk Wasser, et al. 2006) The results of these two studies indicate the potential for high heterogeneity in suitability for A. americanum across Florida and has implications for the distribution of A. americanum borne disease risk. Figure 1 1: Posted safety alert in San Felasco state park warning of elevated exposure risk to ticks, highlighting risk mitigation techniques, and treatment of tick bites.


15 The current thesis research will develop a state wide spatial model t o answer the following two research questions: 1. What environmental and/or climatic factors are associated with the current distribution of adult Amblyomma americanum in mainland Florida? Several recent studies have linked local distributions of A. americanum to environmental and land cover factors. This study consider s statewide associati ons between species occurrence and environmental and climatic co nditions at a sub county resolution 2. Wher e are the areas of high probability of occurrence of A. americanum in Florida? Areas that are likely to harbor A. americanum may also be areas of increased human risk of specific tick borne disease s Understanding the distribution of this species will improve decisions about alerting the public to the risk of exposure to tic ks and tick borne disease. The remainder of this thesis includes two additional chapters. The second chapter will examine the two research questions described above and provide further background, method s results, and conclusions. The third and final ch apter will examine the implications and limitations of this study.


16 CHAPTER 2 CHARACTERIZATION OF ENVIRONMENTAL AND CLIMATIC PREDICTORS OF ADULT AMBLYOMMA AMERICANUM IN FLORIDA 2.1 Background Hard bodied t icks (family: Ixodidae) are obligate, blood feeding ectoparasitic arthropods that are found worldwide (Jongejan and Uilenberg 2004) All Ixodid lifecycles are composed of four stages: eggs, larvae, nymphs, and adults. Oviposition results in a single egg mass in the leaf litter or in animal burrows. Each life stage require s a bl ood meal before detachment from the host and undergoing ecdysis (Anderson and Magnarelli 2008) The time requirement for ticks to complete a life cycle depends on a num ber of factors. The seasonal regulation of generational times may result in northern populations of a species tak ing significantly longer than in southern populations ( 1 3 years as opposed to one or more generations per year) (Oliver Jr 1989, Strickland, et al. 1976) The photoperiod, availability of hosts, temperature, and moisture all impact the diapause period of eggs and molting of instars (Oliver Jr 1989) Sufficient moisture availability modulates both the general time s for many tick species as well as daily behavior patterns. Ticks are prone to dehydration due to their infrequent intake of fluids (during feeding), large surface to volume ratio, and the amo unt of time spent in low moisture environments while questing (Anderson and Magnarelli 2008, Knulle and Rudolph 1981) T o maintain an adequate water balance, most ticks retreat from questing locations on exposed plant surfaces back into moist leaf litter or other locations when ambient temperatures climb or relative humidity decreases (Anderson and Magnarelli 2008, Vail and Smith 2002) Many tick species are host generalists that feed on vari ous species of animals as adults Ixodid ticks most commonly parasitize mammals although at least one species in each genus also can utilize birds or reptiles during one or more life stages (Oliver Jr 1989) Imm ature life


17 stages may preferentially take blood meals from a single vertebrate class. Larval or nymphal ticks from more t han 300 species feed on rodents (Oliver Jr 1989) Most hard bodied ticks are three host species meaning that they feed on three different individual hosts, one during each life stage. Many t hree host species such as Ixodes scapularis Dermacentor variabilis and Amblyomma americanum are important vectors of wildlife and human diseases becaus e the multi host feeding characteristics increase the likelihood of transmitting a pathogen from an infected individual to an uninfected individual via co feeding, transstadial or interstadial transmission (Randolph, Gern and Nuttall 1996, Jaworski, Bowen and Wasala 2013, Bremer, et al. 2005) 2.2 Amblyomma americanum : Species of Public Health Concern Amblyomma americanum commonly referred to as the lone star tick, is a non nidicolous ( non nest dwelli ng ) species whose known regional distribution extends from west central Texas to the eastern seaboard and as far north as Maine (Anderson and Magnarelli 2008) Adult and nymphal ticks are generally most active from April throug h June with activity declining in late summer. Larval ticks frequently become active later in the season than adults or nymphs (Davidson, Siefken and Creekmore 1994, Cilek and Olson 2000) The abundance of A. americanum is highly dependent on the availability of suitable host species (Childs and Paddock 2003) I mmature s (both larvae and nymphs) of the lone star tick feed on various small, medium and large mammals as well as birds w hereas in the adult life stage this species frequently parasitizes white tailed deer, cattle, and feral swine (Anderson and Magnarelli 2008) The general preference for adults to feed on deer and the non specific biti ng habits of other life stages significance as a vector for wildlife and human diseases (Childs and Paddock 2003) White tailed deer ( Odocoileus virginianus ) are a primary reservoir of several Ehrlichia species making infection of feeding ticks likely (Childs and


18 Paddock 2003) Furthermore, a survey of emerging wildlife pathogens by Dobson and Foufopoulos (2001) found that one common trait of many emerging pathogens is the ab ility to infect a wide range of hosts. A non discriminatory vector such as A. americanum would similarly be advantageous for promoting the spread of a pathogen to previously uninfected hosts (Childs and Paddock 2003) Among ot her human pathogens, t his species is known to transmit Ehrlichia chaffeensis the currently undefined agent of southern tick associated rash illness (STARI), and Francisella tularensis (Estrada Pena and Jongejan 1999, Anderson and Magnarelli 2008, Hopla 1955, Hopla 1953, Masters, Grigery and Masters 2008, Anderson, Greene, et al. 1992, Anderson, Sims, et al. 1993) The public health burden of lone star tick assoc iated illness is likely to increase in the future, however the noted rise of A. americanum as a species of major public health concern is confounded by improvements in surveillance and diagnostics for tick borne diseases (Childs and Paddock 2003) Nevertheless, changes in vector reservoir host dynamics and human interactions will potentially increase human risk of tick borne diseases. Increases in white tailed deer densities, expansion of the geographic distribution of A. americ anum and increased human contact with vectors can all influence the landscape of lone star tick transmitted pathogens (Childs and Paddock 2003) Already i n many areas in the southeastern United States, the lone star tick is t he most abundant nuisance tick, and the most frequently encountered by humans (Childs and Paddock 2003) It is notorious for its voracious biting of humans. Studies on parasitism of humans by ticks in the southern US show that bites are overwhelmingly (83 95% in some instances) attributed to A. americanum (Childs and Paddock 2003, Felz, Durden and Oliver Jr. 1996)


19 The abundance of adult A. americanum in Florida is seasonal with con flictin g reports about when peak adult activity occurs. In the northwestern region of the state adult ticks are highly abundant from April through August (Cilek and Olson 2000) although the peak is reported as early as March for Florida (Allan, Simmons and Burridge 2001) Studies on lone star tick abundance in nearby states found similar peaks in adult abundance from April to June and from May to July in Georgia and M issouri, respectively (Semtner and Hair 1973, Kollars Jr, et al. 2000) In central Florida, recent studies have found that temporal patterns of nymphal and larval activity mirror that of northern populations with peak activity in the fall (Allan, Simmons and Burridge 2001) However, no studies were found that have collected adult or immature A. americanum i n numbers great enough to describe temporal activities of southern Florida population s. The local geographic distribution of A. americanum in Florida is unknown beyond crude county level maps. The seminal work on North American tick distributions by Bishopp and Trembley (1945) currently provides the basis for The Centers for Disease Control and Prevention (CDC) published map of the regional distribution of the lone star tick and indicates all o f Florida falls within the species Springer et al. (2014) describe establishment of the species at the county level across the United States using historic data and report occurrence in nearly all counties in northern and central Florida by the 1960 s. Subsequent re cords between 1960 and the 2010 s include additional counties in south central Florida, however the authors caution that apparent recent establishment in counties not previously reported may be due increased surveillance rather t han invasion of previously unoccupied areas (Springer et al. 2014) This is important to note as previous studies indicate that the prevalence of the species diminishes from north to south in the state (Allan, Simmons and Burridge 2 001) with the


20 southernmost reports of adult individuals occurring in Palm Beach Co., Glades Co., and Miami Dade C ounty (Taylor 1951, Greiner, et al. 1984) These findings however appear to be host specific as the southernmost collection of A. americanum from big cats was from ce ntral Florida in Highlands C o. (Wehinger, Roelke and Greiner 1995) while collections from feral swine and white tailed deer in the southern portions of the state were inconsistent. The reports from Palm Beach, Glades, and Miami Dade Counties constitute a single adult collected by Allan et al (2001), 3 by Greiner (1984) and th ose by Taylor (1951) from deer or swine. Other collections from sout h Florida deer have not recovered A. americanum (Smith Jr. 1981) The reasons for presumptively rare occurrences of A. americanum in southern Florida is uncle ar although Allan et al. (2001) posit ed that climate, vegetation, and host abundance may play a roll 2. 3 Species Distribution Model s There are several statistical and pattern matching approaches to estimating the distribution of organisms in the environment These techniques can range from purely theoretical to mechanistic or empirical (Guisan and Zimmermann 2000) Predictive species distribution models (SDMs), are designed to detect non random associations between location specific occurrence observations and environmental conditions ( often as raster based predictors) to estimate a response at unobserved locations (Guisan and Zimmermann 2000, Blackburn 2010) Broadly speaking they are classed as either estimating occurrence of a species or abundance of a species in relation to the environment and rely on presence only, presence absence, or count data Species distribution models are a commonly employed technique in ecology, biogeography, and other fields to estimate the habitat suitability for many species and to describe distributions of disease or disease causing agents (Austin, Nicholls and Margules 1990, Pearson, Raxworthy, et al. 2007, Guisan and Thuiller 2005, Springer, et al. 2015) Species distribution


21 model are frequently divided into two approaches: presence only models, and presence absence models. The prima ry distinction between these two approaches stems from the structure of the data available (Elith, et al. 2011) Commonly used presence only methods include genetic algorithms, and maximum entropy models (Stockwell 1999, Phillips, Anderson and Schapire 2006) A presence absence data structure the combination of occurrence and (thorough) absence data lend themselves to regression techniques such as generalized linear models (GLM). Other m ethods such as classification trees are also used to fit presence absence data, but will not be discussed here (Elith, et al. 2011, Guisan, Edwards and Hastie 2002) 2.3.1 Generalized Linear Model Framework for SDMs Generalized linear models include the well known logistic regression models, for describing simple presence/absence, and Poisson or negative binomial structured count models describing frequency or abundance of occurrence (Guisan, Edwards and Hastie 2002) Broadly, GLMs attempt to apply a single, linear combination of coefficients and corresponding predictor variables to describe a response variable while minimizing unexplainable error (Guisan, Edwards and Hastie 2002, Potts and Elith 2006, Guisan and Zimmermann, Predictive Habitat Distribution Models in Ecology 2000) In ecological terms, this means that the observed response, that is presence or absence, is limited to a fixed directional relationship with each predictor as indicated by the magnitude of the regression coefficient. Conceptually the output of a SDM, a habitat suitability surface, is an estimation of where the combination of environmental conditions are clo (Elith, et al. 2011) In other words, the suitability surface is a probability estimate of presence or absence in a pixel. The geographic distr ibutions of both vagile and non vagile species have been estimated by GLM methods successfully. Guisan et al. (1998) used logistic regression to model the


22 distribution of an alpine plant species; Zimmermann and Breitenmoser (2002) modeled the distribution of the Eurasian Lynx ; t he relationship between landscape pattern and south Australian bird species was evaluated by Westphal et al. (2003); Brown et al. (2011) estimated high risk areas for exposure to the lone star tick ( A. americanum ) in Missouri. Additi onal applications for GLMs include distributions of plant and animal communities (see Guisan et al., 2002 for examples). 2.3. 2 Estimating Geographic Distributions of Ixodid Ticks Many studies have sought to estimate the potential distribution and ecologi cal preferences of Ixodid ticks using climatic or environmental variables at a variety of spatial scales (Springer, et al. 2015, Estrada Pena 1998, James, et al. 2015, Raghavan, et al. 2016) Advances in modeling techniques has spurred the application of a number of methodological approaches or ensembles of multiple modeling approaches (Guisan and Thuiller 2005, Elith, et al. 2011) The general agreement upon variables associated with tick presence and the resulting spatial predictions across many of these studies supports the notion that the models are capturing biologically relevant phenomena (Springer, et al. 2015, O stfeld and Brunner 2015) Due to the direct influence of climatic conditions on the physiological processes and behavior of ticks, variables characterizing temperature and precipitation patterns are frequently included in SDMs for ticks. For example, Porretta et al. (2013) estimated the current and future distribution of Ixodes ricinus in Europe using the MaxEnt modeling algorithm. The presence only data was composed of occurrence records from across the known range of the species. B ioclimatic variable s from the WorldClim database (Hijmans, Cameron, et al. 2005) were used providing a resolution of roughly 5km. These variables are derived from average monthly precipitation and temperature data and capt ure seasonal and annual trends. The minimum temperature of the coldest period and precipitation of the driest quarter provided the greatest


23 contribution to the model. The predicted area of climatic suitability encompasse d the distributions inferred from other studies lending cred ence to the appropriateness of using climatic variables to d etermine geographic ranges for I xodid ticks. Springer et al (2015) applied an ensemble modeling approach to examine the present and future distribution of A. americanum at the county level for the contiguous United States under both current and projected future climate conditions. An ensemble model approach utilizes multiple modeling algorithms to identify areas of agreement or disagreement (Araujo and New 2 006, Springer, et al. 2015) Published presence records for A. americanum in the contiguous US going back to 1898 were aggregated to the county level, representing establishment of the species in 651 counties across 32 states. A total of 66 climate va riables were generated to represent the average climatic conditions in each county. Selection of present day models by each modeling approach: general linear model, boosted regression trees, maximum entropy, random forest, and multivariate adaptive regress ion spline, resulted in between one and nine climate variables showing significant predictive ability. Overwhelmingly, in each model mean vapor pressure in July had the greatest explanatory power. Other included variables were annual precipitation, mean gr owing degree days in October, mean temperature of driest quarter, mean snow days in October, precipitation seasonality, mean temperature of wettest quarter, mean growing degree days in February, and mean diurnal range. The authors concluded that summertime humidity is highly associated with determining habitat suitability. In addition, the inclusion and varied relationships of additional variables in one or more model s suggest that at the continental scale, different ecological pressures may drive the distr ibution of A. americanum outside their core range in the south central and southeastern region of the US


24 The geographic distribution of the American dog tick (ADT) Dermacentor variabilis has also been estimated using the MaxEnt method at the continental scale. The study by James et al. (2015) utilized the same bioclimate variables dataset at Porretta et al. (2013) as well a s additional habitat metrics including monthly average measures o f vegetation greenness (Normalized Difference Vegetation Index, NDVI), and elevation. Presence data were drawn from 317 observations of D. variabilis Several possible distributions were estimating using combinations of uncorrelated variables. The set of c onsidered MaxEnt models indicated that while several climate variables were still importa nt predictor s of suitable habit at, other environmental variables including elevation and summer (July) NDVI produced the greatest contribution to the models. Furthermo re, while climate features may drive the broad distribution patterns, the inclusion of habitat specific characteristics may improve the resolution of the analysis (1km x 1km pixels). For example, the ADT is associated wi th habitat conditions found at lower elevations (Bishopp and Trembley 1945) so the inclusion of climate and elevation topographical variation. One limitation of this study results from the mismatch between the spatial resolution of the geographic prediction and the accuracy or precision of observation locations; observation precision ranged from GPS coordinates to city centroids. The result are p otentially biased associations between observations and environmental variables. An autologistic framework has been used to estimate the distribution of Ixodes scapularis for the coterminous United States (Brownstein, Holford and Fi sh 2003) Known occurrence s of I. scapularis at the county level were regressed against climatological data for minimum, mean, max and standard deviations of monthly minimum, maximum and average temperature averages and vapor pressure at a resolution of 0.5. Higher order polynomial terms for these variables


25 were also considered. Brownstein et al. found I. scapularis was ass ociated with average maximum temperature up to a 4 th order polynomial, and vapor pressure. These findings indicate that temperatu re extremes likely play a part in determining the distribution of the black legged tick in the United States. At the state level, MaxEnt has been used to estimate the distribution of A. americanum in Kansas (~18x18 km) (Raghavan, et al. 2016) Climate and environmental characteristics were evaluated with principal component analysis before being considered in the model. At this scale variables representing temperature and soil moisture prov ided the largest contribution to the model although variables for precipitation were also important (Raghavan, et al. 2016) Notably, the estimated distribution extend ed further west than has been suggested by the CDC, and is i n general agreement with the prediction by Springer et al. (2015) (Raghavan, et al. 2016) Elevated risk to A. americanum adults and nymphs (separately) has also been predicted in Missouri using logistic regression (Brown, et al. 2011) at a resolution of 30m x 30m Elevated risk was defined as collection rate of ticks above the median rate across the study area. Relative humidity and habitat type (forest vs grassland) were significant predictors A. americanum nymphs and adults. Forested area was found to be highly favorable to both nymphs and adult lone star ticks. The purpose of th is current study was to determine the relative importance of environmental and climatic variables in determining the presence of adult Amblyomma americanum in Florida. A species distribution modeling approach developed using a logistic regression framework to estimate the prob able occurrence of A. americanum was implemented to


26 predict the spatial potential distribution of the species across Florida at a resolution of one hectare. 2.4 Materials and Methods 2. 4 .1 Study Area The surveyed area for this study encompasse d mainland Florida and exclud ed the Florida K eys The state includes a wide range of environmental conditions with high climatic variability from north to south which supports numerous habitats The National Oceanic and Atmospheric Administration (NOAA) divides the state into seven distinct regions considered climatically homogeneous (Keim, Fischer and Wilson 2005) B oth the El Nio Southern Oscillation (ENSO), and the Atlantic Multidecadal Oscillation (AMO) impact long term climate trends in Florida however the dominant drivers include latitude (as reflected in the NOAA climate divisions), prevailing winds and pressure systems, as well as ocean currents (Tsai, Southworth and Waylen 2014) N orther ly regions of Florida experience greater seasonal variability and receive a larger proportion of rainfall during the winter months than southern regions The seasonal precipitation patterns are refle cted in patterns of changing greenness (using the Normalized Difference Vegetation Index or NDVI ) wherein greenness tends to increase in the fall and winter and decline in the spring (Tsai, Southworth and Waylen 2014) Sampled habitats ranged from monoculture agricultural cro p fields, seasonal wetlands, hardwood forests and pine scrub. Central and southern regions of the state contain ed a high percentage of seasonal fields and orchards, primarily citrus crops. A set of 33 locations (henceforth called various Florida State Parks, county extension service locations, and University of Florida affiliated Institute of Food and Agricultural Science research and education centers (IFAS REC) (Figure 2 1 ) These sites


27 were selected to be widely distributed geographic ally and to incorporate variability in local environ mental and climatic conditions. 2. 4 .2 Tick Collection and Environmental Data Sources Observed presence or absence of adult Amblyomma americanum was determined from collections performed on transects from each site between September 2015 and September 2016 using flagging. Flags consist ed of a 1 m 2 white flannel cloth attached to a pole which could be detached for storage and cleaning. Al l tick collections were performed in accordance with the Florida Department of Environmental Protection under Research and Collection Permit # 03111610 Each site was visited a minimum of twice during the collection period, with most sites being repeatedly sampled approximately once per month between April and September. At each of the 33 sites pairs of 100m 200m transects were performed through primary areas of homogeneous land cover types. Transects were located by latitude and longitude with the World Ge odetic System 1984 ( WGS84 ) using a GPS enabled Nexus tablet at the start, middle, and end points of the transects using the DoForms application for Android to input data Each GPS tagged location wa s time stamped and accompanied by a photograph of the surr ounding land cover including canopy and understory for ground truthing of land cover coverages A minimum of one set of paired transects were run at any given site although sites with multiple dominant land cover types include d upwards of five pairs of tra nsects (10 total transects at the site). Each transect was given a unique identifier by county, site, date, and transect number. As sites were sampled multiple times throughout the study period transects repeated at the same location were given the same transect number although these transects may not exactly inc lude the same start and stop points Because care was taken to limit each transect to a single land cover type there was variability in the length of transects between sites.


28 The flags were checked for ticks approximately every 10m along a transect and sp ecimens were removed and placed into vial s containing 100% EtOH. Flagging material was placed into plastic bags after each transect so that any remaining tick larvae and nymphs could be removed later without cross conta minating subsequent transects. Ticks were identified morphologically by species and life stage based on standard taxonomic keys (Keirans and Litwick 1989, Strickland, et al. 1976) and adult s and nymph s were stored at 80C. Presence of larval ticks were n oted and identified to suspected genus although no specimens were retained. The environmental and climatic variables included in this analysis were selected primarily based on reported associations with tick presence/absence in the literature (Table 2 1) (Springer, et al. 2015, James, et al. 2015, Fryxell, et al. 2015) Variables describing environmental parameters we re l and cover Normalized Difference Vegetation Index ( NDVI ) elevation, and distance to water bodies These datasets were retrieved from the Florida Geographic Data Library (FGDL) and derived from the Florida Cooperative Land Cover Database (CLC) MODIS NDVI composites, the ASTER Global DEM, and the na tional hydrography dataset respectively. Varia bles describing climate parameters are the 19 (BioClim) calculated by Hijmans et al. (2005 ) from the WorldClim Version 1.4 climate database and use the same naming schema. The CLC Database includes state wide classifications of all major land cover types at a native resolution of 1 0m Classification of land cover types is determined at both the state level and at site level which includes additional classifications for certain areas. Using the existing hie rarchical classification scheme major state level land cover types were reclassified as one of five primary types: Forest, which includes pine and hardwoods; Shrub, encompassed shrub and


29 brush lands; Grasses; Wetlands; and a final general category includi ng all other land types such as water bodies and urban areas (Table 2 2) The five selected categories potentially capture biologically relevant relationships between ticks and their ecological preferences. Areas dominated by forested landscapes provide a multitude of cond itions beneficial for ticks; forests provide habitat and protection for po tential hosts and i ncreased leaf litter and canopy cover produce microclimates with increased humidity and cooler midday te mperatures. Similarly, shrub dominated landscapes provide food sources and suitable microclimates for ticks. A study by Ostfeld et al. (1995) showed that shr ubby areas Table 2 1. Characteristics of variables Variable Dataset Resolution (m) Measure Forest Cover CLC v3.0 10 binary Shrub Cover CLC v3.0 10 binary Grass Cover CLC v3.0 10 binary Wetland Cover CLC v3.0 10 binary Other Cover CLC v3.0 10 binary Elevation ASTER global DEM 30 Continuous Minimum annual NDVI Modis 250 Continuous Maximum annual NDVI Modis 250 Continuous Average annual NDVI Modis 250 Continuous Mean Temperature [Bio1] WORLDCLIM 100 0 Continuous Mean diurnal temperature range (mean(period max min)) (C) [Bio2] WORLDCLIM 100 0 Continuous Isothermality (Bio02 Bio07) [Bio3] WORLDCLIM 100 0 Continuous Temperature seasonality (Coeff of Variation) [Bio4] WORLDCLIM 100 0 Continuous Max temperature of warmest month (C) [Bio5] WORLDCLIM 100 0 Continuous Min temperature of coldest month (C) [Bio6] WORLDCLIM 100 0 Continuous Temperature annual range (Bio05 Bio06) (C) [Bio7] WORLDCLIM 100 0 Continuous Mean temperature of wettest quarter (C) [Bio8] WORLDCLIM 100 0 Continuous Mean temperature of driest quarter (C) [Bio9] WORLDCLIM 100 0 Continuous Mean temperature of warmest quarter (C) [Bio10] WORLDCLIM 100 0 Continuous Mean temperature of coldest quarter (C) [Bio11] WORLDCLIM 100 0 Continuous Annual Precipitation [Bio12] WORLDCLIM 100 0 Continuous Precipitation of Wettest Month [Bio13] WORLDCLIM 100 0 Continuous Precipitation of Driest Month [Bio14] WORLDCLIM 100 0 Continuous Precipitation seasonality (Coeff of Variation) [Bio15] WORLDCLIM 100 0 Continuous Precipitation of wettest quarter (mm) [Bio16] WORLDCLIM 100 0 Continuous Precipitation of driest quarter (mm) [Bio17] WORLDCLIM 100 0 Continuous Precipitation of warmest quarter (mm) [Bio18] WORLDCLIM 100 0 Continuous Precipitation of coldest quarter (mm) [Bio19] WORLDCLIM 100 0 Continuous Distance to water Natl Hydrography Dataset 100 0 Continuous


30 supported high I. scapularis density surpassed only by oak and maple forests. The authors posit that hosts using these areas for food and shelter played a role in tick numbers. Furthermore, a study by Allan et al. (2010) indicates that areas dominated by brushy vegetation (the invasive honeysuckle species Lonicera maackii ) had higher counts of nymphal A. americanum compared to more sparsely growing native vegetation, and that removal of the species redu ced ov erall nymph numbers. Grassland areas likely provide some contrast with areas of moist microclimates. I. scapularis are unlikely to inhabit open grasslands even though it provides a thick layer of moisture providing chaff (Ostfeld, C epeda, et al. 1995) A. americanum are generally considered fairly drought tolerant which may permit colonization of grassland ecosystems that would be otherwise too xeric to support other tick species Wetlands in Florida consist of areas temporarily or permanently inundated with water throughout the year. Permanently submerged areas are unlikely to harbor permanent tick populations however; areas with transient water or seasonal flooding may allow popul ations to persist during times of no water. Weiler et al. (2017) found that after a flood event of the Danube River, Ixodes ricinus Haemaphysalis concinna and Dermacentor reticulatus were recovered after approximately 2 weeks of submersion albeit in lowe r numbers compared to pre flood collections Under laborato ry conditions adult lone star tick has been shown to survive submersion for extended periods (LT50: 3.3 3.5 weeks ) with nymphs survival extending far beyond that time (Koch 1986) The final land cover classification con sidered in the study encompassed all remaining areas that would be inhospitable to lone star tick populations. These areas consist ed primarily of water bodies and urban areas. Several miscellaneous land co ver classifications that encompassed negligible area were included in this category as they were not obvious fits for other categories.


31 Table 2 2. Reclassification scheme for land cover variables CLC Land Cover Type CLC State code Reclassified LC Hardwood Forested Upland 11 Forest High Pine and Scrub 12 Pine Flatwoods and Dry Prairie 13 Mixed Hardwood Coniferous 14 Shrub and Brushland 15 Shrub Coastal Uplands 16 Cu l tural Terrestrial 180 Grass mowed grass 181 rural 183 Palustrine 20 Wetlands Freshwater Non forested Wetlands 21 Freshwater Forested Wetlands 22 Non vegetated Wetland 23 Altered Wetlands 24 Lacustrine 30 Other Natural Lakes/Ponds 31 Artificial Lakes/Ponds 32 Riverine 40 Natural Rivers and Streams 41 Cultural Riverine 42 Estuarine 50 Subtidal 51 Intertidal 52 Cultural Estuarine 53 Marine 60 Surf Zone 61 Exotic Plants 70 Australian Pine 71 Melaleuca 72 Brazilian Pepper 73 Exotic Wetland Hardwoods 74 Unconsolidated Substrate 91 Barren, Sinkhole, Outcrop 17 urban 182 transportation 184 communication 185 utilities 186 extractive 187 Bare Soil/Clear Cut 188


32 Maximum, minimum, and average normalized difference vegetation index (NDVI) were calculated for the study period from MODIS NDVI 16 Day composites. NDVI is a frequently considered metric when estimating the distribution or ecological niche of ticks. It is considered a proxy for moisture availability i.e. high humidity microclimates produced by high vegetation density or leaf litter. However, there is some dispute about the ability of NDVI to produce meaningful predictions of tick presence or population dens ity. Bisanzio (2008) found that the correlation between NDVI and Ixodes ricinus was generally positive although the strength of the correlation varied throughout the year and between years. This variability indicates that the usefulness and specification o f NDVI metrics should be considered carefully. Studies of mosquito ( Aedes and Culex spp. ) populations have also found variable correlations with NDVI, however due to the very different biology between these invertebrates, the underlying reasons are likely very different (Britch, et al. 2008, Linthicum, et al. 1990) Furthermore, the geographic extent of the study area produces significant variability in green up time and NDVI patterns as noted by Tsai et. al. (2014) These factors precludes the use of a single statewide NDVI 16 day composite to represent maximum or minimum NDVI. In or der to address potential spurious geographic or temporal intra annual and inter annual variation or correlation in this study, the absolu te maximum, minimum and average NDVI values were calculated by pixel for a five year period preceding and including the study period. The bioclimatic variables from the WorldClim dataset were calculated from interpolated surfaces of average maximum and mi nimum temperature and precipitation data from 1960 1990 (Hijmans, Cameron, et al. 2005) This dataset is widely used in ENM and SDM applications. Porretta et al. (2013) and James et al. (2015) considered these variables in esti mating the distributions of I. ricinus and D. variabilis respectively. The base variables of maximum, and


33 minimum temperatures and precipitation can have direct biological impacts on biology and physiology (Randolph, et al 2002, Ogden, et al. 2005) High and low tempera tures may reduce questing activity and survivability. High p recipitation especially in low lying areas can impact survivability of eggs (Weiler, et al. 2017, Koch 1986, James, et al. 2015) The bioclimatic variables derived from these base variables may have other biologically significant relationships with tick occurrence (James, et al. 2015) 2.4 .3 Methodology to Accou nt for Spatial Autocorrelation The sampling procedure used to collect ticks at each of the 33 collection sites results in a high density of unique transects within a very small area (<1km). The clustered nature of the presented study design len t itself to two issues that are common in species distribution models : sampling or observation bias and spatial autocorrelation (Guisan and Zimmermann 2000) Spatial autocorrelation is a statistical property inherent in many ecological processes in which paired observations in geographic space are more or less similar than expected for randomly associated observations (Legendre 1993, Dormann, et al. 2007) In regression analyses and statistical testing the presence of positive or negative spatial autocorrelation can violate the assumption of independence among observations (Legendre 1993) The procedure of repeatedly sampling each site over time results in multiple, proximal points with similar counts of adult A. americanum and therefore a high degree of spatial autocorrelation of tick presence or absence between transects. T o evaluate the impa ct of this sampling scheme, and to evaluate spatial autocorrelation semivariogram analysis of A. americanum abundance was performed to establish the distance at which the number of collected adult sp ecimens was independent of fin dings at nearby transects (the r ange of the semivariogram) (Legendre 1993) The avera ge nearest neighbor distance was calculated from the midpoints of all transects and used to set the bin width in the semivariogram. The resulting range


34 of the semivario gram i ndicates independence of observa tions from neighboring observations i.e. where the semivariance between observations stops increasing (Karl and Maurer 2010) To account for spatial autocorrel ation at distances less than the calculated range a ll transects for which the GPS designated midpoint were located closer together than the semivariogram range distance were collapsed to a single central point in ArcGIS. The resulting point, was designated as having presence or abse nce (1 or 0) of A. americanum if adult specimens were collected from one or more transects represented by that point. Transects for which t he midpoint was greater than 111 m from another transect were retained as individual points. All environmental and cli matic data layers were resampled to 100 m x100 m resolution to match the spatial scale of the aggregated transects and to preserve a large enough sample size for modeling Resampling raster data to smaller cell sizes than the native resolution does not provi de additional information from the raster. For example, resampling a 1 km x 1km cell with a value of two to a resolution of 100m x 100m will simply result in 100, 100m x 100m cells each with a resolution of two, which occupy the same extent as the original 1km 2 cell. The discrete values of the land cover data set were resampled using a nearest neighbor method in ArcGIS. This method is well suited for discrete data as the origi nal cell values are retained (Richards 2013) All other data were resampled to the extent and resolution of the land cover layer using bilinear interpolation in R (Richards 2013) All layers and dat a were then projected to an Albers Conic Equal Area pr ojected coordinate system to minimize linear scale and area l distortion (U.S. Geological Survey 1987) ArcGIS and R were used to process spatial data, project the data in projected coordinate systems, perform interpolations and resample raster layers.


35 2.4 .4 Multivariate Logistic Regression In this study, a multivariate logistic regression was used to est imate the relationship between the presence or absence of A. americanum and the set of environmental covariates. The standard form of the multivariate logistic regression is expressed in the form of Equation 2 1 (Hosmer and Lemeshow 1989) : Where (2 1) The dependent variable, Y, is the conditional probability of the outcome being present. In the context of this study, Y is the probability of A. americanum being present at a specific location. The logit function, g(x), is the linear combination of the variables and their coefficients, X j j respectively (Guisan, Edwards and Hastie 2002) Once the associated variables ar e determined and their covariates are estimated, the probability of species presence or absence can be predicted for any location on the landscape using Equation 2 2: (2 2) Due to the relatively small sample size, the lo gistic regression model was validated using k fold cross validation. Procedures for model building, selection, and validation are described below. The model building and selection processes for the log istic regression were performed on centered and standardized continuous variables. The centering and Z score standardization of continuous variables provided a uniform basis by which the magnitude and direction of a


36 variable can be compared across models (Bring 1994) Each continuous variable considered was centered using equation 2 3: (2 3) Where the centered values of covariate X are the differ ence of the column mean from each element of covariate X. The result is a distri bution of values centered on zero. Similarly, the resulting centered variables are scaled using equation 2 4 (Bring 1994) : (2 4) When performed on centered data, as done here, the scaling function results in data centered on zero with a standard deviation of one (Bring 1994) This allows for greater interpretability of regression coefficients (Schielzeth 2010) Selection of variables for consideration in the final model search was performed in two parts. The R code written for model building is included in Appendix A As part of an initi al exploratory data analysis, univariate relationships were evaluated to identify the set of covariates with a significant (p<0.05) association s with the dependent variable (presence/absence of A. americanum ) (Springer, et al. 2015) The restriction of considered variables during this step reduce d the computation time of model searching while also limiting the possibility of misspecification of the final model that included unimportant variables (Chatterj ee and Hadi 2013) M any of the remaining variables considered in this analysis are derived from precipitation and temperature measurements As a result, many of those spatial patterns described by these variables were highly correlated (Springer, et al. 2015) T o reduce the likelihood of introducing collinearity into the model a stepwise procedure was implemented in R


37 e inflation factor (VIF) for the remaining predictors (Naimi, et al. 2014) The VIF is calculated from the square of the multiple correlation coefficient determined by regressing a given variable against all other variables (Chatterjee and Hadi 2013) The higher the VIF, the greater the correlation between the selected response variable and all other covariates (Graham 2003, Craney and Surles 2002, Chatterjee and Hadi 2013) A VIF with a value greater than 10 was considered a sign that the model had a collinearity problem (Chatterjee and Hadi 2013) i.e. the selected response variable was highly collinear with the other predictors. The variable with the highest VIF was excluded and the VIF was recalculated for all remaining variables. This process was repeated until no variables remained with a VIF>10 (Naimi, et al. 2014, Craney and Surles 2002) A major criticism of using stepwise regressio n for selecting model variables is that the order of parameter entry or deletion can affect the parameterization of the final model (Whittingham, et al. 2006, Derksen and Keselman 1992) However, t his VIF stepwise proc edure is largely unconcerned with the predictor variable s relationship with the dependent variable as VIF relates to how much variability in each independent variable is explained by the other independent variables (Craney and Surles 2002, Graham 2003) As a result, the sole purpose of this procedure is the minimization of collinearity among predictors, and not specifically the development of a variable set that best fits the response variable ( Graham 2003) T he remaining uncorrelated variables were evaluated with an exhaustive logistic regression 2 4 .5 Model Evaluation Evaluation of model performance is the most important step in developing a useful model (Symonds and Moussalli 2011) Both the statistical assumptions and predictive performance must be assessed to ensure that model is suitable and well characterized for its intended purpose. Although logistic regression rel axes many of the statistical assumptions of linear regression,


38 remaining assumptions such as absence of collinearity among predictors and independence of error structure must be met (Hosmer and Lemeshow 1989) In addition, the predictive performance must also be evaluated to ensure that it is reliable and discriminatory. The identification of significant, non collinear variables resulted in a subset of continuous variables and binary categorical variables used for final model selection. The final, model se arch of first order variable interactions. Exhaustive model searches fit a set of models using every combination of covariates including a model with every covariate included, and one with no covariates. With n covariates, the result is 2 n possible models (Chatterjee and Hadi 2013) Exhaustive model searches can be time, and computationally intensive: in selecting a best model from a set of 6 possible covariates, 2 6 (64) models must be estimated; if another two covariates are co nsidered, the number of possible model combinations balloons to 256 (2 8 ). The use of standardized variables allowed for evaluation of model quality based on the corrected Aikake information criterion (AICc). This measure compares a model relative to a suit e of other among the set of models (Chatterjee and Hadi 2013, Symonds and Moussalli 2011) These models represent the set of possible variable comb inations that pass the statistical assumption of non collinear predictors The regression residuals of the selected best model were evaluated for spatial and identical d istribution of residuals. To determine if spatial dependence in the underlying data structure was adequately controlled by aggregating the data points the s patial distribution of model residuals was assessed for clustering (Franklin, et al. 2009) The equation for the Global


39 statistic is given by equation 2 5 and can be evaluated by z score and associated p value, provided in equation 2 6 (Anselin 1995) ( 2 5) The associated z score is computed as: (2 6) randomly distributed Interpretation of the p value returned by this statistic using a two sided approach infers negative, positive, or no spatial autocorrelation. A non significant p value indicates the null hypothesis cannot be rejected whereas a significant p value and positive or negative z score indicates positive or negative spatial autocorrelation, respectively (Anselin 1995) If there is positive spatial dependence remaining in the model, as indicated by significant p value and positive z score, it could be i ndicative of a poorly defined model (missing explanatory variables) or that a spatial lag or an autologistic error term should be included (Legendre 1993, Miller, Franklin and Aspinall 2007) The Hosmer Lemeshow test was used to evaluate the goodness of fit for the logistic regression. This goodness of fit test estimates fit based on the values of estimated probabilities (Hosmer and Lemeshow 1989) Grouping values b ased on percentiles is a common strategy employed here. square statistic 7: (2 7) Where n k is the number of covariate patterns in the k th group, o k is the number of responses among the n k k is the average estimated probability. The


40 distribution of the Hosmer Lemesho w statistic approximat es that of the chi square with (g 2) degrees of freedom where g is the number of groups or deciles If the p value determined from the chi square table is large we can accept the assumption of good fit in the model. If the p value is very small e.g less t han 0.1 or 0.05 the model is poorly fit (Hosmer and Lemeshow 1989) Due to the relatively small number of positive points in the dataset (n=23), k fold cross validation was selected to estimate the m performance as an alternative to further reducing the effective size of the dataset by dividing into training and testing sets. Cross validation (CV) and its variations are hailed for their applicability to a variety of algorithms and frameworks because on ly assume an i.i.d. of data, and that the partitioned data are independent (Arlot and Celisse 2010) To improve accuracy measures, r epeated k fold CV was performed using k=10 folds with 100 repetitions. In this form of CV, the p rocedure follows that of simple k fold CV in which data was divided into k subsets (fold s). Model training uses k 1 folds, and predictions are performed on the fold that was withheld (Arlot and Celisse 2010) This procedure was repeated until each of the 10 folds had been withhel d and probability estimates had been determined. This procedure was repeated 100 times in which the folds were resampled each repetition to e nsure unbiased training and testing folds (Arlot and Celisse 2010) It should be noted that when available, the use of a truly independent dataset (i.e. a unique set of locations sampled post model development) is preferable to splitting or folding data, especially in cases where the trai ning data may not be representative of the entire study region (Pearce and Ferrier 2000) The predictions from each fold are compared to the known values and the model is assessed using a measure of overall accuracy and the Ka ppa statistic. The Kappa statist ic calculated using E quation 2 8 measured the agreement b etween the observed data and those predicted by the model.


41 (2 8 ) The statistic is standardized with a range of 1 to 1 where 1 indicates perfect agreement between the observed and predicted data and values less than zero indicate systematic disagreement. A value of zero is what would be expected by random chance (Allouche, Tsoar and Kadmon 2006) Fielding and Bell (1997) modify the Kappa statistic ranges proposed by Landis and Koch (1977) such that poor agreement: K<0.4; moderate, 0.40.75. The model wa s further assessed against an external dataset to wh ich the model was nave spatially and temporally The testing dataset consisted of 316 transects performed during a subsequent field season between October 2016 and M ay 2017 using the same collection procedures and sites (with an additional 4 sites) detailed in section 2.3.2. The model was used to generate probability estimates for this testing dataset and a variation of the binning scheme for the Hosmer Lemeshow GOF test was performed as well as model accuracy metrics (sensitivity,specificity, accuracy), and Kappa values were determined. The variation of on the Hosmer Lemeshow groupings was used to determine if presence or absence could be easily discerned from the ca lculated proba bility estimates. T he bins for Equation 2 7 are defined such that g=10 groups can be based on fixed values corresponding to deciles of estimated probabilities or as done here, determined based on percentiles of estimated probabilitie s such that n 1 =n/10 observations have the smallest estimated probabilities and the n 10 =n/10 observations have the l argest estimated probabilities (Hosmer and Lemeshow 1989) The ratio of positive and negative observations across the bins should ideally follow the curve of the logistic function, however a distribution tending towards dichotomous separation of observations is also indicative of high discrimination by the model (Pearce and Ferrier 2000)


42 The calculation of these metrics on an independent dataset evaluates how reliably the model can make predictions for unseen data and how well it can discern presence and absence in unseen data. 2. 4 6 Spatial Prediction s and Gradients The geographic distribution of A. americanum was estimated by generating a map of predicted probabilities for occurrence based on the output of the logistic regression model. The raster layers for each variable included in the logistic regression model were stacked in R and output values of the lo gistic regression were estimated for every location on the landscape using The values of each pixel in the stack of raster layers at a given location prov ide the inputs for the logistic regression and outputs a probability value for that location with a range from 0 1. The closer the predicted probability is to one, the more likely that the environmental conditions at that pixel would be suitable for A. ame ricanum This method produced a statewide probability surface representing probable occurrence of A. americanum at a resolution of one hectare The potential distribution of A. americanum in Florida was then estimated by reclassifying the map of occurrence probabilities based on a specified cutoff criterion The cutoff value used in this study was the probability value where sensitivity and specificity are equivalent. Sensitivity and speci ficity are measures of the percentage of actual presence values correctly predicted by the model, and the percentage of actual absence values correctly predicted by the model, respectively (Fielding and Bell 1997) By selecting a cutpoint where these two measures are equivalent, implies that positive and negative observations are likely to be predicted correctly with equal chance (Freeman and Moisen 2008, Fielding and Bell 1997) All probability values greater than the sensitivity/specificity cutoff are deemed suitable for presence


43 of adult A. americanum and all values less than the cutoff are deemed unsuitable for adult A. americanum The existence of latitudinal gradients in th e predicted distribution of occurrence was evaluated by both changes in the proportion of suitable habitat for A. americanum and by a reduction in s from north to south Highly suitable was defined as a more conserva tive estimate of suitable habitat. Whereas suitable habitat was defined was determined as probability estimates greater than or equal to the average probability estimate for the entire state. Although these two measures are similar they provide information on two different questions about the spatial distribution of A. americanum in Florida : 1. Is existence of suitable habitat distributed uniformly across the sta te? 2. Is suitability of predicted habitat distributed uniform ly across the state? The answers to these questions provide addition information about the risk of exposure to ticks in Florida. The proportion of area predicted to be suitable for A. americanum using the LPT cutoff was determined in one hectare wide ban ds extending east to west across Florida. However, the trend is presented here as a 10km wide average solely to smooth local variability for visualization purposes The proportional area for each band was calculated from t he summation of the number of cells above the cutoff (suitable areas) divided by the total number of cells within that band. Similarly, the trends in predicted probabilities w ere assessed by determining the percentage suitable area containing high predicted probability value s Using the same 10km bands, the total area with high probabilities greater than the mean occurrence probability of was dete rmined and divided by total suitable area withi n that band


44 2.5 Results A. americanum D variabilis and I scapularis which comprise t hree of the five most common pest ticks in Florida wer e collected during this study ( Table 2 3 ) The target species, A. americanum was the most commonly collected species followed by D variabilis and I scapularis The gulf coast tick, Amblyomma maculatum and the brown dog tick, Rhipicephalus sanguineus were not found. Adult ticks belonging to any of the identified three species were collected from 23 of the 33 (70%) sites. Adult A. americanum were collected from 33 of 328 unique transects (10%) spread across 13 of 33 (40%) sites, ( Figure 2 1 ) Nymphal ticks were collected in far greater numbers than adults, accounting for 63% of all collected ticks. Large numbers of larval ticks were collected at several sites, however they were not counted or included in the results of this study. One site, Faver Dyke state park accounted for approximately 50% of all adult A. americanum collected in this study. San Felasco state park pr oduced the second highest percentage of adults (13%) Anecdotal evidence from correspondences with state park staff and officials at other sites support that these are hot spot areas; officials from other sites indi cated ticks were far less of a problem.


45 Table 2 3. Specimen totals by site. Site Name Ixodes scapularis Dermacentor variabilis Amblyomma americanum Amblyomma maculatum Rhipicephalus sanguineus Nymphs Apopka 0 0 0 0 0 0 Astor 1 1 4 0 0 9 Big Talbot 7 0 2 0 0 15 Colt Creek 0 1 0 0 0 0 Everglades REC 0 0 0 0 0 0 Farles Lake 0 2 6 0 0 3 FaverDyke 0 0 50 0 0 93 Fore Lake 0 1 8 0 0 6 Ft Pierce 0 0 0 0 0 0 Gainesville 0 0 0 0 0 0 Jay 0 0 0 0 0 2 JD MacArthur 0 0 0 0 0 0 Juniper Prairie 0 0 0 0 0 1 Lake Kerr 0 0 0 0 0 1 Lovers Key 0 0 0 0 0 0 Marianna 0 0 0 0 0 0 Monticello 0 0 3 0 0 21 Myakka 0 0 0 0 0 0 Oleta 0 0 0 0 0 0 Ona 0 1 0 0 0 0 Paleo Hammock 0 0 0 0 0 0 Patrick AFB 0 0 0 0 0 0 Pine Castle 0 4 3 0 0 4 Quincy 0 0 1 0 0 0 Rodman 0 0 0 0 0 0 San Felasco 0 0 13 0 0 35 Satsuma 0 0 5 0 0 17 Sellers Lake 0 0 1 0 0 1 St Sebastian 0 0 0 0 0 0 SW REC 0 0 0 0 0 0 Tropical Rec 0 0 0 0 0 0 Tyndall 0 0 1 0 0 0 Yulee 3 4 1 0 0 1 Totals 11 14 98 0 0 209


46 Figure 2 1 : Distribution of sample collection sites from August 2015 August 2016 indicating where adult A. americanum were collected (red), and locations that returned no adult A. americanum (black).


47 The majority of n ymphs were identified as belonging to the genus Amblyomma but were not in cluded in the study as a specific identification could not be made. Amblyomma nymph s were collected from 14 sites including three locations from which no adult A. americanum were repo rted. In total, data from 328 individual transects surveyed from September 2015 through August 2016 were used in this study. Results from the collection effort for this study indicate questing adult A. americanum occurred from April through September with a peak number of adults collec ted per sampling effort in June (Figure 2 2) The largest numbers of adult A. american um were collected in April. Figure 2 2: Monthly totals of adult Amblyomma americanum and collection intensity during the study period of Septeber 2015 through August 2016. No collections took place during December, January, and February. The highest total number of adult ticks was collected during April, however the highest catch per unit effort occurred in June. 0 10 20 30 40 50 60 70 80 90 100 0 5 10 15 20 25 30 35 40 45 Sep-15 Oct-15 Nov-15 Dec-15 Jan-16 Feb-16 Mar-16 Apr-16 May-16 Jun-16 Jul-16 Aug-16 Total Number of Transects Adult Ticks Collected A. americanum Sampling Frequency


48 Mean transect length was 134.2 m eters with an interquartile range of 70.3 meters Adult A. americanum were absent from all sites south of Ocala National Forest with the highest numbers collected from north central Florida and the central Atlantic coast. The results from the semivariogram analysis indicate d that variance between A. americanum counts increase d up to approximately 110m, giving the effective range at which the number of ticks found on neighboring transect are independent of each other. This r esult was used to dictate the aggregation distance for the 328 transects. After collapsing neighboring transects within 110m of one another, the resulting dataset was composed of 13 0 points with 23 points containing positively identified adult A. americanum Due to the relatively small number of positive locations a ll points were used to build the logistic regression model. Results from the univariate analysis of significant re lationships between tick presence and the environmental covariate s in dicate d that 12 of the 28 considered var i a bles provided no significant predictive ability (p> 0.05) These variables were removed from consideration for the remainder of the model building process Table 2 4 Environmental and climatic variables used to model the spatial distribution of Amblyomma americanum in Florida Variable P V alue F orest 0.008204 B io1 0.004293 B io3 0.01259 B io4 0.006253 B io6 0.002727 B io7 0.002762 B io8 0.020901 B io9 0.001816 B io10 0.001376 B io11 0.004418 B io13 0.002655 B io15 0.045435 B io16 0.011945 B io18 0.009604 Max NDVI 0.000421 Mean NDVI 0.001037


49 The remaining 16 variables, which significant ly predicted Amblyomma american um presence in the logistic regression model, are listed in ( Table 2 4) The results of the VIF procedure further reduced the number of variables under consideration from 16 to 10 by limiting the potential for collinearity in the final model ( excluding variables with a VIF>10 ) After excluding collinear variables, the minimum and m aximum correlations between variables are 0.003 and 0.8, respectively. The final set of variables used to build the model ( Table 2 5 ) c onsist ed of one land cover variable (forest), a suite of bioclimatic measures pertaining primarily to precipitation, and two measures of NDVI. Table 2 5 Environmental and climatic variables with low collinearity determined by Variance Inflation Factor, highly collinear variables not shown. Variables listed constitute final variable set considered in logistic regression model selection. Variable VIF F orest 1.104511 B io3_fl 4.587343 B io7_fl 4.507246 B io8_fl 2.272024 B io9_fl 4.087867 B io13_fl 6.758279 B io15_fl 7.523549 B io18_fl 8.014768 NDVI_max 1.982174 NDVI_mean 2.06272 Selection of t he final multivariate logistic r egression model utilized an exhaustive model search on the remaining 10 variables resulting in 1024 models being considered The set of models with the smallest AICc values differed by less than 2.0 units indicating they are of essentially equal quality (Symonds and Moussalli 2011) After careful consideration ultimately the model with the lowest AICc was selected. This model (Table 2 6 ) included 6 of the 10 considered variables each with a significance of p<0.01.


50 Table 2 6 Coefficients and variables included in the best logistic regression model to predict presence of A. americanum across Florida Coefficients: Beta Std. Error Z Value Pr(>|z|) Intercept 3.1486 0.5878 5.356 8.49E 08 *** Forest 1.3328 0.6366 2.094 0.034006 Bio3 1.1053 0.5214 2.12 0.034006 Bio8 1.0896 0.4937 2.207 0.027313 Bio13 1.4302 0.6082 2.352 0.018692 Bio15 1.6852 0.6718 2.508 0.012126 Max NDVI 1.4121 0.387 3.648 0.000264 *** Sign. Codes 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' The variables included in the final model were Forest cover, maximum NDVI, Bio3 (temperature stability or isothermality), Bio8 ( mean temperature of wettest quarter ) Bio13 ( precipitation of wettest month ) and Bio15 ( precipitation seasonality ) Both forested land cover ( which includes predominantly coniferous forest, deciduous forest, and mixed forests) and maximum NDVI (a proxy for vegetation health) were both positively associated with tick presence This indicates that A. americanum is more likely to occur in areas with higher vegetation health (greenness) and in forested areas. Three of the four included bioclimatic variables: Bio3, Bio8, and Bio13 were all negatively associated with A. americanum presence. Thus, as isothermality in creases (Bio3) occurrence of A. americanum is likely to decrease. Similarly, areas that receive the highest amounts of rainfall during the wettest months (Bio8) or are the warmest during the wettest quarter, are associated with a lower probability of occur rence for A. americanum The final bioclimatic variable included in the model, precipitation seasonality (Bio15) was positively associated with tick presence indicating that as intra annual variation in precipitation increases (i.e. areas with more pronoun ced wet and dry seasons) probability of occurrence for A. americanum increases as well.


51 T o test the assumption s of statistical indepe ndence and adequate control of spatial autocorrelation in the final model the spatial distribution o f the standardized re siduals were s I. The null hypothesis of is complete spatial randomness in the distribution of point values. The test returned a n insignificant (p>0.05) p value of 0.83 which indicates the null hypothesis should not be rejected. This result indicates that one of the main assumptions of logistic regression, ind ependence of model residuals, was met and that the model adequately control led for spatial autocorrelation The sp atial distribution of residuals indicates that presence locations f requently had larger positive residuals ( Figure 2 3 ). The assessment of model accuracy via the Hosmer Lemeshow and cross validation (Table 2 7 ) indicate d that the model d id an adequate job of fitting the training data. The model successfully fits the training data as indicated by a non significant p value for the Hosmer Lemeshow GOF test. Further validation via k fold cross validation indicate d that the model did an exceptional job of ident ifying true positive points (sensitivity), however it did a much poorer job of discerning true negativity points (specificity). The Kappa metri c is considered a better measure than overall accuracy for determining how well a model performs as it is a comparison against what could be expected by random chance. The calculated K appa statistic of 0.43 indicated a solid improvement in model performance over what is expected by chance (Fielding and Bell 1997) Table 2 7 Model validation metrics indicating model fit against training and testing datasets indicating a reduction in overall accuracy and sensitivity. Hosmer Lemeshow GOF (p value) Sensitivity Specificity Accuracy Kappa Training 0.087 0.94 0.48 0.86 0.43 Testing 3.27E 12 0.63 0.83 0.81 0.278


52 B y examining the ratio of positive and negative observations in each decile of estimated probabilities it was found that the model produces a nearly dichotomous discrimination between presence and absence (Table 2 8 ). Positive observations are absent until reaching the top 40% of observations and correspond to a minimum estimated probability of 0.1. Table 2 8 Distribution of presence to absence points for each range of estimated probabilities indicating good disc rimination by the model. Each bin corresponds to the range of estimated probabilities for 10% of observations in the testing dataset. Expected Probabilitie s Absence Presence Proportion of Presence (1.3e 06,6.9e 03) 31 0 0 (7.1e 03,1.5e 02) 31 0 0 (1.6e 02,3.0e 02) 31 0 0 (3.0e 02,4.4e 02) 31 0 0 (4.4e 02,9.3e 02) 31 0 0 (9.4e 02,1.8e 01) 28 3 0.11 (1.8e 01,3.2e 01) 28 3 0.11 (3.4e 01,4.9e 01) 27 4 0.15 (5.0e 01,6.0e 01) 27 4 0.15 (6.0e 01,9.4e 01) 24 13 0.54 Further, calculations of model accuracy (sensitivity, specificity, accuracy, and Kappa) indicate d large reductions in Kap pa and sensitivity, although the model continued to perform better than random Specificity of the model was greatly improved against the testing dataset however; the reduced sensitivity produced a slight decrease in overall lower model accuracy.


53 Figure 2 3 : Spatial distribution of model residuals. Legend values indicate the five quantiles: minimum, 25%, median, 75%, maximum. Positive and negative values indicated by green and red, respectively.


54 The resulting spatial prediction based on the logistic regress ion model (Figure 2 4 ) indicates that large portions of northern and central Florida have moderate to high predicted occurrence probabilities for A. americanum. The areas with the highest predicted probabilities for occurrence are the far western panhandle gulf coast and the far northeastern corner extending south along the coast to the Melbourne area and inland to the Ocala National Forest. The spatial prediction indicates that most of s outhern Florida is n ot s uitable for adult Amblyomma americanum. The cutoff value for delineating presence from absence in the model was 0. 186 which is where the model correctly predicts approximately 78% of presence and absence values correctly. This value was used as t he cutoff value to distinguish what could be suit able conditions ( presence location ) from an unsuitable ( absence ) location. In James et al. (2015) this method of determining a cutoff to delineate medium from high suitability was used and produced a value of 0.1. The average probability value of suitable areas based on the LPT in the present study was determined to be 0.42 The spatial prediction was reclassified based on th is cutoff to produce a map of the estimated spatial distribution of adult A. americanum (Figure 2 5 ) in Florida An examination of the spatial gradients in predicted occurrence of A. americanum (Figure 2 6 ) produced obvious directionality with the percentage of land suitable for the species based on the LPT threshold. Habitat suitable for adult A. americanum comprised 30 5 0% o f all areas from the Big Bend northward s Less than 1 0% of land in southern Florida was predicted to be suitable for the species The trend of estimated probabi lity values within the potential distribution area was also examined (Figure 2 7 ) This measur e e xamines if areas of high estimated suitability are uniformly distributed within the areas of predicted occurrence. High


55 suitability was defined as an area with probability values greater than the mean value of suitable areas (areas with probability values greater than the LPT). In this study the average probability valu probability valu ). The trend in highly suitable areas roughly mirrors that for suitable a reas in general. T here is a distinct reduction in the suitability of areas within the estimated distribution in southern areas of the state. Between 30 50% of the estimated distribution in north Florida is highly suitable, whereas in the south, 0 10% of th e estimated distribution would be considered highly suitable for the species. T he far southern regions of the state are not highly suitable for the species at all


56 Figure 2 4 : Probabilities of occurrence of Amblyomma americanum in Florida. Green shading and red shading indicates low and high probability of occurrence respectively.


57 Figure 2 5 : Estimated distribution of Amblyomma americanum based on a lowest presence threshold (LPT) indicating substantial areas of suitable habitat in northern Florida an d increased unsuitable areas in southern Florida.


58 Figure 2 6 : Latitudinal trend indicating an increase in percent of total area suitable for A. americanum from South to North in Florida. Points indicate totals in 10km increments. Latitude (km) from South (L) to North (R), expressed in meters, shown on the X axis. Percentage of land area deemed suitable for A. americanum based on LPT is shown on the Y axis.


59 Figure 2 7 : Latitudinal trend indicating an increase in high predicted suitability in relation to total suitable area from South to North in Florida. Points show totals in 10km increments. Latitude (km) from South (L) to North (R), expressed in mete rs, shown on the X axis. Proportion of area within the estimated distribution of A mblyomma americanum predicted to be highly suitable for A. americanum is shown on the Y axis.


60 2. 6 Discussion The estimated spatial distribution of adult A. americanum pre sented here is the first effort to produce a well resolved spatial distribution at the sub county level for Florida. The spatial prediction from the logistic regression model indicates that a large portion of the state is environmentally favorable to this tick species; however, the favorable areas are not uniformly distributed. Areas of suitable habitat drop off significantly south of Lake Okeechobee (27N). A meridional gradient for both the probability of presence and for percent area of predicted habitat indicates that habitat suitable for the lone star tick decreases from Nor th to South. During the study period adult A. americanum were not collected from any sites south of Ocala national forest, which would support the model prediction of limited habitat suitability farther south in the state. However, additional collection efforts performed between September 2016 and May 2017 have returned adult A. americanum as far south as Lake Okeechobee in areas predicted to be suitable by the model The ensemble model of county level habitat suitability for A. americanum by Springer et al. (2015) similarly shows reduced model agreement in sout hern Florida counties indicating lower certainty in predicted habitat suitability. These results reflect reports in the literature (Springer, et al. 2014) and the findings from collections for this study that adult lone star ti cks in southern Florida are scarce; it is still unclear if this is due solely to lower sampling intensity in the southern portion of the state or because of the underlying distribution of ticks. The results of this study also are support ed by findings in Allan et al. (2001) showing regional proportions of A. americanum from ticks collected from deer and swine in Florida drop ped markedly between northern, central, and southern regions ( d eer: 56.9%, 47.9%, 0%; s wine: 27.3%, 0.6%, 1.5%). In this study, it was shown that the percentage of area suitable for A.


61 americanum decrease d from n orth to s outh. While collections by flagging more closely resemble what could be expected of human exposure during outdoor activities, estimates of tick presence or density deriv ed from free ranging wildlife potentially provide a more accurate representation of the true tick burden on the landscape due to the continuous environmental exposure and wide geographic movements of the animals (Ginsberg and Ewing 1989) It is therefore assumed that the significant reduction in A. americanum prevalence found by Allan et al. (2001) across the three ecoregions is result of underlying ecological conditions and not simply local variability between sampling location s. The divisions used by Allan et al. (2001) to classify the state into three regions based on the dominant ecosystem types relies heavily on differences in dominant vegetation types, many of which overlap multiple regions. This approach provides a distin ct boundary underpinned by biologically relevant parameters by which differences are compared. In this study however, land cover classifications were insignificant predictors of A. americanum presence (forest areas being the one exception) and the applicat ion of such an approach would therefore be arbitrary. By considering latitudinal differences in suitable area based on the resolution of the data a more continuous trend in habitat suitability becomes apparent. Both the reduction in suitability and proport ional area of potential occurrence from north to south at this scale supports the notion that ecological conditions become less favorable for A. americanum in southern Florida. big bend area of Florida, along the gulf coast was predicte d to be especially favorable to A. americanum This region was not sampled during the course of this study and it is unknown if the high probability estimates in this area are substantiated by occurrence of ticks or if the results in this area are anomalou s. Springer et al (2015) do not note particularly high habitat suitability in this region likely due to its prominence being masked by


62 the county level scale of analysis. Collections by Sayler et al (2014 ) returned large numbers of adult A. americanum from several Florida state parks bordering this area. However, visualization of the BioClim climate coverages used in this study reveal sharp transitions in temperature and precipi tation trends at this location indicative of possible interpolation errors. Further consideration of this area is contained in Appendix B The variables included in the final model were Forest cover, maximum NDVI, Bio3 (temperature stability or isothermality), Bio8 ( mean temperature of wettest quarter ) Bio13 ( precipitation of wet test month ) and Bio15 ( precipitation seasonality ) Each of the se variables can be interpreted as biologically relevant information about the ecological preferences of A. americanum Each variable included in the model was statistically significant with gr eater than 95% confidence (Table 2 7). Both forested land cover (which includes predominantly coniferous forest, deciduous forest, and mixed forests) and maximum NDVI (a proxy for vegetation health) were positively associated with tick presence. This indi cates that A. americanum is more likely to occur in areas with higher vegetation health (greenness) and in forested areas. Presence of A. americanum was indi cate ticks utilize the microclimate provided by dense leaf litter and brushy understory to prevent desiccation when environmental conditions become unfavorable and to lay eggs (Oliver Jr 1989) Anderson et al. (2008) state that the highest numbers of adult A. americanum are found on brushy vegetation due to the relatively high humidity. NDVI is a proxy for vegetation health and productivity. The inclusion of a positive association with maximum Healthy vegetation produces necessary high humidity microclimates utilized by questing ticks


63 (Anderson and Magnarell i 2008, Vail and Smith 2002) Vegetation growth will generally be greatest during summer months resulting in maximum NDVI values during a time of the year when temperatures are highest and ticks are most prone to desiccation. Areas with high maximum N DVI are significantly more likely to have adult A. americanum present. The determination of significant associations between A. americanum and both Forest land cover and maximum NDVI is noteworthy. During the variable selection process environmental varia bles were checked agai nst each other for collinearity and the correlation between Forest and maximum NDVI was deemed to be satisfactorily low. And while other land cover types such as shrubs or grasslands alone yielded no significant association with tick presence, maximum NDVI values outside of forested areas must play a role in increasing the probability of tick presence. Three of the four included bioclimatic variables: Bio3, Bio8, and Bio13 were all negatively associated with A. americanum presence. Thus, as isothermality increases (Bio3) occurrence of A. americanum is likely to decrease. Isothermality is a quantification of how large day night temperature oscillations relate to summer winter oscillation. High isothermality is indicative of temperatur e evenness over the course of the year, whereas a low isothermality indicates that the size of temperature swings vary between summer and winter. Tick life cycles are regulated by seasons and temperatures can play a role in development and reproduction. J ames et al. (2015) posits that isothermality impacts immature development for D. variabilis and Oliver (1989) states that lone star tick development rate can be accelerated under laboratory conditions by modulating temperature. The negative relationship re ported in this study between A. americanum probability of A. americanum occurrence, appears to contradict the general biolo gy of the


64 species. Previous studies of the environme ntal preferences of ticks at various geographic extents have considered isothermality but found it to be non significant (Springer, et al. 2015, James, et al. 2015) It may be that the isothermality has a very indirect effect not related to the biology of the species. A reas that receive the highest amounts of rainfall during the wettest month (Bio13 ) or are the warmest during the wettest quarter (Bio8 ) are associate d with a lower probability of occurrence for A. americanum The mean temperature of the wettest quarter ( Bio 8) showed a High temperatures can decrease survivorship in ticks by increasing the like lihood of experiencing desiccation during host seeking activities (Knulle and Rudolph 1981, Vail and Smith 2002, Anderson and Magnarelli 2008) While the availability of moisture can reduce this risk by increa sing relative humidity, laboratory studies on survivorship among multiple larval Boophilus ( Rhipicephalus ) species show that even at high relative humidity (97%) survival is significantly impacted by temperature increases from 20C to 25C (Davey, Cooksey and Despins 1991) In the current study, no ticks were found in southern Florida, which coincidently experiences the highest average temperatures in the state during the wettest months (>25C). Springer et al. (2015) found mean temperature of the wettest quarter to be a significant predictor of A. americanum presence. James et al. (2015) cites precipitation of the wettest month (Bio13) as impacting egg development and larval survivability. In this analysis greater precipitation during the wettest high annual precipitation and a great deal of low lying areas and flat topography makes many season, large precipitation events may result in standing water long enough to reduce


65 survivability of eggs or larvae and ultimately impacting adult tick populations. Weiler et al. (2017) showed that major flood events can reduce the abundance of questing ticks of several species. Adejinmi (2011) also found that prolonged submersion in water decreased hatchability of R. s anguineus and H. leachi eggs. Similarly, the final bioclimatic variable included in the model, precipitation seasonality (Bio15) was positively associated with tick presence indicating that as intra annual variation in precipitation increases (i.e. areas w ith more pronounced wet and dry seasons), probability of occurrence for A. americanum increases as well. T he positive association between A. americanum implies that areas of North Florida with a more p ronounced wet and dry season are more likely to harbor adult ticks. The shift between wet and dry seasons may allow areas to dry out enough for eggs and larvae to survive. Koch (1986) showed in the laboratory A. americanum could survive submersion for seve ral weeks so periodic flooding would have little long term effect on the species. In south Florida where the precipitation is less variable, constant moisture in areas otherwise appropriate for sustaining tick populations result in unsuitable conditions. S pringer et al. 2015) and James et al. (2015) found precipitation seasonality to be a significant predictor of A. americanum and D. variabilis distributions, respectively. T he preliminary examination of significant relationships (via univariate logistic r egression during the variable selection process) between the considered environmental variables and the presence of A. americanum revealed that only a fraction of the considered habitat and vegetation characteristics (Forest cover, and two measures of NDVI ) had any significant association. Distance to water bodies, all other land cover types, and a third measure of NDVI had non significant associations with tick presence Of the considered climate variables, six of 19 considered variables were also found to be not significant predictors. Tick physiology is


66 notably modulated by abiotic conditions so the dominance of climate measures is not surprising Survivorship of all life stages, reproductive success, and development are affected by temperature and moistu re availability (Anderson and Magnarelli 2008, Yoder and Tank 2006, Yoder and Benoit 2003, Randolph, et al. 2002) For this reason, many attempts to estimate the geographic distributions of ticks have focused on associating presence or abundance with temperature and moisture measurements (Atkinson, et al. 2014, Estrada Pena 1998, Porretta, et al. 2013, Springer, et al. 2015, Raghav an, et al. 2016, James, et al. 2015) Furthermore, l imited findings (76 sample sites across 74.5km 2 ) have indicated that habitat and vegetation characteristics including land cover type, NDVI, and distance to water, are not adequate predictors of tick presence or abundance in the southeastern United States (Fryxell, et al. 2015) The observed relationships between presence of A. americanum and the variables considered in this study appear to support the notion that climatic conditions (temperature and moisture) are most important in determining the distribution of the species in Florida.


67 CHAPTER 3 CONCLUSIONS This thesis attempts to identify environmental and climatic factors which contribute to the spatial distribution o f adult A. americanum and identify areas with high probability of infestation. Environmental and climatic data was used to generate a species distribution model using a logistic regression framework. The hig h resolution spatial prediction generated by this study can better inform the public and state and federal agencies about areas with high risk of exposure to A. americanum This study constitutes the first attempt at estimating the state wide geographic distribution o f this this pest a t resolutions below the county level The following are the major findings drawn from this study: 3. The distribution of A. americanum in Florida is significantly associated with a number of climatic variables and land cover characteristics. The general excl usion of most variables associated with habitat and vegetation is in agreement with other studies predicting ecological preferences of ticks in the southeastern United States. 4. Habitat suitability for A. americanum and total area of predicted presence for A. americanum decreases meridionally. W hile A. americanum is commonly collected from northern and central Florida, there are limited reports, both in counts and frequency, of the species from southern counties The findings in this study support the no tion that the species is uncommon in South Florida. There are se veral limitations to this study. This study was limited to estimating the distribution of adult A. americanum as larval and n ymphal ticks were excluded However, all life stages are important to the transmission of disease causing pathogens to humans and an estimated distribution based solely on adults may underestimate risk areas in places where other life stages occur. There were few sites in this study where the presence of nymphal A. ameri canum was noted and adults were absent although their inclusion would improve the interpretation of the model.


68 The size of the study area was a major challenge in determining the locations, number of sites sampled and intensity of sampling at each site The selection of sites was determined based on geographic distribution and variability in local enviro nmental and climatic conditions; however, the available habitat types sampled at various latitudes might contribute to the north south distributio n trend indicated by the model. The location of sampled sites may have also contributed to the anomalous habitat suitability in the Big Bend area as no sites were sampled there. Future improvements on this work should seek to reduce bias in sampling locations. A s patially stratified random design modified from the site selection procedure used in Diuk Wasser (2006) would ensure un biased statewide coverage. Selection of additional independent sites for model validation would also improve estimates of model accuracy Another limitation stems from differences between otherwise similar sites that was not captured by the considered variables. Differences in a ctive management strategies occurs at many of the sites for conservation or other purposes. This ac tive management may bias the variability among where ticks are present by influencing land cover, understory, or other habitat characteristics. For example active management ( controlled burning or canopy thin ning ) result s in significantly reduce d long term tick counts for A. americanum and other species (Gleim, et al. 2014, Stafford, Ward and Magnarelli 1998) However, in some locations recent burning may increase tick abundance due to reintroduction by large herbivore s grazing on re emergent understory vegetation (Cilek and Olson 2000) Inclusion of additional sites, including private properties where active habitat management is less common, would reducing this source of sampling bias. T his thesis intends to identify the geographic distribution of A. americanum in Florida using envi ronmental and climatic factors. The dominance of climatic variables in governing the


69 distribution of adult lone star ticks in Florida reinforces previous reports in the literature that habitat related variables alone are insufficient in predicting tick presence or density Historically, tick collections efforts have under sampled many counties in Florida potentially resulting in under estimates of the speci es distribution and the human risk of exposure to this species and its pathogens The spatial prediction of A. americanum distribution in Florida produced by this research contributes to the current understanding of potential risk of exposure to ticks a nd may inform control measures by state and federal agencies. The importance of the lone star tick as a vector of multiple human pathogens should guide future work on estimating human exposure risk to this vector and its pathogens. The incorporation of ot her aspects of tick ecology, such as host distribution or density, and additional climate variables (such as vapor pressure) or variable int eractions should be considered to improve estimated distribution Consideration of available pathogen prevalence or tick borne disease incidence can further identify tick borne disease risk. Patterns of disease risk have already been described from the prevalence of the Lyme disease pathogen and its primary vector, Ixodes pacificus in California (Eisen, et al. 2006) Future efforts to produce better resolved distribution maps of A. americanum in Florida should attempt to reduce the limitations of a GLM modeling approach. The use of ensemble predictions can reduce the b ias of any single model be leveraging agreement between models.


70 APPENDIX A LOGISTIC REGRESSION IN R The model building process for the l ogistic regression was performed on aggregate data to account for spatial autocorrelation amon g the sampled poin ts. The following script was written in R and uses true presence/absence data that has been aggregated to 100m. Continuous Variables have all been centered and scaled around a mean=0, std dev=1 Code f or generating summary statistics, raster manipulation, model validation or figures is not included. T he structure of the input data is included and must be followed for the code to work as shown This code corresponds to Methods section 2.3.4 ################################################################# ############################ # Build Logistic Regression Model ############################################################################################# str(aggregatedata_presabs_std) #Examine Structure of Dataframe for Modeling data.frame': 130 obs. of 30 variables: $ presabs : num 0 0 1 0 0 0 1 0 0 0 ... $ forest : num 1 0 0 0 0 0 0 1 1 0 ... $ grass : num 0 0 0 0 0 0 0 0 0 0 ... $ other : num 0 1 1 1 1 1 1 0 0 1 ... $ shrub : num 0 0 0 0 0 0 0 0 0 0 ... $ wetlands : num 0 0 0 0 0 0 0 0 0 0 ... $ DEM : num 0.498 0.607 2.031 0.981 0.981 ... $ bio1_fl : num 0.67 0.67 0.67 1.1 1.1 ... $ bio2_fl : n um 1.271 1.255 1.354 0.715 0.715 ... $ bio3_fl : num 0.183 0.183 0.183 1.54 1.54 ... $ bio4_fl : num 0.584 0.586 0.6 1.263 1.263 ... $ bio5_fl : num 0.64 0.64 0.64 1.58 1.58 ... $ bio6_fl : num 0.823 0.823 0.928 0.913 0.913 ... $ bio7_fl : num 0.893 0.893 0.996 0.673 0.673 ... $ bio8_fl : num 0.536 0.536 0.536 0.536 0.536 ... $ bio9_fl : num 0.268 0.257 0.237 0.366 0.366 ... $ bio10_fl : num 0.905 0.905 0.905 0.18 0.18 ... $ bio11_fl : num 0.641 0. 644 0.66 1.133 1.133 ... $ bio12_fl : num 0.115 0.114 0.245 1.433 1.433 ... $ bio13_fl : num 0.464 0.464 0.418 0.556 0.556 ... $ bio14_fl : num 0.0814 0.0892 0.0757 1.5775 1.5775 ... $ bio15_fl : num 0.0546 0.0698 0.0782 0.8928 0.89 28 ... $ bio16_fl : num 0.248 0.244 0.178 0.179 0.179 ... $ bio17_fl : num 0.229 0.227 0.246 1.765 1.765 ... $ bio18_fl : num 0.496 0.492 0.419 0.284 0.284 ... $ bio19_fl : num 0.569 0.57 0.57 1.272 1.272 ... $ dist2water: num 0.4 0.303 0.282 0.144 0.225 ... $ NDVI_max : num 1.0176 1.2166 1.6686 0.0272 0.388 ... $ NDVI_mean : num 1.251 1.251 1.327 0.334 0.432 ... $ NDVI_min : num 0.57 0.242 0.168 1.128 1.176 ... ##2.1 Create Base Model with all variables included # Create l ist of variables names stdcov< c(colnames(aggregatedata_presabs_std)) stdcov< stdcov[ 1] # Remove 'presence' column stdcov< stdcov[ 3] # Remove 'Other' Landcover classification stdvars< paste(stdcov, collapse="+") # Collapse to Linear combination of variab le names


71 # Run Binomial GLM with Logit Link stdaggnaive_model< glm(paste("presabs~", stdvars,sep=""), data=aggregatedata_presabs_std, family=binomial(link = "logit")) summary(stdaggnaive_model) # Examine Model Summary, ##2.2 Begin Model Selection to Generate Valid Model #2.2a Determine Variables with significant predictive capacity # of dependent variable dep_vars< names(aggregatedata_p resabs_std[1]) # Specify the Dependent Variable # pairwise combinations with dep_vars: var_comb < expand.grid(dep_vars, stdcov) # Create Matrix of 1 var combos of dep/ind # formulas for all combinations formula_vec < sprintf("%s ~ %s", var_comb$Var1, var_comb$Var2) # Run each logistic model combination glm_res < lapply( formula_vec, function(f){ fit1 < glm( f, data = aggregatedata_presabs_std, family = binomial("logit")) fit1$coefficients < coef( summary(fit1)) return(fit1) }) names(glm_res) < formula_vec # Name each model based on variable combination # Extract p values for covariates of each model in a data.frame p_values < cbind(stdcov,formula_vec, (, lapply(glm_res, function(x) { coefs < coef(x) rbind(c(coefs[,4])) }) ))) names(p_values)[1:4]< c("var_names","forumula_vec", "Intercept" ,"Covariate") # Name Columns of P value dataframe p_values$var_names< as.character(p_values$var_names) p_values # View dataframe of P values for each model # Variables that have a significant univariate relationship #with presence/absence of A. americanum signvar< subset(p_values,p_values$Covariate<0.05) # Creat e vector of sig Variables p<0.05 nonsignvar< subset(p_values,p_values$Covariate>=0.05) # Create vector of NONsign Variables p>=0.05 myvars< paste(signvar[[1]], sep=",") corr_var< aggregatedata_presabs_std[myvars] # Parse the tick data to only list significant variables


72 #2.2b Select uncorrelated variables using VIF, # cutoff for inclusion is VIF=<10.0 var.cor< vifstep(corr_var, th=10) # Consider collinearity to help limit overfitting/redundancy var.cor # View collinearity results # Extract variable names for use in Logit Model sig.var< as.character(var.cor@results$Variables) # Create vector of names for logistic model ##2.3 Identify 'best' logistic regression model #2.3a Identify n best models from set of all possible models # using exhaustive search model.sigvar.all < glmulti("presabs",xr=sig.var, data=aggregatedata_presabs_std, family=binomial(link="logit"), level=1, method="h", crit="aicc", maxsize=length(sig.var)) # First order evaluate d by AICc #create table of best models tbm < weightable(model.sigvar.all) # Table of all models considered #subset table of best models within a range of minimum AIC+2 tbms < tbm[tbm$aicc <= min(tbm$aicc) +2,] # Table of Candidate 'best' Models tbms # View table of best models #2.3b Best Model sigvarmodelminaicc< glm(paste(tbms[1,1]), data=aggregatedata_presabs_std, family = binomial(link = "logit")) # Best Model=min(AICc) summary(sigvarmodelminaicc) # view Model Summary #2.3c Check statistical assumptions are met # Spat ial Distribution of Residuals scaledresid< scale(sigvarmodelminaicc$residuals, center = TRUE, scale=TRUE) # Extract residuals from best model sigvarmodelminaicc_spat< cbind(aggregatedata_presabs[2:3], sigvarmodelminaicc$residuals, scaledresid) coordinates(sigvarmodelminaicc_spat)< sigvarmodelminaicc_spat[1:2] # Determine coordinates for each residual # Moran's I to test for clustering of residuals points< coordinates(sigvarmodelminaicc_spat) x< sigvarmodelminaicc_spat[[3]] dists< as.matrix(dist(points, method="euclidean")) listw< 1/dists diag(listw)< 0 Moran.I(x,weight=listw,alt ernative = "two.sided", na.rm = TRUE) # Calculate Global Moran's I on residual s # GOODNESS OF FIT TEST anova(stdaggnaive_model,sigvarmodelminaicc,test="Chisq") # IF p<0.05 reject the reduced model h1_sigvar< hoslem.test(sigvarmodelminaicc$y, fitted(sigvarmodelminaicc),g=10) #If p>0.05 accept null hypothesis h1_sigvar # View Results of HS GOF test


73 APPENDIX B CONSIDERATION OF COVARIATE ANOMOLIES ON SPATIAL PREDICTIONS G ridded climate surfaces such as those generated for this study are produced by any number of interpolation algorithms They a re frequently used in environmental and biological studies where fine spatial scales are desired in order to local environmental variability There are severally recently developed datasets commonly used in species distribution models that capture prec ipit ation and temperature measures incl uding the Daymet and WorldClim database s (Fick and Hijmans 2017, Hijmans, Cameron, et al. 2005, Thornton, et al. 2017) The WorldClim dataset implements the ANUSPLIN algorit hm to interpolat e desired variables from land based weather station s and high resolution elevation data Due to distribution bias in station locations, the authors note that uncertainty is not uniform; low station density and low elevation variation result in areas of high uncertainty (Hijmans, Cameron, et al. 2005) The second version of the WorldClim dataset attempts to red uce the uncertainty by using covariates in additional to elevation. The authors note that most surfaces were not significantly improved but this addition are areas of increased uncertainty remain (Fick and Hijmans 2017) The Wo rldClim web portal lists a number of known issues, errors, and problems with earlier versions of the dataset however it is unclear this known errors in older versions have been corrected in version 1.4 or 2.0 (Hijmans, WorldClim Iss ues n.d.) One type of error reported in version 1.2 are where changes in interpolation domains r esult is sudden climate changes The region surrounding the Perry Foley airfield in the Big Bend area of Florida appears to be an area of high uncerta inty or change in interpolation domain A circular area of local maxima or minima in this location results in a veritable mountain s and valleys of values in several biovariable layers (Figure B 1 ). The region is near the coast, and has relatively little


74 variat ion in its topography. Hijmans et al. (2005) cites e ach both these conditions as potential sources of uncertainty in the interpolation.


75 Figure B 1: Examples of potential interpolation error in WorldClim bioclimatic variables. Error appears to originate the annual measured precipitation(B 1a). The error impacts derived variables for precipitation and temperature including: average temperature of driest quarter (B 1b), wettest month (B 1c), and precipitation of driest quarter (B 1d). This abnormality appe ars to influence the estimated distribution of A. americanum presented in this study. The final logistic regression model includes the bioclimate variables B io3, B io8, Bio13 and B io15. Of these, Bio13 (precipitation of wettest month) contains the aberratio n much more strongly than any of the other included variables and results in a pocket of extremely low values. The negative association between occurrence and precipitation during the wettest month determined by the logistic regression results in extremel y high probability values in this area (Figure B 2).


76 Figure B 2: Impact of interpolation error in bioclimatic variable Bio13 on the spatial prediction produced by the logistic regression model. The low values in the Bio13 data and the negative associat 1.43) between Bio13 and occurrence results in extremely high probability values for occurrence of A. americanum


77 LIST OF REFERENCES Adejinmi, Johnson O. 2011. "Effect of Water Flooding on the Oviposition Capacity of Engorged Adult Females and Hatchability of Eggs of Dog Ticks: Rhipicephalus sanguineus and Haemaphysalis leachi leachi." Journal of Parasitology Research. Adelson, Martin E., Raja Venkitesh S. Rao, Richard C. Tilton, Kimberly Cabets, Eugene Eskow, Lesley Fein, James L. Occi, and Eli Mordechai. 2004. "Prevalence of Borrelia burgdorferi, Bartonella spp., Babesia microti, and Anaplasma phagocytophila in Ixodes scapularis Ticks Collected in Northern New Jersey." Journal of Clinical Microbiology 2799 2801. Allan, Brian F., Humberto P. Dutra, Lisa S. Goessling, Kirk Barnett, Jonathan M. Chase, Robert J. Marquis, Genevieve Pang, Gregory A. Storch, Robert E. Thach, and John L. Orrock. 2010. "Invasive Honeysuckle eradication reduces tick borne disease risk by al tering host dynamics." PNAS 18523 18527. Allan, Sandra A., Leigh Anne Simmons, and Michael J. Burridge. 2001. "Ixodid ticks on white tailed deer and feral swine in Florida." Journal of Vector Ecology 93 102. Allouche, Omri, Asaf Tsoar, and Ronen Kadmon. 20 06. "Assessing the Accuracy of Species Distribution Models: Prevalence, Kappa and the True Skill Statistic (TSS)." Journal of Applied Ecology 1223 1232. Anderson, B. E., C. E. Greene, D. C. Jones, and J. E. Dawson. 1992. "Ehrlichia ewingii sp. nov., the et iologica agent of canine granulocytic ehrlichiosis." International Journal of Systematics and Evolutionary Microbiology 299 302. Anderson, B. E., K. G. Sims, J. G. Olson, J. E. Childs, J. F. Piesman, C. M. Happ, G. O. Maupin, and B. J. Johnson. 1993. "Ambl yomma americanum: A Potential Vector of Human Ehrlichiosis." American Journal of Tropical Medicine and Hygiene 239 244. Anderson, John F, and Louis A Magnarelli. 2008. "Biology of Ticks." Infectious Disease Clinics of North America 195 215. Anselin, Luc. 1 995. "Local Indicators of Spatial Association LISA." Geographical Analysis 93 115. Araujo, Miguel B., and Mark New. 2006. "Ensemble Forecasting of Species Distributions." Trends in Ecology and Evolution 42 47. Arlot, Sylvain, and Alain Celisse. 2010. "A S urvey of Cross Validation Procedures for Model Selection." Statistical Surveys 40 79. Armstrong, Philip M., Laura Rosa Brunet, Andrew Spielman, and Sam R. Telford III. 2001. "Risk of Lyme Disease: Perceptions of Residents of a Lone Star Tick Infested Commu nity." Bulletin of the World Health Organization 916 925. Atkinson, Samuel F., Sahotra Sarkar, Aldo Avina, Jim A. Schuermann, and Phillip Williamson. 2014. "A Determination of the Spatial Concordance Between Lyme Disease Incidence and Habitat Probability o f its Primary Vector Ixodes Scapularis (black legged tick)." Geospatial Health 203 212. Austin, M. P. 2002. "Spatial Prediction of Species Distribution: An Interface Between Ecological Theory and Statistical Modelling." Ecological Modeling 101 118.


78 Austin, M. P., A. O. Nicholls, and C. R. Margules. 1990. "Measurement of the Realized Qualitative Niche: Environmental Niches of Five Eucalyptus Species." Ecological Monographs 161 177. Bell, J. Frederick. 1944. "The Infection of Ticks (Dermacentor Variabilis) wi th pasteurella Tularensis." Journal of Infectious Disease 83 95. Bisanzio, Donal, Giuseppina Amore, Charlotte Ragagli, Laura Tomassone, Luigi Bertolotti, and Alessandro Mannelli. 2008. "Temporal Variations in the Usefulness of Normalized Difference Vegetat ion Index as a Predictor for Ixodes ricinus (Acari: Ixodidae) in a Borrelia lusitaniae Focus in Tuscany, Central Italy." Journal of Medical Entomology 547 555. Bishopp, F. C., and Helen Louise Trembley. 1945. "Distribution and Hosts of Certain North Americ an Ticks." The Journal of Parasitology 1 54. Blackburn, Jason K. 2010. "Integrating Geographic Information Systems and Ecological Niche Modeling into Disease Ecology: A Case Study of Bacillus anthracis in the United States and Mexico." In Emerging and Ende mic Pathogens: Advances in Surveillance, Detection, and Identification by K. P. O'Connell, A. Sulakvelidze and L., ed. Bakanidze, 59 88. New York: Springer. Bremer, William G., John J. Schaefer, Elizabeth R. Wagner, S.A. Ewing, Yasuko Rikihisa, Glen R. Ne edham, Sathaporn Jittapalapong, Debra L. Moore, and Roger W. Stich. 2005. "Transstadial and Intrastadial experimental transimssion of Ehrlichia canis by male Rhipicephalus sanguineus." Veterinary Parasitology 95 105. Bring, Johan. 1994. "How to Standardize Regression Coefficients." The American Statistician 209 213. Britch, Seth C., Kenneth J. Linthicum, Assaf Anyamba, Compton J. Tucker, Edwin W. Pak, Francis A. Maloney Jr., Kristin Cobb, et al. 2008. "Satellite Vegetation Index Data as a Tool to Forecast P opulation Dynamics of Medically Important Mosquitoes at Military Installations in the Continental United States." Military Medicine 677 683. Brown, Heidi E., Karen F. Yates, Gabrielle Dietrich, Katherine MacMillan, Christine B. graham, Sara M. Reese, Wm. Steve Helterbrand, et al. 2011. "An Acarologica Survey and Amblyomma americanum Distribution Map with Implications for Tularemia Risk in Misso uri." American Journal of Tropical Medicine and Hygiene 411 419. Brownstein, John S., Theodore R. Holford, and Durland Fish. 2003. "A Climate Based Model Predicts the Spatial Distribution of the Lyme Disease Vector Ixodes scapularis in the United States." Environmental Health Perspectives 1152 1157. Chatterjee, Samprit, and Ali S. Hadi. 2013. Regression Analysis by Example. Somerset: Wiley Childs, James E., and Christopher D. Paddock. 2003. "The Ascendancy of Amblyomma Americanum as a Vector of Pathogens Affecting Humans in the United States." Annual Review of Entomology 307 337. Cilek, J. E., and M. A. Olson. 2000. "Seasonal Distribution and Abundance of Ticks (Acari: Ixodidae) in Northwestern Florida." Journal of Medical Entomology 439 444. Clark, Kerry. 2004. "Borrelia Species in Host Seeking Ticks and Small Mammals in Northern Florida." Journal of Clinical Microbiology 5076 5086.


79 Craney, Trevor A., and James G. Surles. 2002. "Model Dependent Variance Inflation Factor Cutoff Values." Quality Engineering 391 403. Davey, Ronald B., Lynita M. Cooksey, and Joseph L. Despins. 1991. "Survival of larvae of Boophilus annulatus, Boophilus microplus, and Boophilus hybrids (Acari: Ixodidae) in Different Temperature and Humidity Regimes in the Laboratory." Veterinary Parasitology 305 313. Davidson, William R., Debra A. Siefken, and Lynn H. Creekmore. 1994. "Seasonal and Annual Abundance of Amblyomma americanum (Acari: Ixodidae) in Central Georgia." Journal of Medical Entomology 67 71. Demma, Linda J., Marc S. Traeger, William L. Nicholson, Christopher D. Paddock, Dianna M. Blau, Marina E. Eremeeva, Gregory A. Dasch, et al. 2005. "Rocky Mountain Spotted Fever from an Unexpected Tick Vector in Arizona." The New England Journal of Medicine 587 594. Derksen, Shelley, and H. J. Keselman. 1992. "Backward, forward and stepwise automated subset selection algorithms: Frequency of obtaining authentic and noise variables." British Journal of Mathematical and Statistical Psychology 265 282. Diuk Was ser, M. A., A. G. Gatewood, M. R. Cortinas, S. Yaremych Hamer, J. Tsao, U. Kitron, G. Hickling, et al. 2006. "Spatiotemporal Patterns of Host Seeking Ixodes scapularis Nymphs (Acari: Ixodidae) in the United States." Journal of Medical Entomology 166 176. D obson, A., and J. Foufopoulos. 2001. "Emerging Infectious Pathogens of Wildlife." Philosophical Transactions of the Royal Society of London B: Biological Sciences 1001 1012. Dormann, Carsten F., Jana M. McPherson, Miguel B. Araujo, Roger Bivand, Janine Bol liger, Gudrun Carl, Richard G. Davies, et al. 2007. "methods to account for spatial autocorrelation in the analysis of species distributional data: a review." Ecography 609 628. Durden, L. A., J. S. H. Klompen, and J. E. Keirans. 1993. "Parasitic Arthropod s of Sympatric Opossums, Cotton Rats, and Cotton Mice from Merritt Island, Florida." The Journal of Parasitology 283 286. Eisen, Rebecca J, Lars Eisen, and Charles B Beard. 2016. "County Scale Distribution of Ixodes scapularis and Ixodes pacificus (Acari: Ixodidae) in the Continental United States." Journal of Medical Entomology 349 386. Eisen, Rebecca J., Robert S. Lane, Curtis L. Fritz, and Lars Eisen. 2006. "Spatial Patterns of Lyme Disease Risk in California Based on Disease Incidence Data and Modeling of Vector Tick Exposure." American Journal of Tropical Medicine and Hygiene 669 676. Elith, Jane, Steven J. Phillips, Trevor Hastie, Miroslav Dudik, Yung En Chee, and Colin J. Yates. 2011. "A Statistical Explanation of MaxEnt for Ecologists." Diversity and Distributions 43 57. Estrada Pena, Agustin. 1998. "Geostatistics and Remote Sensing as Predictive Tools of Tick Distribution: a Cokriging System to Estimate Ixodes scapularis (Acari: Ixodidae) Habitat Suitability in the United States and Canada from Advan ced Very High Resolution Radiometer Satellite Image." Journal of Medical Entomology 989 995.


80 Estrada Pena, Agustin, and Frans Jongejan. 1999. "Ticks feeding on humans: a review of records on human biting Ixodoidea with special reference to pathogen transmi ssion." Experimental and Applied Acarology 685 715. Felz, Michael W., Lance A. Durden, and James H. Oliver Jr. 1996. "Ticks Parasitizing Humans in Georgia and South Carolina." The Journal of Parasitology 505 508. Fick, S.E., and R.J. Hijmans. 2017. "Worldc lim 2: New 1 km spatial resolution climate surfaces for global land areas." International Journal of Climatology. Fielding, Alan H., and John F. Bell. 1997. "A Review of Methods for the Assessment of Prediction Errors in Conservation Presence/Absence Mode ls." Environmental Conservation 38 49. Forrester, Donald J., Grace S. McLaughlin, Sam R. Telford Jr., Garry W. Foster, and James W. McCown. 1996. "Ectoparasites (Acari, Mallophaga, Anoplura, Dipterea) of White Tailed Deer, Odocoileus virginianus, from Southern Florida." Journal of Medical Entom ology 96 101. Franklin, Janet, Katherine E. Wejnert, Stacie A. Hathaway, Carlton J. Rochester, and Robert N. Fisher. 2009. "Effect of Species Rarity on the Accuracy of Species Distribution Models for Reptiles and Amphibians in Southern California." Diversi ty and Distributions 167 177. Freeman, Elizabeth A., and Gretchen G. Moisen. 2008. "A Comparison of the Performance of Threshold Criteria for Binary Classification in terms of Predicted Prevalence and Kappa." Ecological Modeling 48 58. Fryxell, R.T. Trout, J.E. Moore, M.D. Collins, Y. Kwon, S.R. Jean Philippe, S.M. Schaeffer, A. Odoi, M. Kennedy, and A.E. Houston. 2015. "habitat and Vegetation Variables Are Not Enough When Predicting Tick Populations in the Southeastern United States." PLoS ONE. Ginsberg, Howard S., and Curtis P. Ewing. 1989. "Comparison of Flagging, Walking, Trapping, and Collecting from Hosts as Sampling Methods for Northern Deer Ticks, Ixodes dammini, and Lone Star Ticks, Amblyomma americanum (Acari: Ixodidae)." Experimental and Applied Acarology 313 322. Glass, Gregory E., Brian S. Schwartz, John M. Morgan III, Dale T. Johnson, Peter M. Noy, and Ebenezer Israel. 1995. "Environmental Risk Factors for Lyme Disease Identified with Geographic Information Systems." American Journal of Public Health 944 948. Gleim, Elizabeth R., Mike L. Conner, Roy D. Berghaus, Michael L. Levin, Galina E. Zemtsova, and Michael J. Yabsley. 2014. "The Phenology of Ticks and the Effects of Long Term Prescribed Burning on Tick Population Dynamics in Southwestern Ge orgia and Northwestern Florida." PLoS ONE. Goddard, Jerome, and Andrea S. Varela Stokes. 2009. "Role of the Lone Star Tick, Amblyomma americanum, in Human and Animal Disease." Veterinary Parasitology. Graham, Michael H. 2003. "Confronting Multicollineari ty in Ecological Multiple Regression." Ecology 2809 2815. Gray, J. S. 1998. "The Ecology of Ticks Transmitting Lyme Borreliosis." Experimental and Applied Acarology 249 258.


81 Green, R. G. 1931. "The Occurrence of Bacterium tularense in the Eastern Wood Tick Dermacentor variabilis ." Americanal Journal of Hygiene 600 613. Greiner, Ellis C., Pamela P. Humphrey, Robert C. Belden, William B. Frankenberger, David H. Austin, and E. Paul J. Gibbs. 1984. "Ixodid Ticks on Feral Swine in Florida." Journal of Wildlife Diseases 114 119. Guisan, Antoine, and Niklaus E. Zimmermann. 2000. "Predictive Habitat Distribution Models in Ecology." Ecological Modeling 147 186. Guisan, Antoine, and Wilfried Thuiller. 2005. "Predicting species distribution: offering more than simple habitat models." Ecology Letters 993 1009. Guisan, Antoine, Jean Paul Theurillat, and Felix Kienast. 1998. "Predicting the Potential Distribution of P lant Species in an Alpine Environment." Journal of Vegetation Science 65 74. Guisan, Antoine, Jr, Thomas C. Edwards, and Trevor Hastie. 2002. "Generalized Linear and Generalized Additive Models in Studies of Species Distributions: Setting the Scene." Ecolo gical Modeling 89 100. Heath, A.C.G. 1981. "The Temperature and Humidity Preferencces of Haemaphysalis longiscornis, Ixodes holocyclus and Rhipicephalus sanguineus (Ixodidae): Studies on Engorged Larvae." International Journal for Parasitology 169 175. Hen ry, James A, Kenneth M Portier, and Jan Coyne. 1994. The Climate and Weather of Florida. Sarasota: Pineapple Press. Hijmans, Robert J. n.d. WorldClim Issues. Accessed June 18, 2017. Hijmans, Robert J., Susan E. Cameron, Jua n L. Parra, Peter G. Jones, and Andy Jarvis. 2005. "Very High Resolution Interpolated Climate Surfaces For Global Land Areas." International Journal of Climatology 1965 1978. Hopla, C. E. 1953. "Experimental Studies on Tick Transmission of Tularemia." Amer ican Journl of Hygiene 101 118. Hopla, C. E. 1955. "The Multiplication of Tularemia Organisms in the Lone Star Tick." American Journal of Hygiene 371 380. Hosmer, David W., and Stanley Lemeshow. 1989. Applied Logistic Regression. New York: John Wiley & Sons. James, A. M, C Burdett, M. J McCool, A Fox, and P Riggs. 2015. "The Geographic Distribution and Ecological Preferences of the American Dog Tick, Dermacentor variabilis (Say), in the U.S.A." Medical and Veterinary Entomology 17 8 188. Jaworski, Deborah C., C. Jerry Bowen, and Nalinda B. Wasala. 2013. "A White Tailed Deer/Lone Star Tick Model for Studying Transmission of Ehrlichia chaffeensis." Vector Borne and Zoonotic Diseases 193 195. Jongejan, F., and G. Uilenberg. 2004. "The Global Importance of Ticks." Parasitology S3 S14.


82 Joyner, Timothy Andrew, Larissa Lukhnova, Yerlan Pazilov, Gulnara Temiralyeva, Martin E. Hugh Jones, Alim Aikimbayev, and Jason K. Blackburn. 2010. "Modeling the Potential Distribution of Bacillus anthracis under Multiple Climate Change Scenarios for Kazakhstan." PloS ONE. Karl, Jason W., and Brian A. Maurer. 2010. "Spatial Dependence of Predictions from Image Segmentation: A Variogram based Method to Determine Appropriate Scales for Producting Land Managem ent Information." Ecological Informatics 194 202. Keim, Barry D, Michell Russo Fischer, and Adam M Wilson. 2005. "Are there spurious precipitation tends in the United States Climate Division database?" Geophysical Research Letters. Keirans, James E., and Taina R. Litwick. 1989. "Pictorial Key to the Adulats of Hard Ticks, Family Ixodidae (Ixodida: Ixodoidea), East of the Mississippi River." Journal of Medical Entomology 435 448. Knulle, W., and D. Rudolph. 1981. "Humidity Relationships and Water Balance of Ticks." In Physiology of Ticks by FD. Obenchain and R. Galun, 43 50. Oxford (UK): Pergamon Press. Koch, Henry G. 1986. "Survival of the Lone Star Tick (Acari: Ixodidae) Under Flooding Conditions: A Laboratory Evaluation." Journal of Economic Entomology 1 555 1557. Kollars Jr, Thomas M., James H. Oliver Jr, Lance A. Durden, and Peggy G. Kollars. 2000. "Host Associations and Seasonal Activity of Amblyomma americanum (Acari: Ixodidae) in Missouri." Journal of Parasitology 1156 1159. Kugeler, Kiersten J, Grace M Farley, Joseph D Forrester, and Paul S Mead. 2015. "Geographic Distribution and Expansion of Human Lyme Disease, United States." Emerging Infectious Diseases 1455 1457. Kumar, Sunil, and Thomas J. Stohlgren. 2009. "MaxEnt Modeling for Predicting Suitabl e Habitat for Threatened and Endangered Tree Canacomyrica monticola in New Caledonia." Journal of Ecology and Natural Environment 94 98. Landis, J. Richard, and Gary G. Koch. 1977. "The Measurement of Observer Agreement for Categorical Data." Biometrics 15 9 174. Legendre, Pierre. 1993. "Spatial Autocorrelation: Trouble or New Paradigm." Ecology 1659 1673. Linthicum, K. J., C. L. Bailey, C. J. Tucker, K. D. Mitchell, T. M. Logan, F. G. Davies, C. W. Kamau, P. C. Thande, and J. N. Wagateh. 1990. "Application of Polar Orbiting, Meteorological Satellite Data to Detect Flooding of Rift Valley Fever Virus Vector Mosquito Habitats in Kenya." Medical and Veterinary Entomology 433 438. Masters, Edwin J., Chelsea N. Grigery, and Reid W. Masters. 2008. "STARI, or Maste rs Disease: Lone Star Tick Vectored Lyme like Illness." Infectious Disease Clinics of North America 361 376. Mertins, James W, Stacey L Vigil, and Joseph L Corn. 2017. "Amblyomma auricularium (Ixodida: Ixodidae) in Florida: New Hosts and Distribution Recor ds." Journal of Medical Entomology 132 141. Miller, A., D. Cayan, T. Barnett, N. Graham, and J. Oberhuber. 1994. "The 1976 77 Climate Shift of the Pacific Ocean. ." Oceanography 21 26.


83 Miller, Jennifer, Janet Franklin, and Richard Aspinall. 2007. "Incorporating Spaatial Dependence in Predictive Vegetation Models." Ecological Modeling 225 242. Naimi, B., N. A. S. Hamm, T. A. Groen, A. K. Skidmore, and A. G. Toxopeus. 2014. "Where is posit ional uncertainty a problem for species distribution modelling?" Ecography 191 203. Ogden, N. H., M. Bigras Poulin, C. J. O'Callaghan, I. K. Barker, A. Maarouf, K. E. Smoyer Tomic, D. Waltner Toews, and D. Charron. 2005. "A Dynamic Population Model to Inve stigate Effects of Climate on Geographic Range and Seasonality of the Tick Ixodes Scapularis." International Journal of Parasitology 375 389. Oliver Jr, James H. 1989. "Biology and Systematics of Ticks (Acari: Ixodida)." Annual Review of Ecology and System atics 397 430. Oliver Jr, James H, Marc P Hayes, James E Keirans, and David R Lavender. 1993. "Establishment of the Foreign Parthenogenetic Tick Amblyomma rotundatum (Acari: Ixodidae) in Florida." Journal of Parasitology 786 790. Ostfeld, Richard S., and Jesse L. Brunner. 2015. "Climate Change and Ixodes Tick Borne Diseases of Humans." Philosophical Transactions of the Royal Society B. Ostfeld, Richard S., Gregory E. Glass, and Felicia Keesing. 2005. "Spatial Epidemiology: An Emer ging (or Re Emerging) Discipline." TRENDS in Ecology and Evolution 328 336. Ostfeld, Richard S., Obed M. Cepeda, Kirsten R. Hazler, and Michael C. Miller. 1995. "Ecology of Lyme Disease: Habitat Associations of Ticks (Ixodes Scapularis) In a Rural Landscap e." Ecological Applications 353 361. Pearce, Jennie, and Simon Ferrier. 2000. "Evaluating the Predictive Performance of Habitat Models Developed using Logistic Regression." Ecological Modeling 225 245. Pearson, Richard G., Christopher J. Raxworthy, Miguel Nakamura, and A. Townsend Peterson. 2007. "Predicting Species Distributions from small numbers of occurrence records: a test case using cryptic geckos in Madagascar." Journal of Biogeography 102 117. Pearson, Richard G., Terence P. Dawson, and Canran Liu. 2004. "Modelling Species Distributions in Britain: A Hierarchical Integration of Climate and Land cover data." Ecography 285 298. Peterson, A. Townsend, Monica Papes, and Muir Eaton. 2007. "Transferability and Model Evaluation in Ecological Niche Modeling: A Comparison of GARP and MaxEnt." Ecography 550 560. Phillips, Steven J., Miroslav Dudik, and Robert E. Schapire. 2004. "A Maximum Entropy Approach to Species Distribution Modeling." Proceedings of the 21st International Conference on Machine Learning 655 662. Phillips, Steven J., Miroslav Dudik, Jane Elith, Catherine H. Graham, Anthony Lehmann, John Leathwick, and Simon Ferrier. 2009. "Sample Selection Bias and Presence only Distribution Models: Implications for background and pseudo absence data." Ecolog ical Applications 181 197. Phillips, Steven J., Robert P. Anderson, and Robert E. Schapire. 2006. "Maximum Entropy Modeling of Species Geographic Distributions." Ecological Modeling 231 259.


84 Porretta, Daniele, Valentina Mastrantonio, Sara Amendolia, Stefan o Gaiarsa, Sara Epis, Claudio Genchi, Claudio Bandi, Domenico Otranto, and Sandra Urbanelli. 2013. "Effects of global changes on the climatic niche of the tick Ixodes ricinus inferred by species distribion modeling." Parasites & Vectors 1 8. Potts, Joanne, M., and Jane Elith. 2006. "Comparing Species Abundance Models." Ecological Modelling 153 163. Raghavan, Ram K., Douglas G. Goodin, Gregg A. Hanzlicek, Gregory Zolnerowich, Michael W. Dryden, Gary A. Anderson, and Roman R. Ganta. 2016. "Maximum Entropy Bas ed Ecological Niche Model and Bio Climatic Determinants of Lone Star Tick (Amblyomma americanum) Niche." Vector Borne and Zoonotic Diseases 205 211. Randolph, S. E., R. M. Green, A. N. Hoodless, and M. F. Peacey. 2002. "An empirical quantitative framework for the seasonal population dynamics of the tick Ixodes ricinus." International Journal of Parasitology 979 989. Randolph, S.E., L. Gern, and P.A. Nuttall. 1996. "Co feeding Ticks: Epidemiological Significance for Tick borne Pathogen Transmission." Parasit ology Today 472 479. Reeves, Will K., Amanda D. Loftis, William L. Nicholson, and Alan G. Czarkowski. 2008. "The first report of human illness associated with the Panola Mountain Ehrlichia species: a case report." Journal of Medical Case Reports. 2017. "R eportable Diseases Frequency Report." Tallahassee: Florida Department of Health. Richards, John A. 2013. Remote Sensing Digital Image Analysis: An Introduction. New York: Springer. Sallam, Mohamed F., Azzam M. Al Ahmed, Mahmound S. Abdel Dayem, and Mohamed A. R. Abdullah. 2013. "Ecological Niche Modeling and Land Cover Risk Areas for Rift Valley Fever Vector, Culex tritaeniorhynchus Giles in Jazan, Saudi Arabia." PLoS ONE. Sayler, Katherine A., Anthony F. Barbet, Casey Chamberlain, William L. Clapp, Rick A lleman, Julia C. Loeb, and John A. Lednicky. 2014. "Isolation of Tacaribe Virus, a Caribbean Arenavirus, from Host Seeking Amblyomma americanum Ticks in Florida." PLoS ONE 1 16. Sayler, Katherine, Carisa Boyce, and Samantha Wisely. 2016. Facts about Wildli fe Diseases: South Tick Associated Rash Illness of "STARI". WEC376, Gainesville: University of Florida Institute of Food and Agricultural Sciences. Schielzeth, Holger. 2010. "Simple means to improve the interpretability of regression coefficients." Methods in Ecology and Evolution 103 113. Semtner, P. J., and J. A. Hair. 1973. "The Ecology and Behavior of the lone star tick (Acarina: Ixodidae)." Journal of Medical Entomology 618 628. Smith Jr., H. M. 1981. Parasites of wild swine (Sus scrofa) in the southern United States. University of Georgia. Smith, R. D., D. M. Sells, E. H. Stephenson, M. R. Ristic, and D. L. Huxsoll. 1976. "Development of Ehrlichia canis, causative agent of canine ehrlichiosis, in the tick Rhipicephalus sanguineus and its differe ntiation from a symbiotic Rickettsia." American journal of Veterinary Research 119 126.


85 Springer, Yuri P, Catherine S Jarnevich, David T Barnett, Andrew J Monaghan, and Rebecca J Eisen. 2015. "Modeling the Present and Future Geographic Distribution of the Lone Star Tick Amblyomma americanum (Ixodida: Ixodidae), in the Continental United States." American Journal of Tropical Medicine and Hygiene 875 890. Springer, Yuri P., Lars Eisen, Lorenza, James, Angela M. Beati, and Rebecca J. Eisen. 2014. "Spatial Dist ribution of Counties in the Continental United States with Records of Occurrence of Amblyomma americanum (Ixodida: Ixodidae)." Journal of Medical Entomology 342 351. Stafford, Kirby C., Jeffrey S. Ward, and Louis A. Magnarelli. 1998. "Impact of Controlled Burns on the Abundance of Ixodes scapularis." Journal of Medical Entomology 510 513. Stockwell, David. 1999. "The GARP Modeling System: Problems and Solutions to Automated Spatial Prediction." International Journal of Geographical Information Science 143 1 58. Strickland, Robert K., Robert R. Gerrish, James L. Hourrigan, and Glen O. Schubert. 1976. Ticks of Veterinary Importance. Animal and Plant Health Inspection Service, U.S. Dept of Agriculture. Sumner, John W., Lance A. Durden, Jerome Goddard, Ellen Y. S tromdahl, Kerry L. Clark, Will K. Reeves, and Christopher D. Paddock. 2007. "Gulf Coast Ticks (Amblyomma maculatum) and Rickettsia parkeri, United States." Emerging Infectious Diseases 751 753. Symonds, Matthew R. E., and Adnan Moussalli. 2011. "A Brief gu ide to Model Selection, Multimodel Inference and Model Averaging in Behavioural Ecology using Akaike's Information Criterion." Behavioral Ecology and Sociobiology 13 21. Taylor, Doyle Jackson. 1951. "The Distribution of Ticks in Florida." University of Flo rida, M.S. Thesis. Thornton, P.E., M.M. Thornton, B.W. Mayer, Y. Wei, R. Devarakonda, R.S. Vose, and R.B. Cook. 2017. "Daymet: Daily Surface Weather Data on a 1 km Grid for North America, Version 3." Oak Ridge, Tennessee: ORNL DAAC. Tsai, HuiPing, Jane So uthworth, and Peter Waylen. 2014. "Spatial persistence and temporal patterns in vegetation cover across Florida, 1982 2006." Physical Geography 151 180. U.S. Geological Survey. 1987. Map projections: A working manual. USGS Numbered Series, Washington, D.C. : U.S. Government Printing Office. USGCRP. 2016. The Impacts of Climate Change on Human Health in the United States: A Scientific Assessment. Washington, DC.: U.S. Global Change Research Program. Vail, Stephen G., and Gary Smith. 2002. "Vertical Movement a nd Posture of Blacklegged Tick (Acari: Ixodidae) Nymphs as a Function of Temperature and Relative Humidity in Laboratory Experiments." Journal of Medical Entomology 842 846. Wehinger, K. A., M. E. Roelke, and E. C. Greiner. 1995. "Ixodid ticks from panther s and bobcats in Florida." Journal of Wildlife Disease 480 485. Weiler, Martin, Georg Gerhard Duscher, Monika Wetcher, and Julia Walochnick. 2017. "Tick Abundance: A One Year Study on the Impact of Flood Events Along the Banks of the River Danube, Austria. Experimental and Applied Acarology 151 157.


86 Westphal, Michael I., S. A. Field, A. J. Tyre, D. Paton, and H. P. Possingham. 2003. "Effects of Landscape Pattern on Bird Species Distribution in the Mt. Lofty Ranges, South Australia." Landscape Ecology 413 4 26. Whittingham, Mark J., Philip A. Stephens, Richard B. Bradbury, and Robert P. Freckleton. 2006. "Why Do We Still Use Stepwise Modelling in Ecology and Behaviour." Journal of Animal Ecology 1182 1189. Wintle, Brendan, A., Jane Elith, and Joanne M. Potts. 2005. "Fauna Habitat Modelling and Mapping: A Review and Case Study in the Lower Hunter Central Coast Region of NSW." Austral Ecology 719 738. Yabsley, Michael J, Todd N Nims, Mason Y Savage, and Lance A Durden. 2009. "Ticks and Tick Borne Pathogens and P utative Symbionts of Black Bears (Ursus americanus floridanus) from Georgia and Florida." Journal of Parasitology 1125 1128. Yoder, Jay A, and Justin L Tank. 2006. "Similarity in Critical Transition Temperature for Ticks Adapted for Different Environments: Studies on the Water Balance of Unfed Ixodid Larvae." International Journal of Acarology 323 329. Yoder, Jay A., and Joshua B. Benoit. 2003. "Water Vapor Absorption by Nymphal Lone Star Tick, Amblyomma americanum (Acari: Ixodidae), and its Ecological Sign ificance." International Journal of Acarology 259 264. Zimmermann, Fridolin, and Urs Breitenmoser. 2002. "A Distribution Model for the Eurasian Lynx (Lynx lynx) in the Jura Mountains, Switzerland." Predicting Species Occurrences.


87 BIOGRAPHICAL SKETCH William Kessler received his undergraduate education at the New Mexico Institute of Mining and Technology in 2010. He pursued a bachelor of science in biology and graduated cum laude in 2014. In the f geography at the University of Florida. His research interests include the study of vector borne diseases, infectious academic career as a PhD student studying epid emiology.