Of a Species Distribution Model for Bog Turtles (Glyptemys Muhlenbergii) in the Southeastern United States

Home , Bog turtle

Herpetological Conservation and Biology 11(1):199–213. Submitted: 15 May 2015; Accepted: 23 February 2016; Published: 30 April 2016.

LOCATING SUITABLE HABITAT FOR A RARE SPECIES: EVALUATION OF A SPECIES DISTRIBUTION MODEL FOR BOG TURTLES (GLYPTEMYS MUHLENBERGII) IN THE SOUTHEASTERN UNITED STATES

1,2,4 1 3 THERESA S. M. STRATMANN , KYLE BARRETT , AND THOMAS M. FLOYD

1Department of Forestry and Environmental Conservation, Clemson University, 261 Lehotsky Hall, Clemson, South Carolina 29634, USA 2Current address: Senckenberg Biodiversity and Climate Research Centre (BiK-F) and Department of Biological Sciences at the Goethe University, Senckenberganlage 25, 60325 Frankfurt am Main, Germany 3Wildlife Resources Division, Georgia Department of Natural Resources, 116 Rum Creek Drive, Forsyth, Georgia 31029, USA 4Corresponding author, e-mail: [email protected]

Abstract.—Because rare and cryptic species can be difficult to locate, distribution maps for such species are often inaccurate or incomplete. Bog Turtles (Glyptemys muhlenbergii) are emblematic of this challenge. Conducting surveys of known, historical, and potential Bog Turtle habitat is a specific need stated in the Bog Turtle Northern Population Recovery Plan and in most Comprehensive Wildlife Conservation Strategies of states in the southern population. To address this need, we constructed a species distribution model for the southern population of Bog Turtles and ground- validated the model to assess its ability to locate suitable Bog Turtle habitat. Our final model identified 998,325 ha of potentially suitable habitat. On-the-ground evaluation of habitat identified as potentially suitable was carried out at 113 wetlands in Georgia and 83 in South Carolina. Of these, only nine wetlands met criteria for suitable Bog Turtle habitat in Georgia and 13 in South Carolina. Trapping efforts at the nine Georgia sites and eight of the South Carolina sites showed Bog Turtles to be present at two of the Georgia sites. This ground-validation effort demonstrates that the species distribution model greatly over-predicts the amount of suitable habitat for Bog Turtles. Nonetheless, this manner of searching for rare and cryptic species does avoid the typical biases of haphazard searches and helps identify habitat on private property. Given these findings, the model is most useful when the area of interest is small, such as a county within the range of a species that currently has no known occurrence records.

Key Words.—endangered species; habitat suitability; Maxent; model validation; South Carolina; Georgia; wetland

INTRODUCTION populations on the landscape and individuals within a wetland especially challenging. Conservation efforts for rare and cryptic species are Endemic to the Eastern United States, Bog Turtles are often hampered by an incomplete understanding of their the smallest and rarest species of chelonian in the USA distribution (Guisan et al. 2006). There are the extreme (Zappalorti 1978; Ernst and Lovich 2009). The range of cases, such as Zhou’s Box Turtle (Cuora zhoui), where the species is split into two regions, a Northern range species are known only from specimens that appear in reaching from New York and Massachusetts south to markets, with no wild populations known to scientists Maryland and Delaware and a Southern range (Zhao et al. 1990; McCord and Iverson 1991). Yet the predominantly in the Appalachian region, reaching from challenge of finding populations of rare and cryptic southern Virginia to northern Georgia. In the northern species is not solely a problem for the understudied portion of their range, Bog Turtles are listed as ecosystems of the world. Searches for new populations Threatened under the U.S. Endangered Species Act; of rare and cryptic species and efforts to better whereas, they are listed as Threatened by Similarity of understand their geographic ranges occur regularly Appearance in the southern portion (U.S. Fish and within the United States (Yozzo and Ottmann 2003; Wildlife Service [USFWS] 1997); globally they are Campbell et al. 2010; Graham et al. 2010; Sheldon and considered one of the 40 most endangered turtles (Turtle Grubbs 2014), and many of these searches are for Conservation Coalition 2011). Within their range, Bog herpetofauna (Apodaca et al. 2012; Groff et al. 2014; Turtles are confined to small, isolated, open-canopy, Lindeman 2014; Pierson et al. 2014; Searcy and Schaffer spring-fed wetlands characterized by shallow rivulets 2014). In particular, efforts to study and conserve the flowing over ankle to hip-deep muck (Zappalorti 1976a, Bog Turtle (Glyptemys muhlenbergii) capture the b; Chase et al. 1989; Tryon 1990). These wetlands are challenges faced in these search endeavors, as its use of rare features within the landscape, in part due to natural a rare habitat and its cryptic habits make locating succession and past drainage efforts for farming

(Weakley and Schafale 1994). It is in the muck of these areas for re-introduction (McKenna et al. 2013). rare spring-fed wetlands where Bog Turtles spend most Because SDMs have demonstrated utility for rare of their time, making this small (maximum shell length species, including those with only a few known of 11.5 cm), mud-colored turtle very difficult to find localities, we decided to apply the approach toward more within a wetland (Ernst and Lovich 2009). The fact that efficiently and strategically locating suitable Bog Turtle they tend to occur in small populations (< 50 habitat. individuals) lessens their detection probability even We propose this SDM approach at a time when it is more (Rosenbaum et al. 2007). Because of the more important than ever to survey for new populations difficulties associated with finding Bog Turtle habitat of Bog Turtles. Habitat loss, alteration, and degradation and individuals, the Comprehensive Wildlife are the main threats facing Bog Turtles (USFWS 2001). Conservation Strategies of Georgia, South Carolina, A century of past ditching and draining efforts, North Carolina, and Virginia, and the Bog Turtle encroachment of wetlands by woody vegetation, Northern Population Recovery Plan all emphasize invasive plant species, and loss of habitat connectivity surveying for new populations in unexplored areas as leaves us trying to understand Bog Turtle habitat part of their conservation strategy (USFWS 2001; selection in an altered landscape (Tryon 1990; Buhlmann Georgia Department of Natural Resources [GA DNR] et al. 1997; USFWS 2001; Ernst and Lovich 2009). It is 2005; North Carolina Wildlife Resources Commission not uncommon to see this long-lived turtle persisting in [NCWRC] 2005; South Carolina Department of Natural degraded habitat it selected long-ago. We must therefore Resources [SC DNR] 2005; Virginia Department of search for Bog Turtles before this habitat degradation Game and Inland Fisheries [VA DGIF] 2005). Although becomes prohibitive in identifying habitat and there are guidelines on how to assess whether a site is understanding habitat selection. suitable for Bog Turtles and how to survey for Bog Our objective is to address the problems associated Turtles within potentially suitable sites (USFWS 2001; with Bog Turtle habitat detection by constructing a SDM Somers and Mansfield-Jones 2008), there are no for the species and ground-validating the model to assess suggestions on how to locate potential habitat within the its ability to locate suitable Bog Turtle habitat. This greater landscape. Without guidelines, searches for this represents the first attempt to understand Bog Turtle rare habitat tend to be haphazard and will naturally be distribution and habitat selection at a regional scale, as biased towards roads and easily accessible areas, all previous attempts have been narrowly focused at the providing no way to assess habitat on private lands or far state level or home range scale (Chase et al. 1989; Carter from the road. Thus there is a need to map suitable et al. 1999; Morrow et al. 2001; Pittman and Dorcas habitat for Bog Turtles to efficiently and strategically 2009; Myers and Gibbs 2013). This regional scale achieve goals of state wildlife agencies for locating new approach was applied to the southern portion of the Bog populations (USFWS 2001). Turtle range. This is an area where a regional model Species distribution modeling (SDM) techniques are might be particularly useful as three states (South one way to map species habitat associations, and they Carolina, Georgia, and Tennessee, USA) have too few have greatly increased in popularity as integration with occurrence records to build strong state level models. In user-friendly geographic information systems (GIS) has addition, by using all known occurrence records in the become easier and remotely sensed data have become south, we increase the likelihood of providing a more more available (Johnson et al. 2012). At the root of all complete depiction of environmental characteristics the these techniques is the idea that environmental species tolerates in the south, which should generate a conditions in places a species is known to occur can more accurate model (Elith et al. 2011). provide information on species-specific habitat requirements, and then new areas with similar MATERIALS AND METHODS environmental conditions can then be found on the landscape. The end product is a habitat description or Study site.—Bog Turtles were first scientifically map that highlights potentially suitable areas for the described in 1801 from a Pennsylvania, USA, specimen. species. Such maps are now being used for conservation They were considered a species of northern North planning (Rodríguez et al. 2007; Gogol-Prokurat 2011; America until 1917 when an individual was found in Hunter et al. 2012), to decrease sampling effort when North Carolina, but other states were not added to the searching for rare species (Singh et al. 2009), to locate southern range until the 1960s when Bog Turtles were rare species of plants (Guisan et al. 2006; Williams et al. found in Virginia (Tryon 1990). The range of Bog 2009; Le Lay et al. 2010; Buechling and Tobalske 2011), Turtles in the south did not take its current form until endemic insects (Rinnhofer et al. 2012), a cryptic species 1986 when the species was discovered in Tennessee, the of mole (Jackson and Robertson 2011), a bat (Rebelo last state to be added to the current range (Tryon 1990). and Jones 2010), and salamanders (Apodaca et al. 2012; This late discovery in the south is emblematic that the Chunco et al. 2013; Peterman et al. 2013), and to define task is ongoing to find new populations that will

200

Herpetological Conservation and Biology

FIGURE 1. Number of Bog Turtle (Glyptemys muhlenbergii) localities by county for the southern portion of the Bog Turtle Range (dark gray polygons). The light gray polygon represents all counties within a 25-km buffer around counties known to be occupied. The mapped region was used as the area of interest (i.e., background) for the species distribution model. establish a specific, thorough knowledge of the Bog Choice of species distribution model.—There are Turtle range (USFWS 2001; GADNR 2005; NCWRC many potential modeling techniques to examine species 2005; SCDNR 2005; VADGIF 2005). The task is habitat selection and distribution (Elith et al. 2006; Satter complicated by the fact that we are now searching for et al. 2007; Singh et al. 2009; Buechling and Tobalske populations after extended periods of human alteration 2011; McKenna et al. 2013); however, many of the of landscapes through wetland drainage and other available tools require knowledge of both presence and means. The need to find new populations is most absence at a suite of sites. Because Bog Turtles are rare pressing in the south, where research on the species is and extremely difficult to locate, robust absence data are not as extensive and developmental pressures are high unavailable, and we were restricted to presence-only (Wear and Bolstad 1998). For this reason, we restricted modeling techniques. We chose to use a machine- our study to the southern range of Bog Turtles, which learning approach to species distribution modeling called occurs in northern Georgia, northern South Carolina, maximum entropy (Maxent; Phillips et al. 2006), which western North Carolina, eastern Tennessee, and has been shown to perform very well relative to other southwestern Virginia, USA. For modeling purposes, presence-only options (Elith et al. 2006), is easily we considered the study area to be all counties with interpretable, and works well with small datasets known Bog Turtle localities plus an additional 25 km (Hernandez et al. 2006). Maxent (version 3.3.3k) buffer (Fig. 1); however, due to an error we did not compensates for absence data by randomly selecting buffer Franklin County, Virginia. The one point within points (i.e., background points) from the study area to this county is in close proximity to the western counties characterize the range and variation of the environmental and thus this oversight does not interfere with the variables available within the range of a species. By purpose of the buffer. The purpose of the buffer is to comparing the environment in areas with known Bog allow the SDM to examine areas just outside of the core Turtle populations to the environment available to it (i.e., area. As the omitted county buffer represents a very the background points), Maxent identifies preferences of small part of the overall range it is not likely to greatly species for certain ranges of environmental variables. influence conclusions and has no bearing on ground- The direction and strength of these preferences allows validation efforts which occurred far south of the area in for predictions on the probability of suitable conditions question. in unsurveyed areas. The ultimate product is a map,

201

Stratmann et al.—A species distribution model for Bog Turtles. dividing the study area into suitable and unsuitable data included all anurans, all chelonians, several species patches. of squamata (Worm Snake, Carphophis amoenus, Ring- necked Snake, Diadophis punctatus, Eastern Fence Occurrence records.—All presence data were Lizard, Sceloporus undulatus, and Ribbon Snake, obtained from the Georgia Department of Natural Thamnophis sauritus), and all salamanders in the genera Resources, South Carolina Department of Natural Desmognathus, Plethodon, Eurycea, Gyrinophilus, Resources, North Carolina Wildlife Resources Pseudotriton occurring in the range of interest. Commission, Virginia Department of Conservation and Occurrence records were obtained from HerpNet Recreation, and Tennessee Department of Environment (Available from http://www.herpnet.org/ [Accessed 22 and Conservation through data use agreements. Data November 2013]), BISON (Biodiversity Information were provided as GPS points or polygons. All polygons Serving Our Nation. Available from http://bison. were assigned a point representing their center and then usgs.ornl.gov/#home [Accessed 1 December 2013]), and combined with the point data into a single shapefile GBIF (Global Biodiversity Information Facility. using ArcGIS 10.1 (ESRI, Redlands, California, USA). Available from http://www.gbif.org/ [Accessed 22 This original data set of 172 localities consisted of all November 2013]), and any duplicate points were extant and historical sites cataloged within each state removed. The study area was divided into three equal (Fig.1). areas and points were randomly removed from areas Historical localities, locations in proximity to other until the distribution of points was approximately even known locations, and non-random species surveys can among the three areas. After filtering in this manner, all introduce biases into the assessment of habitat 1,967 background points remained. This is a small but suitability (Phillips et al. 2009; Kramer-Schadt et al. sufficient number of background points (Phillips and 2013). We applied three data filters to reduce these Dudík 2008), but we also built a model using a biases in the data. First, we kept only sites last visited background layer with 10,000 randomly generated points and confirmed as extant in the past 30 y, a time period (Phillips et al. 2006), as Phillips and Dudík (2008) show that approximates the life-span of the species (Ernst and a larger number of points can improve model Lovich 2009). Second, when several localities are performance. Thus, we produced two models, a random clumped in a small area, the habitat characteristics of points (RP) model, and a target group (TG) model. that area become over-represented and can bias the model. For sites that occurred in clusters, we performed Predictor variables.—A Maxent model uses a single data reduction exercise. Specifically, we environmental characteristics related to the natural randomly eliminated sites, until no site was within 5 km history of a species to make predictions about potentially of another (Barrett et al. 2014; Sutton et al. 2015). Boria suitable habitat. We focused on 12 environmental et al. (2014) and Kramer-Schadt et al. (2013) both have variables (Table 1). The first set of environmental shown that addressing this bias can result in large variables, elevation, topographic relief, temperature differences in model performance. The first and second seasonality (standard deviation of monthly filters reduced our data set from 172 to 72 occurrence temperatures), and maximum temperature of warmest records. The third filter addresses the fact that species month, all relate to the montane, ectothermic nature of are rarely sampled randomly. Locality records can this turtle. Topographic relief was calculated with the easily become spatially biased, for example, toward ArcGIS tool Focal Statistics as the standard deviation of roads and certain geopolitical areas that are more elevation within a 1-ha moving window. One hectare sampled than others (Funk and Richardson 2002; was chosen for this and all other moving window Graham et al. 2004; Rondinini et al. 2006; Beck et al. analyses as most bogs are smaller than a hectare (Lee 2014). Bog Turtle records, like records for many other and Nordon 1996; Buhlmann et al. 1997). Developed species, likely suffer this bias. Most sites are close to areas, pasture and hay fields, and wetlands were chosen roads and developed areas, and locality data suggest that as the relevant land cover variables. We specifically some states have invested much more search effort than selected the pasture and hay field category to address the others (Fig. 1). This kind of bias can be accounted for fact that the wetlands these species occupy tend to occur by adjusting how background points are chosen. We in flat areas in the mountains that are often ditched and made this adjustment using a target background drained for agricultural or development purposes (Tryon approach, in which background data are acquired from 1990; Moorhead and Rossell 1998). We hypothesized species that are collected with similar biases as the target that developed areas are likely to represent unsuitable species (Phillips and Dudík 2008; Phillips et al. 2009). habitat, but hay fields and pastures browsed by livestock To build our target group background data set, we often remain wet and can support viable Bog Turtle acquired locality data from other reptiles and amphibians populations (Tesauro and Ehrenfeld 2007). Because to capture the behavior of the sampling agent (i.e., known Bog Turtle localities are represented by points herpetologists). Species we included for background and extracting the data directly below a point can be an

202

Herpetological Conservation and Biology

TABLE 1. The source of data and their biological relevance for 12 environmental characteristics used to build the species distribution model of the Bog Turtle (Glyptemys muhlenbergii). All data were accessed 8 September 2013. An asterisk (*) denotes a variable is characterized within a 1-ha moving window.

Environmental Characteristic Source Biological Relevance

Elevation Topographic Relief (standard deviation of USGS Digital Elevation Model (DEM) - Available elevation)* from http://eros.usgs.gov/#/Guides/dem Montane, ectothermic nature of turtle. Temperature Seasonality (standard deviation of

monthly temperatures) WorldClim database - Available from www.worldclim.org/bioclim Maximum Temperature of Warmest Month

Percentage Developed Land* To rule out flat but unsuitable areas. Percentage Pasture or Hay Fields* National Land Cover Database (NLCD) - Available from www.mrlc.gov/nlcd2011.php Land cover that could contain

Percentage Wetland* suitable habitat.

Distance to Nearest Wetland National Wetlands Inventory (NWI) - Available from http://www.fws.gov/wetlands/ Connectivity to other wetlands. Distance to Nearest Stream USGS National Hydrography Dataset - Available from http://nhd.usgs.gov/

Percentage Organic Matter Survey Spatial and Tabular Data (SSURGO 2.2). Gaps were filled using the larger grain but more Percentage Clay geographically complete U.S. General Soil Map Suitable wetland soils and (STATSGO2) - Available from conducive for burrowing. Percentage Hydric Soils http://websoilsurvey.nrcs.usda.gov/

inaccurate representation of the larger area the wetland the cell size of the smallest available data (30 m × 30 m) covers, we used the Focal Statistics tool to calculate the and we processed them in ArcGIS 10.1. We made a percentage land cover of interest (e.g., percentage Pearson’s correlation test in R version 3.1.1 (R pasture/hay) in the surrounding hectare. This was only a Development Core Team 2014, Vienna, Austria) on all problem for the land cover variables which were at a 12 environmental variables. BIO4 and BIO5 each much finer scale than the average Bog Turtle wetland correlated with elevation (r = ˗0.70 and ˗0.98, size, the other environmental variables were at a much respectively), but not with each other. Despite this we coarser scale and captured the size of the wetland and kept elevation in the model. While temperature is the surrounding area. We did not include the National more likely mechanistic driver of distribution from a Wetland Inventory (NWI) database as a predictor historical standpoint, elevation is a more important variable in the model, as this dataset is notorious for driver in recent times because it reflects places where missing the small wetlands that typify Bog Turtle habitat developmental pressures may accumulate more slowly (Leonard et al. 2012). By not restricting the model to (Napton et al. 2010). No other variables exhibited strong known wetland localities, we hoped to identify potential correlations (all r values < 0.70). habitat in previously unconsidered locales. As a species with high site fidelity that historically Model processing.—Maxent was run using the auto existed in a metapopulation structure (Buhlmann et al. features option, which attempts to fit a range of 1997), connectivity is important, so we also included functions including linear, quadratic, product, threshold, distance to nearest wetland and stream as modeled and hinge features. We withheld 10% of points for environmental variables. Distance to nearest wetland evaluating model quality. We ran the model 10 times, and stream were calculated using Euclidean distance. and all results shown are the average of the10-fold cross- Finally, because this species spends much of its time validation. within the first 10 cm of organic wetland muck (Pittman To select potential Bog Turtle habitat, we examined and Dorcas 2009) and must be able to move through the the Maxent logistic output in the context of 11 different soil, we also used information on soils: percentage default thresholds, used to categorize values as suitable organic matter, percentage clay, and percentage hydric (at or above threshold value) or unsuitable (below soils. We resampled all 12 environmental variables to threshold value). For each of our models (RP and TG),

203

Stratmann et al.—A species distribution model for Bog Turtles. we used the most restrictive threshold (maximum test vegetation, and hydrology, as outlined in the Phase 1 sensitivity plus specificity and equal training sensitivity survey guidelines established by the U.S. Fish and and specificity, respectively) in an effort to restrict the Wildlife Service (USFWS 2006). Characteristics of model to the most promising sites. We then examined interest are: (1) presence of mucky areas of more organic the intersection of the two models to evaluate areas than alluvial characteristics; (2) presence of small considered unsuitable by both, suitable by both, and, rivulets of water flowing over a muddy substrate; (3) suitable by one model but not the other. presence of springs or seep heads; (4) presence of sphagnum (Sphagnum spp.), rushes (Juncus spp.), Model ground-validation.—To evaluate the predictive sedges (Carex spp.), Alder (Alnus serrulata), Red Maple power of the species distribution model, we conducted (Acer rubrum), Bog Rose (Rosa palustris), Multiflora on-the-ground surveys of sites modeled to be potentially Rose (Rosa multiflora), Withe Rod (Viburnum nudum L. suitable by both models of Bog Turtle habitat. var. cassinoides), Royal Fern (Osmunda regalis), Resources were not available to ground validate the Cinnamon Fern (Osmunda cinnamomea), Red model across the entire southern range. We therefore Chokeberry (Aronia arbutifolia), and Turtle Head made a series of decisions to prioritize the areas that we (Chelone glabra); (5) large area of open canopy; (6) would examine on-the-ground. First, our ground- proximity to a stream; and (7) active or historic North validation assessed only errors of commission (i.e., areas American Beaver (Castor canadensis) activity. We incorrectly identified as suitable habitat). Next, we surveyed wetlands that met USFWS standards for Bog narrowed our focus to Georgia and South Carolina, Turtles. where demand to find new populations was highest and Traditional techniques to survey for Bog Turtles resources were readily available. In Georgia, the model include trapping surveys and visual/probing surveys identified 88,038 ha of potentially suitable habitat to (USFWS 2006). To maximize detection probability, we survey; whereas, 12,635 ha were identified in South used a trapping method similar to that described by Carolina. We further narrowed our search to the Blue Somers and Mansfield-Jones (2008) but at a much Ridge physiogeographic region of each state, as all Bog higher trap density (1 trap /25 m2). With this method un- Turtle records except for 17 in North Carolina fall within baited, custom-made traps of galvanized welded-wire are this physiogeographic region. A lack of high-quality set in shallow rivulets of water that could act as potential color infrared imagery made it difficult to assess in travel corridors for Bog Turtles (Stratmann 2015). Traps ArcGIS whether suitable areas were also wet areas. To were open from mid-May to mid-July 2014 and checked ensure we would spend resources traveling to actual every-other day. Based on trapping efforts at other sites wetlands, we clipped the suitability map to all National in the region known to be occupied, we estimate that this Wetlands Inventory (NWI) wetlands in the palustrine trapping duration means there is a very low probability emergent (PEM), palustrine scrub-shrub (PSS), or (average = 0.03) that turtles are present but were not palustrine forested (PFO) categories (following detected (Stratmann 2015). Cowardin et al. 1979). The NWI does not capture all wetlands (Leonard et al. 2012) but was sufficient for a RESULTS first round of ground-validation. Finally, we had to consider accessibility to sites. Therefore, of the sites Both the random points (RP) and target group (TG) remaining, we prioritized those sites on public land background models had high AUC values (RP model = within 1 km of a road and sites on private property 0.88, TG model = 0.91) and correctly classified a within 20 km of a Bog Turtle occurrence record. With majority of the known Bog Turtle localities (Table 2). this strategic search, we were able to test the hypothesis Each model was driven by different environmental that sites modeled as suitable are actually occupied by variables and thus produced different suitability maps Bog Turtles. Given this hypothesis, the decisions we (Fig. 2). In the RP model, distance to wetland, made to sample those areas that could be reasonably maximum temperature of warmest month, elevation, and accessed did not bias our field surveys. With unlimited topographic relief had the highest percentage time and money we would have taken our unbiased contributions (22.8%, 19.3%, 17.4%, and 10.8%, habitat model and assessed both errors of commission respectively; Table 3). In the TG model, distance to and omission, yet given the difficulties of working on wetland, percentage pasture/hay, percentage developed private lands, the costs of driving through a large area, and distance to stream had the highest contributions geographic extent, and the logistical challenges of (30.7%, 29.3%, 17.1%, and 10.8%, respectively; Table trapping remote areas, we made a series of decisions to 3). Jackknife tests show that maximum temperature of take a first pass at validating the model in way that was the warmest month holds the most information by itself practical. for the RP model, whereas in the TG model it is percent At prioritized sites, we evaluated the quality of the pasture/hay. habitat for Bog Turtles based on indicator soils,

204

Herpetological Conservation and Biology

TABLE 2. The Maxent models had high accuracy in correctly sites we trapped in South Carolina, three were on public classifying the known Bog Turtle (Glyptemys muhlenbergii) localities property and five were on private property. In South as suitable habitat. Known Bog Turtle localities (n = 72) and their classification (suitable or unsuitable) are presented for each of the Carolina, because we were denied access to five three different models: random points (RP), target group (TG), and a promising wetlands, we trapped at three of the eight sites model which combines the RP and TG models. only deemed suitable by a single model. These sites were targeted based on their proximity to known No. locations correctly occurrences. Of the 17 sites trapped, we discovered Bog Site Suitability classified (percent) Turtles at two of the sites in Georgia. One site was a Combined Model spring-fed wet meadow with open canopy and active suitable in both models 52 (72.2%) beaver presence and is the only known extant population of Bog Turtles in Rabun County. The other was an suitable in only one model 14 (19.4%) extremely atypical site located on the slope of a hill unsuitable in both models 6 (8.3%) under a power line right of way. Seeps run down the hill and pool in several places to create areas of very shallow Random Points Model muck (< 30 cm). This is only the second Bog Turtle suitable 56 (77.8%) population found in Towns County.

Target Group Model DISCUSSION suitable 62 (86.1%) The objective of our study was to determine if species distribution modeling could be a useful tool to find For the RP model, the most conservative threshold habitat for a rare and cryptic species. Although past was the maximum test sensitivity plus specificity. This studies have demonstrated success with this approach, translates to 1,219,828 ha of suitable habitat and we found that the modeled distribution greatly over- 6,995,533 ha of unsuitable habitat across the area of predicted the amount of suitable habitat for Bog Turtles interest. For the TG model, the most conservative and was thus difficult to use in a resource-efficient threshold was the equal training sensitivity and manner. specificity threshold. This resulted in 3,074,811 ha of suitable habitat and 5,140,550 ha of unsuitable habitat. Model differences.—We built two SDMs to address When both models were combined 998,325 ha were biases in our data. Known Bog Turtle populations are considered suitable (Fig. 3). biased towards roads and easily accessible areas, and the Of the area considered suitable by both models, discovery of one population in an area tends to focus 88,038 ha and 12,635 ha occurred in Georgia and South future search efforts in nearby areas. We dealt with this Carolina, respectively. When limited to PSS, PEM, or bias by filtering our locality points and modeling PFO NWI wetlands in the Blue Ridge physiogeographic available habitat from a target group background of region, 316 wetlands were designated as suitable by both other reptile and amphibian localities, which presumably models and 34 wetlands were designated as suitable by share any biases held by the presence points (Phillips only one model but not the other in Georgia. In South and Dudík 2008). This target group (TG) model was Carolina 109 wetlands were designated as suitable by substantially different from the standard random points both models and 87 wetlands designated as suitable by background (RP) model. As was expected, the RP one model but not the other. Of these wetlands, we model was biased toward areas of known occurrence; examined 113 on the ground in Georgia and 83 in South whereas, the TG model made stronger predictions in Carolina. Nine of the 113 sites in Georgia met USFWS less-sampled areas (Phillips et al. 2009). In the target criteria for suitable Bog Turtle habitat and we trapped group model, pasture/hay held the most information; these. In South Carolina, 13 of the 83 sites met criteria; suitability increased with an increasing percentage of although, we were only able to gain permission to trap at pasture/hay land cover. Consequently, the model eight of the sites in South Carolina. Therefore, for about indicated suitable habitat in areas of pasture/hay in the every 100 wetlands searched, we found that only 10 met Piedmont region that the RP model did not consider. the USFWS criteria for suitable Bog Turtle habitat. During ground-validation, we examined many pastures, Although wetlands deemed unsuitable for Bog Turtles as NWI wetlands often showed up in this land cover according to USFWS guidelines cannot be counted as type. Nonetheless, we believe wetlands in pasture/hay absences, the high number of wetlands not matching areas were over-emphasized in the model because 51% these guidelines would at least suggests that Maxent of Bog Turtle presence localities (37 out of 72) were greatly over predicted the amount of suitable habitat. Of associated with pasture/hay, but only 10% (186 out of the nine sites we trapped in Georgia, five were on public 1967) of the target group points were associated with property and four were on private property. Of the eight this land use type. The importance of pastures and

205

Stratmann et al.—A species distribution model for Bog Turtles.

FIGURE 2. Maxent logistic output for (a) the random points model and (b) the target group model of Bog Turtles (Glyptemys muhlenbergii). Warmer colors indicate higher suitability. Models are based off of the same presence points but the random points model uses 10,000 randomly generated points as background points, while the target group model uses about 2,000 points derived from herpetofauna occurrences as its background points.

206

Herpetological Conservation and Biology

TABLE 3. The percentage contribution and permutation importance of variables used in Maxtent for the Random Points and Target Background models for Bog Turtles (Glyptemys muhlenbergii). Permutation importance is the drop in training AUC when the values of a variable for training presence and background data are randomly permuted and the model is reevaluated using the permutated data. Variables of higher importance result in larger drops in AUC.

Random Points Model Target Background Model Percentage Permutation Percentage Permutation Variable Contribution Importance Contribution Importance Distance to Wetland 22.8 16.0 30.7 34.4 Maximum Temperature of Warmest Month 19.3 29.8 2.0 1.1 Elevation 17.4 9.8 0.4 1.6 Topographic Relief 10.8 25.6 5.6 12.2 Percent Hydric Soil 9.1 0.9 1.3 0.1 Distance to Stream 8.6 8.9 10.8 24.8 Percent Organic Matter 3.6 1.1 0.6 0.4 Percent Pasture/Hay 3.0 1.1 29.3 2.0 Percent Developed Land 2.8 4.6 17.1 20.8 Percent Clay 1.8 1.8 0.8 0.5 Percent Wetland 0.6 0.0 0.6 0.0 Temperature Seasonality 0.2 0.5 0.7 2.0

hayfields to Bog Turtles is well established (Tesauro and Wetlands tend to be small features (1–3 ha) within the Ehrenfeld 2007). Throughout the Appalachians, farmers greater landscape. In our model, 30 m × 30 m was the have converted wetlands to pastures or hay fields finest grain available (NLCD, DEM), with soil and through ditching and draining. When these efforts fail climate environmental layers being particularly coarse. and some part of the wetland remains, the site is kept as These large scale data sets are also minimally ground- an emergent wetland through grazing activities, which validated and have their own errors, which when occasionally provides quality Bog Turtle habitat combined with the lack of fine-scale resolution may (Zappalorti 1976b, 1997). Yet it is not the pasture/hay mean that data are simply not at a fine enough resolution land cover that initially made the site suitable; conditions to pick-up small features like small wetlands within the for wetland formation must be present. This larger landscape. Until finer resolution wetland and complicated mix of geomorphology, historic land environmental data are available, it will be difficult to alteration, and current land use (i.e., grazing) highlights improve this aspect of the SDM. Two other factors that the importance of our duel-modeling approach and might explain the difficulty of building a SDM for Bog ground validation for model output. Turtles were brought to our attention: we have trapped many wetlands that meet established USFWS criteria for Ground validation.—The distribution model high- good Bog Turtle habitat without detecting Bog Turtles, lighted many areas that were not wetlands. There appear but know of populations in sites that would never be to be four main reasons why the SDM was not as considered suitable under USFWS criteria because they restrictive as we had hoped. First, identifying small are so unusual or degraded. This seems to hint at two wetlands across large extents is difficult in general (Pitt things. One, unusual sites might suggest that Bog et al. 2011; Leonard et al. 2012). We did not expect our Turtles are less habitat specialists than thought, with model to only identify wetlands. Instead, we aimed to main requirements being shallow water, winter refugia, determine if a model generated from available remote and open canopy. For example, the new Bog Turtle sensing data would be sufficient to identify Bog Turtle locality in Towns County is very different from typical habitat. It appears key missing data (e.g., better Bog Turtle habitat as it is situated on a slope, has very hydrologic and soil maps) do considerably limit the shallow muck, is very rocky, and is more a series of effectiveness of the SDM. We did incorporate the hillside seeps that occasionally puddle than a typical available NLCD and NWI wetland data to identify wetland. Although the model deemed it as suitable, we wetlands but NWI is notorious for missing small only trapped this area because of claims that a Bog wetlands (Leonard et al. 2012), and many NWI wetlands Turtle had been seen at the site. Radio-telemetry at the we visited no longer have standing water in them and are site has shown that with little mud available, individuals barely distinguishable as wetlands. As remote sensing use cavities under shrub root clumps amid rocky seeps as continues to improve, especially with the availability of alternative refugia. This observation could be explained LiDAR (Light Detection And Ranging), this problem by the fact that moving spring water is more important to may diminish and enable us to restrict the model to hibernating Bog Turtles than the depth of the mucky soil specific areas of interest (e.g., Peterman et al. 2013). (Ernst et al. 1989; Zappalorti 1997). The reporting of Second, high resolution environmental data are lacking. Somers and colleagues (2007) demonstrate that this

207

Stratmann et al.—A species distribution model for Bog Turtles.

FIGURE 3. Binary (suitable/unsuitable) model results for the random points (purple) and the target group (green) models of habitat suitability for Bog Turtles (Glyptemys muhlenbergii) in the southeastern United States. Black areas were considered suitable in both models. Green or purple areas are considered suitable in only one model.

unusual use of habitat is not unprecedented. On the understand the realized distribution of a species (Guisan other hand, for degraded sites, site history (e.g., past and Thuiller 2005). ditching and draining efforts, past beaver influence, canopy cover over the years) may play a larger role than Recommended uses of the model.—When rare and current habitat characteristics in explaining where turtles cryptic species are targeted for management and persist or have sought refugia. Some historically drained conservation action, finding suitable habitat and new wetlands may have had time to recover and now appear populations can represent an overwhelming task, suitable, but a recolonization event would be needed if especially when the species occurs across large spatial those draining efforts initially extirpated Bog Turtles. extents. Species distribution models can highlight a Other wetlands may have been recently or minimally subset of possible search environments and by using drained, enabling Bog Turtles to persist despite apparent such models to guide search efforts, we can strategically habitat degradation. Thus, a SDM could highlight search for populations, prioritizing where to invest wetlands suitable for Bog Turtles based on resources. Using a SDM to guide searches for suitable environmental characteristics, but a detailed knowledge habitat is strategic in the sense that it eliminates typical of the history of the site would be required to determine search bias and examines the entire area of interest. All its true suitability. Other studies have shown that potential areas of interest are known and thus a strategic incorporating information on site history can enable plan can be made to examine as many wetlands as better predictions of species occurrence and patterns of possible. We had to make a series of choices on where biodiversity (Dupouey et al. 2002; Lunt and Spooner to search based on time and budget constraints, but 2005; Piha et al. 2007). Site history would also be one ideally all areas could be examined. Searches for habitat way to parse suitable but unoccupied sites from those driven by information from a SDM would avoid biasing suitable and occupied. Suitable sites may have all the searches along or close to roads or previously searched habitat characteristics required by a species, but may still locales, while drawing attention to difficult to access be unoccupied due to historical disturbance events, loss areas, like those on private property. This is especially of connectivity, extirpation, competition, and predation, important considering 80% of known Bog Turtle which would need to be added to a model to better localities and 91.6% of suitable habitat occur on private

208

Herpetological Conservation and Biology

property. In South Carolina, we were able to examine however, even those improvements may not alter the much of the publically owned land, and subsequently utility of the model if site history drives the distribution concluded if Bog Turtles still exist in the state, they most of the species. Conservation efforts for many species likely occur only on private property. In South Carolina (e.g., flatwoods salamanders, Ambystoma cingulatum and Georgia, the publically owned land is often the most and bishopi), Pine Barren Treefrogs, Hyla andersonii, topographically intense in the state because these areas Gopher Frogs, Lithobates capito) pose the same were not valuable as agricultural lands. Yet Bog Turtles, logistical challenges because the species is rare on the like agriculture, thrive in the valleys, which puts the landscape, difficult to detect even in suitable habitat, and species in direct conflict with agricultural developments. greatly affected by land-use change. For these species The SDM emphasizes these associations and offers a the answer may simply be that until wetland and platform for large-scale conservation planning that will environmental data become more fine scale, finding new undoubtedly need to incorporate a number of populations will be resource intensive, and as long as stakeholders. these data are unavailable, it may be more prudent to A regionally based SDM such as the one we created is prioritize resources for restoration efforts and particularly valuable for Georgia, South Carolina, and conservation of known populations. Tennessee, which have too few Bog Turtle records to build locally-based SDMs for Bog Turtles. In addition, Acknowledgments.—Funding and support was in these states the need to find new populations is the generously provided by the Bern W. Tryon Funding for greatest and the amount of area to search is the smallest. Southern Bog Turtle Population Research, Clemson Collectively, these conditions mean over-prediction is University, the Georgia Department of Natural less of a logistical problem. The large area modeled as Resources, Riverbanks Zoo and Garden, and the South suitable errs on the side of inclusion, which may be wise Carolina Department of Natural Resources. The Georgia given our discovery of a Bog Turtle population in Department of Natural Resources, South Carolina atypical habitat. Practical model application would Department of Natural Resources, North Carolina likely require narrowing the focus to wetlands for Wildlife Resources Commission, Virginia Department of survey, as was done in this study. Once this list is Conservation and Recreation, and Tennessee created, a long-term plan of action based on available Department of Environment and Conservation each resources becomes more realistic. For states like North provided data and state-level insight and support. We Carolina and Virginia where the area modeled as especially thank Mary Bunch, Will Dillman, Gabrielle suitable is much larger, this approach may be less Graeter, J.D. Kleopfer, and Michael Ogel for orienting feasible. Nevertheless, these states may choose to focus us to state-level Bog Turtle knowledge. We thank the on areas where gaps in the range remain to be filled. following people for their help with field work: Grover It should also be noted, that although a SDM may not Brown, Briana Cairco, Maddy Fesite, David Hutto, always identify populations of the target species, it more Meaghan Miranda, Todd Pierson, Danielle Sexton, often identifies suitable but unoccupied habitat. This Claudia Stratmann, Kathy Vause, and Sydney Wells. suitable habitat could be used as a re-introduction site for We are also indebted to all the landowners who let us Bog Turtles, or other mountain fen specialists (Tryon explore the wetlands on their property and taught us 2009). For example, our SDM identified two sites in about the history of the area. We thank Rob Baldwin Georgia that are now being considered as out-planting and Mike Dorcas for feedback on the manuscript and sites for Swamp Pink (Helonias bullata) and pitcher John Maerz, Nate Nibbelink, and Jena Hickey for initial plants (e.g., Sarracenia purpurea). These wetlands help and advice on the species distribution model. All could now also be surveyed for other species of concern work was conducted in compliance with Clemson’s within the state (e.g., Bog Lemmings, Synaptomys Institutional Animal Care and Use Committee (IACUC) cooperi, Four-toed Salamanders, Hemidactylium protocol (#2014-003) and the South Carolina scutatum, and Golden-winged Warblers, Vermivora Department of Natural Resources (Permit # 14-2014). chrysoptera). Work in Georgia was conducted under the authority of The ground-based survey process is extremely time the Georgia Department of Natural Resources in intensive (USFWS 2001; Somers and Mansfield-Jones accordance with the Official Code of Georgia Annotated 2008). Finding new populations of Bog Turtles will § 27-1-22. always represent a significant investment and state agencies must seriously consider if population discovery LITERATURE CITED is where they should direct resources earmarked for this species. Such considerations are especially pertinent Apodaca, J.J., J. Homyack, and L.J. Rissler. 2012. Using because a model that better identifies suitable habitat a species-specific habitat model helps identify will not be available until a complete wetland layer and unprotected populations of the federally threatened finer-grained environmental layers become available;

209

Stratmann et al.—A species distribution model for Bog Turtles.

Red Hills Salamander (Phaeognathus hubrichti). improve prediction of species' distributions from Herpetological Review 43:230–233. occurrence data. Ecography 29:129–151. Barrett, K., N.P. Nibbelink, and J.C. Maerz. 2014. Elith, J., S.J. Phillips, T. Hastie, M. Dudík, Y.E. Chee, Identifying priority species and conservation and C.J. Yates. 2011. A statistical explanation of opportunities under future climate scenarios: MaxEnt for ecologists. Diversity and Distributions amphibians in a biodiversity hotspot. Journal of Fish 17:43–57. and Wildlife Management 5:282–297. Ernst, C.H., R.T. Zappalorti, and J.E. Lovich. 1989. Beck, J., M. Böller, A. Erhardt, and W. Schwanghart. Overwintering sites and thermal relations of 2014. Spatial bias in the GBIF database and its effect hibernating Bog Turtles, Clemmys muhlenbergii. on modeling species’ geographic distributions. Copeia 1989:761–764. Ecological Informatics 19:10–15. Ernst, C.H., and J.E Lovich. 2009. Turtles of the United Boria, R.A., L.E. Olson, S.M. Goodman, and R.P. States and Canada. 2nd Edition. Johns Hopkins Anderson. 2014. Spatial filtering to reduce sampling University Press, Baltimore, Maryland, USA. bias can improve the performance of ecological niche Funk, V.A., and K.S. Richardson. 2002. Systematic data models. Ecological Modelling 275:73–77. in biodiversity studies: use it or lose it. Systematic Buechling, A. and C. Tobalske. 2011. Predictive habitat Biology 51:303–316. modeling of rare plant species in Pacific Northwest Georgia Department of Natural Resources (GADNR). forests. Western Journal of Applied Forestry 26:71– 2005. A Comprehensive Wildlife Conservation 81. Strategy for Georgia. Georgia Department of Natural Buhlmann, K.A., J.C Mitchell, and M.G. Rollins. 1997. Resources, Social Circle, Georgia, USA. New approaches for the conservation of Bog Turtles, Gogol-Prokurat, M. 2011. Predicting habitat suitability Clemmys muhlenbergii, in Virginia. Pp. 359–363 In for rare plants at local spatial scales using a species Proceedings: Conservation, Restoration, and distribution model. Ecological Applications 21:33–47. Management of Tortoises and Turtles - an Graham, C.H., S. Ferrier, F. Huettman, C. Moritz, and International Conference. Van Abbema, J. (Ed.) New A.T. Peterson. 2004. New developments in museum- York Turtle and Tortoise Society, New York, New based informatics and applications in biodiversity York, USA. analysis. Trends in Ecology & Evolution 19:497–503. Campbell, J.W., M.T. Mengak, S.B. Castleberry, and Graham, S.P., D.A. Steen, K.T. Nelson, A.M. Durso, and J.D. Mejia. 2010. Distribution and status of J.C. Maerz. 2010. An overlooked hotspot? Rapid uncommon mammals in the Southern Appalachian biodiversity assessment reveals a region of exceptional Mountains. Southeastern Naturalist 9:275–302. herpetofaunal richness in the Southeastern United Carter, S.L., C.A. Haas, and J.C. Mitchell. 1999. Home States. Southeastern Naturalist 9:19–34. range and habitat selection of Bog Turtles in Groff, L.A., S.B. Marks, and M.P. Hayes. 2014. Using southwestern Virginia. Journal of Wildlife ecological niche models to direct rare amphibian Management 63:853–860. surveys: a case study using the Oregon Spotted Frog Chase, J.D., K.R. Dixon, J.E. Gates, D. Jacobs, and G.J. (Rana pretiosa). Herpetological Conservation and Taylor. 1989. Habitat characteristics, population size, Biology 9:354−368. and home range of the Bog Turtle, Clemmys Guisan, A., and W. Thuiller. 2005. Predicting species muhlenbergii, in Maryland. Journal of Herpetology distribution: offering more than simple habitat models. 23:356–362. Ecology Letters 8:993−1009. Chunco, A.J., S. Phimmachak, N. Sivongxay, and B.L. Guisan, A., O. Broennimann, R. Engler, M. Vust, N.G. Stuart. 2013. Predicting environmental suitability for a Yoccoz, A. Lehmann, and N.E. Zimmermann. 2006. rare and threatened species (Lao Newt, Laotriton Using niche-based models to improve the sampling of laoensis) using validated species distribution models. rare species. Conservation Biology 20:501−511. PLoS ONE 8:e59853. Hernandez, P.A., C.H. Graham, L.L. Master, and D.L. doi:10.1371/journal.pone.0059853 Albert. 2006. The effect of sample size and species Cowardin, L.M., V. Carter, F.C. Golet, and E.T. LaRoe. characteristics on performance of different species 1979. Classification of wetlands and deepwater distribution modeling methods. Ecography 29:773– habitats of the United States. U.S. Fish and Wildlife 785. Service. Washington, D.C. 142p. Hunter, E.A., P.A. Raney, J.P. Gibbs, and D.J. Leopold. Dupouey, J.L., E. Dambrine, J.D. Laffite, and C. 2012. Improving wetland mitigation site identification Moares. 2002. Irreversible impact of past land use on through community distribution modeling and a patch- forest soils and biodiversity. Ecology 83:2978–2984. based ranking scheme. Wetlands 32:841–850. Elith, J., C.H. Graham, R.P. Anderson, M. Dudik, S. Jackson, C.R. and M.P. Robertson. 2011. Predicting the Ferrier, A. Guisan, R.J. Hijmans, F. Huettmann, J.R. potential distribution of an endangered cryptic Leathwick, A. Lehmann, et al. 2006. Novel methods

210

Herpetological Conservation and Biology

subterranean mammal from few occurrence records. distribution in southeastern New York State. Journal Journal for Nature Conservation 19:87–94. of Fish and Wildlife Management 4:255–266. Johnson, C.J., M. Hurley, E. Rapaport, and M. Pullinger. Napton, D.E., R.F. Auch, R. Headley, and J.L. Taylor. 2012. Using expert knowledge effectively: lessons 2010. Land changes and their driving forces in the from species distribution models for wildlife Southeastern United States. Regional Environmental conservation and management. Pp. 153–171 In Expert Change 10:37–53. Knowledge and Its Application in Landscape Ecology. North Carolina Wildlife Resources Commission Perera, A.H., C.A. Drew, and C.J. Johnson (Eds.). (NCWRC). 2005. North Carolina Wildlife Action Springer, New York, New York, USA. Plan. North Carolina Wildlife Resources Commission, Kramer‐Schadt, S., J. Niedballa, J.D. Pilgrim, B. Raleigh, North Carolina, USA. Schröder, J. Lindenborn, V. Reinfelder, M. Stillfried, Peterman, W.E., J.A. Crawford, and A.R. Kuhns. 2013. I. Heckmann, A.K. Scharf, and D.M. Augeri. 2013. Using species distribution and occupancy modeling to The importance of correcting for sampling bias in guide survey efforts and assess species status. Journal MaxEnt species distribution models. Diversity and for Nature Conservation 21:114–121. Distributions 19:1366–1379 Phillips, S.J., R.P. Anderson, and R.E. Schapire. 2006. Le Lay, G., R. Engler, E. Franc, and A. Guisan. 2010. Maximum entropy modeling of species geographic Prospective sampling based on model ensembles distributions. Ecological Modelling 190:231–259. improves the detection of rare species. Ecography Phillips, S.J., and M. Dudík. 2008. Modeling of species 33:1015–1027. distributions with Maxent: new extensions and a Lee, D.S. and A.W. Norton. 1996. The distribution, comprehensive evaluation. Ecography 31:161–175. ecology, and conservation needs of Bog Turtles, with Phillips, S.J., M. Dudik, J. Elith, C.H. Graham, A. special emphasis on Maryland. Maryland Naturalist Lehmann, J. Leathwick, and S. Ferrier. 2009. Sample 40:7–46. selection bias and presence-only distribution models: Leonard, P.B., R.F. Baldwin, J.A. Homyack, and T.B. implications for background and pseudo-absence data. Wigley. 2012. Remote detection of small wetlands in Ecological Applications 19:181–197. the Atlantic coastal plain of North America: local Pierson, T., T. Stratmann, E. White, A. Clause, C. relief models, ground validation, and high-throughput Carter, M. Herr, A. Jenkins, H. Vogel, M. Knoerr, and computing. Forest Ecology and Management B. Folt. 2014. New county records of amphibians and 284:107–115. reptiles resulting from a bioblitz competition in North- Lindeman, P. 2014. A significant range extension for the Central Georgia, USA. Herpetological Review Texas Map Turtle (Graptemys versa) and the inertia of 45:296–297. an incomplete literature. Herpetological Conservation Piha, H., M. Luoto, and J. Merilä. 2007. Amphibian and Biology 9:334−341. occurrence is influenced by current and historic Lunt, I.D., and P.G. Spooner. 2005. Using historical landscape characteristics. Ecological Applications ecology to understand patterns of biodiversity in 17:2298–2309. fragmented agricultural landscapes. Journal of Pitt, A.L., R.F. Baldwin, D.J. Lipscomb, B.L. Brown, Biogeography 32:1859–1873. J.E. Hawley, C.M. Allard-Keese, and P.B. Leonard. McCord, W.P., and J.B. Iverson. 1991. A new box turtle 2011. The missing wetlands: using local ecological of the genus Cuora (Testudines: Emydidae) with knowledge to find cryptic ecosystems. Biodiversity taxonomic notes and a key to the species. and Conservation 21:51–63. Herpetologica 47:407–420. Pittman, S.E. and M.E. Dorcas. 2009. Movements, McKenna, J.E., D.M. Carlson, and M.L. Payne-Wynne. habitat use, and thermal ecology of an isolated 2013. Predicting locations of rare aquatic species’ population of Bog Turtles (Glyptemys muhlenbergii). habitat with a combination of species-specific and Copeia 2009:781–790. assemblage-based models. Diversity and Distributions Rebelo, H., and G. Jones. 2010. Ground validation of 19:503–517. presence-only modelling with rare species: a case Moorhead K.K., and Rossell, I.M. 1998. Southern study on Barbastelles Barbastella barbastellus mountain fens. Pp. 379–404 In Southern Forested (Chiroptera: Vespertilionidae). Journal of Applied Wetlands. Messina, M.G., and W.H. Conner (Eds.). Ecology 47:410–420. Lewis Publishers, New York, New York, USA. Rinnhofer, L.J., N. Roura-Pascual, W. Arthofer, T. Morrow, J.L., J.H. Howard, S.A. Smith, and D.K. Dejaco, B. Thaler-Knoflach, G.A. Wachter, E. Poppel. 2001. Habitat selection and habitat use by the Christian, F.M. Steiner, and B.C. Schlick-Steiner. Bog Turtle (Clemmys muhlenbergii) in Maryland. 2012. Iterative species distribution modelling and Journal of Herpetology 35:545–552. ground validation in endemism research: an Alpine Myers, A.T., and J.P. Gibbs. 2013. Landscape-level Jumping Bristletail example. Biodiversity and factors influencing Bog Turtle persistence and Conservation 21:2845–2863.

211

Stratmann et al.—A species distribution model for Bog Turtles.

Rodriguez, J.P., L. Brotons, J. Bustamante, and J. Tryon, B. 2009. Egg-harvesting, hatching and release: a Seoane. 2007. The application of predictive modelling population augmentation tool for Bog Turtles in of species distribution to biodiversity conservation. Tennessee. Turtle Survival Alliance Newsletter. Diversity and Distributions 13:243–251. August:44–45. Rondinini, C., K.A. Wilson, L. Boitani, H. Grantham, Turtle Conservation Coalition (Rhodin, A.G.J., A.D. and H.P. Possingham. 2006. Tradeoffs of different Walde, B.D. Horne, P.P. van Dijk, T. Blanck, and R. types of species occurrence data for use in systematic Hudson [Eds.]). 2011. Turtles in Trouble: The World’s conservation planning. Ecology Letters 9:1136–1145. 25+ Most Endangered Tortoises and Freshwater Rosenbaum, P.A., J.M. Robertson, and K.R. Zamudio. Turtles - 2011. IUCN/SSC Tortoise and Freshwater 2007. Unexpectedly low genetic divergences among Turtle Specialist Group, Turtle Conservation Fund, populations of the threatened Bog Turtle (Glyptemys Turtle Survival Alliance, Turtle Conservancy, muhlenbergii). Conservation Genetics 8:331–342. Chelonian Research Foundation, Conservation Sattler, T., F. Bontadina, A. H. Hirzel, and R. Arlettaz. International, Wildlife Conservation Society, and San 2007. Ecological niche modelling of two cryptic bat Diego Zoo Global, Lunenburg, Massachusetts, USA. species calls for a reassessment of their conservation 54 pp. status. Journal of Applied Ecology 44:1188–1199. U.S. Fish and Wildlife Service (USFWS). 1997. Searcy, C.A., and H.B. Shaffer. 2014. Field validation Endangered and threatened wildlife and plants; final supports novel niche modeling strategies in a cryptic rule to list the northern population of the Bog Turtle as endangered amphibian. Ecography 37:983–992. threatened and the southern population as threatened Sheldon, A.L., and S.A. Grubbs. 2014. Distributional due to similarity of appearance. Federal Register ecology of a rare, endemic Stonefly. Freshwater 62:59605–59623. Science 33:1119–1126. U.S. Fish and Wildlife Service (USFWS). 2001. Bog Singh, N.J., N.G. Yoccoz, Y.V. Bhatnagar, and J.L. Fox. Turtle (Clemmys muhlenbergii), northern population 2009. Using habitat suitability models to sample rare recovery plan. U.S. Fish and Wildlife Service, Hadley, species in high-altitude ecosystems: a case study with Massachusetts, USA. Tibetan Argali. Biodiversity and Conservation U.S. Fish and Wildlife Service (USFWS). 2006. 18:2893–2908. Guidelines for Bog Turtle Surveys. U.S. Fish and Somers, A.B., J. Mansfield-Jones, and J. Braswell. 2007. Wildlife Service, State College, Pennsylvania, USA. In stream, streamside, and under stream bank Virginia Department of Game and Inland Fisheries movements of a Bog Turtle (Glyptemys muhlenbergii). (VADGIF). 2005. Virginia's Comprehensive Wildlife Chelonian Conservation and Biology 6:286–288. Conservation Strategy. Virginia Department of Game Somers, A.B., and J. Mansfield-Jones. 2008. Role of and Inland Fisheries, Richmond, Virginia, USA. trapping in detection of a small Bog Turtle (Glyptemys Weakley, A.S., and M.P. Schafale. 1994. Non-alluvial muhlenbergii) population. Chelonian Conservation wetlands of the southern blue ridge - diversity in a and Biology 7:149–155. threatened ecosystem. Water, Air, and Soil Pollution South Carolina Department of Natural Resources (SC 77:359–383. DNR). 2005. South Carolina’s Comprehensive Wear, D.N., and P. Bolstad. 1998. Land-use changes in Wildlife Conservation Strategy. South Carolina Southern Appalachian landscapes: spatial analysis and Department of Natural Resources, Columbia, South forecast evaluation. Ecosystems 1:575–594. Carolina, USA. Williams, J.N., C.W. Seo, J. Thorne, J.K. Nelson, S. Stratmann, T.S.M. 2015. Finding the needle and the Erwin, J.M. O'Brien, and M.W. Schwartz. 2009. Using haystack: new insights into locating Bog Turtles species distribution models to predict new occurrences (Glyptemys muhlenbergii) and their habitat in the for rare plants. Diversity and Distributions 15:565– southeastern United States. M.Sc. Thesis, Clemson 576. University, Clemson, South Carolina, USA. 76 p. Yozzo, D.J., and F. Ottman. 2003. New distribution Sutton, W.B., K. Barrett, A.T. Moody, C.S. Loftin, P.G. records for the Spotfin Killifish, Fundulus luciae deMaynadier, and P. Nanjappa. 2015. Changes in (Baird), in the lower Hudson River estuary and climatic niche and climate refugia of conservation adjacent waters. Northeastern Naturalist 10:399–408. priority salamander species in the northeastern United Zappalorti, R.T. 1976a. The Amateur Zoologist's Guide States. Forests 6:1–26. to Turtles and Crocodilians. Stackpole Books, Tesauro, J., and D. Ehrenfeld. 2007. The effects of Harrisburg, Pennsylvania, USA. livestock grazing on the Bog Turtle (Glyptemys (= Zappa1orti, R. T. 1976b. The status of the Bog Turtle Clemmys) muhlenbergii). Herpetologica 63:293–300. (Clemmys muhlenbergii), in North Carolina. Bulletin Tryon, B.W. 1990. Bog Turtles (Clemmys muhlenbergii) of the Maryland Herpetological Society12:69. in the south - a question of survival. Bulletin of the Chicago Herpetological Society 25:57–66.

212

Herpetological Conservation and Biology

Zappalorti, R.T. 1978. The Bog Turtle, smallest of North History Book Distributors, Lanesboro, Minnesota, American Turtles. Bulletin of the Chicago USA. Herpetological Society 13:75–81. Zhao, E., T. Zhou, and P. Ye. 1990. A new Chinese Box Zappalorti, R.T. 1997. Turtles of New Jersey and the Turtle (Testudinata: Emydidae) - Cuora zhoui. Pp. Bog Turtle in the Northeast. Pp. 15–22 In Status and 213–216 In From Water onto Land. Zhao, E. (Ed.). Conservation of Turtles of the Northeastern United China Forestry Press, Beijing, China. States. Tyning, T.F. (Ed.). Serpent’s Tale Natural

THERESA STRATMANN is currently a Ph.D. student at the Senckenberg Biodiversity and Climate Research Centre (BiK-F) and Department of Biological Sciences at the Goethe University in Frankfurt, Germany. This work was part of her undergraduate research at the University of Georgia and her Master’s project at Clemson University. Theresa is interested in how we can combine ecology, math, and modeling to better approach today’s challenging conservation problems. She is a former co-chair of the Southeast Partners in Amphibian and Reptile Conservation (SEPARC). (Photographed by Grover Brown).

KYLE BARRETT is an Ecologist in the Department of Forestry and Environmental Conservation at Clemson University. He is broadly interested in the influence of large-scale anthropogenic disturbances, such as urbanization and climate change, on the diversity and distribution of vertebrates. Kyle enjoys working on projects that have relevance to basic ecology and management/policy decision making. He received his Ph.D. from Auburn University. (Photographed by Shannon Barrett).

THOMAS M. FLOYD is a Wildlife Biologist in the Nongame Conservation Section of the Wildlife Resources Division, Georgia Department of Natural Resources. He received his BSFR from the University of Georgia and his MS from Clemson University (2003). Thomas is involved in the conservation of reptiles and amphibians including species of conservation concern within the State of Georgia, USA. Conservation efforts include population monitoring and species status assessments of rare herpetofauna, and habitat restoration and management including the implementation of prescribed fire. Thomas’s research pursuits include species biogeography, techniques for the survey of elusive species, effects of land management and habitat degradation on herpetofauna, human dimensions in wildlife, and natural resources policy including policy analysis, drafting of laws and regulations for wildlife species conservation, and the development of incentives for private landowners in the conservation of endangered species. (Photographed by Matthew Kaurt).

213