The Influence of Altered Habitat: Landscape Ecology of Freshwater Turtles in Rhode Island

Beginning several centuries ago, European settlement brought great change to the landscape of New England. Vast deforestation, subsequent reforestation, and rapid development that continues to this day have had a profound impact on wildlife populations. Elements of this legacy of landscape change have been studied, but the ways in which past and current landscape patterns affect freshwater turtles remains poorly understood. The primary objective of this research was to better understand the influence of the landscape of Rhode Island on populations of freshwater turtles. It is in many ways a work in landscape ecology, but with components of spatial ecology, habitat selection, and population genetics. Chapter 1 is a stand-alone, opportunistic study of the effects of a small forest clear-cut on a population of spotted turtles (Clemmys guttata). We conducted a radiotelemetry study for one year before, and one year after a 3-ha forest clear-cut in close proximity to wetlands known to contain a resident population of the species. The annual home range size of turtles was 18.5% larger post-cut, possibly due to changes in the distribution of resources and suitable habitat after the harvest. However, turtles exhibited fidelity to hibernacula and communal hibernation despite nearby disturbance, and patterns of activity and habitat use were similar in both years and were generally consistent with those of other spotted turtle populations. Our results suggest that timber harvesting of this spatial scale and management approach may not have any short-term effects on the spatial ecology or habitat use of populations of spotted turtles. It is a strong caveat though, that further research is needed to understand longer-term effects. Chapters 2, 3 and 4 consist of data collected during a state-wide sampling effort of freshwater turtles in small, non-riparian wetlands across a gradient of forest cover. By systematically and intensively sampling 88 randomly selected wetlands across this gradient, we intended to capture the variability in landscape composition and configuration found in Rhode Island and determine how this variability is related to species distribution, abundance, demography, and population genetic structure. In Chapter 2 we report abundances based on capture per unit effort, and use occupancy analysis to determine which among a suite of environmental and within-wetland covariates are driving species occurrence. Eastern painted turtles (Chrysemys p. picta) and snapping turtles (Chelydra serpentina) were widespread (occurring in 83% and 63% of wetlands, respectively), relatively abundant, and exhibited wide niche breadth. Spotted turtles were far less common, occurring in 8% of wetlands, and exhibited a strong association with forested, shallow, natural (i.e., not manmade or heavily modified) wetlands. Non-native red-eared sliders (Trachemys scripta) occurred in 10% of wetlands and exhibited a strong, positive association with road density, likely as a function of human population density. In Chapter 3 we further examine eastern painted turtle abundance, demography, and the relationship between sex ratio and road density. There was no difference in abundance or any demographic trait between natural and manmade wetlands. A negative relationship between abundance and forest cover surrounding wetlands emerged as the best model, but explained very little variation. Contrary to expectations, there was a significant, but weak relationship between increasing road density and the proportion of females in a population. Collectively, these results suggest that eastern painted turtles are exhibiting little to no detectable variation in population demography across the range of landscapes found in Rhode Island and are resilient in the face of human-induced landscape change. Finally, in Chapter 4 we used microsatellite markers to compare the population genetic structure between the common and widespread eastern painted turtle, and the rare and more specialized spotted turtle. Due to their relative rarity and smaller populations, we predicted that spotted turtles were more likely to have experienced the detrimental effects of habitat loss and fragmentation associated with landscape change, and that these effects would manifest in the form of more inbreeding, reduced genetic diversity, and greater population genetic structure. As expected, eastern painted turtles exhibited little population genetic structure, showing no evidence of inbreeding or strong differentiation among sampling sites. For spotted turtles however, results were consistent with certain predictions and inconsistent with others. We found tentative evidence of recent population declines in spotted turtles, as well as a greater degree of inbreeding in the species when compared to eastern painted turtles. Genetic diversity and differentiation among sites were comparable between species, however. As our results do not suggest any major signals of genetic degradation in C. guttata, the southern region of Rhode Island may serve as a regional conservation reserve network where the maintenance of population viability and connectivity is prioritized. Globally, turtles are among the most threatened of vertebrate taxa. Information on how populations respond to human-induced landscape change has important implications for conservation. The work herein will provide wildlife biologists with a better understanding of the current state of populations of freshwater turtles in the state and region, and allow for more informed management decisions.


Introduction
Habitat alteration can be an important component of wildlife management (Russell et al. 1999;Degraaf et al. 2006). The maintenance or creation of early successional habitat via mowing, prescribed burns, and clear-cuts is commonly employed by natural resource managers to benefit native wildlife (Greenberg et al. 1994, Van Dyke et al. 2004), including some birds (Degraaf and Yamasaki 2003), mammals (Litvaitis 2001;Fuller and DeStefano 2003), and reptiles (Dovĉiak et al. 2013). In southern New England of the United States, the abandonment of agricultural fields that occurred in the first half of the 20 th Century led to an increase in early successional habitat. The gradual process of forest succession that followed however, has greatly reduced the amount of early successional habitat on the landscape Buffum et al. 2011). State wildlife agencies and conservation groups have made the creation of early successional habitat a priority in the region because of its benefits to many species of wildlife including shrubland birds and particularly to the New England Cottontail (Sylvilagus transitionalis; Schlossberg and King 2007;Buffum et al. 2014;Fuller and Tur 2015).
However, questions remain regarding the effects of early successional habitat creation on certain species, especially those that are associated with mature, forested habitats.
Although several studies have reported impacts of timber harvesting on reptiles (Enge and Marion 1986; Todd and Andrews 2008; Moorman et al. 2011), including turtles (Currylow et al. 2012), to our knowledge none have focused on how freshwater turtles respond to forest clear-cutting. This may be less important for highly aquatic turtles that make only occasional upland movements, for example, to an open area to nest. However, some freshwater turtle species, including the Spotted Turtle (Clemmys guttata) and the Wood Turtle (Glyptemys insculpta), move frequently between ephemeral and permanent wetlands and are known to estivate terrestrially, with some Spotted Turtles spending as much as 30% of their time on land  and Wood Turtles as much as 40% of their time (Arvisais et al. 2004). Use of upland habitats by some forest and wetland-associated turtle species may make them vulnerable to forest alteration if habitat is destroyed or fragmented. Alternatively, the removal of the forest canopy for the creation of early successional habitat may create new microhabitats suitable for thermoregulation and nesting.
The Spotted Turtle is a species of increasing conservation concern. The International Union for the Conservation of Nature (IUCN) reviewed the species in 2013 and upgraded its status from Vulnerable to Endangered (van Dijk 2013). In five of the six New England states where it occurs, the Spotted Turtle has been designated with some type of conservation protection and the status of the species is currently under review by the U.S. Fish and Wildlife Service (USFWS) for federal listing under the U.S. Endangered Species Act (USFWS 2015). Spotted Turtles are relatively small (carapace length up to 142.5 mm) freshwater turtles that are native to the eastern United States and Great Lakes regions of North America. They occur in a variety of wetland types throughout their range and have sometimes been described as habitat generalists ). However, Spotted Turtles have also been shown to exhibit strong habitat selection based on the physical and biological conditions of their environment Anthonysamy et al. 2014). This selection is detectable at multiple spatial scales and can vary with season and by sex (Litzgus and Mousseau 2004;. Spotted Turtles are often described as semi-aquatic 5 because they use both wetland and upland habitats. They spend the majority of their time in wetlands and depend on these habitats for overwintering, foraging, thermoregulation, and mating Ernst and Lovitch 2009). Most individuals exhibit high fidelity to wetlands, often overwintering in the same hibernaculum each year . Spotted Turtles use uplands for nesting and moving between wetlands, and both sexes spend extended periods of time in upland habitat estivating in shallow forms or underneath leaf litter during the warmest periods of the summer Gibbs et al. 2007). Thus, uplands are essential to this species and concern is raised when these habitats are to undergo significant alteration. In Rhode Island, USA, Spotted Turtles are a strongly forest-associated species (Chapter 2), but the implications for the removal of forest surrounding wetlands where they occur is unknown.
We investigated the potential impacts of a clear-cut timber harvest that took place within close proximity to a complex of wetlands in southern Rhode Island that is known to contain a population of Spotted Turtles. We radio-tagged individuals in this population for one year prior to, and one year after, a clear-cut that was implemented to create early successional habitat for wildlife. Our objectives were to examine the effects of forest clear-cutting on Spotted Turtle spatial ecology, activity, and habitat use.

Materials and Methods
Study site.-Our study took place in Washington County, Rhode Island, USA.
We have withheld specifics of the location out of concern for making this population of Beginning in December of 2013 and concluding in February of 2014, while turtles were inactive in aquatic hibernacula, approximately 3 ha of mature forest was harvested to create early successional habitat using a Clear-cut with Reserves approach (Miller et al. 2006). The cut retained approximately eight residual trees per hectare to serve as seed trees and sources of food for wildlife. Large amounts of coarse woody debris were left on the ground to reduce deer browse and six large brush piles were created for wildlife habitat. No herbicides were applied after the cut and no rutting or erosion was observed 7 after the cut. The shape of the cut was irregular and a buffer of at least 15.2 m (50 feet) was retained around all wetland habitat ( Fig. 1.1 and 1

.2).
Radiotelemetry and data collection.-We captured Spotted Turtles using baited hoop traps and by hand. We attached RI-2B 6g radio transmitters (Holohil Systems Ltd., Carp, Ontario, Canada) with waterproof putty epoxy to the right-posterior of the carapace. The combined mass of transmitter and epoxy averaged approximately 6% of body mass and did not exceed 8%. Following transmitter attachment, we released all individuals at their original points of capture. We used an ATS R410 receiver (Advanced Telemetry Systems, Isanti, Minnesota, USA) and a three-point Yagi antenna to track turtles. We recorded geographic coordinates (Universal Transverse Mercator; North American Datum of 1983) for each turtle radio-location using a Garmin Oregon 450 handheld global positioning system receiver (Garmin International Inc., Olathe, Kansas, USA). We conducted radio-telemetry for one season before (2013) and one season after (2014) the implementation of the clear-cut. We radio-tracked turtles approximately once every 5 d (mean = 5.35 ± 0.11 [SE] d, n = 655 intervals) between 1 April and 31 October, and less frequently in the early spring and late fall. We classified radio-locations into one of three categories based on the precision of the detection of the turtle. If we found a turtle and actually saw it, we classified the radio-location as Visual. If we obtained a signal and identified the location to a small area (a few square meters) without use of the telemetry antenna (i.e., using just the receiver), we classified the radio-location as Exact.
If we obtained a signal and we estimated the location using the telemetry antenna, we considered the radio-location as Approximate, in which case we used triangulation to confirm that turtles were within wetlands. 8 We measured midline carapace length (mm) using analog calipers and we measured initial body weight (g) using a digital scale. We obtained daily maximum temperature (° C) and precipitation (mm) data for 1 April to 31 October in both years from a representative weather station (Kingston, Rhode Island; NOAA, National Centers for Environmental Information. op. cit.). We used these data to obtain annual means (for temperature) or sums (for precipitation) and we determined averages to compile weekly means over the course of the activity season. We conducted an initial forest inventory of the clear-cut area in October 2013 after the clear-cut area had been delineated but before logging operations began, and a second inventory after the logging was complete in October 2014. In both cases we assessed the vegetation at 56 locations along parallel transects spaced equal distances apart. We used 2 m 2 fixed area plots to record frequency of occurrence of understory vegetation, and variable area plots to measure diameter at breast height and density of overstory vegetation. We measured overstory tree canopy cover at each point using a spherical densiometer.
Home range and habitat use.-We categorized all turtle radio-locations as occurring in either wetland or upland habitat. We calculated percentage wetland use by dividing the number of radio-locations that occurred in a wetland by the total number of radio-locations. For all upland radio-locations, we calculated the distance to the nearest wetland edge. The lack of consistent, precise radio-locations (particularly when turtles were in wetlands) made it impossible to calculate distance between radio-locations throughout the activity season, but did not preclude the calculation of home range size estimates. We estimated home range sizes using minimum convex polygons (MCPs).
MCPs are widely used in home range analyses of reptiles and have been used in multiple 9 studies of Spotted Turtles making them the most useful for comparison with other studies (Litzgus and Mousseau 2004;Row and Blouin-Demers 2006). We included Approximate radio-locations in the construction of MCPs, as these were the majority of locations because many turtles were located in the interior of a wetland and their precise location could not be determined. The majority of these points fell within the interiors of constructed polygons and did not influence MCP size. We also inspected all radiolocations for each turtle and manually removed points from the home range analysis that were ambiguous or erroneous due to transcription errors (n = 7 points).
We calculated overall home range size and overall percentage wetland use by combining all available data from both years. In addition, to examine both home range fidelity and potential differences pre-and post-clear-cut, we calculated annual home range size and annual percentage wetland use in 2013 and 2014 and compared these data between years. To maximize the comparability of these metrics between years, we also calculated constrained post-clear-cut values by constricting the radio-locations used to the range of dates when turtles were tracked in both years. We estimated annual home range percentage overlap between years to compare potential changes in resource use overall and between sexes. For all turtles tracked in both years, we divided the common area of both MCP polygons (one for each year) by the total merged area of both polygons. We used all available radio-locations to estimate annual home range overlap. We also determined all instances in which an individual used any of the area inside the delineation of the clear-cut, in a given year, by identifying all the instances in which an annual MCP (constrained MCP for post-cut) overlapped the area of the clear-cut.
Statistical analyses.-We assessed normality using Shapiro-Wilk tests and equality of variances using Levene's tests. All data were normally distributed and homoscedastic. We used paired t-tests to compare home range sizes and percentage use of wetlands pre-and post-clear-cut. We used an independent samples t-test to determine if home range sizes differed by sex pre-and post-clear-cut, using the difference between pre-and post-clear-cut MCP as the dependent variable. The paired t-tests and the independent samples t-test used observations only from individuals tracked in both years (n = 9), and the post clear-cut observations were constrained to the dates when turtles were tracked in the previous year. We compared overall home range size and annual home range percentage overlap between males and females using independent samples ttests. We used linear regression to examine the relationships of body size (midline carapace length) and the number of radio-locations, and of body size and overall home range size. We compared overall percentage of wetland use between sexes with an independent samples t-test. For descriptive statistics, we report means ± one standard error (SE), and we defined statistical significance as P ≤ 0.05. We calculated MCPs and distance to nearest wetland using Geospatial Modeling Environment (version 0.7.3.0, www.spatialecology.com/gme, Accessed 15 January 2013) and ArcGIS 10.2. All other statistical analyses were performed using R (R Core Team 2013).

Results
Radiotelemetry and data collection.-We tracked 12 turtles over the 2 y (six females, six males), nine of which (four females, five males) were tracked in both years ( (143/712) of radio-locations, and estimated locations using triangulation (Approximate) for 56% of radio-locations. Approximate radio-locations occurred almost exclusively when turtles were in interior sections of a wetland.
Turtles moved from hibernacula in mid-to late-March and appeared to congregate in nearby vernal pools. Seven of 12 (58 %) tracked turtles were found in the same small vernal pool (~0.05 ha) in the same two-week period of May 2014 and as many as five turtles were found in the vernal pool on the same day. Annual home range overlapped with the clear-cut delineation in 4 of 9 (44.4%) instances in 2013, and with the clear-cut Turtles hibernated exclusively in wetlands. Several individuals exhibited fidelity 13 to hibernacula and we observed the use of communal hibernacula in both years (Table   1.2). Of the six instances in which individuals were tracked to hibernacula in both years, four individuals (67 %) used the same hibernaculum. Another individual hibernated in different locations within the same wetland. All turtles occupied hibernacula by 12 November in 2013, and by 28 October in 2014. Turtles remained in the uplands as late as 31 October in 2013, and as late as 14 October in 2014. Hibernation sites were all associated with Sphagnum hummocks and/or the roots of woody shrubs. An untracked Spotted Turtle was found dead in the adjacent mowed field on 28 October 2014, suggesting use of the field at some time of the year. The turtle was decomposed, so it was not clear how long the turtle had been dead, but the shell remnants were found in many pieces suggesting that it had been crushed.
From 1 April to 31 October, mean daily maximum temperature was 21.5° C (range = 5.8-33.4° C) in 2013 and 21.0°C (range = 6.9-29.7° C) in 2014. Total  (Table 1.3). Average canopy cover of the area was 76 % in 2013, and 35 % in 2014. Clear-cut border trees and a few remaining seed trees contributed to post-clear-cut estimates of canopy cover.

Discussion
Home range and movements.-The duration and timing of the activity season was consistent with other observations of Spotted Turtles at the northern portion of their range (Haxton and Berrill 2001;; but see . Surface activity began in mid-to late-March and ceased in late October or early November, after 14 which turtles entered wetland hibernacula. Overlap of annual MCPs with the delineation of the clear-cut in both years suggests that turtles used the area both before and after the cut took place. Spotted turtle home range size was nearly 20% larger post-clear-cut, but lack of a statistical difference precludes a clear interpretation of this result, particularly given our relatively small sample size. Habitat alteration can cause wildlife to travel greater distances to locate necessary resources, which for turtles may include food items, mates, thermoregulatory habitat, nesting habitat, and overwintering habitat. However, the creation of early-successional habitat (such as a clear-cut) could also create new opportunities for thermoregulation and nesting, thereby reducing the distance required to locate these habitat types. Open areas including power line rights of way and recent clear-cuts have been used by Spotted Turtles for nesting (Litzgus and Mousseau 2004).
Whether a habitat alteration serves to expand or reduce home range size probably depends on the proximity of the alteration to established home ranges as well as the nature of the alteration itself. Spotted Turtle home range size increased after disturbance in the form of flooding by Beaver (Castor canadensis) dams, but probably because the turtles were using newly available aquatic habitat (Yagi and Litzgus 2012); the flooding was interpreted as beneficial to this population of Spotted Turtles in Ontario.
We detected a difference in annual home range overlap between sexes. Male turtles exhibited greater overlap between years, suggesting a higher fidelity to specific sites. If males can reliably locate females for mating during early spring congregations, the additional distances a male must travel are potentially limited to those where it can find food, thermoregulatory habitat (e.g., for basking and estivation) and hibernation habitat. In addition to these types of movements, females must also locate nesting 15 habitat. As a proportion of female Spotted Turtles in a population do not breed every year , differences in reproductive condition between years may explain the observed differences in annual home range overlap in females. Alternatively, the clear-cut may have influenced female movements by altering habitat selection. The clear-cut could have created new areas that had potential to serve as nesting and thermoregulatory habitat. Females may have moved greater distances while seeking out these newly available sites. Determining the proximate effects of a given habitat alteration is difficult. Our inference is limited in this case due to insufficient information (e.g., reproductive condition of females), the lack of additional treatment and control sites, and the fact that our data are limited to one year before, and one year after the clear-cut.
Spotted Turtles exhibited smaller home range sizes at our study site in Rhode Island than those from populations of Spotted Turtles in Massachusetts , South Carolina (Litzgus and Mousseau 2004), and Ontario (Rasmussen and Litzgus 2010), but were larger or comparable to those of other studies (Ernst 1970;Wilson 1994;Graham 1995). Differences in home range size among studies are usually attributed to distribution and density of resources (i.e., food items, critical habitat, and mates) on the landscape. Intermediate home range sizes suggest a moderate density of resources at our study site. Males and females exhibited similar overall home range size.
In turtles, males generally engage in larger movements during the mating season to locate mates, and females exhibit larger movements during the nesting season to locate nest sites (Morreale et al. 1984;Parker 1984). Movements of Spotted Turtles do not always follow this pattern, though. Early season congregations in Spotted Turtles appear to be common (Ernst 1967; and likely take place for breeding purposes (Litzgus and Mousseau 2004), thus limiting the distance that males must travel to actively search for mates. Larger home range sizes were observed for gravid females in South Carolina (Litzgus and Mousseau 2004), and results of other studies support the idea that gravid females exhibit larger home ranges because they must find appropriate nesting habitat but see Rasmussen and Litzgus 2010). The fact that we did not observe a difference in home range size between sexes may be due to an absence of gravid females, or the fact that appropriate nesting habitat existed in close proximity to wetlands used throughout the activity season.
We suspect that, among populations, the location and configuration of appropriate nesting habitat plays a large role in the home range sizes of females.
Habitat use.-Turtles used wetlands with much greater frequency than uplands.
Most likely, the majority of observations of upland use were associated with summer estivation, possibly influenced by water levels in ephemeral wetlands . Vernal pools in the area dry in late June through late July, and increased use of upland areas may reflect decreases in available wetland area. Overall wetland use was consistent between years, but the shift from wetland use to persistent use of uplands occurred about three weeks later in 2014. Total precipitation was greater in 2013 though, and data from a different study confirms that 2014 was a drier year in small wetlands state-wide (Scott Buchanan, unpubl. data). Thus, the timing of wetland drying does not explain the difference in timing of upland use between years, which remains unexplained. Future studies should investigate what factors influence the shift between wetland use and upland use for this species.
Upland areas surrounding wetlands, often termed buffer zones or core terrestrial habitat, are important for ensuring the protection of wetland fauna that use both habitat types. Use of upland areas appears to be variable among populations of Spotted Turtles.
In 12 instances (approximately 11 % of upland radio-locations), turtles in our study were found in upland areas beyond the protected buffer of 50 ft (15.2 m) required for 'perimeter wetlands' (pond area > 0.10 ha [0.25 ac] and standing water for ≥ 6 mo/y) under the Rhode Island Fresh Water Wetlands Act (Rhode Island Department of Environmental Management 1998). In addition, there were many instances in which individuals moved from one wetland to another, and in doing so used upland habitat outside of the regulatory buffer zone. In our study population, current RI regulations would not be adequate to ensure that upland habitat used by Spotted Turtles was protected from development projects or other activities that would result in the destruction or fragmentation of upland habitat. In Massachusetts, > 90% of Spotted Turtles nested or estivated outside the 30 and 60 m upland buffer zones (for palustrine and permanently flowing wetlands, respectively) stipulated by Massachusetts wetlands regulations at the time of study . In Ontario, one population of Spotted Turtles nested between 2-139 m from a wetland .
In contrast, individuals in another population in Ontario were described as rarely observed farther than 2 m from a wetland except in instances of nesting or movements between areas ; the study did not quantify these distances. A review of aquatic turtle nesting data estimated that a core area of 127 m surrounding wetlands would be required to protect 95% of Spotted Turtle nests ).
Our results and those of other studies of Spotted Turtle habitat use suggest that protection 18 of upland habitat around wetlands is important to ensure that habitat used for nesting, thermoregulation, and movement between sites is not compromised.
Spotted turtles hibernate in wetlands, hibernate communally, and show fidelity to overwintering sites . Most (66%) of the individuals tracked to hibernacula in both years exhibited fidelity to hibernacula. This level of fidelity is comparable to other studies of Spotted Turtles at undisturbed sites in Ontario , and suggests that turtles were able to navigate to and from specific wetlands, even after the dramatic alterations to our study site associated with the clear-cut. Wetland habitat is critical to this species and, from the perspective of conservation, the protection of wetlands containing Spotted Turtle hibernacula is of preeminent importance. question. Nonetheless, our data suggest that timber harvesting of this intensity (i.e., percent of forest removal and management practices carried out) and spatial scale may be compatible with maintaining populations of Spotted Turtles, even when the harvest takes place in close proximity to wetlands where the species occurs. However, the spatial configuration of the clear-cut relative to wetland habitat is probably an important factor to consider. Although the clear-cut did come very close to several wetlands containing Spotted Turtles, the continuity of forest north of the rights of way (where turtles spent the majority of their time) remained largely intact (see Figure 1.1) and no wetlands were completely fragmented. A larger cut or a cut that completely fragmented individual wetlands may have had a more dramatic effect on turtle movements. In addition, the availability of longer-hydroperiod wetlands at our study site may have ameliorated some of the effects of the clear-cut. The study site contains several vernal pools, which dry nearly every year, and one permanent wetland on the site and another just off-site.
Permanent wetlands in the area of the study provide refugia for turtles as vernal pools dry, probably reducing the need for long-term estivation in upland sites, as has been documented in other populations (Litzgus and Brooks 2000;). Thus, a clear-cut similar to this one is probably less likely to impact Spotted Turtle populations where turtles are able to move from ephemeral into permanent wetlands during the hottest and driest parts of the activity season.
To our knowledge, this is the first study to investigate responses of Spotted Turtles to creation of early successional habitat. Given that more than 3,300 ha of early successional habitat was created for New England Cottontail in six northeast states in (Fuller and Tur 2015, we are encouraged that we did not detect major 20 impacts of this activity on the turtle population in our study. However, we strongly recommend that the spatial arrangement and hydroperiods of wetlands near a proposed clear-cut area be investigated prior to commencement of operations and that the entire harvesting process take place during months when turtles remain in or near wetland hibernacula. In the Northeast, this would generally be between mid-November and early March, but may vary depending on weather conditions in a given year. Additionally, care must be given to avoid any significant disturbance to wetlands that contain Spotted Turtles at any point in the year, especially those containing hibernacula. Spotted Turtles are a species of increasing conservation concern. Habitat destruction and modification, vehicular mortality (i.e., automobiles and agricultural equipment), and personal and commercial collection are considered the greatest threats to the species van Dijk 2013). An improved understanding of how early successional habitat creation affects populations of Spotted Turtles will allow resource managers to identify instances in which the implementation of the practice is consistent with the site-specific conservation goals for the species. Clemmys guttata, a species of special concern in Massachusetts. Chelonian Conservation Biology 1:207-214.

Abstract
Turtles are one of the most threatened groups of vertebrates worldwide. In the northeastern United States, a legacy of centuries of dramatic landscape alteration has impacted freshwater turtle populations, but the relationships between the current landscape and distributions and abundances of freshwater turtles remain poorly understood. We used a stratified random approach to select small, isolated wetlands across a gradient of forest cover throughout Rhode Island and systematically sampled freshwater turtles in these wetlands. We performed occupancy analysis to determine which environmental variables drive the occurrence and probability of detection of different species. We report naïve estimates of abundance and generated estimates of niche breadth for each species and partitioning among species. Eastern painted turtles (Chrysemys p. picta) and snapping turtles (Chelydra serpentina) were widespread (occurring in 83% and 63% of wetlands, respectively), relatively abundant, and exhibited wide niche breadth. Spotted turtles (Clemmys guttata) were far less common, occurring in 8% of wetlands, and exhibited a strong association with forested, shallow, natural (i.e., not manmade or heavily modified) wetlands. Non-native red-eared sliders occurred in 10% of wetlands and exhibited a strong, positive association with road density, likely as a function of human population density. Identifying landscape-scale habitat features that are associated with the occurrence of sensitive species can improve the ability of biologists to identify and protect populations of that species.

Introduction
Human-induced landscape alteration is often implicated in compromising vertebrate biodiversity, with habitat loss and degradation widely recognized as the leading contributor to a loss of population stability across taxa Brooks et al. 2002;Johnson et al. 2011). New England, in the northeastern United States, has experienced dramatic shifts in landscape composition since the time of European settlement. Deforestation associated with agriculture and logging peaked in the midnineteenth century when as much as 80% of the landscape had been cleared. Beginning around 1850 agriculture shifted to states farther west, ushering in a period of reforestation lasting approximately 100 years . In Rhode Island, a small state in southern New England, this period was followed by another phase of deforestation for urban and suburban development. Total forested land area in Rhode Island has been decreasing since at least 1953 (RIDEM 2010; Butler and Payton 2011; but see Butler 2013) and was recently estimated as 147,000 ha, approximately 54% of the total land area of the state . This extreme landscape alteration in a relatively short period of time has certainly led to changes in the distribution and abundance of wildlife, but the legacy of this change is poorly understood for many species, including freshwater turtles.
As a vertebrate group, turtles have an extremely high risk of population extirpation and extinction (Bohm et al. 2013). In the United States, freshwater turtles are of particular conservation concern largely due to pervasive wetland draining and filling that has resulted in a significant loss in wetland area beginning in the eighteenth century (Dahl , 2000. Additional factors putting freshwater turtle populations at risk include the loss of meta-population structure associated with terrestrial habitat loss and 38 degradation (Dodd 1990;Gibbs 2000), collection for pet, food, and medicine trades (Shiping et al. 2006;Luiselli et al. 2016), and life history characteristics that include delayed sexual maturity and low recruitment (Congdon et al. 1993;Congdon et al. 1994;Heppell 1998) All freshwater turtle species use terrestrial habitats to some extent, but the proportion of time spent on land varies. Freshwater turtles use uplands to nest and to move between wetlands, and some species spend substantial periods of time estivating in uplands . Spotted turtles are known to move frequently between temporary and permanent wetlands and to estivate terrestrially, spending as much as 30% of their time on land . The landscape adjacent to and between wetlands is directly linked to many ecological processes of freshwater turtles (Bodie and Semlitsch 2000; and occupancy parameters that can be modeled with environmental covariates. We conducted a three-year field study into the associations between freshwater turtles and the landscape by sampling small (0.1 -1.8 ha) wetlands along a gradient of forest-cover. We targeted our sampling scheme on a subset of wetland types that could potentially contain spotted turtles. We used single-species, single-season occupancy models to elucidate relationships between landscape-and wetland-scale variables, and the occurrence of four species of freshwater turtles in Rhode Island. Our intent was to: (1) characterize the distribution and abundance of freshwater turtles across an urban gradient while testing the prediction that spotted turtles are a forest-associated species, (2) determine what landscape-and wetland-scale features and conditions freshwater turtles are selecting, and (3) improve our understanding of the conservation implications of landscape management for these species, especially spotted turtles.

Study area and species
Our study was conducted throughout the state of Rhode Island (

Site selection
We selected sites using a stratified random design to capture the state-wide variability in landscape composition and configuration. To minimize confounding factors among sites, we focused our sampling on relatively small (0.1-1.8 ha; as measured via GIS polygons), isolated (i.e., discrete, non-riparian) wetlands. The minimum size was selected to ensure that the majority of wetlands had a hydroperiod that persisted throughout the turtle activity season in most years (Skidds and Golet 2005 We grouped retained wetlands as either small (0.1-0.4 ha) or large (< 0.4-1.8 ha) wetlands. The 0.4 ha breakpoint was the approximate median of the distribution of wetland size for all retained wetlands. We calculated percent forest cover within buffers of 300 m and 1 km from the wetland edge of all retained wetlands. At the 300-m scale, we grouped wetlands into eight 10% increments of forest cover (excluding 0-10% and 70-80%), and at the 1 km scale we grouped wetlands into four, partially overlapping, larger increments of forest cover (0-40%, 20-60%, 40-80%, 80-100%). These groups created a near-continuous gradient of sites from different forest cover conditions which captured the state-wide variation in landscape conditions. We assigned each retained wetland a random number, sorted them by random number in ascending order, and contacted property owners/managers in that order until we received permission to sample the desired number of wetlands in each stratification, with approximately equal numbers of small and large wetlands. We carried out this process in each of three consecutive years.

43
Turtle sampling and data collection We carried out turtle sampling from May to October in 2013-2015 (Appendix 1).

Statistical analysis
We made naïve estimates of abundance by calculating the total number of unique individuals caught divided by the total number of trap nights, for each forest cover class.
Abundance estimates were compared only for common snapping turtles (hereafter, snapping turtles), eastern painted turtles (hereafter, painted turtles), and spotted turtles because of low sample sizes for the other species.
We used principal components analysis (PCA) to summarize relationships between presence of freshwater turtle species in wetlands and environmental covariates by reducing the dimensionality of our covariate dataset. We were primarily interested in using PCA as an exploratory technique to identify potential differences in explained variation among species (Everitt and Hothorn 2011). We built a data matrix of all sitespecific covariates (Table 2.1; excluding geographic location) consisting of all instances in which a species was detected at a wetland, for each species (i.e., if two species were detected at the same wetland those data were entered twice in the matrix). Data were scaled and principal components were extracted from this correlation matrix using the 'stats' package in R (R Core Team). We constructed a graphical representation of the first two components using the R package 'ggbiplot.' Ellipses were drawn around mean values for each species encompassing one standard deviation of the variation along each axis.
We modeled heterogeneous detection probabilities using covariates that changed between surveys, including Julian date (day two of survey), survey number, temperature, and precipitation ( where U covariates are associated with site i and the U + 1 coefficients to be estimated (i.e., β0 and U regression coefficients for each covariate). Using the same principles, the probability of detection can also be modeled as a function of covariates at site i during survey j, expressed as: where xi1 ,…, xiU represent the U site-specific covariates associated with site i, and yij1 ,…., yijv are the V survey-specific covariates associated with survey j of site i (MacKenzie et al. 2006). We used a simulated annealing optimization process for all models. We used the R package 'MuMIn' to carry out model selection procedures and used the Bayesian Information Criterion (BIC) to select supported models from sets of candidate models . Models with the lowest BIC score and fewest number of parameters within 2 BIC units of the lowest BIC score were considered most supported.
All covariates were treated as continuous data and were standardized to a mean of zero and standard deviation of one prior to modeling (MacKenzie et al. 2006).
We conducted the following procedure for each species. We first modeled the probability of detection by keeping the occupancy parameter constant and allowing detection to vary as a function of the survey-specific covariates. For each covariate, we considered both a linear and quadratic functional form when building models. For model selection, we considered all subsets and used BIC to identify the most supported model.
We retained the most supported model to serve as the detection parameter for all subsequent models for that species. Next, to model the probability of occupancy, we built an 'initial' additive global model consisting of the retained detection parameter and linear terms for each site-specific covariate (for landscape covariates these included only the 300-m scale). We considered all subsets and identified the most supported models using BIC. When subsetting, we limited the number of occupancy parameters (excluding the intercept) in any one model to five to limit the ratio of parameters to sample size. We

Results
We sampled a total of 88 wetlands over three years ( Fig. 2.1 Painted turtle abundance was highest at the lowest forest cover class and generally decreased with increasing forest cover. Spotted turtle abundance was substantially higher 49 in the highest forest cover class and only one individual was detected below the 60-70% forest cover class. Snapping turtle abundance exhibited relatively minor variation across most of the gradient of forest cover ( Fig. 2.2). Non-native red-eared sliders did not occur in cover classes >50-60% forest cover.
We retained the first four principal component axes based on a scree plot of component variances (Cattell 1966;Everitt and Hothorn 2011). Collectively, these accounted for 55.6% of the variation in our data (Appendix 4). We modeled occupancy for four species of freshwater turtles. We did not consider musk turtle occupancy as detection probability fell below 5% (Cunningham and Lindermayer 2005;MacKenzie et al. 2006). One wetland, which yielded no turtle detections, was not included in occupancy models because of incomplete covariate data.
There was evidence for lack of fit (p < 0.05) and overdispersion ( ̂ > 1) in the top model for painted turtles, but all top models for other species exhibited evidence of model fit (p 50 > 0.05; Table 2

Discussion
Spotted turtles and red-eared sliders were encountered far less frequently than painted turtles and snapping turtles. The fact that the introduced red-eared slider was found in a greater number of wetlands than the native spotted turtle is concerning.
Principal component ellipses for these two species exhibited minimal overlap demonstrating strong differences in the types of habitats where they are found. There was 51 strong evidence of an association between spotted turtles and forest cover. Spotted turtles were completely absent, except for a single individual, from wetlands surrounded by less than 60% forest cover, and abundance increased dramatically in wetlands with 90-100% forest cover. Similarly, the top spotted turtle occupancy model indicated a positive relationship with forest cover at the 1-km scale. The relatively low state-wide occupancy rate of spotted turtles is consistent with the idea that populations of this species are rare and that they are disproportionately affected by human disturbance (Enneson and Litzgus 2008; Anthonysamy et al. 2014). Spotted turtles are vulnerable to a variety of human impacts including habitat loss and fragmentation, road mortality, and collection (Ernst andLovich 2009, van Dijk 2011). Forest cover at the 1-km scale was negatively correlated with road density (Pearson r = -0.889) and development (Pearson r = -0.901), indicating that human disturbances are generally reduced in areas of higher forest cover.
Furthermore, all wetlands in which spotted turtles were detected belonged to the oldest age class (pre-1939), wetlands that are less likely to have been created or significantly altered by people. Our occupancy models also indicated that spotted turtles prefer shallow wetlands with abundant woody vegetation, results that are consistent with other studies of spotted turtle habitat selection .
In southern New England, forest succession has greatly reduced the amount of early successional habitat on the landscape Aber 2004, Buffum et al. 2011).
The creation and maintenance of early successional habitat, primarily via clear-cutting and fire, is a management priority in the region (Buffum et al. 2014 (Christens and Bider 1987;Janzen 1994;Kolbe and Janzen 2002) and it is likely that nesting habitat becomes more limited with increasing forest cover ). Interestingly, wetlands on or immediately adjacent to golf courses, a highly altered landscape, produced some of the highest abundance estimates for painted turtles. Four of our sites occurred on golf courses and three of these were ranked in the top 12.5% of sites for painted turtle abundance. Recent studies have indicated that golf courses provide potentially important habitat for painted turtles (Failey et al. 2007;Foley et al. 2012) and that painted turtle abundance on golf courses is comparable to that in agricultural and conservation areas . In North Carolina, freshwater turtle species richness was higher in golf course wetlands than in urban or rural wetlands, and the researchers concluded that the maintenance of green space connectivity (including golf courses) would be beneficial for freshwater turtle diversity in urban areas (Guzy et al. 54 2013). Additionally, connectedness of green spaces within 500 m of wetlands had no effect on painted turtle occupancy, but very high painted turtle occupancy rates may have precluded the detection of any relationship between occupancy and the covariates used in the models  Illinois, connectivity to other wetlands increased the probability of both occupancy and colonization of wetlands by painted turtles . Heavily modified habitat types (i.e., urban, suburban and agriculture) in southern New England may be beneficial to painted turtles, even at the extreme end of the gradient, by providing enhanced nesting habitat, basking habitat, and increased aquatic plant production resulting from nutrient runoff (Brinson et al. 1981, Marchand and. There was no apparent trend in snapping turtle abundance across the gradient of forest cover, but unlike painted turtles, snapping turtle abundance decreased in the lowest forest cover class. Snapping turtles are also widespread and considered capable of occupying almost every kind of freshwater habitat . In Indiana, snapping turtle abundance was greatest in impoundments and marshes and was not strongly affected by fragmentation . In North Carolina however, snapping turtle occupancy increased with connectedness of green spaces . In Rhode Island, there is a strong negative relationship between forest cover and road density (Pearson r = -0.889 at 1-km scale). Roads pose a substantial threat to freshwater turtles due to vehicle strikes and this mortality has the potential to disproportionately kill nesting females, resulting in male-skewed sex ratios Aresco 2005;. Larger body size is thought to increase risk of road mortality in turtles, especially in the Northeast given above average traffic density (Gibbs and Shriver 2002). The sharp decline in snapping turtle abundance observed in the lowest forest cover class may come as a result of increased risk of road mortality due to their larger body size. 56 We found that probability of snapping turtle occupancy increased with nitrate levels. Sources of nitrates include fertilizers, human waste, and industrial pollution (Bouchard et al. 1992). Elevated levels of nitrates increase plant and algae production, which can lead to anoxic conditions after this material decays. Snapping turtles and painted turtles are highly tolerant of anoxic conditions when overwintering (Ultsch 2006 We did not model naïve abundance with environmental covariates because these estimates of abundance were associated with a high degree of variation. Precise estimates of abundances of aquatic turtles are considered very difficult to obtain, without longer term mark-recapture studies, due to inherent variation in catchability and observability (Dorland et al. 2014). Moreover, when sampled sites include non-permanent wetlands it can be difficult to define the meaning of an abundance estimate even in the context of a meta-population. Although we marked individuals, recapture rates for most species (except for painted turtles) were too low to yield estimates of abundance via markrecapture modeling, particularly because each wetland was sampled for only one season.
Nonetheless, we report naïve abundance estimates for descriptive purposes and to compare to other studies. Occupancy modeling is more robust to these issues and can be interpreted in the context of presence/absence and habitat selection. Although the utility of occupancy modeling is limited in that it does not permit estimation of important population parameters such as density, survival, or recruitment, the technique contributes to knowledge of distribution and allows for the identification of habitat features associated with a particular species, when multiple species are compared (Nielsen et al.

2010
). It is possible that there was some violation of the closure assumption of occupancy modeling, but as we sampled each wetland for only one season that concern is minimized.
Identifying habitat features at the landscape scale that are associated with species occurrence is a common goal in landscape ecology. Doing so can improve the ability of biologists to predict where sensitive species occur within a state or region and inform management decisions for those species. Amassing herpetological occurrence records, through herpetological atlases or natural heritage programs, is a priority among state biologists in the Northeast, and these occupancy models may be used by biologists to target areas for surveys.  Percent of forest within buffers of 300 m and 1 km wetland (300, 1000)* Percent of wetland within buffers of 300 m and 1 km esh (300, 1000)* Percent of early successional habitat (agriculture, grassland, upland shrubland) within buffers of 300 m and 1 km develop (300, 1000)* Percent of human development within buffers of 300 m and 1 km road.dens (300, 1000)* Road density (m/ha) within buffers of 300 m and 1 km * indicates that both a linear and quadratic relationship were considered.   values inside parentheses are SEs; p is detection probability; Ψ is occupancy probability; k is number of parameters in the model; BIC is Bayesian information criterion; GOF is goodness-of-fit; c-hat is the overdispersion parameter; covariate definitions can be found in Table

ABSTRACT
Painted turtles (Chrysemys picta) are one of the most well-studied species of freshwater turtle, but our understanding of the ways in which populations respond to human-induced landscape alteration remains lacking. We sampled eastern painted turtles (Chrysemys p. picta) from 88 randomly selected wetlands across a range of landscape conditions in Rhode Island, USA. Turtles were systematically and intensively sampled for one year at each wetland to estimate abundance, sex ratio, juvenile ratio, and body mass index. We compared demographic traits between natural and manmade wetlands, used model selection to determine which environmental and within-wetland covariates best explained abundance, and tested whether increasing road density surrounding wetlands resulted in more male-skewed populations. There was no difference in abundance or any demographic trait between natural and manmade wetlands. A negative relationship between abundance and forest cover surrounding wetlands emerged as the most parsimonious model, but explained exceedingly little variation. Contrary to expectations, there was a significant, but very weak relationship between increasing road density and the proportion of females in a population. Collectively, these results suggest that eastern painted turtles are exhibiting little to no detectable variation in population demography across the range of landscapes found in Rhode Island and are resilient in the face of human-induced landscape change.

77
Habitat degradation and loss are considered leading global causes of population declines in turtles , van Dijk 2000. Therefore, human-induced landscape alteration is usually associated with instability in turtle populations (Dudgeon et al. 2006, Bohm et al. 2013. New England, in the northeastern United States, has experienced tremendous changes to the landscape since the time of European settlement  Roads are one of the most ubiquitous ways in which humans alter the landscape (Laurance and Balmford 2013). Aside from being implicated in direct population declines (Nafus et al. 2013), roads are thought to skew sex ratios in turtle populations by disproportionately affecting females as they cross roads while seeking nesting habitat (Steen et al. 2006). Direct observations of turtles dead on roads have supported this idea (Wood andHerlands 1997, Haxton 2000), and studies examining this question experimentally have documented proportionally fewer females in wetlands surrounded by more roads , Aresco 2005a That some turtle populations are threatened by road mortality is not in question (Ashley and Robinson 1996, Wood and Herlands 1997, Aresco 2005b). This problem is compounded by delayed sexual maturity and low recruitment rates in turtles, making populations particularly susceptible to the removal of adult females (Brooks et al. 1991, Doak et al. 1994 (2) modeled abundance using a number of landscape and within-wetland covariates, and (3) tested the prediction that sex ratio would be male-biased in wetlands with higher surrounding road density.

STUDY AREA
Our study was conducted throughout the state of Rhode Island located in

METHODS
In an effort to capture the state-wide variability in landscape composition and configuration, we used a stratified random design to select sites. To minimize confounding factors among sites, we focused our sampling on relatively small (0. At the 300-m scale, we grouped wetlands into eight classes representing 10% increments of forest cover (excluding 0-10% and 70-80%), and at the 1 km scale we grouped wetlands into four, partially overlapping, larger stratifications of forest cover (0-40%, 20-60%, 40-80%, 80-100%). These groups created a near-continuous gradient of sites from different forest cover conditions. We assigned each retained wetland a random number, sorted them by random number, and contacted property owners/managers in ascending order until we received permission to sample the desired number of wetlands in each forest cover and size stratification. We carried out this process in each of three consecutive years.

Turtle Sampling and Data Collection
We  ).

Statistical Analysis
We estimated abundance for each wetland as captures per unit effort by calculating the total number of unique individuals caught divided by the total number of trap nights. To estimate sex ratios at each wetland, we calculated the proportion of females among all unique adults captured. To explore the possibility of temporal bias in estimates of sex ratio, we also calculated the proportion of female captures by each sampling occasion, separately for newly caught individuals and recaptures. We calculated the proportion of juveniles among all unique individuals caught for each wetland. We calculated a body mass index (BMI) for all turtles by dividing weight (g) by straight-line carapace length (mm). We classified all wetlands as either "natural" (i.e., ≥ 77 years old) or "manmade" (i.e., < 77 years old) and compared each demographic measure described above between both groups. We used two-sample independent t-tests to compare population means of each demographic measure between both groups.
We used linear regression to estimate the effect of road density on sex ratio. As a response variable, we used wetland-specific sex ratio estimates and performed identical tests on a 'full' dataset including all wetlands, and a 'restricted' dataset limited to wetlands where at least 10 individuals were caught. Shapiro-Wilk tests were used to assess normality in these response variables.
We used generalized linear models (GLM) to estimate the effect of environmental covariates on unique individuals per trap night as a measure of abundance. We used a Gamma distribution to model error in the abundance distribution, adding nominal values (0.0001) to wetlands with no eastern painted turtle detections to address large numbers of zeros in the dataset that would otherwise limit analyses. We used an information criterion 85 framework to compare models composed of different combinations of covariates with the aim of identifying the model(s) that most parsimoniously described abundance . As a means of variable reduction, for landscape covariates, we first compared single-covariate models at both spatial scales (i.e., 300 m and 1 km) and retained the term from the most supported model based on a Bayesian Information We report means ± one standard error (SE), and we defined statistical significance as P ≤ 0.05. Statistical analyses were performed using R (version 3.4.1, www.rproject.org, accessed 1 October 2017).

RESULTS
We  Table 2). The first was the null model that contained only an intercept and no covariates. The second was a model that included a single coefficient for 87 forest cover within 300 m (Figure 3). The pseudo-R 2 for the forest cover model was 0.027.

DISCUSSION
We found no evidence that eastern painted turtle abundance was influenced strongly by any landscape or within-wetland covariate. Nor was there evidence that sex ratio was male-skewed in wetlands surrounded by more roads. In fact, contrary to expectations we found a very slight, but significant positive relationship between femaleskewed sex ratios and road density. Furthermore, all demographic measures, including abundance, sex ratio, proportion of juveniles, and BMIs, were similar between natural and manmade wetlands. Collectively, these results suggest that eastern painted turtles are cosmopolitan in Rhode Island and are exhibiting little to no detectable variation in population demography related to covariates measured along a gradient of landscape types.
The trend of decreasing abundance with greater forest cover was significant in the top model, but the amount of variation explained was very low (< 3%). There was a recognizable pattern of reduced abundance at the highest end of the forest cover gradient, however. We detected no painted turtles at six of 11 wetlands surrounded by > 90 % forest cover, and four of the five remaining wetlands contained values of abundance that were below the mean. Painted turtles prefer open areas for nesting and the limited availability of nesting habitat in fully forested areas may reduce abundance . Painted turtles are often found in high abundance relative to other species of freshwater turtles, even in areas of major human disturbance Gibbons 1996, Gamble and. Other studies have found significant relationships between painted turtle abundance and a variety of different factors. In New Hampshire, greater abundance was explained by decreasing distance to the nearest wetland and less herbaceous vegetation in wetlands, among other factors . These authors concluded that populations of painted turtles were under threat by habitat alterations related to human development, namely increased road density and a greater density of generalist predators like raccoons (Procyon lotor). Our results do not support a similar conclusion for our study area.
We chose not to model other demographic measures with our suite of covariates as preliminary analyses suggested very little pattern in the data. Instead, we classified each wetland as natural or manmade based on historic aerial imagery that dates back to 1939 and inspected these groups for differences. We found no measurable differences in abundance, sex ratio, proportion of juveniles, or body mass indices between groups. All manmade wetlands were either created by excavation, or formed by restricting the flow of a stream in the last 77 years. The fact that there were no differences in populations of eastern painted turtles between these sites and natural sites suggests that turtles actively colonized the novel habitats shortly after it was created and began successfully breeding.
Painted turtles are known to readily disperse from one wetland to another via terrestrial movements on the order of kilometers ) and readily colonize wetlands where they do not occur ).
Colonization and extinction of small wetlands by painted turtles is a dynamic process that is heavily influenced by wetland size, hydroperiod, and landscape connectivity to other wetlands ).

89
Our overall estimate of sex ratio was near 1:1, but detectability varied substantially over the course of the activity season. The sex ratio detected in the first sampling occasion (May-June) may be most reflective of the actual sex ratio within the population. Alternatively, the first sampling occasion may have occurred during the portion of the activity season when females are most active and most likely to enter traps, thus representing an inflated estimate of sex ratio. Similarly, the reduced proportion of new females detected in subsequent sampling occasions may have resulted from reduced activity, or may be more reflective of the actual sex ratio. Either way, it is clear that there is a strong seasonal effect on the detection of females, which has major implications for the estimate of sex ratios. Without knowing the characteristics of this trapping bias, only by censusing an entire population can we confidently estimate sex ratio. In Long Island, an 18-year study of painted turtles at a complex of small ponds found that estimates of adult sex-ratio varied greatly from year to year, but averaged close to 1:1 over the course of the study . It remains common for many ecologists to ignore both stochasticity and temporal shifts in detection when sampling reptiles and estimating demographic measures like sex ratio, even though these issues have been recognized for decades.
Our results make clear that there is no pattern of male-skewed sex ratios with increasing road density for eastern painted turtles in Rhode Island. Given our large sample size of 88 wetlands distributed broadly across the state and intensive sampling regime, we believe this is a robust result. However, we must consider our scope of inference and place this result in the proper context. The location and traffic density of a road are important factors to be considered as well. When choosing sites, we selectively 90 excluded wetlands within 300 m of federal or state highways. Collectively, these roads account for a very small percentage of total linear distance in the state, but these are the roads with some of the highest traffic volumes. This is likely to have excluded a number of wetlands where turtles experience the highest rates of road mortality Shriver 2002, Litvaitis andTash 2008). A number of studies have examined the effects of high traffic volume roads immediately adjacent to wetlands and there is evidence that the majority of turtle road mortality occurs at severe hotspots with these characteristics (Aresco 2005b, Langen et al. 2012) and that mortality spikes at particular times of year (Ashley and Robinson 1996, Glista et al. 2007). In New York, the proportion of painted turtle females decreased with higher surrounding density of high volume roads, but overall sex ratio was not different than 1:1 .
The disproportionate susceptibility of females to road mortality is thought to be more pronounced in freshwater turtles than terrestrial turtles, as male freshwater turtles are expected to spend more of their time in the water and less exposed to the dangers of roads. However, male painted turtles are known to move across land between wetlands and are certainly susceptible to road mortality as well. Road density alone may be an insufficient predictor of sex ratio in painted turtle populations. Future studies should integrate information on traffic volume, when possible.
Our sampling was limited to one year at each wetland making our estimates of abundance subject to error associated with inter-annual variation. However, eastern painted turtles are highly aquatic and so inter-annual abundance is likely less variable in this species than in a more terrestrial freshwater species such as the spotted turtle (Clemmys guttata). In addition, we made a substantial effort to limit potential bias. By distributing traps in a spatially homogenous way at each wetland, we were able to ensure coverage of the entire wetland and not rely on the assumption that turtles are distributed evenly throughout the wetland at all times. By sampling intensively up to four times, with sampling occasions fairly evenly spaced across the activity season, we were able to limit temporal bias associated with peaks of activity throughout the year. Surprisingly few studies take both of these things into account. Future studies should be sure to sample in a systematic way and, when possible, collect data for multiple years.

MANAGEMENT IMPLICATIONS
At least two factors give painted turtles an advantage over sympatric species of freshwater turtles when it comes to the impacts of road mortality and other disruptive forces that occur in areas of greater human disturbance. Small body size reduces the probability of vehicle strikes when crossing roads (Gibbs and Shriver 2002), and relatively rapid reproductive parameters (i.e., shorter generation time and higher fecundity) allow populations to rebound from declines in shorter periods of time. If we can identify which species are most susceptible to road mortality and which are not, based on these parameters, this could greatly help to steer limited management resources to those species that are most susceptible.

Introduction
Intensive and large-scale landscape alteration by European settlers dates back several centuries in Rhode Island. Clearing of the land for timber and agriculture began in the 17 th century and peaked in the mid-19 th century, when approximately 70% of the state was deforested . Only after agriculture from western states began to outcompete farms in New England did the landscape begin to regenerate to early successional habitat and secondary forest. Today, approximately 54% of the state is forested, with pine, oak, and maple forests dominating the western part of the state . Freshwater wetlands have undergone immense alteration in the previous centuries as well. Drainage, filling, damming, and channelization occurred for centuries without regulation, resulting in the loss of approximately 37% of the wetlands in Rhode Island between 1780 -1980 . The creation of novel wetlands for drinking water and agriculture has further altered the landscape. More recently, post-World War II economic growth led to a construction boom and the creation of the interstate highway system, both of which consumed and fragmented large areas of the landscape in Rhode Island. Undoubtedly, these human activities have had major impacts on the abundance, demography, and connectivity of populations of wildlife throughout the state and region, but for most species the legacy of landscape change remains largely anecdotal or completely unexplored.
Populations of freshwater turtles in the region have certainly been impacted by these alterations, but not necessarily in a uniform fashion across species. Certainly, some species have experienced declines due primarily to historic habitat loss and fragmentation. True habitat specialists, like the bog turtle (Glyptemys muhlenbergii), have probably experienced the most dramatic declines (USFWS 2001, Rosenbaum et al. 2007). Habitat generalists however, that have the ability not only to acclimate to new conditions, but subsist in heavily altered wetlands or colonize newly-created wetlands, have maintained comparable distributions and abundances, and in some cases may have benefited from changes (Price et al. 2013, Winchell and. The common snapping turtle (Chelydra serpentina) is an example of a generalist species that remains abundant throughout most of its range (Paterson et al. 2012, Anthonysamy et al. 2014).
The eastern painted turtle (Chrysemys p. picta) and the spotted turtle (Clemmys guttata) are two species that are thought to have experienced very different responses to recent anthropogenic landscape change, with the former having remained abundant, and the later having experienced substantial declines.
Chrysemys p. picta often occurs in high abundance even in areas of major human disturbance Gibbons 1996, Gamble andSimons 2004, Chapter 3) and has been shown to occur in much higher densities than C. guttata, where they co-occur (Ernst 1976). Chrysemys p. picta is one of four recognized subspecies of C. picta, a small (carapace length up to 25.4 cm) freshwater turtle with a large geographic range that spans across North America . Chrysemys p. picta occupies the eastern part of this range, stretching from Georgia, USA to New Brunswick, Canada along the Atlantic seaboard. They occur in all types of freshwater wetlands including riparian systems. Sexual maturity usually occurs in 2-4 years in males, and 6-10 years in females . Precise data are limited and variable across the range, but generation time is thought to be in the range of 10-20 years (Wilbur 1975, Ernst and Lovich 2009). They are known to readily disperse from one wetland to another via terrestrial movements on the order of kilometers ) and readily colonize uninhabited wetlands .
In contrast to C. picta, C. guttata is believed to have experienced severe population declines throughout its range in the last two centuries due primarily to habitat loss, alteration, and fragmentation , Lewis et al. 2004, van Dijk 2011. Clemmys guttata is a small (carapace length up to 14.3 cm) freshwater turtle native to the eastern United States and Great Lakes region ).
Sexual maturity usually occurs between 7-15 years ) and tends to be at the higher end of this range in northern populations .
Estimates of generation time are usually considered to be between 20-30 years, but may be as high as 40 years in the northern latitudes (Ernst andLovich 2009, COSEWIC 2014). They are often described as semi-aquatic because they use both wetland and upland habitats for extended periods ). Throughout their range, C.
guttata occur in a variety of wetland types, but do exhibit habitat selection for bog-like wetlands Melvin 2001, Rasmussen and. In Rhode Island, C.
guttata are rare relative to other species of freshwater turtles and are strongly forestassociated (Chapter 2). Dispersal is limited and fidelity to wetlands is high, with individuals often overwintering in the same hibernaculum each year (Haxton andBerrill 1999, Litzgus et al. 1999, Chapter 1). They are a species of increasing conservation concern, especially in the northeastern United States where six of the seven states in  (USFWS 2015). In this study, we characterize and compare population genetic diversity and population genetic structure and diversity of this relatively rare species with that of the more widespread and abundant C. p. picta.
The population genetic structure of endangered species is of fundamental interest to conservation biologists. Genetic diversity and inbreeding have implications for a population's vulnerability to environmental and demographic stochasticity, thus affecting the probability of extinction (Brook 2008, Frankham et al. 2010. A loss of genetic diversity can reduce the ability of a population to adapt to changing environmental conditions, and inbreeding depression can have deleterious effects on the reproductive fitness of offspring (Ralls et al. 1988, Frankham 2005, O'Grady et al. 2006. Genetic differentiation among subpopulations is in part a product of gene flow, and measures of differentiation can help identify subpopulations that may be genetically isolated due to barriers associated with habitat fragmentation. Maintaining gene flow to counteract the loss of genetic diversity due to inbreeding and genetic drift is important to ensure genetic viability, especially for species that occur in small, isolated subpopulations (Frankham et al. 2010).
Our primary objective was to assess whether C. guttata is experiencing elevated risk of extirpation due to increased levels of inbreeding, reduced genetic diversity, and increased population genetic structure due to isolation, which collectively we refer to as genetic degradation. We made several predictions based on the insight that C. guttata occur in smaller, more isolated populations, and that they probably exhibit reduced rates of gene flow compared to C. p. picta. We predicted that C. guttata would 1) exhibit less genetic diversity, 2) exhibit more inbreeding, 3) exhibit more differentiation among sites and 4) were more likely to have undergone recent reductions in effective population size (i.e., a population bottleneck), as compared to C. p. picta.

Study area and sampling
Our study was conducted throughout the state of Rhode Island located in 16,000 years before present (Sirkin 1996, Uchupi et al. 2001 (Uchupi et al. 2001).
From 2013-2015, small (0.1 -1.8 ha), hydrologically isolated (i.e., discrete, nonriparian) wetlands throughout the state were randomly selected across a gradient of forest cover for a mark-recapture study focusing on occupancy and demography (Chapters 2 110 and 3). Genetic tissue collection took place concurrently at a subset of these wetlands.
Because C. p. picta were relatively common, tissue was collected only at wetlands with high densities of turtles that would ensure an adequate number of individuals for population genetics analysis (Hale et al. 2012 For all individuals, less than 1 ml of blood was collected from the sub-carapacial vein using a 25-gauge sterile needle and a 3 ml syringe, and placed immediately on a Whatman FTA sample collection card (GE Healthcare, Buckinghamshire, United Kingdom). These cards were stored at room temperature and used for subsequent DNA extraction. All individuals were released at the site of capture.

Genotyping
We used the DNEasy Blood and Tissue Kit (Qiagen Corporation, Valencia, CA, USA) to extract DNA using the standard protocol. For both species, we amplified previously described microsatellite loci (Pearse et al. 2001, King andJulian 2004). We amplified 18 loci for C. p. picta and 17 loci for C. guttata, organizing these into 6 and 5 multiplexes, respectively. We carried out polymerase chain reaction (PCR) using the Qiagen Type-it Microsatellite PCR Kit under conditions recommended in King and Julian (2004), but with a modified initial denaturing step of 95 C for 5 minutes. We used negative controls on PCR plates to identify any potential contamination.  (Adamack and Gruber 2014) to estimate the frequency of null alleles for each locus (Brookfield 1996), private alleles per site, and mean allelic richness per site using the rarefaction method to correct for variation in sample size (Kalinowski 2004). We calculated expected heterozygosity (He), observed heterozygosity (Ho), and inbreeding coefficients (FIS) for each site, and calculated 95% confidence intervals for FIS estimates using 10,000 bootstrap iterations, all using the diveRsity package (Keenan et al. 2013).
We used the diveRsity package to calculate global measures of FIT, FIS, and FST, and to calculate pairwise FST values for all sites. All F-statistics used the bias-corrected formulation of Weir and Cockerham (1984). As an alternative measure of population differentiation and to maximize comparability with other studies, we also used the diveRsity package to calculate pairwise values of the bias-corrected Jost's Dest (Jost 2008, Gerlach et al. 2010). The diveRsity package was used to estimate 95% confidence intervals for all measures of differentiation using 10,000 bootstrap iterations. We used the poppr package to perform an analysis of molecular variance (AMOVA). We conducted the test with two stratifications such that variance of allele frequencies was partitioned within sites and among sites (Excoffier et al. 1992). For the global F-statistics and AMOVA analyses, we excluded the Block Island site for C. p. picta, and included only the five C. guttata sites with sample sizes >4 to limit confounding factors such as outliers and small sample size (Kalinowski 2005), and thereby maximize the comparative inference between the two species.

Population structure
We used the ade4 package to perform a Mantel test with 10,000 permutations to test for genetic isolation by distance. We used Nei's (1972) measure of genetic distance to create the genetic matrix, and geographic locations centered on individual wetlands or on a geographic mean when turtles were sampled from multiple wetlands, to create the Euclidean distance matrix. For C. p. picta, we did not include the Block Island site, and for C. guttata included only the five sites with sample sizes >4 to avoid falsely inflating measures of genetic distance.
We used program STRUCTURE v.2.3.4 (Pritchard et al. 2000) to characterize population genetic structure for both species (Porras-Hurtado et al. 2013) and to test our prediction of a greater degree of subpopulation structure in C. guttata. STRUCTURE allows for the identification of genetic clusters within a dataset by detecting differences in allele frequencies and assigning individuals to those clusters based on analysis of likelihood. For all runs, we assumed an admixture model with correlated allele frequencies and employed the LOCPRIOR parameter using sampling location as the additional sample information. The LOCPRIOR parameter is informative in situations of weak population structure such as that to be expected given the spatial scale of our study (Hubisz et al. 2009, Porras-Hurtado et al. 2013. In all cases, we performed 20 independent iterations of runs consisting of a burn-in of 200,000, followed by 500,000 MCMC repetitions, which was sufficient for all runs to reach convergence. For C. p. picta we ran an initial analysis with all individuals included (hereafter complete analysis) and a second analysis with a maximum of 25 individuals selected randomly (hereafter subset analysis) from each site to ensure that sample size unevenness was not influencing results (Puechmaille 2016). We specified the range of K as 1-10 for both runs. For C.
guttata we ran an initial analysis with all individuals from all sites (i.e., complete analysis), and a second analysis with only sites with more than 9 individuals, while also limiting site 29 to only 30 randomly selected individuals (i.e., subset analysis). We specified the range of K as 1-11 for the complete analysis, and 1-4 for the subset analysis.
We considered both the ln Pr(X|K) and the ΔK method (Evanno et al. 2005) with STRUCTURE Harvester (Earl and vonHoldt 2012) to evaluate the most likely number of clusters. We used CLUMPP v.1.1.2 (Jakobsson and Rosenberg 2007) and distruct v.1.1 (Rosenberg 2004) software for post-hoc data processing and visualization.

Population bottleneck
We used program BOTTLENECK v.1.2.02 (Piry et al. 1999) to test the prediction that C. guttata were more likely than C. p. picta to have undergone recent reductions in effective population size. To test for the signature of heterozygosity excess, we considered results from both a two-tailed sign test (Luikart and Cornuet 1998) and a onetailed Wilcoxon signed-rank test using the two-phase mutation model (TPM) with 10,000 iterations used to generate a distribution of expected equilibrium heterozygosity (Heq).
The vast majority of genetic variance occurred within sites for both species

Population bottleneck
None of the C. p. picta sites showed evidence of recent genetic bottlenecks, but site 18 did yield significant results in the two-tailed sign test (α < 0.05) at more than one TPM level. The one-tailed Wilcoxon test returned a P-value approaching one, suggesting heterozygous deficiency, the signal of a recent population expansion ( guttata results had reduced P-values in the Wilcoxon tests suggesting heterozygous excess, the signal of a recent population decline.

Discussion
Overall, C. p. picta exhibited weak population genetic structure. We found no evidence of isolation by distance, and global FIT and FIS both overlapped zero, suggesting a lack of overall population structure, and a lack of inbreeding, respectively. In C.
guttata, there was no evidence of isolation by distance or strong differentiation among sites, but global FIS and the consistency of private alleles among sites suggested the presence of some population structure consistent with inbreeding.
Results were consistent with some predictions and inconsistent with others, which we interpret as limited evidence that C. guttata is at greater risk due to genetic degradation in our study area. In line with predictions, C. guttata exhibited a greater degree of inbreeding than C. p. picta and there was tentative evidence of recent population declines in C. guttata, whereas there was no evidence for declines and even some evidence for population expansion in C. p. picta. Genetic diversity and differentiation among sites were similar for both species, however.

Genetic diversity
A lower mean allelic richness in C. guttata suggested less genetic diversity compared to C. p. picta, but mean expected heterozygosity was higher in C. guttata. For both species, estimates of observed and expected heterozygosity and allelic richness were comparable to those from other studies of turtles using microsatellites (Vargas-Ramirez et al. 2012, see Table 4) and probably do not indicate significant depletion of genetic diversity. However, long-lived species can mask declines in genetic diversity even after prolonged population declines, making interpretation difficult (Kuo and Janzen 2004). A comparison of genetic diversity in fragmented populations of C. guttata and C. picta marginata in Indiana found lower diversity in C. guttata (Parker and Whiteman 1993).
The authors identify smaller habitat patch size, lower population density, and greater isolation of C. guttata populations as potential factors, but low sample sizes and the possibility of different mutation rates of the genetic markers used for the different species limit strong conclusions from this study. An investigation of population genetic structure found that genetic diversity was highest in C. opportunity for gene flow with the mainland since the Pleistocene (Sirkin 1996). The post-glacial colonization of the northeastern United States by C. p. picta occurred as populations expanded from southern refugia after glaciers retreated (Starkey et al. 2003).
Chrysemys picta are physiologically well adapted to cold climates (Storey et al. 1988, Churchill andStorey 1992) and, along with C. serpentina, were the first turtles to expand northward into formerly glaciated areas (Holman and Andrews 1994). The exact time at which these species first colonized what is now Block Island and mainland Rhode Island is not known, but it probably took place between 10,000 -15,000 years ago (Holman andAndrews 1994, Starkey et al. 2003). A characteristic reduction in genetic diversity associated with this relatively recent post-glacial range expansion (Hewitt 2000, Weisrock andJanzen 2000), along with high rates of contemporary gene flow, may be responsible for the lack of pronounced population genetic structure.
STRUCTURE results indicated that the majority of the mainland sites were assigned to multiple genetic clusters, a common signature of weak population structure 21 predates the earliest available aerial imagery (>77 years old), but is clearly a pool that formed when a former stream was bisected by a road. In total, at least 6/22 C. p. picta sites were manmade or heavily modified. The three sites also contain plentiful nesting habitat immediately adjacent to the wetland. Site 12 is located at the end of a 121 commercial/military airport where much of the grounds next to the wetland are maintained as grassland, site 15 is located on an urban golf course with manicured lawns and sand traps, and site 21 is on private property with sandy soils maintained as lawn ( Figure 4.2). Together, the recent creation of novel habitat and beneficial habitat characteristics could have facilitated recent population expansions at these sites, causing allele frequencies to differ from the population at-large. However, there is no signature of a reduction in heterozygosity in any of the sites as would be expected after a founder event (Frankham et al. 2010). An alternative explanation could be limited gene flow due to isolation. This is plausible at sites 12 and 15 that both occur in highly developed landscapes, but unlikely at site 21 where nearby riparian and permanent wetlands located in a relatively undisturbed landscape are likely to contain C. p. picta. Ultimately, we cannot say with certainty what is causing the observed differentiation.
Contrary to predictions, we detected similarly modest differentiation among sites of C. guttata. In fact, a smaller percentage of sites exhibited differentiation compared to C. p. picta, but direct comparison is difficult because of the disparity in sample size (C. guttata = 20 comparisons, C. p. picta = 420 comparisons). All significant C. guttata pairwise comparisons included site 29 and this site was also differentiated in the complete STRUCTURE analysis. Adults from site 29 were radiotracked for two years as part of another study and were found to exhibit limited movements and high levels of home range fidelity (Chapter 1). Given that dispersal is a requisiste process for gene flow, if dispersal rates to neighboring wetlands are indeed low, limited gene flow could explain the higher differentiation. The C. guttata STRUCTURE subset analysis resulted in a more ambiguous pattern of differentiation and the fact that the ΔK and ln Pr(X|K) 122 methods resulted in disparate results make this difficult to interpret. Overall, there is little evidence that C. guttata exhibits appreciably greater population genetic structure than C.

Population bottleneck
We documented tentative evidence of recent population declines in C. guttata.
Bottleneck tests can be difficult to interpret, but results comparable to ours have been interpreted in a similar way as those for other species of turtles (Kuo and Janzen 2004).
We ran multiple tests under a range of different multi-step mutation model proportions to assess the robustness of results (Peery et al. 2012

Scope and limitations
For both species, the genetic structure that we detected was modest. Given the limited spatial scale of our study (especially for C. guttata) and the fact that we expected these sampling sites to be admixing to some degree, it should be emphasized that we were indeed seeking fine-scale genetic structure. Moreover, in our study area the impact of human activities on turtle populations has occurred in the evolutionarily recent past (~250 years) and has intensified only in the last ~75 years. The number of C. p. picta generations since the more intense period of human influence began is probably between 4-7 generations, and between 12-25 generations for the longer period. The number of C.
guttata generations is probably between 2-4 for the shorter period, and 8-12 for the longer period or nearly half the number of generations of C. p. picta. As it can be difficult to detect the effects of genetic drift in long-lived organisms, the spatial and temporal scales (i.e., time since habitat loss and fragmentation) of our investigation may have limited our ability to detect genetic differentiation and demographic events that have occurred in the recent past, particularly for C. guttata. Simulation studies have demonstrated that FST is relatively insensitive to disruptions to gene flow, especially when dispersal is limited in the organism of study, and that other population-based metrics may be superior in detecting changes that have occurred in the recent past (Landguth et al. 2010). Compounding the issue, turtle DNA evolves slowly relative to that of other vertebrates (Avise et al. 1992, Shaffer et al. 2013. Other studies of population genetics in freshwater turtles have failed to detect predicted genetic structure, even when there is strong empirical evidence of the effects of historic habitat fragmentation (Bennett et al. 2010, Anthonysamy 2012 Clemmys guttata has undergone severe declines in many parts of its range, but their status in Rhode Island is unclear. In the recent revision of the Rhode Island Wildlife Action Plan, C. guttata is listed as a "Species of Greatest Conservation Need," but is given an S5 ranking indicating it is considered widespread and abundant in the state (RIDEM 2015). While a recent state-wide sampling effort disputes the idea that they are common (Chapter 2), it is possible that C. guttata in our study area have not experienced population declines at the same level of severity as in other places throughout its range, and are not representative of the experiences of the species at large. Moreover, it is possible the sites we sampled were inherently biased towards being those least affected 125 by population declines and isolation given that they are places where they were known to occur in relatively robust numbers and where our sampling yielded the most individuals.

Concluding remarks and conservation implications
Chrysemys p. picta is one of the most well-studied freshwater turtle species, largely because they are widespread and abundant. Our analysis confirms that they exhibit little population genetic structure across Rhode Island, making for an appropriate contrast with a far less abundant species. Some sites do exhibit modest genetic differentiation, but the reasons why remain elusive and warrant further investigation. Our analysis provides some evidence that C. guttata exhibit a greater degree of inbreeding and may have experienced population declines in the recent past. Overall, population genetic structure remains comparable to that of C. p. picta though, suggesting the sites studied have not experienced serious genetic degradation at this point. As we were unable to find strong evidence of genetic degradation in C. guttata, the southern region of Rhode Island may be well suited to serve as a regional conservation reserve network where the maintenance of gene flow among wetlands occupied by the species is prioritized (Kautz et al. 2006, Shoemaker andGibbs 2013). Relatively little is known about C. guttata population genetics and how genetic structure varies range-wide. Much of what has been inferred is derived from studies of different species of freshwater turtles considered ecologically similar. Understanding the legacy of habitat loss and fragmentation on population genetic structure is critical for effective management and conservation of this species, at both regional and local spatial scales. Additional population genetics studies of C. guttata at multiple spatial scales will help improve our understanding of the potential vulnerabilities to environmental and genetic stochasticity in this declining species.     Bolded loci are those that were excluded from population analyses. Asterisks (*) denote loci that significantly deviated from HWE (after Bonferroni correction) in an exact test based on 10,000 Monte Carlo permutations of alleles. He is expected heterozygosity and Ho is observed heterozygosity.