The Florida box turtle (Terrapene bauri) ranges from southeastern Georgia across Peninsular Florida to the Florida Keys (Dodd 2001; Liu et al. 2004; Jones et al. 2016; Mays et al. 2017; Turtle Taxonomy Working Group 2017) and has declined at sites across Florida (Dodd and Griffey 2005). Most studies of North American box turtles (Terrapene spp.) have been of relatively short duration (fewer than 5 yrs), representing only a snapshot in time compared with the species' potential lifespan, which can exceed 50 yrs (Kiester and Willey 2015). Consequently, the effects of disturbance processes on long-term demographic trends in box turtle populations are not well understood. Most published, long-term studies of North American box turtle populations have revealed declining populations, even in nature preserves and protected areas (e.g., Stickel 1978; Schwartz et al. 1984; Williams and Parker 1987; Hall et al. 1999; Nazdrowicz et al. 2008). Similar trends have been reported for other turtle species on protected lands (Browne and Hecnar 2007; Markle et al. 2018; Howell et al. 2019), suggesting that efforts to protect such long-lived vertebrate species may require expanding both the spatial and temporal scale of conservation efforts (Kiester and Willey 2015). These species should be a management priority where potentially viable populations occur on land protected for biodiversity (Dodd 2006). Indeed, global research suggests the need to conserve and manage specifically for turtle species if we are to ensure the evolutionary potential of representative lineages of this threatened group (Roll et al. 2017; Rhodin et al. 2018). By continuing to evaluate long-term, demographic change in box turtle populations at multiple sites under a variety of landscape contexts and disturbance regimes, we can better inform sustainable conservation and management efforts for this and related species. At the very least, turtle populations that appear to be of continental significance because of their size, their estimated viability, or their potential to elucidate long-term ecological patterns should be monitored and managed adaptively (Dodd 2006). Nonlisted yet vulnerable and declining species should be considered management priorities on protected sites, particularly national wildlife refuges and other areas with a mission—at least in part—to conserve biodiversity. Further, when turtle populations of national or continental significance occur on protected lands, they offer a rare opportunity for both conservation and insight into long-term demographic changes due to environmental change.

The Florida box turtle is a terrestrial emydid turtle nearly endemic to the Florida peninsula, with an estimated divergence from the more widespread Terrapene carolina of 10.3 million years ago and a limited area of introgression with other lineages of Terrapene in the Florida Panhandle (Martin et al. 2013). The Florida box turtle population on Egmont Key, Hillsborough County, Florida, was studied intensively from 1991 to 2006 using visual encounter surveys and capture-mark-recapture techniques (Dodd 1997a, 1997b, 1998; Dodd et al. 2006, 2012). Egmont Key's box turtle population was estimated to have increased at an average rate of 5% per year during the 15 yrs of study to a peak of more than 1500 turtles in 2002 (Dodd et al. 2012). The long-term body of quantitative research allowed the authors to evaluate the effects of observed disturbances on population parameters and behavioral patterns (such as habitat use) and to develop predictions about how the population might change over time under different disturbance scenarios. In the years since studies ended on Egmont Key, no larger population of T. bauri has been documented (e.g., Verdon and Donnelly 2005; Jones et al. 2016). We consider the Egmont Key population to be of continental significance because it is one of the longest studied Terrapene populations in the southern United States. Only a handful of North American box turtle populations have been studied for longer durations (e.g., Stickel 1978; Schwartz et al. 1984; Williams and Parker 1987; Nazdrowicz et al. 2008). Egmont Key's box turtle population is also the only long-term study of T. bauri. Box turtles have been declining throughout the United States and likely in Florida, with its massive human population explosion and associated habitat loss.

Quantitative data on the effects of disturbances on this important population will help to inform management at this and other protected box turtle sites across the range. To assess the effects of the 2016 wildfire and evaluate the current status and condition of this regionally significant population, we conducted a reassessment of Egmont Key's box turtle population during March 2017, November–December 2017, and March 2018.

METHODS

Study Area. — Egmont Key is a roughly 114-ha island of continental origin at the mouth of Tampa Bay, Hillsborough County, Florida (Fig. 1). Approximately 96 ha were vegetated during this study. The island is part of the US Fish and Wildlife Service (USFWS) National Wildlife Refuge system known as Egmont Key National Wildlife Refuge, but it is also leased and managed by Florida's Department of Environmental Protection (DEP) as a state park. In addition to the Egmont Key Lighthouse and the ruins of 19th-century Fort Dade, the Tampa Bay Pilots Association (TBP) manages a 3.3-ha inholding of houses on the island's eastern shore. Historically, Egmont Key was free of resident mammalian predators, although a stray raccoon (Procyon lotor) found its way to the island and was subsequently trapped and removed in April 1992 (Franz and Dodd 1993).

View Figure

• Download as Powerpoint
• Download Figure

Several significant landscape changes and disturbances have occurred on the island since the box turtle population was last studied in 2006. Roof rats (Rattus rattus) were unintentionally introduced to the island in 2006 (USFWS 2010) and initial trapping in 2008 resulted in the capture of 760 rats, suggesting a large and widespread population (Witmer et al. 2010). An eradication program with the rodenticide diphacinone was undertaken in 2008 and 2009 (USFWS 2008; Witmer et al. 2010) with some preliminary considerations to avoid impacts to gopher tortoises (Gopherus polyphemus), and rats are reportedly now absent from the island (S. Garner, pers. comm., 2019). Additionally, a 16.2-ha prescribed burn was conducted in the southern portion of the island in 2009 (Witmer et al. 2010) to control understory vegetation. Management of invasive vegetation, predominantly Brazilian pepper (Schinus terebinthifolius), has occurred throughout the island (Dodd 2006; USFWS 2010). In addition, visitation to Florida state parks has increased statewide in recent years. For example, Egmont Key's visitation has increased from 130,000–170,000 in 2004–2006 (USFWS 2009) to 248,638 in 2016–2017 (Cutshaw 2017), possibly facilitating high levels of incidental or targeted collection. Finally, and most significantly, a lightning-caused wildfire burned approximately 26.1 ha (roughly 34.2% of the forested habitats on the island) in July 2016 (Fig. 2), distinct from the area that burned in 2009. The 2016 wildfire consumed approximately 34% of the box turtle habitat on the island. The burned area near the center of the island represented some of the best remaining box turtle habitat because the northern end of the island had been heavily disrupted by tourism, the restoration of Fort Dade, and exotic plant removal in the early 2000s. The southern portion of the island was also negatively affected by exotic plant removal (C.K.D., unpubl. data, 1996–2006) and a fire ca. 2009. It is unknown to what extent these changes and trends have influenced the box turtle population.

View Figure

• Download as Powerpoint
• Download Figure

Survey Methods. — We surveyed the 96 ha of vegetated habitats on Egmont Key (Fig. 2) during daylight hours between sunrise and sunset via visual encounter surveys in March 2017, November–December 2017, and March 2018, following methods and searching areas consistent with those surveys conducted previously by Dodd and colleagues. Individual turtles found alive were identified by their marginal notch code (adapted from Cagle 1939) if already marked, or assigned a new notch code beginning at number 2402 if unmarked, resuming the notch code series where it was left off by Dodd et al. (2012). Each turtle was measured using dial calipers, weighed, photographed, assigned an age category, evaluated for shell damage, and released at the capture location. Following Dodd et al. (1994), air and substrate temperature, relative humidity, weather conditions, habitat type, and turtle activity were recorded at each location of a live turtle encounter. All turtles found dead were photographed and processed in the same way as live turtles, except that environmental conditions and mass were not recorded, and shell measurements were only recorded when the shell was complete. Dead turtles were not removed, but were clearly marked so that they were not recounted in subsequent surveys. Turtles were handled and processed for a maximum of 15 min and remained within a few meters of their immediate capture location. Location data were collected at each detection location using handheld global positioning system (GPS) units. We also photographed and GPS-referenced any evidence of mammalian predators. These methods were consistent with, and informed by, protocols utilized at other T. bauri study sites in the Ten Thousand Islands region (Jones et al. 2016).

During our surveys from 8 to 11 March 2017, when evidence of the 2016 wildfire was still visibly fresh, we conducted a total of 85.75 hrs of visual encounter surveys between sunrise and sunset. We used the number of turtles observed alive and dead in random plots (both burned and unburned areas) during the first survey bout (March 2017) to estimate the total number of detectable turtles in each of these sections of the island, respectively. We conducted 3 types of visual surveys during the March 2017 sampling period: meandering upland surveys; randomized, point-centered circular 0.25-ha plots; and transects.

Meandering Upland Surveys. — On each of 4 days, 3 experienced observers (M.T.J., L.L.W., J.D.M.) walked through upland habitats across the entire island, actively searching palm litter, depressions, coarse woody debris, shrub thickets, structural foundations, fence lines, herbaceous communities, and graminoid areas for turtles. Each observer carried a GPS unit and saved the track each day. A GPS point was stored and recorded for each turtle found alive or dead.

Randomized, Point-Centered Circular 0.25-ha Plots. — We generated a database of random points across Egmont Key. At each of the 12 points (8 within the 2016 wildfire burn and 4 outside of the 2016 wildfire burn; Fig. 2), 3 observers conducted 15-min (for a total of 0.75 person-hrs) intensive searches of the area within 28.2 m of the random point (i.e., a 0.25-ha circular plot). A single GPS point was recorded for each turtle found alive or dead.

Transect. — As means of direct comparison with the circular plot results, we surveyed one 206-m east–west transect within the 2016 wildfire area using a randomly generated starting location. Walking a straight line in single file, all turtles observed were recorded, processed, and GPS-referenced. Although the straight-line and right-angle distance off the line was measured for each turtle, the purpose of this transect was only to provide another basis for extrapolation.

Additional Surveys to Estimate Population Size. — In order to estimate the size of the current box turtle population for comparison with previous estimates, we conducted 2 additional sampling bouts following our initial surveys in March 2017. In November and December 2017, M.T.J. and J.D.M. again sampled the population across the island with assistance from USFWS and DEP staff. We repeated all of our random survey plots to establish a pattern of standardized effort in addition to meandering and targeted searches. We added another 6 random plots, for a total of eighteen 0.25-ha circular plots. In March 2018, M.T.J. and L.L.W. again sampled the population across the island with assistance from USFWS and DEP staff utilizing meandering surveys.

Follow-Up Visual Surveys to Estimate Population Size. — We summarized the number of live and dead turtles as well as the number of previously marked and new or unmarked turtles for each survey bout, and across all surveys, and compared our results with Dodd et al. (2016). Using all live turtle detections from our 3 survey bouts (March 2017, November–December 2017, and March 2018), we built log-linear models to estimate the total population size in 2017–2018. To do so, we used the closed population functions of the R package wiqid (Meredith 2020). We fit a total of 4 models: 3 models that account for variation in capture probabilities from different sources, as well as a null model that does not account for variation (model M₀). The 3 sources of variation incorporated were 1) variation in each sampling bout (a temporal effect, model M_t), 2) variation between individual turtles (model M_h), and 3) variation due to behavior (capture probability changes after the first time the turtle is captured, model M_b) (Otis et al. 1978; Rivest and Baillargeon 2014; Meredith 2020). Of these, we selected the model resulting in the best fit (using the difference in the Akaike information criterion with a correction for small sample sizes [AICc], ΔAICc). Several models resulted in a warning that the asymptotic bias was too large, so for comparison, we refit models using bias correction using the closed.bc function, which implements bias correction via general estimating equations or frequency modification as described by Rivest and Levesque (2001).

We fit closed population models rather than open models because we did not have sufficient data for open population models. Given that the 3 sampling periods were within 1 yr, mortality and recruitment was likely a low proportion of the population, causing a relatively minor violation of the closed population assumption. To evaluate the sensitivity of the result to the different sampling periods, we also fit models using just the first 2 sampling bouts and just the second 2 sampling bouts. For these, we used the Petersen estimator with Chapman's (1951) modification as implemented in the closed.bc function (Rivest and Baillargeon 2014).

RESULTS

Postfire Sampling Results. — During the March 2017 surveys, we detected 11 live box turtles in 85.75 person-hrs of searching, or 0.13 box turtles/hr (5 female, 5 male, and 1 subadult male). Live box turtles were found in habitats across the island, but over half (n = 6, 54.5%) were found on the TBP, which did not burn. We marked a series of live turtles with notches 2402 to 2411. Of the live turtles detected, only 1 (F844) had been marked during earlier surveys by Dodd et al. between 1991 and 2006.

During the March 2017 surveys, we detected 259 dead box turtles, of which 148 (57.1%) were either too badly burned or deteriorated—or both—to estimate the likely cause of death (Fig. 3). At least 65 dead individuals (25.1% of the number found dead during March 2018) appeared to have died coincident with the 2016 fire (e.g., complete skeletons with burned skin or keratin; Fig. 4). At least 43 (16.6% of the total number found dead during March 2017) box turtles were found in unburned areas; these individuals appeared to have been predated by or scavenged relatively recently in a manner consistent with earlier descriptions of raccoon depredation (Fig. 5). Additionally, 1 dead juvenile box turtle was found stuck in a chain-link fence, and 1 adult male was found apparently crushed by machinery along a powerline right-of-way. Of the dead turtles detected, 115 (44.4%) were too badly deteriorated (marginal scute bones burned or missing) to determine with confidence whether the turtle had been previously marked. Sixty-six (25.5%) had been marked by Dodd and colleagues during the 1991–2006 sampling period; 77 (29.7%) were not marked prior to our surveys in March 2017. Four gopher tortoises also were observed dead within the 2016 wildfire area.

View Figure

• Download as Powerpoint
• Download Figure

View Figure

• Download as Powerpoint
• Download Figure

View Figure

• Download as Powerpoint
• Download Figure

Follow-Up Surveys in 2017 and 2018. — In November–December 2017, we found 54 box turtles (23 live; 31 new, dead individuals), and most of the live turtles found were in the TBP. A single live turtle was observed in a random plot within the 2016 wildfire area, with the rest observed in unburned areas at both the north and south ends of the island. Of the turtles found alive, 3 were recaptures from March 2017; 3 were recaptures of marked animals from 1991 to 2006; and 17 were new, unmarked turtles. Of the turtles found dead for the first time in November–December 2017, 3 were clearly identifiable from marks from 1991 to 2006; 6 were clearly marked during 1991–2006 but not identifiable to individual ID number; and 22 were apparently not previously marked. In March 2018, we found 30 box turtles (3 live; 27 new, dead individuals). All live turtles were observed in the TBP. Of the 3 turtles found alive, 1 was a recapture from both March 2017 and November 2017.

Living Turtles Recaptured From Earlier Studies. — We recaptured 3 living box turtles that had been marked during the original 1991–2006 study. Female 427 was initially captured as an adult on 23 April 1992 (straight carapace length [SCL] = 140 mm) and was recaptured on 29 November 2017 (SCL = 139 mm). Based on an average age of maturity of 7 yrs (Dodd 1997b), we estimate her age upon recapture to be at least 32.6 yrs. Female 844 was initially captured as an adult on 9 June 1994 (SCL = 124 mm) and was recaptured 29 November 2017 (SCL = 124 mm). We estimate her to be at least 30.5 yrs when recaptured. Male 2115 was initially captured as a juvenile on 24 August 1997 (SCL = 117 mm) and last observed on 24 May 2001 (SCL = 134 mm) before he was recaptured on 30 November 2017 as an adult male (SCL = 139 mm). Based on an estimated initial age of 5 yrs, we estimate his current age to be 25.3 yrs. The first 2 turtles exhibited no significant growth over the more than 16 yrs since their last capture.

Extrapolations from Sampling in the 2016 Wildfire Area. — In each of the eight 0.25-ha random plots within the extent of the 2016 wildfire sampled in March 2017, we detected 6–12 dead box turtles (mean = 9.63, 95% CI = 7.53–11.72) per 0.25-ha plot and 0–1 live box turtles (mean = 0.13, 95% CI = –0.13–0.38) per 0.25-ha plot during 15-min time-constrained surveys. Extrapolation across the entire 26.1-ha 2016 wildfire area, assuming an equal detection and distribution throughout, indicates there may have been 1004.9 detectable (95% CI = 786.4–1223.3), dead box turtles and 13 live, detectable box turtles (95% CI = –13.4–39.5). We walked along one 206-m, randomized east–west transect across the southern end of the 2016 wildfire area and detected 15 dead box turtles 0–7 m from the transect. Assuming equal detection to a distance of 7 m from the transect centerline, this equates roughly to 0.29 ha of search area, or 53.6 detectable turtles per hectare. The results from this limited transect survey extrapolates to 1356.6 detectable, dead box turtles and 0 live box turtles within the extent of the 2016 wildfire.

Extrapolations from Sampling in Other Areas. — In each of the 4 random, 0.25-ha plots situated within the roughly 19.4-ha forested area south of the 2016 wildfire, we detected 1–4 dead box turtles per plot (mean = 2.75, 95% CI = 1.05–4.45) and 0–3 live box turtles per plot (mean = 0.75, 95% CI = –0.95–2.44). Extrapolated across the entire 19.4-ha forested area south of the 2016 wildfire, we estimate 213.4 detectable, dead box turtles and 58.2 detectable, live box turtles.

Population Estimate. — The closed-population loglinear model from all 3 survey bouts that fit best based on AICc (Table 1) was the model that allows for different capture events to have different capture rates (M_t) (Rivest and Baillargeon 2014), with which we cautiously estimate a population size of 65.5 (95% CI = 41.6–149.1) live box turtles across all suitable, forested habitats on Egmont Key, including the 2016 wildfire area. Regardless of the model used, population estimates were all well under 120 animals (Table 1). Petersen estimates using only the first 2 or last 2 bouts yielded estimates of 56.6 (SE = 16) and 47 (SE = 18.8), respectively.

Table 1. Comparison of closed population log-linear abundance models for the box turtle population on Egmont Key, Florida, 2017–2018.^a

View Fullscreen

DISCUSSION

Live Turtles. — Our detection rate for live turtles in March 2017 of 0.13 turtles/hr is less than one-fifth of the lowest detection rate (0.69 turtles/hr) reported by Jones et al. (2016) from 4 long-term sites in Collier County, Florida, all of which supported lower box turtle population densities than the estimates for Egmont Key provided by Dodd et al. (2012). Our November–December 2017 live turtle detection rate was nearly 0.5 turtles/hr, closer to the detection rate observed Collier County. Our closed population estimate of 65.5 live box turtles based on capture–recapture results is consistent with estimates based on the extrapolation of live survey returns from random plots and represents approximately 4.3% of the maximum population estimate made by Dodd et al. (2012) for 2002 (Fig. 6). Without additional, repeated surveys under varying environmental conditions, both the live turtle detection rate and population estimate should be interpreted cautiously; however, these values are consistent with the observed numbers of dead turtles and are strongly suggestive of a significant and substantial population decline. Our study design and capture histories do not permit a quantitative evaluation of detectability. Still, to properly evaluate our findings, it is important to evaluate at least 2 aspects related to detection, including whether detection rates of live or dead turtles differed in burned and unburned areas, and also whether detection rates in 2017–2018 were intrinsically different from those in 1991–2006. In the first case, we expect that detection rates of both live and dead turtles are comparatively high in the recently burned areas, where most cover has been burned away and the mineral soil is exposed. We also expect that surviving turtles would have moved into unburned areas following the fire, elevating the densities in those areas and consequently increasing the number of detectable turtles. In the second case (i.e., other than the speculative changes in detectability associated with the recent fire) we do not see a reason to expect that intrinsic detection rates in unburned areas would be different in 2017–2018 as experienced by earlier researchers from 1991 to 2006. If they are different, the difference must be due in part to ongoing vegetation changes or to long-term climatic trends that our capture–recapture sample cannot adequately illuminate.

Dead Turtles. — Our extrapolated estimate of 1004.9 detectable dead box turtles within the extent of the 2016 wildfire represents approximately 67% of the island-wide peak population estimate in 2002 provided by Dodd et al. (2012) of about 1500 turtles. Many of the turtles that were found within the 2016 wildfire area died during the burn, but it is likely (and supported by our observations of dead turtles outside the wildfire area) that some undetermined number were already dead before the fire occurred, and burning exposed their shells. Including our extrapolated estimate of 213.4 dead box turtles in areas south of the 2016 wildfire equals a total estimate of 1218.3 detectable dead box turtles in the southern two-thirds of the island in March 2017. In 15 yrs of studying the Egmont Key box turtle population, Dodd and colleagues detected only 216 dead box turtles (although they did not search in severely burned areas). We observed more than this in our first 4 days of searching, and estimate that nearly 6 times this total number were detectable on the island in the spring of 2017, a full 6 mo following the July 2016 fire. Box turtle shells appear to persist for 2 yrs or more in this environment (Dodd 1995; C.K.D., unpubl. data, 1991–2006), and clearly for several years in the subtropical environments of Collier County, to the south (Jones et al. 2016; M.T.J., unpubl. data, 2006–2017).

Fire. — Fire is a demonstrated source of mortality throughout the range of T. bauri and its relative T. carolina (Bigham et al. 1965; Verdon and Donnelly 2005; Platt et al. 2010; Howey and Roosenburg 2013; Harris et al. 2020), although the portion of the population estimated to be affected is generally lower (e.g., 10%–22% during wet-season fires; Platt et al. 2010) than estimated here for the 2016 fire on Egmont Key. As evidenced by live, fire-scarred individuals in populations (Dodd et al. 1997; Howey and Roosenburg 2013; M.T.J., L.L.W., and J.D.M., unpubl. data, 2013–2019), as well as fire-scarred individuals in museum collections (Ernst et al. 1995), box turtles are capable of surviving low-intensity fires, and fires, particularly in Florida, are locally an important part of their ecology (Ernst et al. 1995; Platt et al. 2010). Low-intensity and small-scale burns probably exert limited influence on survivorship and population viability; Dodd et al. (1997) observed that 3.5% of the population on Egmont Key exhibited fire scars from previous fires, some of which may have occurred during a fire on the island in 1987. However, the scale and intensity of the 2016 fire appears to have exceeded what box turtles are capable of surviving; thus, the fire had a population-level effect. Therefore, managers should take into account size and intensity of fires, as well as the distribution of optimal box turtle habitat, when creating habitat management plans in areas where box turtles may be active or affected by fires year round (Platt et al. 2010).

Predators. — Franz and Dodd (1993) reported that a raccoon reached Egmont Key in 1991 and began to kill box turtles of all age classes. At least 26 box turtles were killed before the raccoon was removed in April 1992, and subsequently, evidence of mammalian depredation was unknown until the end of the study in 2006 (C.K.D., unpubl. data, 1991–2006). We found raccoon tracks in 3 distinct areas of the island, and 43 box turtles found dead in March 2017 appeared to have been killed or scavenged by raccoons. Several dead, unburned box turtles detected in portions of the 2016 wildfire area were likely killed by raccoons, suggesting that raccoons may have been hunting box turtles in the previous 8 mo.

Management Implications. — Our reassessment documents a substantial decline of a well-studied, relatively large, and historically stable box turtle population. We recommend precaution in future prescribed burns to avoid further catastrophic losses to this population. The intensity of the 2016 wildfire likely led to higher mortality rates than a lower-intensity burn might cause. Consequently, vegetation management to reduce fuels and prevent future catastrophic fires is necessary; however, doing so in the near term before the 2016 burned area has regenerated would likely result in even further losses to the box turtle population, because surviving box turtles appeared to be concentrated in unburned areas. In the future, prescribed burns would ideally be done within smaller treatment areas in order to prevent large portions of the island (and therefore the box turtle population) from being exposed at one time. In addition, seasonal timing and other management considerations might be appropriate to reduce mortality during prescribed fire events (Platt et al. 2010).

Although it is likely that the depredated turtles were eaten by raccoons, it is possible that some or all were eaten by an avian predator such as fish crows (Corvus ossifragus). However, raccoons seem to be the more likely predators because we found evidence of raccoons throughout the island, representing a substantial change from the earlier studies when raccoons were thought to be absent. However, corvids are well known to prey or scavenge on emydid turtles (Marchand 2019). Targeted research should confirm the source of depredation, as it likely played a synergistic role in the decline.

Gopher tortoise burrows would seem to offer a potential refugium for box turtles during severe wildfire. Indeed, box turtles elsewhere in Florida have been observed to use deep mammal burrows similar in structure to tortoise burrows (M.T.J. and L.L.W., unpubl. data, 2013–2016; Meck et al. 2020). Tortoise burrows are widely distributed across Egmont Key and would seem to be readily available to box turtles seeking escape from fire. Without direct behavioral observations during a fire, we cannot speculate whether tortoise burrows were utilized, or whether their density on the landscape would influence box turtle mortality rates, but it is a noteworthy area for future research.

Finally, collection of Florida box turtles has been documented from other Florida counties in recent years (J.D.M., unpubl. data, 2018–2019), and it is plausible that Egmont Key has been targeted by collectors, which may be yet another cause of the population decline on the island. Further studies should evaluate whether the mortality rates inferred from extrapolative random sampling is supported by additional capture-mark-recapture analyses of living turtles. Studies should also evaluate the demographic, behavioral, morphological, and ecological responses to this decline in population density. Specifically, considering the magnitude of rapid decline but evidence of a small remaining core of adult turtles, we suggest that it is important to measure and document the demographic response at intervals over the coming decades. It will be particularly interesting to evaluate whether the population exhibits density dependence in its demographic response to the population decline, for example, through an increase in recruitment.

Our findings are yet another reminder that ostensibly protected populations of freshwater turtles may decline as a result of stochastic, chronic, or synergistic sources of mortality, including both internal processes such as flooding or wildfire and external sources such as collection (Seburn 2003; Browne and Hecnar 2007; Enneson and Litzgus 2009; Howell et al. 2019). Additionally, populations at 4 protected Terrapene research sites exhibited declines over multiple decades (Williams and Parker 1987; Kiester et al. 1991; Hall et al. 1999; Nazdrowicz et al. 2008). Although these sites were protected, the surrounding landscape matrix had been compromised in each case. Egmont Key presents a unique dimension to this recurring narrative: as an island, it is less affected by some of the expected effects of surrounding habitat fragmentation, such as roadkill (although still affected by mesopredators and collection). Further, Egmont Key is a small, low island in the open Gulf of Mexico that has lost more than 40% of its total area to erosion in recorded history (Stott and Davis 2003), constraining the total area of available habitat. In the current context, the total box turtle population may be influenced by stochastic sources of mortality without the benefit of metapopulation dynamics or connectivity to other populations. Our results further illustrate the need for populations of nonlisted, yet vulnerable, freshwater turtles to be prioritized in protected reserves and monitored to detect the effects of stochastic, chronic, and synergistic sources of mortality.