Evidence for Recruitment-Mediated Decline in an Eastern Box Turtle (Terrapene carolina carolina) Population Based on a 30-Year Capture–Recapture Data Set from Maryland
Abstract
The eastern box turtle (Terrapene carolina carolina) population at the Jug Bay Wetlands Sanctuary, Lothian, Maryland, has been monitored continuously for 29 yr (1995–2023). We used open-population capture–recapture models (Jolly–Seber) to estimate annual population size, survival probability, and recruitment rate. The model allows for unknown sex of individuals and includes information on individuals found dead. Our analysis documented a long-term decline of approximately 67% in box turtle population size at the Sanctuary over this nearly 3-decade period. We estimated annual survival for both males and females, which does not show a systematic increase or decrease over time, averaging about 0.90 (95% CI: 0.86, 0.93) for females and 0.97 (95% CI: 0.94, 0.98) for males. Conversely, per-capita recruitment showed a marked decline over the first 15 yr of record, suggesting that population declines may be due to reduced recruitment. Conservation efforts for the species could benefit from a formal population viability analysis to understand the relative effects of survival and recruitment on changes in population size for this long-lived species.
The eastern box turtle (Terrapene carolina carolina) is a charismatic species that is widespread in eastern North America, even in the fragmented habitat matrix surrounding large cities. Although perceived to be relatively common (Roberts et al. 2024), it is considered a species of concern or species of greatest conservation need throughout its range (Erb and Roberts 2023) and is listed as Vulnerable by the IUCN (van Dijk 2011) due to habitat loss, road mortality, collecting for the pet trade, and other factors (Erb and Roberts 2023). Increasing attention is devoted to monitoring box turtle populations (Erb et al. 2015; Roberts and Erb 2023) and understanding causes and mechanisms of population declines (Nazdrowicz et al. 2008; Jones et al. 2021; Roberts et al. 2024).
Although some studies suggest substantial population declines over recent decades (Hall et al. 1999; Kemp et al. 2022), few contemporary studies document long-term trends in eastern box turtle populations. Kemp et al. (2022) is the most recent in the literature. They reported a decrease in population size of almost 75% over a 42-yr period for a population from southeastern Pennsylvania. However, their data were sparse over most of the period, which necessitated a relatively coarse-grained estimation of population size over 2 periods, the early (1978–1982) and later (2015–2022) parts of the record. Hall et al. (1999) documented significant long-term declines of about 80% in the number of individuals encountered in an intensively studied population (Stickel 1950) over a 50-yr period ending with a survey conducted in 1995. Their analysis did not use capture–recapture models, and they did not report population size or survival estimates. Nazdrowicz et al. (2008) reported a decline of about 76% (from 91 individuals to 22) in a small Delaware population over the period 1968 to 2002. Conversely, a recent study from North Carolina (Roe et al. 2021) found “no evidence of population decline at any site over a ten-year period (2008–2017)” across small, local populations, based on an analysis of mostly opportunistic data from a large number of sites. Because the site-specific datasets upon which their analysis was based are short-term, small, and heterogeneous, the analysis should be interpreted with caution. Dodd et al. (2012) analyzed a 16-yr study of the Florida box turtle (T. carolina bauri), showing a 5% per year increase in population size (Jones et al. 2021). Their study applied to a population that was isolated on an island with few predators (“lack of mammalian predators”; Langtimm et al. 1996) and thus not representative of typical mainland populations. That island population has since undergone a catastrophic decline due to fire and predation (Jones et al. 2021).
In this paper, we document the long-term population trend in a 29-yr capture–recapture dataset (1995–2023) of box turtles from Jug Bay Wetlands Sanctuary, Lothian, Maryland, using the Jolly–Seber model (Schwarz 2001). Our model uses encounters of individuals obtained by a combination of organized surveys and opportunistic or incidental encounters, and it includes information derived from individuals “found dead” by any means. The Jug Bay data represent consistent sampling of the same population, producing sufficient sample sizes to obtain annual population size, survival, and recruitment estimates. The long-term and contemporary nature of this dataset and the use of the Jolly–Seber model to directly estimate population size and demographic parameters provide a level of detail and statistical rigor lacking in many analyses of box turtle monitoring data.
METHODS
Study Area. —
The Jug Bay Wetlands Sanctuary (JBWS) covers about 688 ha on the Coastal Plain in Anne Arundel County, Maryland, along the eastern edge of the Patuxent River (Fig. 1). The population of box turtles at JBWS has been studied for 30 yr (Smithberger and Swarth 1993; Marchand et al. 2004; Swarth 2005, 2018; Savva et al. 2010; Therres et al. 2015). The Sanctuary is well staffed and protected: visitors must check in at the Wetland Center and, if needed, pay an entrance fee, thus eliminating the likelihood of illegal collecting. The Sanctuary is 1 of 8 other large parks and nature reserves that protect more than 4000 ha in this section of the Patuxent River (Fig. 1). The diverse habitat consists of extensive freshwater tidal wetlands, mixed upland deciduous forest, riparian forest, stream valleys, managed meadows, and adjacent agricultural fields. The area is drained by 3 semipermanent streams that flow into the river, including the respective floodplains. The “core study area” within which box turtle capture–recapture efforts were focused is 100 ha in size.
Citation: Chelonian Conservation and Biology: Celebrating 25 Years as the World's Turtle and Tortoise Journal 24, 1; 10.2744/CCB-1644.1
Lab and Field Methods. —
Turtles were encountered during organized visual searches (referred to here as “surveys”) and opportunistically by staff and visitors (referred to as “incidental”); we combine all such encounters together and refer to them as “visual encounters.” A survey involved walking slowly through habitats to inspect closely the ground cover, leaf litter, shrubbery, downed logs, and other places where turtles could be found. The survey data are not publicly available. A total of 175 organized surveys (an average of 22 each season) were also conducted from 2000 to 2008 by an average of 6 searchers in 7 different 1-ha plots within the study area. These surveys involved between 4 and 23 individuals intensively searching habitat. Incidental encounters of turtles were by staff and visitors conducting other activities (not specifically looking for turtles) and during telemetry relocation of turtles with VHF units. To accommodate such variability in search effort, we modeled capture probability to be year specific (see below).
Turtles captured in the study area were brought to the Wetland Center lab or processed in the field, where they were inspected, marked with a unique code using a triangular file (after Cagle 1939), measured, and sexed. A photocopy or photographic image was taken of each turtle’s plastron as a secondary identification technique to confirm the identity upon recapture. The straight-line carapace length was measured to the nearest millimeter with a pair of Haglof Mantax aluminum calipers. Sex was determined by assessing several secondary sex characteristics: presence or absence of a plastron depression, eye color, distance of cloaca from rear edge of carapace, and shape of the hind claws (Dodd 2001). After processing, each turtle was returned to the location where it was captured.
The dataset consists of 8038 total encounter records. We discarded 365 encounter records of individuals not in the “core study area” and 18 encounters of individuals < 80 mm carapace length, producing a dataset containing 7655 individual encounters, of which 4581 were telemetry encounters and the remainder were visual encounters by means other than telemetry (i.e., incidental and from organized searching). Starting in 1998, we attached radio transmitters to 108 different turtles to study their home ranges. As a result of tracking these “telemetry turtles” every season, we also encountered many other marked turtles that were not part of the telemetry study. Such encounters were used in our analysis as incidental encounters. Visual encounter records were used to create individual encounter histories, producing encounter histories of 572 individuals captured 1998 times by visual encounter. Of these individuals, 189 were female, 308 were male, and 75 were unknown sex (including 26 recorded as juveniles). A total of 74 previously marked individuals were found dead during the 29-yr study. This total includes 3 individuals found dead during telemetry work and 71 found incidentally. We include information from these “found dead” encounters to include the fate “dead” in the encounter history. Telemetry encounters were not used in the construction of encounter histories because nearly all individuals captured by telemetry in any given year were also encountered by visual search. In such cases, telemetry encounters do not provide additional information about Jolly–Seber model parameters.
The Jolly–Seber Model. —
Joint estimation of population size and demographic parameters (survival, recruitment) from discrete-time encounter history data when individuals cannot be detected perfectly is usually done using the Jolly–Seber (JS) model (Schwarz 2001). Because searching was done regularly between April and November each year, individuals may have been captured multiple times within the same year. For purposes of fitting discrete time Jolly–Seber models, we reduced encounter frequencies to binary events, yi,t, where yi,t = 1 represents a capture in a given year and yi,t = 0 represents not captured. Because survey effort was not recorded, we used statistical models for capture–recapture data that included year-specific capture probability, helping accommodate the variation in survey effort over time.
We use a Bayesian hierarchical formulation of the model (Royle and Dorazio 2008) using an R-based (R Core Team 2019) code template from Kéry and Schaub (2011) in which the model is fit in the JAGS software (Plummer 2003) using the jagsUI R package (Kellner 2015). This hierarchical or state-space formulation of the model (Royle 2008) describes individual survival, recruitment, and detection of individuals in terms of distinct submodels for the binary encounter observations yi,t and the individual binary state variables zi,t representing the “status” (alive or not alive) of each individual during each year or primary occasion.
A key idea of the Bayesian hierarchical formulation of the JS model that allows it to be conveniently fit in popular Bayesian analysis software such as JAGS is the use of data augmentation (Royle and Dorazio 2008) in which the observed encounter histories are embedded in a larger super-population of individuals including those that were not captured. Let M denote the size of this super-population. For each individual in the super-population that was not captured, the data are implied to be all-zero encounter histories, yi,t = 0 for all t and then the model is defined for all i = 1,2,…,M as follows. The encounter observations are assumed to be independent Bernoulli trials, conditional on the latent alive states, according to where pi,t is the probability of detecting individual i in year (or primary occasion) t, which may depend on individual- or year-specific covariates. Naturally, if an individual is not alive at time t, so that zi,t = 0, then the observation yi,t with probability 1, i.e., it is a deterministic 0. The alive states are binary (z = 1 for alive, z = 0 for not alive) and assumed to be Markovian for t = 2,…,T according to
For t = 1, , where ψ1 represents the fraction of the super-population of individuals that are alive at time t = 1. The variable Ai,t is a deterministic indicator of whether an individual in the super-population is available to be recruited just prior to primary occasion t, and it is a function of whether an individual has ever been alive, whereas the parameters γt are pseudo-recruitment parameters, representing the fraction of individuals available to be recruited that are recruited in year t. This value can be converted to a per-capita recruitment by tabulating the number of recruits each year as a step in the Markov chain Monte Carlo (MCMC) analysis and standardizing by the population size.
The pseudo-recruitment parameters were assumed to vary by year as fixed effects. We assumed an additive “year + sex” model for capture probability and survival probability so that and where I() is a dummy variable that equals 1 if individual i is male and 0 otherwise. Thus, α0 and β0 are the “male effect” on detection and survival, respectively. In addition, we assumed that αt (year effects for capture probability, on the logit-scale) were random effects, having a normal distribution: αt ∼ Normal(μα,σα). In this parameterization, μα is therefore the “female effect” on survival (on the logit scale). We could have chosen to model the yearly survival parameters as random effects also, but we believe the data can support estimation of yearly survival as fixed effects and we believe there is some biological interest in having “pure” estimates of survival not overly influenced by superfluous model structure.
Several other modifications to the basic model were made. First, we assumed p is year specific to accommodate variations in total effort over time and induced a weak stochastic constraint that the values of p are random effects from a common distribution by assuming that the logit-transformed yearly capture probabilities are normally distributed with mean μp and standard deviation σp. Imposing this prior distribution on detection probabilities allows the first and last detection parameters to be estimated. Second, because sex was missing for 75 individuals (26 juveniles and 49 unknown-sex adults), we assumed a prior distribution for the sex variable: . We coded the variable sex as a binary variable with sex = 1 representing a male and sex = 0 a female. Unknowns were coded in the data using the standard R representation of missing values sex = NA. Finally, to account for observed mortalities, individuals that were found dead, regardless of cause (74 in total), were regarded as deterministic encounters and contributed information only to the latent z states of the model. Specifically, during the year of dead encounter, subsequent z states are fixed at 0. As such, if an individual is found dead at some time, say, t, then zi,t+1 and all subsequent states are set to 0, so that mortality occurred either in year t or some previous year.
The 29-yr trend in population size was estimated as a derived parameter by computing the posterior distribution of the linear trend through the posterior samples of annual population size. Average survival probability for males and females was computed as a derived parameter by averaging the 28 annual values of survival for males and females. Per-capita recruitment was estimated from the model as a derived parameter, by summing up the number of recruits at time t (individuals that entered the modeled population at time t but were not alive at time t − 1) and dividing by the population size at time t − 1. We further averaged these annual estimates over 4-yr periods to reduce variation.
We fit the model in the JAGS software (Plummer 2003) using the R package jagsUI (Kellner 2015). Posterior summaries were based on 8 Markov chains run for 5000 iterations after a 1000-sample burn-in, with a thinning rate of 2, producing 20,000 total posterior samples. Convergence was assessed using the Gelman-Rubin “Rhat” statistic (Gelman et al. 2004). Values near 1 indicate convergence, and, in practice, values < 1.1 are usually regarded as satisfactory. See Royle and Dorazio (2008) and Kéry and Schaub (2011) for details on Bayesian analysis, MCMC, and the JS model.
RESULTS
The number of unique individuals captured each year is depicted in Fig. 2, which shows 1) a large increase in total counts over the first 5–6 years of the monitoring effort and 2) a large decrease in observed counts over the subsequent 20+ yr. The total count from around 120 to 40 in recent years shows a decline of roughly 67%. However, this pattern does not account for variation in survey effort across years. In particular, the large increase in number of turtles encountered over the first 5 yr (1995–2000) is probably a result of increasing effort as the study was initiated, although survey effort was not recorded. The least-squares linear trend (solid line in Fig. 2) fitted to the observed count produces a slope of −3.0685 captured individuals per year (t = −4.82, p < 0.0001), which is about −2.7% per year relative to the y-intercept of 111.85.
Citation: Chelonian Conservation and Biology: Celebrating 25 Years as the World's Turtle and Tortoise Journal 24, 1; 10.2744/CCB-1644.1
Population Size and Trend. —
The estimated annual population size of box turtles each year from 1995 through 2023 is shown in Fig. 3 along with the upper and lower bounds of the 95% Bayesian credible interval for population size, N, for each year. The estimated population size trajectory shows a marked decline from the range N = 300 to N = 330 in the early years to close to N = 100 individuals by 2023. Using N = 317 as the initial population size and N = 107 as the estimated population size in the last year, the population decline is about 66% over the 29-yr period. To account for the high level of uncertainty in the estimates for the early years, the 317 initial population size figure was based on the weighted mean of the first 5 yearly estimates with weights chosen to be inverse to the posterior variance. The posterior distribution of the least-squares fit of a linear trend through the annual population size estimates (Fig. 3) indicates approximately a 2.3% annual decline in population size (posterior mean = −0.023, SD = 0.0016). Thus, there is evidence of a long-term decline in the population size of box turtles at Jug Bay.
Citation: Chelonian Conservation and Biology: Celebrating 25 Years as the World's Turtle and Tortoise Journal 24, 1; 10.2744/CCB-1644.1
Model Structural Parameters. —
The main time-constant scalar parameters of the model are summarized in Table 1. The parameters male.surv and male.p are the effect of “being male” on survival and detection, respectively. We see a very large positive effect for males on survival (they survive at a higher rate) and a slightly negative effect on capture probability. The estimate of ψsex = 0.524, implying about 1.1 males per female enter the surveyed population (having a carapace length of > 80 mm given our data subsetting convention). Conversely, the sample proportion of males is 0.62 (308 males vs. 189 females in the sample), which is consistent with typical observed sex ratios. The sample sex ratio favors males because although they are slightly less detectable (male.p is negative), they survive at a much higher rate (male.surv is positive).
Capture Probability. —
The estimated annual capture probability estimates (posterior means and 95% credible intervals) for males and females over the 29-yr period (Fig. 4) show an increase in the early part of the monitoring effort, but generally annual capture probability fluctuates between 0.15 and 0.35.
Citation: Chelonian Conservation and Biology: Celebrating 25 Years as the World's Turtle and Tortoise Journal 24, 1; 10.2744/CCB-1644.1
Survival and Recruitment. —
Annual survival probabilities (Fig. 5, posterior means and 95% credible intervals) suggest low precision at the beginning and end of the series (an artifact of the Markovian structure of the model), and there is a visual appearance of a slight decrease in female survival over the last 15 yr. Average survival over the 29-yr period was 0.968 (95% CI: 0.942, 0.983) for males and 0.896 (95% CI: 0.855, 0.933) for females (Table 1; Fig. 6). The posterior distributions of per-capita recruitment summarized into 4-yr periods (Fig. 7, left panel) show higher per-capita recruitment in the early years of the study before declining rapidly, consistent with the pattern in the observed number of captured individuals having a carapace length between 80 and 120 mm (Fig. 7, right panel).
Citation: Chelonian Conservation and Biology: Celebrating 25 Years as the World's Turtle and Tortoise Journal 24, 1; 10.2744/CCB-1644.1
Citation: Chelonian Conservation and Biology: Celebrating 25 Years as the World's Turtle and Tortoise Journal 24, 1; 10.2744/CCB-1644.1
Citation: Chelonian Conservation and Biology: Celebrating 25 Years as the World's Turtle and Tortoise Journal 24, 1; 10.2744/CCB-1644.1
DISCUSSION
We analyzed a nearly 3-decade-long capture–recapture dataset on eastern box turtles from a protected site on the Coastal Plain of Maryland to estimate annual population size, trend, and demographic parameters (survival and recruitment). The estimated decline in box turtle population size that we found, about 66% over 29 yr in the Jug Bay Wetlands Sanctuary, is noteworthy because, to the best of our knowledge, our time series of population size estimates and survival probabilities are the longest and most complete for any population of eastern box turtles. Our observed population decline of about 66% is consistent with other reported long-term trend estimates: about 80% over 50 yr (Hall et al. 1999), 75% over 42 yr (Kemp et al. 2022), and 76% over 34 yr (Nazdrowicz et al. 2008). These long-term estimates are remarkably consistent, and, coincidentally, all represent populations in close proximity to one another in the mid-Atlantic region (Maryland, Pennsylvania, and Delaware, respectively). Taken together, our analysis and the aforementioned studies provide strong evidence of systemic declines in eastern box turtle populations in the mid-Atlantic region. Furthermore, these observed long-term declines are substantial and rapid declines and, at least for the population we studied, ongoing (Fig. 3).
Over the 29-yr period, we estimated that female survival averaged 0.896 (95% CI: 0.855, 0.933) and male survival averaged 0.968 (95% CI: 0.942, 0.983). We note that the posterior uncertainty (wider credible interval) of females is expected due to the lower sample size of females. There have been many estimates of eastern box turtle survival reported in the literature, from both telemetry and capture–recapture studies. Most studies report “overall” survival, which is constant over time and sometimes averaged over sexes, so our results may not be directly comparable to many published estimates, although they are reasonably consistent in magnitude for both sexes for comparable reports. Roe et al. (2021) reported 0.907–0.968 survival depending on the demographic group and lower survival for females consistent with our results. Nazdrowicz et al. (2008) reported an average survival (both sexes) of 0.813, 0.945, 0.951, 0.977 for 4 small populations; Currylow et al. (2011) reported overall survival of 0.962 for all adults; Brisbin et al. (2008) reported overall apparent survival of 0.954 (males and females). Our observed sex difference in survival (higher for males) is typical for the species and consistent with increased mortality due to nesting activity and larger space use of females (Habeck et al. 2019; Bulté et al. unpublished manuscript), which may result in some females crossing roads (Meck et al. 2024).
Survival probability appeared relatively stable over the study period, although there is some indication of a slight decrease in female survival since about 2010. We note that survival estimates are highly imprecise, which produces unstable point estimates with very diffuse posterior distributions (Fig. 5). Thus, the posterior means which show an apparent slight decline since 2010 may be artifacts of the sensitivity of the posterior mean as a summary of skewed posterior distributions. Although we believe this pattern is artifactual, adult female survival is known to be a limiting factor in turtle populations (Heppell 1998), and the loss of only a few females in small populations can negatively impact population viability (Howell and Seigel 2019).
Our estimates of effective per-capita recruitment show a steep decline from the start of monitoring in 1995 through the period 2008–2015 before rebounding slightly. The pattern is consistent with the observed frequency of capture of small-sized individuals between 80 and 120 mm carapace length. The pattern in recruitment, together with the lack of a systematic decline in survival, suggest that the overall population decline over this period may be due to reduced recruitment. The site is protected from development and unrestricted access and had a relatively stable habitat structure over the period of data collection. The causes for reduced recruitment are unclear, and we are unable to provide insight into possible causes for reduced recruitment levels in our study population. Researchers have identified many factors or threats that impact eastern box turtle populations, such as predation and road mortality (see Erb and Roberts 2023), but evidence for the relative importance of these factors is scant. Of the causes that might operate in our study site, predation on nests and young and disease are possible. Over the course of this study, 3 marked turtles were discovered killed by vehicles on roads. Finally, the Sanctuary is open on a limited basis to the public; however, we have never encountered a visitor collecting a turtle. Confirmation of this pattern in recruitment for other populations could enhance our understanding of possible causes of reduced recruitment in the species.
Our estimates of survival and recruitment parameters could be useful in aiding the conservation of this species. For example, they could be used to train models for population status assessments or population viability analysis (e.g., Moore et al. 2022). Additionally, consistent long-term data collection from other box turtle populations could strengthen the model. A challenge in monitoring populations of the eastern box turtle is that monitoring has to be long term to detect trends and patterns in demographic parameters, due to the high survival, low recruitment, and delayed sexual maturity of the species. While the eastern box turtle may be one of the most monitored reptile species in North America, with many efforts initiated at local sanctuaries, schools, and communities, few studies persist long enough to generate quality long-term data with sufficient sample sizes to estimate demographic parameters.
We applied the JS model to the long-term data from Jug Bay Wetlands Sanctuary. This model is used for inference about population demography from animal population studies in which individuals are detected imperfectly (capture probability < 1). Our results emphasize the importance of using statistical methods such as the JS model to interpret monitoring data. We were able to produce statistically qualified estimates of recruitment, sex-structured survival, and population size over a period of record that, to the best of our knowledge, is unmatched for this species. Accounting for imperfect and sex-structured capture probability shows the expected “adjustment” in sex ratio from near 1:1 “at recruitment” (80 mm carapace length) to strongly favor males (about 1.6:1) in the adult population at large, which is typical in box turtle populations (Dodd 2001). Allowing for capture probability to vary over time is important for interpretation of changes in counts when environmental conditions and effort also vary over time, as is most likely the case in the Jug Bay dataset, where we note that annual capture probability shows an increase from the initiation of the effort in 1995 for about 7 yr, remains relatively high until 2007, and then decreases again until 2021. This increase in capture probability from 2000–2007 corresponds to a period in which a number of organized survey efforts were made in which a group of 6–8 people intensively surveyed 1-ha plots within the core study area (some surveys were also done in 2008). Our estimates of population size are “free” from the biasing effects of this temporal variability in capture probability and thus interpreted directly as changes in actual population size.
Despite all the advantages of formal analysis using the JS model, a number of improvements in the data collection and analysis could be made. For example, we treated each year as a single sampling occasion, and thus an individual was captured (y = 1) if it had 1 or more captures in that year. This approach disregards many recaptures (some individuals were recaptured dozens of times in a year), but it is difficult to accommodate the recapture information without having more precise and curated information about spatial and temporal effort of sampling. One might consider dividing a season up into smaller subsamples within a year; this is called the “robust design” (Kendall and Pollock 1992). Although this approach may lead to more statistically efficient estimation, it could lead to additional problems in modeling temporal variation in capture probability, because effort is likely to be more heterogeneous within a year. In addition, the Jolly–Seber model used here is a nonspatial model, which was necessitated by the absence of spatially explicit information to describe the search effort. One consequence is that the population being sampled is not well defined spatially, because some individuals that might be captured may have home ranges on the boundary of the core study area (Royle et al. 2013). Spatial information about search effort and capture locations could be used in spatially explicit capture–recapture models (Royle and Turner 2022), which could greatly increase the precision of capture probability estimates and hence demographic parameters. However, for the Jug Bay dataset, there is not consistent information about search effort, as many of the encounters were opportunistic. In general, we recommend that monitoring efforts for turtles should attempt to record survey effort as much as possible, including both spatial and temporal effort.
Map of Jug Bay Wetlands Sanctuary core study area (thick polygon outline) in Maryland (created by Jug Bay Wetlands Sanctuary staff). Inset map shows the geographic region of the Sanctuary and core study area.
Number of unique individuals captured in each year from 1995 to 2023. Solid line is the least-squares fit.
Estimated population size 1995–2023 (black dots) and 95% posterior intervals (vertical lines). The number of unique individuals captured in each year is shown by the plus signs.
Estimated capture probability of box turtles at Jug Bay Wetlands Sanctuary for each year 1995–2023 with 95% Bayesian credible intervals (vertical lines) for females (black dot symbols) and males (plus signs). Male and female trajectories have a slight horizontal offset for visual clarity.
Estimated annual survival probability 1995–2023 of box turtles at Jug Bay Wetlands Sanctuary for males (triangles) and females (solid dots) with 95% Bayesian credible intervals (vertical lines). Horizontal black line is the value ϕ = 0.95. Credible intervals suggest low precision at the beginning and end of the series and higher precision for interior years, which is a consequence of the Markovian structure of the model and the “borrowing” of information through time.
Posterior distribution of 29-yr average survival of eastern box turtles over the period 1995–2023 at Jug Bay Wetlands Sanctuary for females (left) and males (right). Vertical line is the value ϕ = 0.95.
Left: Posterior distributions (shown as box plots) of average per capita recruitment per 4-yr period under the Jolly–Seber model. Estimates were averaged over 4-yr periods to reduce variation. Horizontal lines mark values of 0.10, 0.09, 0.08, and 0.07 recruits per adult individual. Right: Observed number of captures of individuals with carapace length >80 mm and <120 mm.
Contributor Notes
Handling Editor: Peter V. Lindeman
