Long-Term Trends in Ringed Sawback (Graptemys oculifera) Growth, Survivorship, Sex Ratios, and Population Sizes in the Pearl River, Mississippi
Abstract
Effective management of long-lived species requires demographic and life-history data that are best acquired from long-term studies. The ringed sawback (Graptemys oculifera), endemic to the Pearl River watershed of Mississippi and Louisiana, is a species of management concern at both the state and federal levels. Population sizes, trapping success, basking counts, sex ratios, survivorship, and growth of this species were investigated at 5 sites on the Pearl River in Mississippi over a 25-yr period. Estimates of age at maturity were 4.6 yrs for males and 9.1 yrs for females. Mean annual survivorship estimates for males, females, and juveniles were 0.88, 0.93, and 0.69, respectively. Maximum longevity estimates were 48.8 yrs for males and 76.4 yrs for females. Average longevity estimates were 8.5 yrs for males and 13.9 yrs for females. The sex ratio of captured turtles was male-biased before 2000 but unbiased after 2000. Realized population growth estimates indicated that 4 populations were stable over the 25-yr period and 1 population had declined. Population estimates and basking counts trended downward through time at most sites. Trapping success after 2000 for all sites combined declined by 77%, 45%, and 25% for juveniles, males, and females, respectively. Taken together, these data indicate that 1 population of G. oculifera has declined, 3 appear to be in the initial stages of decline, and 1 is relatively stable. Additional monitoring of these populations will be necessary to determine if these trends continue into the future.
Aspects of the population ecology and demography of long-lived species may only be understood through analysis of information from studies of sufficient duration (Likens 1983; Strayer et al. 1986; Pelton and van Manen 1996; Gibbons et al. 2000; Armstrong and Ewen 2013), as research conducted on a time scale that is short relative to a species' life span may generate data that lead to inaccurate estimates of the traits under study (Zweifel 1989; Pelton and van Manen 1996; Madsen and Shine 2001). Data from long-term studies are important not only for accurately estimating demographic parameters of long-lived species but also for their effective management (Lindenmayer et al. 2012; East et al. 2013). Given that many turtle species are in decline (Gibbons et al. 2000) and that demographic data, including adult survival estimates and population growth rates, are essential in making management decisions for those species (Converse et al. 2005), long-term studies are particularly important for turtles, which as a group may include some of the longest-lived vertebrates (Gibbons and Semlitsch 1982; Gibbons 1987).
The ringed sawback, Graptemys oculifera, is endemic to the Pearl River system of Mississippi and Louisiana, where it occurs primarily in the main channel of both the Pearl River and its largest tributary, the Bogue Chitto River. The species was listed by the US Fish and Wildlife Service as Threatened in 1986. Jones and Hartfield (1995) studied population size, growth, and age at maturity of this species at 5 sites on the Pearl River in Mississippi from 1988 to 1990, and Jones (2006) documented the reproductive biology of G. oculifera at one of these sites, Ratliff Ferry, in 1995−1996. Trapping at the 5 study sites continued periodically from 1994 to 2014 to investigate G. oculifera population size changes through time. These efforts also provided data on trapping success, growth, morphology, and survivorship of G. oculifera and on trapping success for Graptemys pearlensis, a poorly known species also endemic to the Pearl River watershed (Selman and Jones 2017). These data allowed a reassessment of growth and age at maturity of G. oculifera, provided estimates of survivorship and longevity, and permitted an examination of population size, sex ratio, and body size variation through time at each of the 5 sites. This information should not only help wildlife officials make informed decisions about the management of this species but will also provide a baseline for comparative studies in the future.
METHODS
The 5 study areas were described in Jones and Hartfield (1995) and Selman and Jones (2017). The most upriver site is near Carthage and the most downriver site is near Columbia (Fig. 1). Sampling occurred at Carthage and Columbia in 1989−1990, 1994, 2002, 2009, and 2014. Trapping at the other 3 sites took place in 1988−1990, 1994, 2002, 2008, and 2013, so each site was sampled over a period of 25 yrs.
Citation: Chelonian Conservation and Biology 16, 2; 10.2744/CCB-1268.1
Turtles were captured in open-topped, rectangular basking traps constructed of 2.54-cm hex wire poultry netting (chicken wire) or poultry netting coated with plastic (crayfish wire). Traps varied in size from 100 × 50 × 25 cm to 150 × 65 × 45 cm and were attached to limbs and logs used as basking sites by the turtles. Traps were monitored frequently during the day and at the approach of the monitoring boat, basking turtles usually jumped from basking structure into the trap. A basking trap of this nature requires frequent monitoring to be successful, because turtles that fall into the traps can escape within a few seconds by swimming out of the open top, in contrast to treadle basking traps (e.g., Ream and Ream 1966), which are more difficult to escape and thus do not require continuous monitoring.
The number of basking traps deployed per day varied from 28 to 30, so the midpoint of this range (29) was used to calculate capture effort. The number of hours trapped per day sometimes varied owing to the onset of inclement weather, rising water levels, or mechanical difficulties with the trapping boat, so a full day of trapping was considered to last for a minimum of at least 6 hrs. Days with < 6 hrs of trapping were eliminated from analyses of capture success to avoid biasing the results. Traps were moved periodically to minimize trap avoidance and to sample as much of the available basking structure as possible at a particular study site.
Trapping sessions during 1988−1990 are described in Jones and Hartfield (1995) and for the Ratliff Ferry reproductive study by Jones (2006). Sampling for the reproductive study was directed toward females as traps were attached to larger logs and limbs normally used by that sex as basking sites. Males and juveniles were likely underrepresented in these samples, so data from the 1995−1996 reproductive study are included in overall body size, growth, and survivorship analyses, but were not used in estimates of sex ratios, capture success, or population sizes.
Trapping from 1994 to 2014 occurred over a 5-d period at each study site. Captured turtles were sexed, measured, and permanently marked by drilling holes in the marginal scutes using the coding system of Cagle (1939) as was done in earlier studies (Jones and Hartfield 1995; Jones 2006; Selman and Jones 2017). If a captured turtle had been previously marked, the mark was recorded and the turtle was remeasured. Sex was determined based on the lengths of the foreclaws and the tail (Lindeman 2013). Small turtles that could not be sexed were classified as juveniles, but if sex could be determined upon recapture, they were included with the appropriate sex in subsequent analyses. Straight-line maximum carapace length (CL) was measured along the midline of the shell to the nearest millimeter with either tree or dial calipers depending upon the size of the turtle. Midline plastron length (MPL) was measured to the nearest 0.1 mm with dial calipers. Turtles were temporarily marked for population estimates by applying a spot of nontoxic red or orange waterproof paint to both sides of the carapace. The paint dried in less than 2 min and remained visible for up to 3 mo. There was no evidence that paint marks were lost prior to counts of turtles.
Recaptured turtles were divided into 4 classes based on their sizes at capture and at recapture. Class 1 turtles were first captured as juveniles (not sexable) or immatures (males, < 65 mm MPL; females, < 120 mm MPL) and recaptured as immatures, Class 2 were first captured as juveniles or immatures and recaptured as adults, Class 3 were adults smaller than the hypothesized asymptotic MPL (males, 78.21 mm; females, 134.08; Jones and Hartfield 1995) at capture, and Class 4 were adults larger than the hypothesized asymptotic MPL at capture. Only 3 males met the requirements for Class 1, and as these were within < 1 mm of 65 mm MPL, they were added to Class 2.
From 1988 to 1990, 2 basking counts/site were conducted to estimate population sizes at Carthage, Ratliff Ferry, and Columbia, and 4/site at Lakeland and Monticello (Jones and Hartfield 1995). From 1994 to 2014, 5 basking counts were conducted for each sample period at all sites. In all basking counts, the numbers of painted and unmarked G. oculifera were recorded. Turtles were counted primarily from concealed points on the river bank using binoculars or a spotting scope or from a slow-moving boat in the center of the channel when the river banks were steep or otherwise inaccessible. Population sizes for each site were estimated using both the Schnabel mark–recapture model (SM) and the Lincoln-Peterson model (LP) calculated as in Tanner (1978). SM and LP population estimates were made at Ratliff Ferry, Lakeland, and Monticello in 1988, at Carthage and Columbia in 1989, at all 5 sites in 1994 and 2002, at the first 3 sites in 2008 and 2013, and at the last 2 sites in 2009 and 2014. One LP estimate was generated per basking count but only 1 SM estimate was generated per study period per site, resulting in 5 SM and 22−24 LP estimates per site.
Kofron (1991) stated that male G. oculifera matured at an MPL of 60−70 mm and Jones and Hartfield (1995) used the midpoint of that range (65 mm) as an estimate of size at maturity. Females with an MPL ≥ 120 mm were considered adults based upon both a growth curve (Jones and Hartfield 1995) and reproductive data (Jones 2006). Sexable individuals smaller than these sizes were considered immature.
In an earlier study of G. oculifera growth using MPL as the body size variable, the von Bertalanffy (VB) growth model provided the best fit to recapture data (Jones and Hartfield 1995). Growth curves in that study were estimated for each sex by fitting data from turtles recaptured approximately 1 yr or more after their initial capture to the VB growth equation (Table 1) as defined in Frazer and Ehrhart (1985). Few juvenile turtles < 1 yr old at initial capture were recaptured during that study and the absence of data from very young individuals resulted in a poor fit of the recapture data to the VB growth model. To remedy the problem, growth increments in MPL were back-calculated using Sergeev's formula (Moll and Legler 1971) for 10 juveniles to provide growth data from hatching to age 1, which resulted in an increase in the coefficient of determination (R2) and thus a better fit to the VB growth model.
In the present study, CL and MPL were each used to generate separate growth equations for each sex. Both a and k were estimated by nonlinear regression of MPL and CL on the VB growth interval equation, and the hatchling length parameter b (Table 1) was estimated using hatchling length data from Jones (2006). As in the earlier study, few recaptured G. oculifera (2 males and 2 females) were in their first year of growth when initially captured. Back-calculated MPL growth increments were again generated from 10 juveniles and those MPLs were then used to calculate CLs for those individuals at age 1 using the equation CL = 1.092(MPL) + 5.04, based on the relationship between MPL and CL in juveniles of this species (F1,151 = 3685.14, p < 0.001). Growth curves were estimated for each sex using CL and MPL separately, first with recapture data only (FI and MI models) and then with recapture data combined with back-calculated growth increments (FII and MII models).
Survivorship (ϕ), recapture probability (ρ), and realized population growth (λ) of both sexes were calculated using recapture data with Cormack-Jolly-Seber (CJS) models for the first 2 parameters and a Pradel model (Pradel 1996) for the third (Program Mark; White and Burnham 1999, version 8.0). Both the CJS and Pradel models assume that ρ is equal for all animals in a population and that marked animals have equal ϕ from sample to sample (Pollock et al. 1990). These assumptions may be violated when data is pooled across different populations or sex/age classes within the same population (Lebreton et al. 1992), so to minimize potential bias in the estimates of both ρ and ϕ, these parameters were calculated for males and females separately in each of the 5 study populations.
Four models to estimate ϕ and ρ were evaluated for each sex at each site. Both ϕ and ρ were either held constant or were allowed to vary with sampling period. Models for each sex-by-site combination were evaluated using the quasi-Akaike's information criterion (QAICC; Hurvich and Tsai 1989), which estimates the best fit of the data using the fewest parameters. I used Program Mark to calculate QAICC weights, which provided an index of the likelihood of a model relative to other models in the analysis and can be interpreted as model probabilities (Burnham and Anderson 2004). Program Mark was also used to calculate median ĉ, a variance inflation factor that quantifies overdispersion of the data and is analogous to a goodness-of-fit test (Burnham et al. 1987). A median ĉ of 1.0 indicates a model that fits the data well while a ĉ > 3.0 indicates that the model does not provide an adequate fit to the data (Lebreton et al. 1992). Both the QAICC weights and median ĉ values were used to evaluate models. Parameter estimates were averaged using the model averaging routine of Program Mark. Models that exhibited parameters which were either poorly supported (low QAICC weights), confounded, or poorly estimated by the data (e.g., zero standard error), were eliminated from the analyses.
The Pradel model was used to estimate λ for each sex at each site. This model also estimates ϕ and ρ, and all 3 parameters were either held constant or allowed to vary with time, resulting in 8 models for each site-by-sex combination. Model suitability was evaluated by Akaike's information criterion (AICC) and AICC weights. The estimates of λ for the first interval in several models were problematic because of exceptionally high or low λ values or standard errors, so all first-interval estimates of λ from all models were eliminated from further consideration because of the potential for bias and confounding as suggested by White and Burnham (1999). The estimates of λ were averaged for all sites using the model averaging routine of Program Mark and the geometric mean of these estimates, their standard deviations, and their 95% confidence intervals (CI) were calculated.
Maximum longevity (ML) for each sex at each site was estimated using the following equation (modified from Litzgus 2006):
where ϕ = annual survival probability, and n = estimated population size. Average population size at each site over the course of the study was calculated from the SM estimates, but these represented both sexes combined. The sex ratios of males and females at each site, exclusive of turtles captured during the reproductive study at Ratliff Ferry, were calculated and were used with the population estimates to determine the number of each sex at each site. Average longevity (AL) for each sex at each site was estimated using the following equation (van der Toorn 1997):
where ϕ = annual survival probability. The instantaneous mortality rate (the inverse of AL) was estimated using the following equation (modified from Tucker et al. 2001):
where MR = instantaneous mortality rate and ϕ = annual survival probability.
Few G. oculifera first captured as juveniles were recaptured, which precluded an estimate of ϕ for juveniles from recapture data. This parameter can be calculated, however, if ϕ of adult females, mean age at female maturity, mean annual clutch frequency, and mean clutch size are known, by using the following equation (modified from Pike et al. 2008):
where ϕJ = juvenile annual survival rate, ϕA = adult female annual survival rate, CF = average number of clutches per year, CS = average clutch size, and FM = female age at first reproduction.
All statistical tests unless otherwise indicated were performed by Systat® (Systat 2009), and the figure was produced using DIVA-GIS (Hijmans et al. 2012) and Microsoft Paint. Sequential Bonferroni adjustments for multiple comparisons were calculated as in Rice (1989) with an experiment-wise error rate set at α = 0.10 (Chandler 1995).
RESULTS
Recaptures.
Almost 400 G. oculifera were recaptured at least 1 yr or more after their initial capture (Table 2). Most males (147, 86.5%) and females (168, 82.0%) were recaptured < 10 yrs after initial capture, 8 males (4.7%) and 25 females (12.2%) were recaptured between 10 and 20 yrs after initial capture, and 12 males (7.1%) and 8 females (3.9%) were recaptured ≥ 20 yrs after initial capture. There were no significant correlations between date and the number of recaptures (males: r = −0.174, p = 0.116; females: r = 0.140, p = 0.149; sexes combined: r = −0.129, p = 0.138), implying that the overall rate of recapture for all areas combined was relatively constant for the duration of the study.
Growth.
Growth rate as measured by change in CL or MPL was greater in the smaller size classes of both sexes (Table 2). Some individuals in Classes 3 and 4 apparently did not grow between capture and recapture, although this occurred more frequently in males. Four females in Class 2 did not grow and may have had either a lower than normal growth rate or matured at a smaller than typical size, as none had reached the hypothesized size at maturity (120 mm MPL) upon recapture. Although overall growth apparently declined in larger turtles, variability in growth appeared to increase with size. The coefficient of variation of change per year in CL and MPL was small in both sexes for Classes 1 and 2 but increased in the larger size classes, implying that variation in growth was substantially higher for larger turtles relative to smaller and presumably younger G. oculifera (Table 2).
Recapture data used with back-calculated CL or MPL growth increments provided a better fit to the VB growth model for both sexes than did recapture data alone, as regressions using only recapture data (MI, FI) had lower R2 values than those using recapture plus back-calculated data (MII, FII) in the analyses (Table 3). The CI for estimates of a for both sexes and both morphological variables using only recapture data and those using recapture data combined with back-calculated growth increments overlapped (Table 3), implying that the estimates did not differ regardless of the type of data used. Similar results occurred when estimating k in females but not in males (Table 3) as the CI for the estimates of k in the latter did not overlap, implying that they were significantly different.
Ages at Maturity.
Estimated age at maturity differed based upon which body measurement was used (Table 4), as those based on CL generated greater values in both sexes. Estimates of female age at maturity averaged slightly over 9 yrs and the CI of the estimates overlapped regardless of whether recapture data or recapture data combined with back-calculated growth increments were used (Table 4). Estimated male ages at maturity using only recapture data, however, were over 3 yrs greater than those from recapture data combined with back-calculated growth increments (Table 4) and their CIs did not overlap, indicating that they were significantly different.
CJS Models.
The model that best fit recapture data for both sexes at most study sites was one in which estimates of both ϕ and ρ were constant through time (Table 5). The best model for both sexes at Ratliff Ferry and for females at Columbia, however, was one where ϕ was constant but ρ varied with sampling period. The mean annual ϕ for males was approximately 88% and for females approximately 93%. The mean ρ for both sexes was generally low, at approximately 8% for males and 5% for females (Table 5).
Maximum and Average Longevity.
Estimated mean ML was almost 50 yrs for males and ranged from approximately 29 to 68 yrs (Table 6). Mean AL of males was much lower at almost 8.5 yrs and ranged from approximately 5 to 12 yrs. Estimated average ML for females was over 76 yrs and ranged from approximately 59 to 88 yrs, while mean AL was almost 14 yrs and ranged from approximately 10 to 16 yrs (Table 6). The longest period between capture and recapture for turtles first captured as adults was approximately 25 yrs in both sexes. If males mature at 4.6 yrs and females at 9.1 yrs (Table 4), these males at recapture were a minimum of almost 30 yrs old and the females were older than 34 yrs. These turtles may have been considerably older as there was no indication at initial capture in either sex that they had recently become mature.
Using an average adult female survival rate of 0.93 (Table 5), an average of 1.19 clutches/yr and an average clutch size of 3.17 (both from Jones 2006), and female age of maturity of 9.14 yrs (Table 4), estimated annual juvenile ϕ was calculated for each site (Table 6). Assuming that these parameters are similar throughout the Pearl River, mean estimated annual juvenile ϕ was 0.686 and mean AL was almost 2.7 yrs (Table 6), i.e., on average, a hatchling G. oculifera would survive 2.7 yrs. The mean instantaneous mortality rate for females was slightly more than half that of males, while the rate for juveniles was about 5 times that of females and almost 3 times that of males (Table 6).
Body Size Variation over Time.
For combined study sites, there were significant correlations between body size and date of capture for adult males (CL: r = 0.103, p < 0.001; MPL: r = 0.114, p < 0.001), adult females (CL: r = 0.152, p < 0.001; MPL: r = 0.178, p < 0.001), and immature females (CL: r = 0.134, p <0.001; MPL: r = 0.135, p < 0.001) but not for immature males (CL: r = −0.030, p = 0.794; MPL: r = −0.091, p = 0.425) or juveniles (CL: r = −0.073, p = 0.370; MPL r = −0.097, p = 0.235). All significant correlations were positive, indicating an increase in body size over time. Considered by study site, correlations were significant (p < 0.001) for all 3 groups at Ratliff Ferry (adult males: CL r = 0.166, MPL r = 0.145; adult females: CL r = 0.229, MPL r = 0.212; subadult females: CL r = 0.348, MPL r = 0.343) and for adult males at Columbia (CL: r = 0.262, p < 0.001; MPL: r = 0.212, p < 0.001) but not at any of the other sites. Adult G. oculifera of both sexes from Ratliff Ferry and adult males from Columbia captured after 2000 (the approximate midpoint of the study) were significantly larger than those captured earlier (Table 7). Immature females captured at Ratliff Ferry before 2000 did not differ in size from those captured after 2000, but this may have been due in part to the small sample size of immature females after 2000.
Sex Ratios.
The sex ratio for G. oculifera from all sites, excluding recaptures and those trapped during the 1995−1996 reproductive sampling, was male biased, as was the sex ratio for turtles captured before 2000, but for turtles captured after 2000 the sex ratio was 1:1 (Table 8). There was no difference between sexes in the numbers captured by month (Kruskal-Wallis H = 6.0, p = 0.423), so there did not appear to be a seasonal correlation between captures and sex ratio. Considered individually, 4 of the 5 populations had unbiased sex ratios before 2000 but all 5 populations had biased sex ratios after 2000 (Table 8).
Population Sizes, Basking Counts, and Trapping Success.
Both SM and LP population size estimates were negatively correlated with date at 4 of the 5 study sites (Table 9), implying that the numbers of G. oculifera at these sites declined through time. None of the SM correlations were significant, most likely because of small sample sizes (n = 5 in all cases) but 3 of the LP correlations were. Mean population size estimates were lower by as much as 31% after 2000 except at Lakeland, where estimates were over 50% higher after 2000 (Table 9).
The average density of basking G. oculifera observed during the study varied from as few as 14 individuals/km to as many as 90 individuals/km (Table 10). Average numbers of basking G. oculifera observed during the study were correlated with date at 2 sites but not at the others (Table 10). Basking counts before and after 2000 did not change or increased slightly at 2 sites, more than doubled at a third, but declined 25%−35% at 2 others (Table 10).
Captures of all G. oculifera combined declined by over 36% after 2000 (Table 11). The largest decline (> 77%) was among juveniles, followed by males (> 45%) and females (almost 25%). The largest overall decline was at Carthage, where there were over 50% fewer captures/day after 2000, much of this owing to a 66% decline in the number of captured males. Captures at both Ratliff Ferry and Columbia declined by about one-third, but at Lakeland there was a small increase in the number of turtles captured after 2000 (Table 12).
Pradel Models.
The λ estimates averaged slightly > 1.0 for both sexes separately and combined (Table 13), implying that on average, populations of both males and females and for all sites in the Pearl River were stable to slightly increasing over the duration of the study. Considered individually, all sites except Carthage, which appeared to have declined (Table 13), were relatively stable over the study, with λ estimates very close to 1. These changes were not necessarily concordant for the 2 sexes. Male populations appeared to be stable at all sites except Carthage, which decreased, while Monticello appeared to exhibit a slight increase (Table 13). Female populations at all sites were relatively stable, although Carthage, Ratliff Ferry, and Lakeland all had λ estimates slightly less than 1. The annual percentage of change in λ at each site and for each sex was < 1%/yr (Table 13).
DISCUSSION
Growth.
The VB growth model has been used in a number of studies, leading to a variety of opinions about how much and what types of recapture data are necessary to produce reliable estimates of the asymptotic value (a), the intrinsic growth factor (k), and age at maturity. Dunham (1978) stated that mark–recapture data were not appropriate for the VB model if all size classes were not well represented, including smaller and presumably younger individuals. Martins and Souza (2008) indicated that the absence of juvenile growth data when using the VB model resulted in underestimates of both a and k, which then led to incorrect estimates of age at maturity. Kulmiye and Mavuti (2005) hypothesized that the 2 parameters are inversely proportional, such that an underestimate of one results in an overestimate of the other. However, Frazer et al. (1990) found that although the omission of larger individuals of Trachemys scripta underestimated a and overestimated k, the absence of smaller individuals had little effect on those 2 parameters. The addition of data derived from juveniles based on their growth rings in the present study effectively added data from turtles in their first year of growth. As in an earlier study (Jones and Hartfield 1995), the added data improved the fit to the VB growth model for both males and females and had almost no effect on the estimate of a, but had a significant effect on the estimate of k in males. Without these added data, k was underestimated, resulting in an overestimate of age at maturity for that sex. These results may indicate that the VB growth model is more sensitive to the absence of recapture data from very young individuals in either those species or sexes that mature earlier and at smaller sizes than those that mature later and at larger sizes.
The estimates of k found here are similar to those of other species of Graptemys, which ranged from 0.264 to 0.498 for males and from 0.110 to 0.182 for females (Lindeman 1999). The recapture data, however, did not fit the growth model in the present study as well as in the earlier study (Jones and Hartfield 1995). This result likely was due to greater individual variation in growth rate among recaptured turtles over time in the present study (25 yrs) relative to the earlier one (3 yrs) and because of greater variation in growth among larger as opposed to smaller size classes.
Survivorship and Recapture Probabilities.
The CJS model that best fit G. oculifera recapture data for both sexes and from all study areas was one in which survival was constant, relatively high, and generally lower in males than females. Similar results have been found in other turtle studies (e.g., Tucker et al. 2001; Bowen et al. 2004; Eskew et al. 2010; Dinkelacker and Hilzinger 2014). The estimate of ρ at all sites, however, was generally lower for G. oculifera than for other species (e.g., Tucker et al. 2001; Ayaz et al. 2008; Martins and Souza 2008; Germano and Riedle 2015), although Páez et al. (2015) reported similar low ρ estimates for adult Podocnemis lewyana. Why recapture probabilities were so low in the present study is unclear. One possibility is that over time, G. oculifera, particularly marked turtles, may have learned to avoid traps. This seems unlikely as there were no significant correlations between date and the number of recaptured males, recaptured females, or all recaptured turtles combined. Chicken wire traps were used exclusively in the early years of the study and crayfish wire later as the latter is more durable and easier to use than the former. A crayfish wire trap, however, appears to be more visible when deployed than a chicken wire trap, particularly in clear water, and thus may have been recognizable as something to avoid, particularly to marked turtles. Davis and Burghardt (2012) documented long-term memory capabilities involving visual discrimination tasks in Pseudemys nelsoni and T. scripta and Soldati et al. (2017) found that Chelonoidis carbonarius were able to discriminate between visual stimuli representing differential reward values and retained that ability for at least 18 mo, so it is possible that marked G. oculifera may have learned to recognize and avoid traps.
Another possibility is that some of the turtles captured and marked were transients in the study areas and were not available for recapture in subsequent sampling periods. If that were the case, it is likely that their presence would have been reflected in low survival probabilities (Sasso et al. 2006) in each of the populations. Marked turtles might also have periodically moved short distances away from and then back to the study areas and thus were only periodically available for recapture. Shealy (1976) found what he described as significant movement of marked Graptemys ernsti away from a favorable trapping area but implied that these large-scale movements were to other favorable areas only a few hundred meters away. Although there are no published data on movement in G. oculifera, a radiotelemetry study of the closely related Graptemys flavimaculata found that mean home range lengths of males and females were 1.8 and 1.5 river km, respectively (Jones 1996). If G. oculifera has home range lengths similar to those of G. flavimaculata, then it is unlikely that substantial numbers of marked individuals moved outside of the 5 study areas because all were 4.8 km in length except that at Ratliff Ferry, which was 3.2 km long. Although some Graptemys have been recorded moving > 5 km (summarized in Lindeman 2013), there is no evidence that a large proportion of a population of any species periodically moves to new areas. It should be noted, however, that there are no movement studies in Graptemys that have lasted for longer than 1−2 yrs, so average movements over a period of time that is short relative to the lifespan of an individual may differ from movements over a longer time period.
The best CJS model for females at Columbia and for both sexes at Ratliff Ferry had ρ values that varied with sampling period, while in all other populations ρ values were constant. For Ratliff Ferry females, this likely was a result of the intensive sampling during the reproductive study at that site in 1995−1996 (Jones 2006). During that study, trapping directed at females was conducted 4 d/wk for 10 wks/yr, resulting in almost as many females captured and marked (559) as were captured during the rest of the study (608). Of the 559 captured in 1995−1996, 99 (18%) were recaptured, while only 57 (9%) of the 608 females captured during the rest of the study were recaptured. The large number of females captured during the reproductive study and later recaptured appears to have resulted in a CJS best fit model for females at Ratliff Ferry with a variable ρ. Removing females captured in 1995−1996 from the Ratliff Ferry analysis resulted in a best fit CJS model with a constant ρ, further supporting this hypothesis.
The variable ρ values for females at Columbia and males at Ratliff Ferry appear to have resulted from very low numbers recaptured at those sites during the latter part of the study. No marked males at Ratliff Ferry were recaptured after 1994, even though 504 males were marked there before 2002, 548 before 2008, and 577 before 2013. At Columbia, no marked females were recaptured in 2009 and only 1 marked female was recaptured in 2014, even though there were 191 and 201 marked females in that population in 2009 and 2014, respectively.
Many turtle demographic studies, including the present study, do not have recapture data suitable for reliable estimates of juvenile survival rates and longevity (e.g., Tucker et al. 2001; Martins and Souza 2008). In those studies that dealt specifically with juvenile survival (summarized in Iverson 1991; Shoemaker et al. 2013; Dinkelacker and Hilzinger 2014), estimates ranged from 0.48 to 0.70, so the average estimate calculated here (0.69) is within the range of estimates from those earlier studies.
Longevity.
The estimate of ML for male G. oculifera was approximately 64% of that of females and was similar to estimates calculated in the same way for Clemmys guttata, in which the ML of males was approximately 59% of that of females (Litzgus 2006). The variation in ML estimates among populations in the present study appears to have resulted from variation in survival estimates rather than in population sizes, as calculations of longevity were more sensitive to variation in the former than in the latter. Estimates of AL were overall much lower than ML estimates and support Gibbons' (1987) statement that although some individuals in a population may live a long time, most do not.
Sex Ratios.
Gibbons (1990) hypothesized that most variation in turtle sex ratios could be ascribed to 1 of 4 causes: 1) the initial sex ratio of hatchlings, 2) differential mortality between the sexes, 3) differential emigration or immigration rates of the sexes, or 4) differences in ages at maturity. He considered the last to be a primary cause of biased sex ratios in some turtles, including species like G. oculifera, in which males mature at an earlier age and would thus be expected to outnumber females, other factors being equal, in a population sample. Prior to 2000, both the overall sex ratio and that of 1 population in the present study were male biased while the other 4 populations were unbiased. After 2000, the overall sex ratio became equal and individual study area ratios were female biased (3 populations) or male biased (2 populations), implying a relative increase in females in some areas later in the study.
Differences in ages of maturity between the sexes may have been partly responsible for some of the observed sex ratios, particularly at Monticello, which was male biased both before and after 2000. However, the changes in all populations after 2000 were likely influenced by additional factors. Although not yet experimentally confirmed, it is probable that the sex of hatchling G. oculifera is determined by incubation temperature. In most species of Graptemys, cooler incubation temperatures produce males and warmer temperatures produce females (Bull et al. 1982; Ewert and Nelson 1991; Ewert et al. 1994). If warmer incubation temperatures occurred later in the study, more female than male hatchlings may have been produced, thus altering observed sex ratios after 2000. However, there were few hatchlings captured after 2000, implying either a decrease in nesting success or in hatchling survival, and without an influx of younger turtles, it appears unlikely that increased incubation temperatures could have influenced sex ratios in these populations, nor does it explain the relative increase in the number of males at Columbia. It is also unlikely that the sexes had differential emigration or immigration rates based on what is known about movement in the closely related G. flavimaculata as discussed earlier. The CJS models indicated that male G. oculifera generally had a lower survival rate than females, which in the absence of recruitment, could result in a female-biased sex ratio over time. However, given the relatively long potential lifespan of male G. oculifera relative to the length of the study, it is unclear whether a simple difference in average survival rates between the sexes would have acted quickly enough to have accounted for the changes in sex ratios after 2000. It seems more likely that these changes resulted from either a decrease in the numbers of males or an increase in the numbers of females in these populations after 2000. The numbers of both sexes captured per day declined during the study, so it is unlikely that female-biased sex ratios resulted from an increase in female numbers. The largest declines among captured sexable turtles overall and in most populations during the study were among males, so it appears that the sex ratio changes were more likely owing to fewer males present after 2000. Dorcas et al. (2007) found a change to a female-biased sex ratio in Malaclemys terrapin that was attributed to the effects of differential bycatch by crab traps selective for the smaller males and differential predation on females was suggested as the cause of biased sex ratios in Gopherus agassizii (Esque et al. 2010). In the present study, the changes in sex ratios may have resulted from greater rates of predation on the smaller male G. oculifera.
If, as hypothesized, increased predation on male G. oculifera was responsible for the changes in sex ratios, what species may have been responsible? It is unlikely that any fish in the Pearl River would have been involved, as none of the known species in the watershed are large enough to present a significant threat to immature or adult male G. oculifera. Males are rarely found on land and are not usually exposed to predation from terrestrial mammals. The only aquatic mammal that occurs in the Pearl River watershed which is large enough to be a predator of G. oculifera is the northern river otter (Lutra canadensis). Otters are known predators of turtles (e.g., Liers 1951; Lanszki et al. 2006; Ligon and Reasor 2007; Platt and Rainwater 2011; Stacy et al. 2014); their impact on populations may range from very low (e.g., Greer 1955; Noordhuis 2002; Roberts et al. 2008) to moderate (e.g., Manning 1990; Brown et al. 1994; Roberts et al. 2008) to substantial (Brooks et al. 1991). River otters, however, are not common in the Pearl River and neither they nor their tracks, scat, or slides were observed in any of the study areas, so it is unlikely that they caused the decline in male G. oculifera. Bald eagles (Haliaeetus leucocephalus) also prey on turtles (Clark 1982; Grubb 1995), including Graptemys (Mabie et al. 1995) but there are fewer than 10 breeding pairs of this species in the entire Pearl River watershed in Mississippi (N. Winstead, pers. comm., August 2017), so it is unlikely that they had an impact on G. oculifera populations.
A predator that did increase during the study was the American alligator (Alligator mississippiensis). When the study began in the late 1980s, alligators were listed as endangered in Mississippi and were infrequently encountered during sampling. Populations of A. mississippiensis have increased substantially since then to the point that the species was removed from the state endangered species list and a sport hunting season was initiated in 2005. Lindeman (2008) hypothesized that alligators prefer more lentic waters than those normally occupied by Graptemys, but in the Pearl River both species were frequently observed in the same riverine habitats during the latter part of the study. Alligators are known predators of turtles (Valentine et al. 1972; Delany and Abercrombie 1986) and would presumably capture and eat G. oculifera if they were available. This would likely impact not only males but immature females and juveniles as well, but would not impact adult females to the same degree, as the latter are larger and thus would be more difficult to eat for smaller alligators. Females also appear to be more wary and able to swim more rapidly than either males or juveniles (R.L.J., pers. obs.). Although there are no studies of their diets in the Pearl River, alligators may have had a significant impact on the population dynamics and life history of G. oculifera, as hypothesized for other species of turtles by Semlitsch and Gibbons (1989).
The number of female G. oculifera captured per day at Columbia declined by almost 60% after 2000, which was over twice the rate of decline in male captures. This was contrary to what occurred in other populations, where male decline was greater than that of females. The reason for this is unclear. It is unlikely that predation was involved because that would require a predator that would substantially reduce the female population at Columbia but not at the other study sites. The low numbers of females captured at the Columbia site, as discussed for female G. pearlensis (Selman and Jones 2017), may have resulted from geomorphic changes to the river channel resulting in less than suitable habitat for females, through increased recreational usage of this part of the Pearl River, or because there may have been illegal collecting directed primarily toward females at this site.
The decline in captured juveniles was greater than that in either males or females, which may have resulted not only from increased direct predation but also from the effects of increased nest predation. Major nest predators of G. oculifera are raccoons (Procyon lotor), armadillos (Dasypus novemcinctus), and fish crows (Corvus ossifragus; Jones 2006). These are all subsidized predators, i.e., predators whose populations have benefited directly or indirectly from human activities resulting in population densities higher than natural levels (Gompper and Vanak 2008). Fish crows and armadillos expanded their ranges in Mississippi prior to 1995 (Jones 2006) and both of these species as well as raccoons appear to have increased their populations since then, so it is possible that they are now having greater impacts on G. oculifera nesting success. Increased recreational activity at some sites, particularly Ratliff Ferry, may also have impacted nesting success and juvenile recruitment. Sandbars are the primary nesting habitats of G. oculifera (Jones 2006), but are also prime recreational sites on the Pearl River, and increased human use of sandbars has been associated with reduced nesting attempts by G. oculifera (Jones 2006). Recreational use of sandbars often results in the deposition of food refuse from picnicking and camping. These sources of anthropogenic foods may attract fish crows and raccoons, so increased recreational use of the Pearl River sandbars may not only result in reduced nesting attempts but also increased nest predation by attracting additional predators to nesting habitat.
Population Trends.
Based on the λ estimates, the Pearl River population of G. oculifera as a whole remained relatively stable over the course of the 25-yr study, as have populations at 4 of the 5 study sites. It should be noted that λ estimates are summaries of what happened in the past rather than forecasts of what will happen in the future. Other lines of evidence, including those parameters that were statistically different over time and those that, although not statistically different, exhibited trends that were consistent over time and across sites, may be necessary to predict the fate of Pearl River populations of G. oculifera. Population estimates and basking counts, for example, generally trended downward at all sites except Lakeland. Trapping success also trended downward at all sites, including Lakeland, but the decrease there was relatively small compared with other sites. The downward trend in trapping success was not uniform across age and sex classes, as the largest decreases were in juveniles, followed by males, and then by females, nor were these trends concordant among sites. Additionally, males and females at Ratliff Ferry and males at Columbia captured after 2000 were significantly larger than those trapped before that date, implying that those 2 populations, particularly at Ratliff Ferry, are composed of older individuals.
Based on these trends, it appears that 4 of 5 study populations of G. oculifera in the Pearl River have either already declined (Carthage) or are in the early stages of decline (Ratliff Ferry, Monticello, Columbia), while the population at Lakeland appears to be stable to increasing. Dorcas et al. (2007) found an overall decline in several populations of M. terrapin, a shift to larger body sizes and older turtles, and a change to a female-biased sex ratio, all changes that are very similar to what was found in the present study. In both studies, the primary cause appears to be differential loss of individuals related to body size, in the first case based on trapping mortality and in the present case because of predation, presumably by alligators, which were once nearly extirpated in the Pearl River but which now have recovered and are assumed to be approaching historical levels. There are no data, however, on population levels of G. oculifera prior to the decline of the alligator in Mississippi. The assumption that G. oculifera population sizes observed in the late 1980s were normal for the species may be a hypothesis analogous to the shifting baseline syndrome in fisheries (Pauly 1995; Pinnegar and Engelhard 2008), where base stocks observed at the beginning of a fisheries scientist's career are assumed to be normal for a species and are used to evaluate subsequent changes. In the absence of alligator predation prior to the 1980s, did G. oculifera populations increase to higher than normal levels? Is the decline of G. oculifera observed in the present study evidence of population levels adjusting to the increased numbers of alligators, and if so, will those populations stabilize in the future?
Although the present study resulted in a refinement of the estimates of ages at maturity for G. oculifera and provided information on survivorship and changes in population size over time, there are still questions about the biology, behavior, and life history of G. oculifera that remain unanswered. Periodic trapping would be helpful in determining how long marked turtles survive in the Pearl River and should be initiated 1−5 km above and below the designated study areas to see if marked turtles may have moved away from those areas over time. Tests of different types of basking trap construction would be useful in trying to determine if turtles have learned to identify and avoid certain types of traps. A radiotelemetry study of G. oculifera would assist in determining home range sizes and short-term movements in the species. Nesting at Ratliff Ferry should be reinvestigated to determine if nest predation rates have remained the same or, as suspected, increased because of an increase in subsidized predators correlated with increased recreational use of sandbars. Reproduction should also be monitored at one or more of the other populations to determine if and by how much nesting success differs from that at Ratliff Ferry. Food habits of alligators in the Pearl River should also be investigated to determine if they actually do eat significant numbers of G. oculifera.
Graptemys oculifera, although still relatively abundant, is slowly declining over much of its range in the Pearl River. One of the difficulties in managing a species like G. oculifera is that a long lifespan in concert with a slow decline may mean that population and demographic changes are not conspicuously evident. Wildlife managers, particularly those accustomed to working with relatively short-lived species, may not immediately notice these changes and assume that the species is stable. Continued monitoring of Pearl River G. oculifera over the long term will be important in determining whether the apparent decline continues or if populations stabilize at some point in the future.
Locations of the study areas on the Pearl River, Mississippi.
Contributor Notes
Handling Editor: Peter V. Lindeman
