Life-history studies and demographic data provide important information necessary to develop and assess conservation plans for declining reptile species (e.g., Congdon et al. 1993; Heppell 1998). Basic information about population size is important, but so are data on population growth, juvenile survival, age of sexual maturation, and individual growth rates. When combined with knowledge about habitat and environmental conditions, knowledge of individual growth rates enhances an overall perspective of a species' ecology.

The first thorough study of growth and maturation of ornate box turtles (Terrapene ornata ornata) was by Legler (1960), who stated that growth in Kansas was variable and size was of little value as an indicator of age after the first year. Legler's (1960) concern was partially based on variation in size within age classes of turtles, and he noted that ornate box turtle growth was positively correlated with average monthly precipitation in the growing season along with relative abundance of food. Ornate box turtles also grew slower in years when cool springs delayed emergence from overwintering (Legler 1960). Rhen and Lang (1995) also documented that hatchling growth was affected by incubation temperature in the common snapping turtle (Chelydra serpentina). In addition, Dodd and Dreslik (2008) noted that habitat disturbances differentially affected individual growth as related to sex and age for the Florida box turtle (Terrapene bauri) and the effect of food as a density-dependent factor affecting turtle growth, even after maturity, has been noted in several studies (e.g., Legler 1960; Gibbons 1967; Bjorndal et al. 2000; Chaloupka et al. 2004; Kubis et al. 2009).

For ornate box turtles, monthly temperature and precipitation vary yearly as well as over the species' range (Dodd and Dreslik 2008). While largely a Great Plains species associated with Kansas and Nebraska, ornate box turtles are found from southern South Dakota and southern Wisconsin, east to Indiana, west to eastern Colorado, and south to Louisiana and eastern Texas (Dodd 2001; Redder et al. 2006). Given this geographic range, in which most states provide protection for this species (Redder et al. 2006), differences in growth and maturation are of interest for comparative studies and conservation efforts.

Since the work of Legler (1960), several studies reported size and/or growth of adult ornate box turtles (e.g., Metcalf and Metcalf 1985 [Kansas]; St. Clair 1998 [Oklahoma]; Converse et al. 2002 [Nebraska]; Quinn et al. 2014 [South Dakota]) and the closely related desert box turtle (T. o. luteola [Germano 1988, 2014; Nieuwolt-Dacanay 1997]). In Iowa, ornate box turtles are a state-threatened species with only 2 large, reproductive populations (Richtsmeier et al. 2008). There is a need for growth and maturation data to evaluate the long-term population demographics of ornate box turtles in Iowa from hatching to adult. The species is difficult to detect (Refsnider et al. 2011), and we found hatchling and small turtles to be hard to locate because of dispersal from nest to cover, predation, low detectability due to small size, disruptive coloration, and a tendency to remain still while buried in vegetation and soil (see also Dodd 2001; Converse et al. 2002).

By focusing efforts on capturing and measuring hatchling and juvenile turtles and incorporating our data on adult turtles, we modeled growth of turtles captured from 1993 to 1996 and 2008 to 2016 and compared our results with other studies on ornate box turtle growth and maturation. Further, we also compared our growth estimates with those of ornate box turtles from South Dakota (Quinn et al. 2014) and desert box turtles from New Mexico (Germano 2014). Along with comparative data, we hope the models can be used by land and preserve managers to better assess age structure and environmental impacts on individual growth and maturation in ornate box turtle populations as related to changing or deteriorating habitats, information that can be used in developing long-term management plans to conserve the species.

METHODS

We captured and measured ornate box turtles at the Hawkeye Wildlife Area in Johnson County, Iowa, with most measurements occurring within 2 subpopulations: Mallard Pools and Greencastle. Mallard Pools and Greencastle are < 1 km apart with no apparent barriers to migration. Although both areas consist of sandy soils that accumulated south of the Iowa River during postglacial flow, no turtles have been found moving between the areas (Bernstein et al. 2007; Richtsmeier et al. 2008). Box turtles were first marked from 1993 to 1996 (R. Rhodes II, unpubl. data, 1993–1996) and we continued marking turtles at this site in 1997 and 1998, although we did not measure turtles from 1999 to 2007. We resumed random searches in 2008 and from 2013 to 2016, we dug approximately 600-m drift-net fences of metal hardware cloth (approx. 1.5-mm mesh) 4–6 cm into the ground in areas of known turtle nesting and overwintering (Bernstein and Black 2005). To stabilize the hardware cloth 15−20 cm above the ground, we stapled or wrapped the cloth around sticks and propped sticks along the fence. We dug 19-l plastic buckets into the ground approximately every 5–10 m so that the lip of the bucket was flush with the surface and an edge of the fence. We checked the buckets daily from mid-May to early July, at which time we removed the fence and buckets. From 2010 to 2016, we also collected data on spring-emerging hatchlings from nests protected from predators during the previous nesting season. We notched marginal carapace scutes with a hacksaw blade using a system adapted from Cagle (1939) and determined sex by a combination of examination of cloacal location and thickness of inner claw on the hind foot. Sex could not be discerned in hatchlings and small turtles (Quinn et al. 2014). We used Vernier calipers (± 0.1 mm) to measure midline carapace length (CL), maximum carapace width (CW), midline length of both front and rear plastra (FP and RP, respectively), width of plastral hinge (PW), and maximum shell depth (D), and counted major plastral growth rings on the least worn scute. Legler and Vogt (2013) recently addressed the difference between major and minor growth rings. Rhodes counted > 20 growth rings by examining the margins of scutes with a hand lens. Our approach was more conservative and we rarely recorded turtles with > 20 growth rings because older turtles displayed well-worn plastra. From 2013 to 2016, we brought hatchlings and juveniles into the laboratory for attachment of radiotransmitters and photography for other studies. Transmitters were Holohils LB-2X (0.34−0.37 g), BD-2X (0.39−0.45 g), or BD-2 (0.95−1.6 g); or Advanced Telemetry Systems F1514 (0.3 g), R1614 (0.36 g), R1655 (1.1 g), or R1680 (3.6 g). Transmitters were affixed with 5-min epoxy and never exceeded 2.0% of body weight. We measured these turtles in the laboratory, examined their growth rings with a 10–15× dissecting microscope, and returned them to their last field location within 24 hrs. While in captivity, we housed turtles separately, provided water and food, and kept them at room temperature (American Society of Ichthyologists and Herpetologists 2004).

Two major climatic disturbances occurred during our study. Rhodes measured 77 adults in autumn 1993 after the Greencastle area flooded during the summer. Greencastle also flooded in summer 2008, but we had fewer than 10 adult measurements from Greencastle after that time because poachers had collected most of that population. Hatchling and juvenile measurements were from Mallard Pools, which did not experience flooding, but adult data were from both Mallard Pools and Greencastle subpopulations.

We calculated a correlation matrix for all parameters except growth ring count and mass. Pearson's product-moment correlations indicated CL was the best predictor for all measurements (0.93 < r < 0.99); therefore, to avoid autocorrelation, we used only CL in subsequent growth models. Over the years, multiple researchers measured turtles, transcribed field data sheets to the computer, and counted growth rings; as a result, there were errors and inconsistencies as to what constituted a major growth ring and data entry errors. Some inconsistencies could be corrected from photographs taken on the day of capture, but before growth modeling, we examined measurement errors for CL by assuming that (FP + RP)/CL ∼ 1, plotted (FP + RP)/CL vs. CL, and eliminated 36 measurements with ratios outside of 0.8–1.2 (0.6–0.7; 1.3–10) or measurements for individuals without PL measurements. The final data set consisted of 606 measurements of 510 individuals (231 males, 226 females, and 53 unknown sex). We could not be certain of sex for individuals with ≤ 4 growth rings, so these unknowns were included in both the female and male growth models. If there were multiple measurements for the same turtle in a year, we used the first measurement because the carapace of some hatchlings and juveniles recaptured in the same year exhibited growth. We used Mathematica (Wolfram Research 2017) to fit 7 nonlinear growth models of CL vs. growth rings: 3-parameter logistic, 4-parameter logistic, 3-parameter Gompertz, 4-parameter Gompertz, von Bertalanffy, Richards, and exponential (Fabens 1965; Ricklefs 1967; Brisbin et al. 1987; Chen et al. 1992; Henderson and Seaby 2006; Tjørve and Tjørve 2010, 2017). We examined Akaike Information Criterion (AIC) analysis for the most parsimonious/best-fit model and compared the fit of the nonlinear models with a linear model to calculate deviance. To address concerns raised by Wilson et al. (2003) regarding variance for age estimates based on counts of growth rings, we calculated 95% confidence bands (confidence in curve fit) as well as 95% prediction bands (uncertainty of value of new data) on the best-fit curve.

Based on CL and the ratio of length of abdominal scutes to length of plastron, Legler (1960) identified 4 growth stages: 2 juvenile stages based on size (CL < 50 mm and 50 ≤ CL ≤ 69 mm, respectively), subadults (70 ≤ CL ≤ 100 mm), and adults (CL > 100 mm). Males and females were distinguished only for adults. Using our best-fit model for males and females, separately, we defined the same four growth states according to properties of the growth rate of CL vs. growth rings (i.e., the derivative of the model function). Subsequently, we used these growth stages on the male and female growth models, separately, and compared growth stages with those of Legler (1960). Unlike Legler (1960), who noted hatchlings emerging from nests and feeding in the same season that eggs were laid, we considered hatchlings as individuals who hatched from eggs in the autumn, overwintered in the nest, and emerged to the surface in the following spring in late May to early June. Converse et al. (2002) also speculated that some ornate box turtle hatchlings in Nebraska may have emerged from the nest in autumn. We have few complete data on hatchling growth over a season because of transmitter failure, transmitters falling off, or predation. In 2016, we were able to analyze first-year plastral and carapace growth every 20–25 d from spring to mid-September for 8 hatchlings.

We plotted change in CL for 131 recaptured individuals (45 males, 57 females, and 29 unknown sex) for which we had ≥ 2 measurements from different years. Similar to Germano (2014), age was estimated from turtles initially captured in the 1990s and recaptured after 2010 by adding number of elapsed years to original growth ring count. We averaged the CL of young turtles with 0−3 growth rings for which we had repeated measurements every 20−25 d for the year.

RESULTS

Of the 7 different growth models we used to examine the relationship of CL to number of growth rings, the Richards growth model resulted in the lowest AIC value for females and males (Tables 1 and 2). The Richards models provided a good fit, with all but 17 measurements for females and unknowns (8%; Fig. 1a) and 9 measurements for males and unknowns (4%; Fig. 1b) within a 95% prediction band of the models. Of the female outliers, 6 were recorded for turtles with fewer than 10 growth rings and the remaining 11 in older turtles (Fig. 1a). For males, 4 of the outliers were recorded for turtles with fewer than 10 growth rings and the remaining 2 in older turtles (Fig. 1b). For both females and males, the 95% prediction band expanded between 8 and 14 growth rings, indicating more variation in growth; however, there were low CL measurements for both males and females of individuals with > 15 growth rings that were potential errors, which we could not remove objectively (Figs. 1a, 1b). With multiple individuals measuring shells and counting growth rings, the source of some outliers could have been human error, but there also could be individual variation in growth. However, for the entire data set, including the 24 measurements outside the 95% prediction range, there was a strong correlation between CL and growth rings (r = 0.76, r_se = 0.007, p < 0.0001).

Table 1. Akaike Information Criteria (AIC) analysis, including ΔAIC = difference between model AIC and lowest AIC in the model set, ω = Akaike model weight, K = number of estimable parameters (including mean and variance), and Deviance = measure of model fit of 7 nonlinear growth models compared with the linear model, as applied to carapace length vs. growth rings count for 226 females and 53 unsexed juvenile ornate box turtles (Terrapene ornata ornata) in eastern Iowa.

View Fullscreen

Table 2. Akaike Information Criteria (AIC) analysis, including ΔAIC = difference between model AIC and lowest AIC in the model set, ω = Akaike model weight, K = number of estimable parameters (including mean and variance), and Deviance = measure of model fit of 7 nonlinear growth models compared with the linear model, as applied to carapace length vs. growth rings count for 231 males and 53 unsexed juvenile ornate box turtles (Terrapene ornata ornata) in eastern Iowa.

View Fullscreen

View Figure

• Download as Powerpoint
• Download Figure

Legler (1960) did not distinguish sexes when defining growth states prior to the adult stage; to facilitate comparison, we used the properties of the derivative to define 4 growth stages (juvenile 1, juvenile 2, subadult, and adult), separately, in our male and female growth models. For females, the first and third delineations for the model were defined at inflection points of the derivative of the model (2.39 and 9.33 growth rings, respectively) and the second delineation occurred at the maximum of the derivative of the model (5.86 growth rings; Fig. 2a). For males, the first and third delineations for the model occurred at inflection points of the derivative of the model (0.41 and 8.48 growth rings, respectively) and the second delineation occurred at the maximum of the derivative of the model (4.46 growth rings; Fig. 2b).

View Figure

• Download as Powerpoint
• Download Figure

Applying these delineations to the CL measurements Legler (1960) used to define 4 stages of growth, female growth rate increased through juvenile 1 (0.00–2.39 growth rings) and juvenile 2 (2.40–5.86 growth rings; Fig. 1a). Growth rate decreased during the latter part of the subadult stage (5.87–9.33 growth rings) and continued into the adult stage (9.34+ growth rings), during which growth rate slowed until stopping between 18 and 20 growth rings. For males, growth rate increased through juvenile 1 (0.00–0.40 growth rings) and juvenile 2 (0.41–4.44 growth rings) and decreased during the latter part of the subadult stage (4.45−8.48 growth rings; Fig. 1b). As with females, growth rate in adults (8.49+ growth rings) slowed until growth stopped between 18 and 20 growth rings.

Based on the derivative of our model for females, we defined juvenile 1 at CL ≤ 43.5 mm (≤ 3 growth rings), juvenile 2 at 43.6 ≤ CL ≤ 69.7 mm (4–8 growth rings), subadults at 69.8 ≤ CL ≤ 94.5 mm (8–9 growth rings), and adults at CL ≥ 94.6 mm (> 9 growth rings). For males, juvenile 1 was defined earlier at CL ≤ 31.9 mm (< 1 growth ring), juvenile 2 turtles at 32.0 ≤ CL ≤ 62.2 mm (1−4 growth rings), subadults at 62.3 ≤ CL ≤ 91.8 mm (4–9 growth rings), and adults at CL ≥ 91.9 mm (> 9 growth rings). Once CL reached approximately 92−95 mm, all our female and male turtles were within the 95% prediction band for adults (Fig. 1a–b). For females and males, growth slowed after CL = 100 mm (9−10 growth rings) and little growth occurred after CL reached 105 mm, after which size was no longer an accurate predictor of age (Fig. 1a–b).

Adult males (x̄ = 109.53 mm CL, SD = 5.15 mm) were significantly larger than adult females (x̄ = 107.72 mm CL, SD = 5.23 mm; t₅₁₀ = −3.95, p < 0.0001). The CL for recaptured turtles showed a similar pattern to the larger data set in that growth began to slow at approximately 8 growth rings (Fig. 3).

View Figure

• Download as Powerpoint
• Download Figure

For the entire data set, examination of a graph of CL vs. (FP + RP)/CL revealed that the CL was relatively longer than FP + RP prior to a CL of approximately 87.3 mm; as CL increased, FP + RP grew relatively larger (Fig. 4). A CL of 87.3 mm corresponded to approximately 8.1 growth rings for females and 7.7 growth rings for males, approximate inflection points for growth toward the latter quarter of the subadult stage (Fig. 1a–b).

View Figure

• Download as Powerpoint
• Download Figure

Two hatchlings measured in 2016 showed no growth in CL, FP, and RP and another turtle had zero growth of the plastron with no data recorded for the carapace. Removing data for turtles that did not grow, 5 hatchlings' plastra grew an average of 26% (SD = 14%) the first year, and the carapaces grew an average of 15% (SD = 7%). We were able to document growth rings during the first 3 yrs of life for only 10 individuals. One individual was recorded as developing 2 rings in 1 yr, but the others only developed 1 ring/yr. For 6 other juveniles or subadults with 9 or fewer growth rings, 1 ring was added per year (1−3 yrs of growth since first capture) except for 1 individual that was recorded to develop only 1 ring over 2 yrs.

DISCUSSION

While the von Bertalanffy equation has been used to model turtle growth (e.g., Fabens 1965; Frazer et al. 1990; Spencer 2002; Plummer and Mills 2015), based on AIC and deviation from a linear model, we determined that Richards growth model fit our data better. However, the Richards growth model was almost indistinguishable from the 4-parameter logistic for females (Table 1). In addition, the 4-parameter Gompertz model was similar to the 3-parameter logistic model for females; both yielded only slightly higher AIC values compared with the 4-parameter logistic and Richards growth models (Table 1). For males, AIC values indicated that the Richards, 4-parameter logistic, 3-parameter logistic, and 4-parameter Gompertz growth models provided similar fits for the data, but Richards was still the best model (Table 2). Potentially, other turtle species and different data sets would produce a more parsimonious fit with other growth models. Chen et al. (1992) noted that fish growth models could vary between species. However, our overall conclusions were the same with all models tested. Our data and models compared favorably to the conclusions of Legler (1960), who defined smaller juveniles at CL < 50 mm, larger juveniles at 50 ≤ CL ≤ 69 mm, subadults at 70 ≤ CL ≤ 100 mm, and adults at CL > 100 mm. However, our demarcations between the growth stages occurred at slightly smaller CLs, especially with regard to what size to classify individuals as adult. In the Richards model, once female CL reached 95 mm and male CL reached 92 mm, all our individuals were within the 95% confidence band of the models. Legler (1960) would have classified individuals at these sizes as subadults. Therefore, though similar to the growth categories of Legler (1960), our age-related categories are defined by smaller turtles than Legler's (1960) categories.

Legler (1960) stated that males grew faster than females. Our models indicated that male carapace growth was faster than that of females when combined with our unsexed young turtles. However, the first derivative of CL for the male model was before the turtle had a single growth ring; this could be an artifact of the model of our data. St. Clair (1998) stated that male eastern box turtles (T. carolina) grew faster than females. Although Quinn et al. (2014) noted ornate box turtle males were larger than females until age 8, they also stated that male and female growth was not statistically different. Quinn et al. (2014) also defined adults at a slightly larger size than we did; as a result, their estimate of adult CL for males and females was larger than our estimates. Legler (1960) found that adult females were larger than males, as did Quinn et al. (2014), who found that adult females were significantly larger than males. Therefore, based on our models, adult males were significantly larger than females, which contradicts both studies, although a comparison of our data with those of Quinn et al. (2014) showed that the mean CL difference between adult male and female turtles was only 2.5−3.0 mm. Germano (2014), who also used a Richards growth model on CL vs. age (growth rings), stated that there were no differences in growth between males and females for the desert box turtle. While Germano (2014) did not define age categories, his models for males and females combined produced similar growth patterns to our models, including fast growth in the early years.

Legler (1960) also noted differential growth of the carapace and plastron as turtles matured. Younger turtles had longer carapaces relative to plastra, but adult turtles had longer plastra relative to carapaces. Our results were similar and the inflection point of the (FP + RP)/CL was near the transition between our subadult and adult stages in our models. These predictions are consistent with other studies in which growth in long-lived turtles was biphasic and asymptotic (e.g., Congdon et al. 2001; Armstrong and Brooks 2014).

Dodd (2001) reviewed growth studies and attempts to estimate age in box turtles (Terrapene sp.) and Germano and Bury (1998) and Wilson et al. (2003) both critically reviewed the growth literature for all turtles and derived different conclusions as to whether growth rings were accurate predictors of age. In our study, several individuals counted growth rings, and discrepancies in counting occurred, especially for counts in the field. However, although our data are limited, we concur with Legler (1960) and Germano (2014) that growth rings are a relatively good predictor of age in years for ornate box turtles, at least prior to the subadult stage. This assertion is reinforced in our data on recaptured turtles since the 1990s, in which the growth curve derived from recaptured individuals resembled the curves modeled from static measurements (Fig. 4).

Legler (1960) reported 1.8%–66% (x̄ = 17.5%) plastral growth in 36 hatchings that emerged in the same season that the nest was laid and found that, under favorable conditions, hatchlings could have up to 8 wks of growth before overwintering. These early hatching turtles were as large as 1- or 2-yr-old turtles by the following summer and continued to grow rapidly. Legler (1960) further stated that turtles grew 68% in the first year after hatching, 24% in the second year, and 18% in the third year. We had no evidence of hatchling emergence from the nest in the same year that eggs were laid. Our hatchlings came from two sources: 1) nests caged immediately after spring oviposition with the cages not removed until the subsequent spring, or 2) spring pitfall traps that were removed after 1 July. Caged nests were checked at end of September for signs of disturbance; none were noted. Therefore, our data on first-year hatchling growth were very different and more similar to studies in South Dakota (Quinn et al. 2014). Legler (1960) measured plastral growth; however, in our model, a 68% growth rate from the first spring CL measurement for an average hatchling would correspond with 3.6 growth rings, and our estimate of first-year CL growth from 5 individuals was 26%, only slightly lower than the 28% plastral growth measurements of Quinn et al. (2014), whose criterion for first-year growth was the same as ours.

Some of the differences in the populations could be attributed to local climate. Compared with our Iowa population, ornate box turtles in Kansas are active above ground for a longer period of time (Legler 1960; Metcalf and Metcalf 1970; Bernstein and Black 2005). Although the overwintering period for ornate box turtles in Nebraska was similar to ours, Converse et al. (2002) noted that some hatchlings in the Sandhills of Nebraska potentially emerged from the nest in the autumn after oviposition and overwintered in a different location; however, this behavior was not noted in the same area by Costanzo et al. (1995). Closer to Iowa, Doroff and Keith (1990) did not report hatchling dispersal away from the nest before overwintering in southcentral Wisconsin, nor did Quinn et al. (2014) in South Dakota. Lovich et al. (2014) discussed selection pressures regarding whether hatchling turtles remain in the nest during their first winter or disperse before winter for several freshwater turtle species. Certainly, growth before overwintering would lessen some of the physiologic stresses and ornate box turtle hatchlings are freeze-tolerant (Costanzo et al. 1995, 2008). A rapid increase in growth would also be adaptive for predator protection, competition, and dispersal, but other than a warmer climate with shorter winters, we have no explanation for the discrepancy between our data and those of Legler (1960). We also have no data regarding food eaten by young or adult turtles that could influence growth. Therefore, assuming that growth rings approximate age, size can be used to estimate age in ornate box turtles up to 60−90 mm, or up to about 8 growth rings. Between 8 and 12 growth rings, size and age are also correlated, but the accuracy decreases as indicated by the widening 95% confidence band in our male and female models. Germano (2014) estimated age of desert box turtles up to 14 growth rings, but both our model and his model indicated growth leveling off toward an asymptote by that time. This agrees with Dodd (2001), who stated that growth in box turtles was determinate.

In Iowa, there are only 2 large populations of ornate box turtles with several isolated, small populations (Richtsmeier et al. 2008) and at least 1 of the 2 large populations was greatly reduced by a poacher. In determining the relative reproductive viability of a population, age structure is one important element for consideration, especially for some of the smaller populations, and even stable populations are vulnerable to declines (Congdon et al. 1993; Bowen et al. 2004). Head-starting hatchling turtles is one strategy to boost declining turtle populations (e.g., Heppell et al. 1996) and translocation of adults from healthy populations to declining populations is also a conservation strategy (e.g., Kuo and Janzen 2004; Rittenhouse et al. 2007). Growth comparisons of translocated or head-started turtles with undisturbed populations can help to assess viability of efforts to boost isolated populations, which may not offer adequate shelter or food for ornate box turtles. In addition, hatchling survival in ornate box turtles is not certain (Bernstein et al., unpubl. data, 2011–2016). Hatchling and juvenile turtles suffer higher mortality rates than do adults (Legler 1960; Belzer et al. 2000; Dodd 2001); monitoring age and growth of young turtles is important in areas where the threat of predation is high. As mentioned, even yearly changes in the environment can affect growth (Dodd and Dreslik 2008). Therefore, along with other life-history information (Shine and Iverson 1995; Wikelski and Cooke 2005), growth is important for conservation planning and management of this Iowa threatened species.