Demographics, reproduction, growth, and abundance of Jollyville Plateau salamanders (Eurycea tonkawae)

Abstract Insights into the ecology and natural history of the neotenic salamander, Eurycea tonkawae, are provided from eight years of capture‐recapture data from 10,041 captures of 7,315 individuals at 16 sites. Eurycea tonkawae exhibits seasonal reproduction, with peak gravidity occurring in the fall and winter. Size frequency data indicated recruitment occurred in the spring and summer. Open‐population capture‐recapture models revealed a similar seasonal pattern at two of three sites, while recruitment was dependent on flow at the third site. Females can reach sexual maturity within one year, and oviposition likely takes place below ground. The asymptotic body length of 1,290 individuals was estimated as 31.73 mm (at ca. two years of age), although there was substantial heterogeneity among growth trajectories. Longevity was approximately eight years, and the median age for a recaptured adult was 2.3 years. Abundance estimated from closed‐population and robust‐design capture‐recapture models varied widely within and among sites (range 41–834), although, surprisingly, dramatic changes in abundance were not observed following prolonged dry periods. Seasonal migration patterns of second‐year and older adults may help explain lower ratios of large individuals and higher temporary emigration during the latter half of the year, but further study is required. Low numbers of captures and recaptures precluded the use of open‐population models to estimate demographic parameters at several sites; therefore, closed‐population (or robust‐design) methods are generally recommended. Based on observations of their life history and population demographics, E. tonkawae seems well adapted to conditions where spring flow is variable and surface habitat periodically goes dry.

Jollyville Plateau salamanders (Eurycea tonkawae; Figure 1) are restricted to northwestern Travis and southern Williamson counties, Austin, Texas (Chippindale et al., 2000), and are federally listed as threatened (US Fish and Wildlife Service 2013). The first published account on the ecology of E. tonkawae was by Bowles, Sanders, and Hansen (2006), who examined monthly count data from a two-year period to assess relationships between relative abundance, habitat characteristics, and urbanization. Recent studies have extended this work, including estimation of demographic parameters (O'Donnell & Gluesenkamp, 2009), relative abundance in the context of land-use patterns (Bendik, Sissel, Fields, O'Donnell, & Sanders, 2014), and movement and occupancy within headwater streams (Bendik, McEntire, & Sissel, 2016).
In this study, I examined the natural history of the E. tonkawae by summarizing capture-recapture and body size data from an 8-year period from 16 sites across its geographic range. I analyzed body size frequency distributions to characterize demographic and reproductive patterns and compared these results to other central Texas Eurycea.
From the capture-recapture data, I modeled individual growth and generated estimates of age to determine the timing of life history events. I also estimated demographic parameters using open-and closed-population capture-recapture models. Collectively, this work provides new and important insights into the ecology and natural history of this microendemic species.

| Study sites and data collection
Sites were located within and around the Jollyville Plateau physiographic region of central Texas and were distributed across the range of E. tonkawae ( Figure 2). All of the study sites and survey areas were F I G U R E 1 Adult Eurycea tonkawae in Bull Creek, Travis County, Texas F I G U R E 2 Eurycea tonkawae localities and study sites adopted from prior studies (Bendik et al., 2014, Bowles et al., 2006City of Austin 2001;O'Donnell & Gluesenkamp, 2009). Fixed survey areas were delineated within and around springs, although survey areas varied substantially among sites (Table 1) due to differences in spring and stream flow (and thus, available habitat) and judgments made by the original investigators. These sites varied in quality and were distributed throughout the range of the species, from heavily urbanized areas to large preserves (Bendik et al., 2014).
Surveys performed during this study represent a continuation of several population monitoring efforts of E. tonkawae at 16 sites which were subsequently extended to obtain multiyear capture-recapture data. The sampling followed either a closedpopulation capture-recapture protocol following O'Donnell and Gluesenkamp (2009) or an open-population capture-recapture protocol at sites where sampling initially consisted of population counts (Bendik et al., 2014, Bowles et al., 2006City of Austin 2001). The closed-population method (Otis, Burnham, White, & Anderson, 1978) entailed successive sampling (three days in a row) during a short enough period (the "primary" period) where demographic closure (no births, deaths, or migration) of the population was assumed.
Sampling under the open-population method (Nichols, 2016) occurred over longer time intervals (generally > three months), during which the population was open to demographic change. When multiple closed-population samples are combined to include intervals where the population is open to demographic change, this is referred to as the robust-design (Pollock, 1982).
From 2008 to 2012, surveys were performed at three sites within the Balcones Canyonlands Preserve using closed-population capture-recapture sampling (Table 1). Although quarterly sampling events were desired, severe drought hampered the initial attempt to collect data at evenly spaced intervals. Thus, sampling for these sites was generally opportunistic during the study period. From 2012 to 2015, open-population surveys were performed at 15 sites (including two sites from the closed-population sampling). This includes surveys that were either opportunistic, performed as part of another study, or was part of a prescribed monitoring plan. Opportunistic sampling includes single surveys (e.g., MacDonald Well; Table 1) or surveys to collect additional demographic and individual information on previously marked animals. Data are also included from other studies where individual salamanders were identified. This includes a four-month openpopulation capture-recapture study of movement  and a study of hormone levels (Gabor et al., 2016). Finally, nine sites were regularly sampled once per quarter during February/March (winter), May/June (spring), August/September (summer), and November/ December (fall). However, springs periodically ceased flowing during our study which resulted in ragged survey intervals and incomplete sampling for some years and sites. Please refer to Table 1 for additional site-specific information.
Salamanders were found by exhaustively searching available cover objects including gravel, cobble, leaves, and woody debris. Surveyors also searched through vegetation and disturbed fine sediments, using small aquarium nets to capture salamanders. Salamanders were flushed from dense leaf litter or vegetation into large pool nets. Individuals were placed in mesh containers in the water before and after processing.

| 5005
BENDIK methanesulfonate (MS-222)/L of naturally buffered spring water and then marked using VIE tags. Sterile 28-gauge syringes were used to inject small amounts (2-20 μl) of elastomer just underneath the skin to form a bead. Each salamander was given three to four unique VIE tags using a combination of seven different colors in five locations on the body. All captured salamanders were photographed on a standardized grid background in a water-filled tray using a DSLR and wireless flash. Marking with VIEs was replaced from 2012 onward with photographic identification using the software Wild-ID (Bolger, Morrison, Vance, Lee, & Farid, 2012) due to the decreased cost and invasiveness, as well as increased speed and accuracy of photographic identification for this species (Bendik, Morrison, Gluesenkamp, Sanders, & O'Donnell, 2013). This savings in time and cost facilitated an expansion of capture-recapture monitoring to include more sites.
Gravid females were considered sexually mature. Otherwise, individuals were not sexed due to the difficulty of candling for determining presence or absence of testes. Body length (BL) was quantified as the mid-vertebral distance from the tip of the snout to the posterior insertion of the hindlimbs. Total length (TL) was quantified as the distance from the tip of the snout to the tip of the tail. BL and TL were measured to the nearest 0.1 mm using ImageJ software (Rasband, 1997).
To allow for comparisons across different studies, BL measurements were converted to SVL with the following linear regression formula: SVL = 1.032 × BL + 0.896 (n = 14, r 2 = .99). I summarized body length and gravidity data to assess patterns in population demographics and reproduction.
For each primary period, I first determined the best model structure for the conditional capture and recapture probabilities from closedpopulation models. I included the best structure for that period in the robust-design model based on Akaike's information criterion corrected for small sample size (AICc; Burnham & Anderson, 2002 (γ k and γ k−1 ) and apparent survival parameters (φ k ) are confounded (Kendall et al., 1997) and were therefore excluded from the results.
Survival was assumed to be the same for animals regardless of their probability of capture. Data from Lower Ribelin and Wheless were excluded from the robust-design analysis due to small sample sizes.
Instead, for these populations, I used closed-population models (as above) to generate estimates of abundance and conditional capture probability for each period.
Jolly-Seber (JS) models allow estimation of apparent survival and effective capture probability (p 0 ) in addition to population entry probabilities (Pollock, Nichols, Brownie, & Hines, 1990). I used the Link-Barker JS formulation (Link & Barker, 2005), which parameterizes population entry as per capita recruitment (f). This parameter quantifies entrants to the population, either from immigration or births.
Preliminary analyses revealed that models with full time variation of parameters were unsupported by the data, so constraints were placed (e.g., . The ecology of these oligotrophic headwater streams (Mabe, 2007) is also influenced by seasonal variability in water temperature and nutrient input (e.g., via leaf fall).
I used JS models to examine demographic patterns at three sites (Franklin, Hill Marsh, and Troll). Low rates of recapture and low overall abundance (in particular at several highly urbanized sites; Table 1 Jolly-Seber models require several assumptions, which if not met could result in biased parameter estimates (Pollock et al., 1990). These assumptions include the following: (1) marked animals are recognized accurately and marks are not lost, (2) sampling is instantaneous, (3) marked animals have the same capture probability as unmarked animals, (4) every animal in the population has the same probability of capture in a given sampling period, (5) every marked animal has the same probability of survival between sampling periods, and (6) emigration is permanent. Based on the study design and the use of photographs instead of physical marks, assumptions 1 through 3 are unlikely to be violated. There is some evidence of size-biased capture and survival heterogeneity in E. tonkawae (assumptions 4 and 5) from a previous study , although that study included very small juveniles (<16 mm SVL). Small juveniles can be particularly difficult to capture and identify reliably, and have lower survival rates compared to larger animals ; therefore, I excluded those individuals from the capture-recapture analysis. The last assumption is the most difficult to meet given prior evidence of movement and temporary emigration at some sites O'Donnell & Gluesenkamp, 2009). If temporary emigration is random, parameters may be unbiased, while Markovian movement can result in biased parameter estimates (Kendall et al., 1997;Nichols, 2016). Given these points, it is important to consider the potential consequences of violating assumption 6 and interpret the JS estimates with caution.
I used program MARK (v8.1; White, 2015) to fit all capture-recapture models and estimate parameter means and 95% confidence intervals (CI).
I checked parameters estimated at the boundary (0 or 1) using the data cloning option in MARK to verify they were actually being estimated (at 0 or 1). I used model-averaged estimates if the top model had an AICc weight <0.9.  Among recaptured individuals that were gravid on more than one occasion, 35 were recaptured within a 100-day period. Of these, 74% were gravid during both occasions.  (Table 2). Temporary emigration followed a Markovian pattern whereby the availability of individuals within the survey area was dependent upon whether they were available or not during the prior survey. Based on higher estimates of ̂′ ′ , individuals were more likely to have temporarily emigrated following a summer interval compared to winter ( Table 3).

| Capture-recapture models
The best Link-Barker JS models supported by the data included either season-or flow-dependent survival, per capita recruitment, and capture probability (Table 4) Figure 6a). In contrast, models with seasonal variation in apparent survival were best supported for all sites (Table 4). Survival peaked in the winter at Franklin and Troll but was highest in the late summer and fall at Hill Marsh (Figure 6b). Capture probability was negatively correlated with flow at Troll and Franklin, but was seasonal at Hill Marsh (Table 4).
Capture probability was generally the lowest at Troll (Figure 6c); this, combined with substantial model selection uncertainty, resulted in high uncertainty of demographic parameter estimates for this site ( Figure 6). Hill Marsh had the highest capture probabilities, which varied by season (Figure 6c).
Although goodness-of-fit tests for robust-design and Link-Barker JS models are unavailable, one qualitative way to assess the potential consequences of lack-of-fit is to assess the sensitivity of model rankings to changes in ĉ, a variance inflation factor. Values of ĉ from 1.25 to 2.00 did not result in changes to the top capture-recapture models at Hill Marsh or Franklin. For Troll Spring, models with flow as a covariate were more heavily favored over those with season as ĉ increased.

| Growth and longevity
The  Parameter γ″ k represents the probability an animal in the study area during period k−1 moves out of the study area in period k; γ′ i is the probability that an animal stays away from the study area in period k, given that it was a temporary migration in period k−1 (Kendall et al., 1997). φ(t) = yearly apparent survival probability from time t−1 to t. Superscripts indicate where the interval (4-7 months) from period k−1 to period k included either summer (s) or winter (w). LCL and UCL = lower and upper bounds, respectively, for the 95% confidence interval.

T A B L E 3 Temporary emigration and apparent survival estimates for Lanier Spring, 2008-2012, for model φ(t) γ″(t) γ′(t) p*(best) c(best)
The expected size (mm SVL) of a typical individual from the data at 1, 1.5, and 2 years of age was 28. (≥1 year old), the median age was 2.3 years, the 75th percentile was 4.0 years, and the maximum was 7.9 years.
The approximate age for young-of-year from the size distribution data was 0.9 months (CRI = 0.79-1.02) for the 11.3 mm mode in February/March (Figure 3a). This was followed by a larger 18.7 mm peak in May/June at 3.6 months (CRI = 3.2-4.1), which corresponds to an average age difference of 2.7 months between the first two quarters. This is roughly equivalent to the average survey interval (approximately three months). Similarly, growth from 18.7 mm to the following mode of 23.7 mm (mean age = 6.5, CRI = 5.7-7.3) in August/ September corresponds to a 2.9-month age difference.

| DISCUSSION
Population demographics of E. tonkawae follow a predominantly seasonal pattern. Gravid females first appeared in September and were most abundant during the late fall and winter. The proportion of gravid individuals observed also increased as a function of body size. Larger females may be more abundant than large males (due to sex-biased survival or growth) or may be gravid more frequently than smaller females (higher fecundity). For example, numerous individuals were gravid more than once during a season, indicating E. tonkawae may lay multiple clutches per year. Oviposition likely occurs below the surface throughout the fall and winter, as I observed only a single egg during this study. Peak gravidity in December followed by peak hatchling (<12 mm SVL) abundance in February is consistent with the timing of egg incubation (1 month) and yolk sac absorption (2 weeks) in the closely related E. sosorum (Cantu, Crow, & Ostrand, 2016).
While seasonal reproduction is common among Eurycea and many other salamanders in temperate regions (Bruce, 2005;Petranka, 1998), reproductive phenology is not well documented for most central Texas Eurycea. Seasonality in reproduction was suggested for E. tonkawae (Bowles et al., 2006), E. neotenes (Bogart, 1967), and transforming populations of E. troglodytes (Sweet, 1977), but is well documented for E. naufragia (Pierce et al., 2014). These species occupy smaller springs compared to those inhabited by E. nana and E. sosorum. Eurycea nana is restricted to one of the largest springs in Texas (San Marcos Springs) and reproduces year-round (Nelson, 1993;Tupa & Davis, 1976). Based on juvenile (individuals <25 mm TL) abundance data, reproduction in E. sosorum (at Barton Springs) also appears to be nonseasonal (Dries, 2012;Gillespie, 2011). Differences in reproductive phenology and recruitment may be a function of spring size and reliability. For example, per capita recruitment of E. tonkawae varied seasonally at Hill Marsh, which is an isolated, low-discharge spring. This contrasts with Franklin, where spring and stream flow from its large watershed was a better predictor of recruitment compared to seasonality. Large spring systems may be less susceptible to variation in environmental conditions caused by terrestrial seasonality compared to smaller, intermittent springs and stream habitats.
The VB model indicated substantial variation in growth of E. tonkawae. This is partly a consequence of combining data from T A B L E 4 Model selection results of Jolly-Seber models using the Link-Barker formulation. f = per capita recruitment, the number of individuals at time t compared to t−1; = quarterly apparent survival; p 0 = effective capture probability multiple populations and across a range of environmental conditions, although it is possible to directly account for environmental stochasticity in the hierarchical VB model (e.g., Connette, Crawford, & Peterman, 2015). Nevertheless, the VB model used here is useful for approximating average age-at-length with two primary assumptions: (1) variance in initial size is negligible; (2) generalizations from the data are applicable to the species at large. Based on the minimum estimated age at maturity for females as well as time required for oviposition and incubation in similar species (Cantu et al., 2016), the generation time for E. tonkawae is probably between 1 and 1.5 years. Female size at maturity (ca. 24-28 mm) was within the range of that reported for E.
troglodytes (25 mm SVL; Bruce, 1976;Sweet, 1977 Temporary emigration can result in biased parameter estimates of the JS model when the pattern of emigration is Markovian (Kendall et al., 1997). In one study, JS estimates of population size were found to be generally unbiased when animals had a high probability of returning to the study area, but were negatively biased when they had a low probability of returning (Zehfuss, Hightower, & Pollock, 1999). Because recruitment is a function of population size, the direction of bias may be similar. At Lanier Spring, temporary emigration was Markovian, although estimates from a prior single-year study with shorter, monthly sampling intervals at Lanier, Lower Ribelin, and Wheless indicated random emigration (O'Donnell & Gluesenkamp, 2009). In the case of random emigration, JS capture probability is the product of the probability of capture and the probability of temporary emigration, while the other parameters are unbiased (Kendall et al., 1997;Nichols, 2016). Thus, it is unclear the degree to which per capita recruitment estimates may be affected in this study because the form (Markovian or random), and the magnitude of temporary emigration is unknown. Another consideration in the interpretation of recruitment from the JS models is that it includes both births (in this case, entrants of small juveniles) as well as immigration (e.g., the return of temporary migrants). Despite these challenges, it is encouraging that timing and pattern of recruitment predicted by the JS models was generally consistent with the size frequency distribution, gravidity, and growth data.
For many sites it was difficult to obtain open-population JS estimates due to low recapture probabilities. This was partly a consequence of decisions regarding the study design, such as survey interval (quarterly at most sites) and the area surveyed (limited to areas near spring outlets). However, the tendency for salamanders to migrate from the vicinity of springs (this study, Bendik et al., 2016), their ability to retreat to subterranean habitats , and  These temporal dynamics are consistent with the high temporal variation in counts of E. tonkawae (Bendik et al., 2014;Bowles et al., 2006). This may be driven by variation in reproductive output, migration patterns, or mortality due to adverse environmental conditions. However, abundance following long dry periods was not disproportionate to estimates from other periods as might be expected given the negative impacts of drought on body condition of E. tonkawae . Remarkably, some recruitment still occurred during drought, as several juveniles were observed only a few weeks following a ten month period of completely dry surface conditions (March 2009, Lanier Spring;. It is worth noting, however, on several occasions I observed stranded juvenile salamanders under rocks immediately following the cessation of spring flow; a similar observation was made by Sweet (1977). Thus, the apparent population structure may be determined in part by spring flow (this study; Sweet, 1977) and a size bias in the ability of individuals to migrate into subterranean habitats as springs cease flowing.
Collectively, the results presented here suggest that life history characteristics of E. tonkawae help facilitate a resiliency to drought and an ability to respond quickly to favorable environmental conditions (e.g., via short generation times and multiple clutching). Adaptations to temporally variable flow are likely common among epigean central Texas Eurycea populations, as spring reliability varies widely throughout the Edwards Plateau (Sweet, 1982). Although urbanization remains the primary near-term threat to many populations of E.
tonkawae (Bendik et al., 2014), adaptations that allow them to cope with variable flow conditions may become more important as a drier future looms over central Texas aquifers (Stamm et al., 2015).

ACKNOWLEDGMENTS
Blake Sissel, Jacob Owen, and Brad Nissen were instrumental in obtaining many of the salamander measurements. Grant Connette gave insight and sample code for the growth modeling. Rob Clayton and