AMPHIBIANS are experiencing declines globally due

Transcription

1 Copeia 2009, No. 2, Survival and Breeding Frequency in Marbled Salamanders (Ambystoma opacum): Implications for Spatio-temporal Population Dynamics Lloyd R. Gamble 1, Kevin McGarigal 1, Douglas B. Sigourney 1, and Brad C. Timm 1 Despite known population-level sensitivity to adult vital rates, there is a shortage of robust estimates of adult survival and breeding frequency for pond-breeding amphibians. Evaluating the applicability of metapopulation principles to amphibians has also been constrained by the lack of demographic studies scaled beyond individual breeding populations. We investigate spatial and temporal demographic variability among six proximate breeding populations of Marbled Salamanders (Ambystoma opacum) in western Massachusetts, USA, focusing on the terrestrial adult life stage. Between 1999 and 2004, we captured and photographed approximately 1700 breeding adults, each between one and 12 times, at drift fences encircling breeding sites. After constructing individual capture histories from field data, we applied maximum likelihood approaches to estimate parameters for survival in the pond basins during breeding periods, survival in upland habitat during breeding and non-breeding periods, and both breeding and capture probabilities. Models selected using Akaike s Information Criteria suggested that there was moderate variability in pond survival across years, but that generally, pond and upland survival varied little among populations within years. This suggests that asynchronous variability indicative of metapopulation dynamics is unlikely to be significant in the adult stage, at least among nearby populations in similar upland forests. An integrated measure of annual survival was 0.66 (SE = 0.02) for males and 0.62 (SE = 0.01) for females. Average breeding probabilities were very high (0.96; SE = 0.01) for males and high (0.67; SE = 0.02) for females, resulting in estimates of 2.5 and 2.0 lifetime breeding attempts, respectively. These parameter estimates will be used to develop spatially explicit population models to guide conservation and forest management strategies for A. opacum and other pond-breeding amphibians with similar life histories. Additional empirical work that estimates the effects of alternative management strategies on these vital rates will greatly improve the utility of these models. AMPHIBIANS are experiencing declines globally due to numerous and interacting factors including habitat loss and degradation, climate change, emerging diseases and exotic species (Alford and Richards, 1999; Stuart et al., 2004). In densely populated regions such as the eastern United States, the widespread conversion of forested areas for development and the loss of seasonal wetlands that serve as essential breeding sites appear to pose the most immediate threats (Semlitsch and Bodie, 1998; Gibbs, 2000). To develop effective conservation strategies in the face of rapid and widespread landscape changes, it is essential that we understand the primary factors regulating amphibian populations, the life stages that are most sensitive to changing environments, and the functional scales at which populations operate and achieve high probabilities for long-term persistence. Much of the field-based research on pond-breeding amphibians has focused on the aquatic larval stage (e.g., community ecology, density dependence [Pechmann, 1995]) or on aspects of breeding migrations (e.g., migration cues, timing, and orientation [Paton and Crouch, 2002; Timm et al., 2007]). This bias has resulted in part from logistical barriers to studying fossorial animals in their terrestrial habitats in particular, the difficulty of identifying and tracking many individuals and the extended time frames necessary to evaluate demography. Efforts to estimate terrestrial survival probabilities have also faced analytical challenges, including the need to differentiate dispersal or skipped breeding years from mortality (Bailey et al., 2004). Despite these challenges, a growing literature suggests that adult vital rates are often disproportionately important in affecting population-level change in amphibians and other taxa (Pfister, 1998). In one example, an elasticity analysis of Western Toads (Anaxyrus [5 Bufo ] boreas) found that a 15% reduction in terrestrial juvenile and adult survival rates would have a greater impact on population lambda than a 49% reduction in embryo survival (Biek et al., 2002). Recent interest in amphibian conservation has also focused on spatio-temporal population dynamics and the question of whether some species occur as metapopulations. A strict view of metapopulation theory describes assemblages of local populations that are individually prone to extinction but collectively persistent as a result of dispersal and recolonizations (Hanski and Gilpin, 1997). Many pondbreeding amphibians have been considered likely candidates for metapopulation models because they typically have high breeding site fidelity and are characterized by high levels of population variability (Semlitsch, 2000). The relevance of metapopulation models to these species ultimately depends on the level of demographic variability experienced by local populations and the level of independence in this variability over space and time. The potential for dramatic temporal variability (Semlitsch et al., 1996; Meyer et al., 1998) and local extinctions (Blaustein et al., 1994; Skelly et al., 1999) has been well established; however, most research on population regulation in pond-breeding amphibians has been limited in scale to individual breeding populations (Semlitsch, 2002) and therefore insufficient to evaluate issues of synchrony or spatial autocorrelation. As part of a long-term, landscape-level field study in western Massachusetts, we evaluated demography in Marbled Salamander (Ambystoma opacum) populations distributed around 14 seasonal ponds in a continuously forested landscape. In this analysis, we examine demographic variability in the terrestrial adult life stage. Our objectives 1 Department of Natural Resources Conservation, University of Massachusetts, Amherst, Massachusetts 01003; (LRG) lloydgamble2@ gmail.com; (KM) mcgarigalk@nrc.umass.edu; (DBS) dsigourn@nrc.umass.edu; and (BCT) amphib_guy@yahoo.com. Send reprint requests to LRG. Submitted: 8 November Accepted: 20 January Associate Editor: T. W. Reeder. F 2009 by the American Society of Ichthyologists and Herpetologists DOI: /CH

2 Gamble et al. Survival in Ambystoma opacum 395 are: 1) to assess the spatial and temporal distribution of variability in adult demographic traits including upland survival (during breeding and non-breeding periods), pond survival, and breeding frequency; 2) to derive unbiased estimates of these parameters that will enable us to develop spatially explicit population models in future studies; 3) to consider life history and ecological implications of our findings and how they compare to those for other pondbreeding amphibians; and 4) to evaluate the implications of our findings for local and regional population dynamics in A. opacum and other pond-breeding amphibians. MATERIALS AND METHODS Study organism. The range of A. opacum includes much of the eastern United States from southern New England to northern Florida and extends westward from southern Illinois to eastern Texas (Petranka, 1998). In late summer and early fall, adults migrate from terrestrial refugia in upland or floodplain forests to receded or dry seasonal pond basins to breed. After courtship, females deposit eggs under cover objects and in shallow leaf litter. In favorable conditions, the eggs are inundated by rising pond water in the subsequent weeks or months, and soon thereafter hatch into aquatic larvae. The larvae overwinter in the ponds and metamorphose into terrestrial juveniles during the following spring and summer. After a sub-adult stage of one to five years (Scott, 1994), the majority of individuals return to natal ponds to breed, but some disperse several hundred meters or farther to breed in new ponds (Gamble et al., 2007). Though locally abundant in portions of its range, A. opacum is considered vulnerable to decline due to the widespread loss of seasonal ponds that remain largely unprotected by existing state and federal wetlands regulations (Scott, 2005). In Massachusetts, they approach the northern limits of their natural range and are listed as a state- Threatened species (M.G.L c.131a and regulations 321 CMR 10.00). Study area. The study area encompasses approximately 300 hectares of mixed-deciduous hardwood forests on the Holyoke Range in western Massachusetts, USA (Fig. 1). The site is mostly undeveloped, but is bisected by a 50-m wide powerline corridor and contains numerous carriage roads and trails. Ten seasonal ponds are clustered tightly in the western section of the study area, and four are distributed more widely to the east, with interpond distances ranging from 50 to 1500 m. The ponds vary considerably in structure, including shrub-dominated, open-deep water, and shallow (open and/or vegetated) ponds. Most ponds typically dry between June and September, but some occasionally hold water continuously throughout the year. Average numbers of breeding individuals at each of these ponds range from 0 to approximately 150 breeding females with male-biased sex ratios (Gamble, 2004). The nearest seasonal ponds outside of the study area are approximately 800 m away, and we have not detected A. opacum at any ponds within 1250 m of the study ponds. Field methods. To monitor A. opacum movements, we completely encircled all 14 seasonal ponds with continuous drift fences and pitfall traps (see Jenkins et al., 2003 for details). We checked traps daily from May through November of each year from 1999 to 2004, fully encompassing the emergence and breeding periods, and released captured animals on the opposite sides of the fences. The sex of adults was determined by inspecting the cloacal region for swelling. Adults were then digitally photographed for individual identification based on their unique dorsal patterns. We later matched these photographs with a computer-based pattern recognition algorithm (Gamble et al., 2008) and matching results were cross-referenced to original data to construct individual capture histories. During the off-season, we closed all traps and created frequent openings along all fences to allow passage of non-target animals. Capture recapture (CR) analysis. Our objective to derive unbiased survival estimates for A. opacum was complicated by several issues, including the following: 1) some individuals may pass (i.e., trespass ) drift fences without being captured, 2) some individuals may skip breeding seasons and therefore be unobservable in some years, and 3) our opportunities for capturing individuals were limited logistically to periods when salamanders migrate to and from breeding pond basins, and the duration of breeding and non-breeding seasons is known to vary from year to year and between females and males (Jenkins et al., 2006). Capture recapture (CR) methods use maximum likelihood estimators to address the first of these issues by simultaneously estimating survival and capture probabilities, but have only recently been adapted to accommodate the second issue, year skipping. In an analysis of Tiger Salamanders (A. tigrinum), Bailey et al. (2004) developed a modified open robust design that enables estimation of survival parameters for both breeding and non-breeding intervals and allows transitions to an unobservable state to reflect the possibility of skipped breeding years. Survival of unobservable individuals (presumed to remain in upland habitat rather than breeding) is estimated as a function of survival of observable animals in upland habitat during the subsequent non-breeding season (Church et al., 2007). We used this approach to model our female A. opacum data; however, we were not able to overcome estimation problems in its application to the male data. Some exploratory models suggested that males rarely skipped breeding seasons, possibly explaining the shortage of data to estimate these parameters, and that mortality of non-breeding males during their short breeding periods was negligible. With this knowledge, we fixed non-breeding male survival parameters to 1.0 during the breeding periods and modeled the male data separately from the female data. This also addressed the third issue listed above, because we directly input lengths of all intervals into the models based on median immigration and emigration dates. Survival probabilities are the only non-instantaneous parameters and are estimated on a biweekly unit and subsequently adjusted to these realized time intervals (Bailey et al., 2004). We included the six largest breeding populations (n. 20 breeding individuals annually) from our study area in the CR analysis. The duration of our study included 12 sampling occasions, or two occasions (i.e., immigration and emigration) in each of six years. Our data were insufficient for numeric optimization procedures in the most general configurations of the modified open robust design (i.e., with most parameters that are allowed to vary; n parameters for each sex), so it was necessary to take several steps to simplify the parameterization in order to achieve working general models (hereafter, global models ) for both sexes (Appendix 1). First, we were unable to incorporate dispersal

3 396 Copeia 2009, No. 2 Fig. 1. Study area located on the Holyoke Range in western Massachusetts, USA. Fourteen seasonal ponds occur in the study area. Those depicted in black support the largest breeding populations of Ambystoma opacum and were included in the capture recapture analysis. Scale bar is 1 km.

4 Gamble et al. Survival in Ambystoma opacum 397 Table 1. Alternative a priori Model Configurations for Capture, Transition, and Survival Probabilities of Ambystoma opacum Evaluated through Model Selection. Configuration p(t,pd) p(.) p(.hurr) psi-oo(t) psi(.dry) S-UP(t,pd) S-UP(.) S-PD(t,pd) S-PD(t) S-PD(.hurr) Description Global. Allows capture probabilities to vary by year and population. Constrains all capture probabilities to be equal. Expected. Capture probabilities are constrained to be equal across all populations and years except for the first, hurricane year during which higher trespass was expected. Global. Observable to observable transitions are allowed to vary by year, but not by population. Expected. Observable to observable transitions are constrained across both years and populations with the exception of the driest year, 2001, which is estimated independently. Global. Upland survival is allowed to vary by year and population. Expected. Upland survival is fully constrained, in agreement with our expectations that 1) adult individuals in familiar upland habitats with terrestrial refugia would be resilient to temporal environmental variation and 2) upland habitat quality is consistent across populations. Global. Pond survival is allowed to vary by year and population. Expected. Pond survival may vary from yearly in response to variable precipitation and temperatures, but similar variation is expected across populations. Pond survival does not vary by population, but may vary in response to dramatic weather years, such as the 1999 hurricane. among ponds due to model complexity and data limitations; however, a separate analysis indicated that fewer than 4% of experienced breeders dispersed to different breeding ponds in subsequent years (Gamble et al., 2007). Therefore, we removed the few individual capture histories that included more than one pond and treated remaining individuals at each pond as separate groups without the possibility of dispersal. Second, exploratory models suggested that virtually all males who skipped a breeding year would breed in the subsequent year, so we fixed transitions from unobservable to observable states equal to 1. For females, these transitions were constrained to be equal for all ponds and years, resulting in a single parameter estimate. Third, for both males and females, data were insufficient to independently estimate observable to observable transitions (i.e., animals that remain in a breeding state for consecutive seasons) for all ponds and years. We expected that the probability of breeding in a given year was most likely to be affected by environmental variables such as precipitation that would vary more by year than by population in our study area; therefore, we constrained these transitions to be equal across populations but allowed them to vary with time. Fourth, a hurricane and a subsequent storm caused several drift fences to flood during the first emigration period and may have changed both capture and survival probabilities. To address estimation problems occurring in the first interval and second occasion (CR models are conditioned on first capture), we constrained capture probabilities to be equal across populations in the female data and pond survival probabilities equal across all populations for both sexes. Lastly, the populations at ponds 5 and 6 had the lowest sample sizes in our study. To overcome estimation problems in these two populations, breeding season (BS) and non-breeding season (NBS) survival probabilities were constrained to be equal over time for all intervals except the first (i.e., hurricane) interval. Capture probabilities were also constrained to be equal across these occasions in the female data, but could be estimated independently from the male data. Upon reaching a functioning global model, we configured a set of more constrained models representing specific hypotheses developed a priori about how these parameters may vary in space and time. These alternative models and our expectations for model selection are described in Table 1. With three model configurations for capture probabilities, two each for transitions and upland survival, and three for pond survival, we built a total of 36 models each for males and females and ran them in a modified version of the program MSSURVIV (Hines, 1994). We used Akaike s Information Criteria adjusted for overdispersion and small sample size (QAICc) to compare models and calculated Akaike s weights to estimate the relative level of support in the data for each model (Burnham and Anderson, 2002). Weights for the best-supported models (combined weights. 95%) were recalculated and used to derive modelaveraged parameter estimates and standard errors. A final consideration in the CR analysis was related to the original construction of capture histories from the imagebased individual identification method. Testing of this method revealed that false-positives (i.e., mismatches) were exceedingly rare; however, false negatives (i.e., missed matches) did occur occasionally (Gamble et al., 2008). Similar to a lost or misread tag, the result of a missed match would be a fragmented capture history in which one or more captures of an individual are interpreted to be a different individual, resulting in two or more incomplete capture histories. Given the size of our data set, a perfect identification technique was unattainable (i.e., toe-clips and pit tags are also prone to regeneration or loss); however, we conducted a trial to estimate the effects of missed matches on parameter estimates. To do this, we first estimated the frequency at which missed matches occur in the data by selecting a random sample of images and exhaustively searching the data for matches. We then created a simulated data set for female A. opacum with known parameter values comparable to those found in our real data. We simulated the fragmentation of this data set by randomly removing individual capture events from capture histories with two or more captures at frequencies estimated for the real data set and treating these removed events as separate individuals with single captures. Lastly, we derived parameter estimates in MSSURVIV for these paired (unfragmented and fragment-

5 398 Copeia 2009, No. 2 Table 2. Numbers of Capture Events of Female and Male Ambystoma opacum Immigrating to and Emigrating from Six Pond Basins between 1999 and The first and second occasions of each year represent immigration and emigration captures, respectively. In rare cases where an individual was captured immigrating or emigrating more than once, only the first immigration event and last emigration event were counted in the total Total Pond Sex individuals a 2 f f f f f f Total f m m m m m m Total m a This is the estimated total number of unique individuals captured in all years, as determined from image-based individual identification (see text in Materials and Methods). ed), simulated data sets using the top model structure selected from our real data and compared the results. Simulation models to estimate longevity and breeding frequency. We used parameter estimates from the CR analysis to model longevity and total lifetime breeding events in adult A. opacum. To accomplish this, for each sex we constructed an individual-based simulation with 30 time steps (breeding and non-breeding intervals for each of 15 years) and 1000 individuals, each starting at the point of first breeding capture. Survival and transition probabilities for each time step were randomly drawn from a normal distribution derived from our parameter estimates (Church et al., 2007). Then, at each time step, each individual was subjected to these probabilities using a random number generator between 0 and 1 to determine whether it survived or died, or remained in a breeding state or not. We ran 100 iterations of this model and estimated the average number of years that salamanders lived (beginning with their first breeding season) and the average number of times that salamanders bred in their lifetime. This analysis required two significant assumptions: first, that environmental conditions affecting survival and transition probabilities through the duration of our study represent their natural range of variability over longer time periods and were adequately represented by the normal distribution used in the simulation, and second, that survival probabilities do not vary with age after first breeding. The second assumption was unavoidable because we did not have the means to age individual salamanders in our study and therefore pooled adult salamanders into a single age class. RESULTS Capture recapture analysis. From 1999 to 2004, we recorded 1890 and 3245 captures of breeding female and male A. opacum from all 14 ponds combined. Using the image-based individual recognition process, these were organized into capture histories for 761 and 1086 females and males, respectively. The six breeding populations included in the CR analysis accounted for 98% of the total captures at the study site (Table 2). Total numbers of individual females through the six years of the study varied from 0 to 15 for the eight smallest populations and up to 391 in the largest population. Most breeding populations had male-biased sex ratios; the cumulative sex ratio (from total numbers of individuals captured in all years) for the six large populations was 1.4 to 1. Approximately 40% of females and 50% of males were captured in more than one year of the study, and small numbers of individuals of each sex (,1.5%) were captured in all six years. The duration of the breeding season, calculated as the number of days between median immigration and median emigration dates, averaged 33 days for females (min 5 10, max 5 59) and 19 days for males (min 5 13, max 5 28). Subject to the constraints detailed in Appendix 1, the global model for females estimated 68 parameters and fit the data based on a Pearson s goodness-of-fit test after pooling cells with small expected values (x , df 5 101, P , ĉ ). The comparable model for males estimated 75 parameters and also fit the data (x , df 5 96, P , ĉ5 1.09). In the female CR analysis, the top four models combined carried.96% of the QAICc weights (Table 3) and were used to derive model-averaged parameter estimates. In all four of these models, upland survival (S-UP) was constrained to be equal among ponds and years and pond survival (S-PD) was constrained to be equal among ponds, but varied across years. Both the time-general (psi-oo(t)) and the dry year (psi(.dry)) state transition configurations (see Table 1) occurred in these models, suggesting some support for the hypothesis that fewer females may attempt to breed in dry years. The top two models (combined QAICc. 0.83) constrained capture probabilities to be equal across years and ponds with the exception of the first year when a hurricane inundated many traps and drift fences. However, there was some support for models with capture probabilities varying by year and by pond (combined QAICc ).

6 Gamble et al. Survival in Ambystoma opacum 399 Table 3. List of a priori Model Configurations, DQAICc Values, Model Weights, and Numbers of Parameters (K) for Female and Male Ambystoma opacum Captured at Six Ponds from 1999 to Models are ordered by female DQAICc values with highest-ranked models (top 95% of model weights) displayed in boldface type. The most general model, noted by asterisk, allows capture probabilities (p) to vary by time (t) and pond (pd), state transitions (psi) to vary by time, and upland survival (S-UP) and pond survival (S-PD) to vary by time and population. Note this is not a fully saturated model, as several constraints were imposed to overcome estimation problems (see text). Females Model configuration DQAICc Weight K DQAICc Weight K p(.hurr)psi(t)s-up(.)s-pd(t) p(.hurr)psi(.dry)s-up(.)s-pd(t) p(t,pd)psi(t)s-up(.)s-pd(t) p(t,pd)psi(.dry)s-up(.)s-pd(t) p(t,pd)psi(t)s-up(.)s-pd(.hurr) p(t,pd)psi(.dry)s-up(.)s-pd(.hurr) p(.)psi(t)s-up(.)s-pd(t) p(.hurr)psi(t)s-up(.)s-pd(t,pd) p(.hurr)psi(.dry)s-up(.)s-pd(t,pd) p(.hurr)psi(t)s-up(t,pd)s-pd(t) p(t,pd)psi(.dry)s-up(.)s-pd(t,pd) p(t,pd)psi(t)s-up(.)s-pd(t,pd) p(.hurr)psi(.dry)s-up(t,pd)s-pd(t) p(.)psi(.dry)s-up(.)s-pd(t,pd) p(.)psi(t)s-up(.)s-pd(t,pd) p(t,pd)psi(t)s-up(t,pd)s-pd(.hurr) p(t,pd)psi(t)s-up(t,pd)s-pd(t) p(t,pd)psi(.dry)s-up(t,pd)s-pd(.hurr) p(t,pd)psi(.dry)s-up(t,pd)s-pd(t) p(.)psi(t)s-up(t,pd)s-pd(t) p(.)psi(.dry)s-up(t,pd)s-pd(t) p(.hurr)psi(t)s-up(t,pd)s-pd(t,pd) p(.hurr)psi(.dry)s-up(t,pd)s-pd(t,pd) p(.hurr)psi(t)s-up(.)s-pd(.hurr) p(.hurr)psi(.dry)s-up(.)s-pd(.hurr) p(t,pd)psi(t)s-up(t,pd)s-pd(t,pd)* p(t,pd)psi(.dry)s-up(t,pd)s-pd(t,pd) p(.)psi(t)s-up(t,pd)s-pd(t,pd) p(.)psi(.dry)s-up(.)s-pd(.hurr) p(.)psi(t)s-up(.)s-pd(.hurr) p(.)psi(.dry)s-up(t,pd)s-pd(t,pd) p(.hurr)psi(t)s-up(t,pd)s-pd(.hurr) p(.hurr)psi(.dry)s-up(t,pd)s-pd(.hurr) p(.)psi(.dry)s-up(.)s-pd(t) p(.)psi(t)s-up(t,pd)s-pd(.hurr) p(.)psi(.dry)s-up(t,pd)s-pd(.hurr) Males In the male CR analysis, the top two models carried.97% of the QAICc weights and were used for model averaging. As with the female data, model selection favored configurations where pond survival was constrained to be equal across populations and varied with time; however, in the bestsupported male models, upland survival varied both across populations and among years. The dry year transition model had significantly more support than the time general model, but this appears to have resulted from the aggregation of other years (see Transition probabilities, below). In contrast to the female analysis, both top models allowed capture probabilities to vary by population and year. Pond survival. Pond survival for females in the aggregated populations (ponds 2, 3, 4, and 12) ranged from 0.71 (SE ) in 2001 to 0.95 (SE ) in 2003 (Fig. 2). Due to small sample sizes, two populations (pond 5 and 6) were configured separately in all models, each with timeconstrained pond survival to facilitate model convergence. Survival estimates for one of these populations, pond 5, were consistently and substantially lower than for the aggregated populations, with probabilities less than 0.60 in three breeding seasons. Pond survival estimates for breeding males followed very similar patterns to those of females, with survival probabilities estimated between 0.75 (SE ) and 0.93 (SE , Fig. 3). As with the female data, populations 5 and 6, respectively, had consistently lower and higher estimates than the aggregated populations. Upland survival. Survival probabilities of females in uplands during the non-breeding period (spanning winter, spring, and much of the summer) varied little across years with an average of 0.68 (average SE , Table 4). The average of all male estimates was 0.66; however, there was much more variability among these parameter estimates (min , max ) that did not appear to be explained consistently

7 400 Copeia 2009, No. 2 Fig. 2. Model-averaged survival probabilities for female Ambystoma opacum in ponds (breeding individuals) and uplands (non-breeding individuals) during the breeding seasons from 1999 to Ponds 5 and 6 were treated separately in all models to address estimation problems. Error bars reflect 6 1 SE, but are omitted for pond 6 for visual clarity. At this pond, standard errors range from 0.10 to by population or year effects. The one exception to this was the population of males at pond 5, which was characterized by dramatically lower estimates averaging 0.31 for all years combined. With pond 5 estimates removed from the calculation, average upland survival (non-breeding) in the remaining populations was Upland survival of nonbreeding females (during the breeding period) varied little across years and was consistently above 0.95, suggesting the females remaining in upland habitat incurred little mortality during these periods. Transition probabilities. Estimated probabilities of breeding in consecutive years differed between females and males. Females that bred in the previous year attempted to breed again at probabilities between 0.44 (SE ) and 0.65 (SE ), revealing that year-skipping is common, but also that females are capable of breeding in consecutive years (Fig. 4). Females that skipped breeding in one year were very likely to breed in the subsequent year with a probability of 0.86 (SE ). The year in which females had the lowest probability of breeding coincided with the year of lowest total precipitation (measured as total rainfall from May through September); however, precipitation was not explicitly tested as a linear covariate. In contrast to the females, males had a very high probability of breeding in all years, with observable to observable transitions estimated between 0.95 to 0.98 (SE for all estimates ). The dry year configuration had more support of the data than the time-general configuration; however, this probably represented an AIC benefit from aggregation of the remaining transition parameters rather than a specific dry year effect. An exploratory comparison with a fully time-constrained model for males suggested it would be selected over the dry year model. Capture probabilities. Female capture probabilities between 2000 and 2004 ranged from 0.78 to 1.0 (average ), but were considerably lower in the hurricane year (estimated at 0.62; Appendix 2). Male capture probabilities were more variable, ranging from 0.39 to 1.0 (average ). Three of the four lowest capture probabilities for males occurred at pond 5, and were much lower than corresponding estimates for females at this pond. Fig. 3. Model-averaged pond survival probabilities for male Ambystoma opacum during the breeding seasons from 1999 to Populations 5 and 6 were treated separately in years 2000 to 2004, with instantaneous survival constrained to be equal across years to address estimation problems. Variability reflected in yearly survival in these two populations results from different breeding interval lengths among years. Error bars reflect 6 1 SE. Simulation of data fragmentation. We estimated that approximately 7.3% of the total capture events were incorrectly fragmented from longer capture histories and therefore were treated as single capture individuals in the CR analysis. Almost all instances of missed matches were between one image (usually with poor image quality) and one or more remaining images of an individual; that is, missed matches between groups of images of the same individual were rare or non-existent. In the CR analysis of the known parameter data set with simulated fragmentation at these levels, estimates of transition, capture, and upland survival probabilities appeared unbiased and fell within 2% of paired estimates from unfragmented data. However, estimates of pond survival were 1 to 10% (x 5 4.6%) below corresponding unfragmented parameters, appearing to be systematically biased low. Simulation models. Based on the assumption that annual survival probabilities after first breeding are not agedependent, simulations estimated that females and males had similar patterns of longevity. Of the individuals that survived to breed, approximately 55%, 30%, and 20% lived at least one, two, or three additional years, respectively, but fewer than 10% lived for five or more additional years. The average female and male (i.e., that survived to breed) survived 1.3 and 1.4 years beyond first breeding, respectively. Given higher probabilities of skipping breeding seasons, average lifetime breeding events for females (x 5 1.9) were lower than for males (x 5 2.3). The male distribution was weighted slightly more to the tail, with approximately 21% of individuals breeding more than three times as compared to 13% of the females. Adjustments to the simulation parameters to correct for bias estimated from data fragmentation increased these estimates only slightly to 2.0 breeding attempts for females and 2.5 attempts for males. DISCUSSION Spatio-temporal variation in survival and implications for metapopulation dynamics. In this study, we focused on adult survival and breeding probabilities to derive unbiased estimates of these parameters and to assess how variability in this life stage is distributed over time and among populations. Our results generally agreed with our expectations that in a forested terrestrial environment with minimal anthropogenic disturbances, adult survival proba-

8 Gamble et al. Survival in Ambystoma opacum 401 Table 4. Upland Survival Estimates and Standard Errors for Female and Male Ambystoma opacum during Non-Breeding Intervals from 1999 to Intervals extend from median date of emigration in listed breeding year to median date of immigration in following year (e.g., typically about 49 weeks for males and 47 weeks for females). The occasion number representing the beginning of the interval is listed in parentheses. Females Interval All ponds a Pond 2 Pond 3 Pond 4 Pond 5 b Pond 6 b Pond 12 Average 1999(2) (4) (6) (8) Fixed (10) Average a Models constraining instantaneous survival probabilities to be equal across populations carried.99% of weights given the data (Table 3), so these estimates apply to all populations. b To overcome estimation problems, instantaneous survival probabilities were constrained to be equal across all years (see text in Materials and Methods), so yearly variation in these values results from slight differences in interval lengths among years. Males bilities would not be a significant source of variability among proximate populations. For example, in both the female and male analyses of pond survival, model selection favored models that allowed variability across years, but not among populations. When aggregated in dry pond basins, breeding salamanders were probably more susceptible to environmental fluctuations that occurred over time (e.g., dry spells or extreme temperatures) than were non-breeding individuals in familiar upland territories with better opportunities for cover; however, these fluctuations were experienced similarly among populations. Pond-level variability in habitat quality (e.g., unmeasured factors such as predator abundance, availability of refugia) was not significant enough to result in support for models with pond-specific estimates. Similarly, models fit to the female data with upland survival constrained over time and populations were favored over fully general models. Resulting parameter estimates for female upland survival deviated little from an average of 0.68 through the duration of the study period, suggesting that terrestrial refugia away from the pond basins were sufficient to buffer extremes in both temperature and moisture that may occur across years and seasons. We can only speculate on what particular climate extremes may affect adult survival; however, this finding is noteworthy given the range of climatic variability that was apparent over Fig. 4. Estimated probabilities of remaining in a breeding (observable) state in two consecutive breeding seasons for male and female Ambystoma opacum from 2000 to The line displays total precipitation amounts from May through September in each year. Error bars reflect 6 1 SE. this time period. For example, the 1999 breeding season was exceptionally hot with a monthly average temperature 5.4uC above normal in September, and January 2004 was unusually cold (monthly average temperature 2.6uC below normal) with minimal snow cover (National Oceanic and Atmospheric Administration, 2008). It is important to note, however, that any resilience to climatic variation apparent in these data may not be generalizable to areas with poor or more varied habitat quality. We did find two notable exceptions to the general pattern of spatial homogeneity of survival. First, the models with the most support in the male data allowed upland survival to vary across populations and years. A closer look at these parameter estimates showed that three of the six populations (including two with the largest sample sizes and lowest standard errors) experienced similar upland survival probabilities (during non-breeding periods) around 0.75 through much of the study period (Table 4). Ponds 2 and 3, on the other hand, exhibited significant year-to-year variability in upland survival estimates, conveying support to the population-variable models. Estimates at these two populations were accompanied by large and overlapping confidence intervals, however, suggesting they should be interpreted cautiously with regard to spatial or temporal trends. Second, pond survival estimates for both females and males and upland survival estimates for males during the non-breeding periods were consistently and substantially lower for pond 5 than for the remaining populations. Reasonably low variances in these estimates suggest that they reflect real differences in this population; however, we are unable to offer any definitive explanation. A 50-m wide powerline corridor traversing the study area in close proximity to pond 5 is regularly cleared of woody vegetation and may act as a barrier to movement or potentially as an area of high predation risk; however, upland habitats appear accessible around the remainder of this pond and similar to those in the rest of the study area. Despite these two inconsistencies, our results generally indicate some variation in adult survival over time, particularly during breeding periods, but minimal variation among populations. Similar to these results, A. tigrinum adults in Virginia showed interannual variation in pond survival and very limited temporal variation in upland survival (Church et al., 2007); however, data in the A. tigrinum study suggested that population-level variability occurred in both pond and

9 402 Copeia 2009, No. 2 Table 5. Summary of Estimated Annual Survival and Breeding Frequency for Selected Pond-Breeding Ambystomatids. When available, standard error was provided in parentheses with the parameter estimate for annual survival, unless footnoted otherwise. Species Location Sex a survival Annual Proportion breeding b Lifetime breeding attempts c Source A. californiense CA f 0.60(0.31) d 1.36 (Trenham et al., 2000) m 0.50(0.20) d 1.59 CA f/m 0.67(0.03) e (P. Trenham, pers. comm.) 0.87(0.07) f A. maculatum AL f/m 0.63(0.032) 0.70(0.04) (Blackwell et al., 2004) MI f 0.63/0.80 g 0.32 (Husting, 1965) m 0.79/0.94 g 0.36 RI f/m 0.90 (Whitford and Vinegar, 1966) A. opacum SC f 0.70 h (Taylor and Scott, 1997) f 0.80 h (Taylor et al., 2006) f (Taylor and Scott, 1997) MA f 0.62(0.01)/0.60(0.01) i 0.67(0.02) 2.0/1.9 f (This study) m 0.66(0.02)/0.59(0.01) i 0.96(0.01) 2.5/2.3 f A. talpoideum LA f/m 0.74 (Raymond and Hardy, 1990) A. tigrinum VA f j,0.30,1.5 (Church et al., 2007) m j,0.50 a f female; m male; f/m females and males pooled. b Proportion of adult population estimated to be breeding in a year. c Estimated lifetime breeding attempts for individuals that survive to breed at least once. d One standard deviation included in parentheses. e Estimated using maximum likelihood based approach without accounting for dispersing individuals. f Estimated using maximum likelihood based approach and accounting for dispersing individuals. g First estimate from Method 2 and second from Method 1 adjusted, both in source appendix. h Estimated indirectly from simulation models with parameter adjusted to level necessary to explain observed population trend. i First estimate excludes population at Pond 5, an apparent outlier; second includes all populations in study. Both were calculated as lambda of matrix to integrate survival in breeding and non-breeding years (see Appendix 3). j Calculated as breeding season survival * non-breeding season survival from averages from three populations for breeders (first estimate in range) and non-breeders (second estimate in range). Standard errors of original parameter estimates available in source. upland survival. A. tigrinum are aquatic breeders and have a significantly longer breeding season than A. opacum. Persistent differences in hydrologic regimes among breeding ponds (e.g., ponds that have short hydroperiods on average as compared to those with long hydroperiods) may account for this population-level variability as compared to terrestrial nesting A. opacum. As with our study, population-level variability in upland survival estimates was largely unexplained, highlighting the common difficulty (and lack of understanding) in identifying meaningful differences in upland habitat quality for terrestrial salamanders. As we consider the implications of our findings for population dynamics at broad spatial scales, Hanski and Gilpin (1997) suggest than an essential component of strict or classic metapopulation models (Levins, 1969, 1970) is some level of asynchrony in local population trends. These theoretical metapopulation models are characterized by many ephemeral habitat patches blinking off and on over the long term as the result of local extinctions and recolonizations. Our results suggest that at scales comparable to our study area (and with similarly consistent upland habitat conditions), adult survival is unlikely to act as a highly heterogeneous parameter promoting asynchrony among populations. Further, if potential breeding wetlands are fairly stable geological features persisting in landscapes over decades to centuries, then demographic synchrony or lack thereof is likely to be largely driven by spatio-temporal variability in reproductive success (Gamble, 2004) and/or survival in the pre-adult stages. The potential for dramatic temporal variability in amphibian populations is well established. An A. opacum population in South Carolina, for example, was monitored from apparent colonization to thousands of breeding individuals in a period of less than ten years (Semlitsch et al., 1996). Success in the egg and larval stages can vary in response to hydroperiod, aquatic community dynamics, or catastrophic disease, among other factors (Wilbur, 1987; Gamble, 2004). A critical remaining issue will be to determine how these sources of variability are distributed among local populations and over time. A high level of independence in local variability is likely to promote regional dynamics approximating classic metapopulation models; however, if the most significant environmental variation occurs at regional scales, then less dynamic source sink population models may better represent these populations. Adult survival and breeding frequency in pond-breeding ambystomatids. Annual survival estimates in the literature for pond-breeding ambystomatids vary widely from 0.50 (Trenham et al., 2000) to 0.90 (Whitford and Vinegar, 1966) and do not indicate consistent differences between males and females (Table 5). For comparison purposes, we generated steady-state annual survival probabilities for A. opacum

10 Gamble et al. Survival in Ambystoma opacum 403 at our study site using a transition matrix that integrates breeding state transitions and associated survival probabilities (see Appendix 3 for details). Resulting annual survival estimates were 0.62 (SE ) and 0.66 (SE ) for females and males, respectively (0.60 [SE ] and 0.58 [SE ] with pond 5 included), and occur at the low end of published estimates for other pond-breeding ambystomatids. Though multiple confounding variables limit meaningful comparisons of these estimates across species (see discussion of error, below), our estimates were notably lower than the A. opacum population at Rainbow Bay, South Carolina. Using a stochastic population model, Taylor et al. (2006) indirectly estimated that salamanders at this site must survive terrestrial stages at 80% annually to explain observed trends in this population, and at 60% annually to persist in the long term. If survival in the egg and larval stages is similar between our study populations and Rainbow Bay, our populations may exist at levels very close to this persistence threshold, in which case they would be particularly vulnerable to any negative impact on adult survival. These populations differ, of course, in their locations relative to range boundaries for this species. If density is a significant determinant of aquatic survival as has been indicated in previous work (Scott, 1990), the likelihood of lower larval densities in the study populations, which approach the northern limits of the species range, may improve survival in this stage, perhaps compensating partially for differences in adult survival. Skipping potential breeding years is also reported for many pond-breeding ambystomatids (Table 5) as well as for other amphibians. This has been attributed to energetic readiness (Scott and Fore, 1995; Harris and Ludwig, 2004), to avoidance of risks associated with breeding in drought years (Church et al., 2007), and/or to selection of optimal conditions for migration (Douglas, 1979; Semlitsch, 1985). None of these factors appeared to generate substantial variability among years for males in our study, as greater than 95% were likely to breed in any given year, and migrations were occasionally observed without rainfall. For females in our study, the lowest transition to breeding was experienced during the lowest precipitation year (Fig. 4); however, most females still bred in most years, and very few skipped consecutive years. The higher probability of breeding in individual females who skipped a previous season suggests there is also an energetic recovery component for females in these populations. It is also possible that selective pressures toward breeding cues are weighted differently in A. opacum than they are for springbreeding ambystomatids. That is, weather-based cues may have more predictive value to spring breeders, indicating favorable aquatic egg-laying and larval conditions (i.e., hydroperiod) as well as migration conditions. If late summer climate variables are less informative to A. opacum in these ways, there may be a greater fitness advantage to investing less resources in each breeding event and breeding more often (i.e., spreading the risk). The finding of Pechmann et al. (1991) that breeding numbers of A. tigrinum and A. talpoideum were correlated with breeding season rainfall, while those of A. opacum were not, is generally consistent with this argument; however, experimental work (e.g., interspecific comparisons at the same geographic locations) that addresses numerous potential confounding variables would be necessary to further explore this hypothesis. Female A. opacum that survived to breed in our study area averaged 2.0 total lifetime breeding attempts as compared to estimates of less than 1.5 in A. californiense (Trenham et al., 2000) and A. tigrinum (Church et al., 2007). Error and sensitivity. The wide range of survival estimates reported in Table 5 represents variability among species, geographic regions, and/or habitat quality at different study locations as well as estimation error. Even studies conducted on the same species at the same location resulted in very different estimates when different methods were used. For example, Husting (1965) reported annual survival for female A. maculatum at 0.63 or 0.80 using two different methods. Similarly, Trenham et al. (2000) reported annual survival of 0.60 and 0.50 for female and male A. californiense; however, later use of maximum-likelihood based approaches resulted in a parameter estimate for both sexes of 0.67 (SE ), or 0.87 (SE ) when taking dispersal into account (P. Trenham, pers. comm.). These differences are striking, particularly given the known sensitivity of population models to small changes in adult vital rates (Biek et al., 2002; Vonesh and De la Cruz, 2002; Harper et al., 2008) and the related importance of the adult stage as a storage stage that may offset frequent reproductive failures in ephemeral habitats (Taylor et al., 2006). For these reasons, it is essential that robust analytical approaches are used to minimize bias that could result from the occurrence of unobservable states, varied capture probabilities, or other confounding variables and to provide estimates of error in survival parameters (Lebreton et al., 1992). The modified open robust design (Bailey et al., 2004) applied in this study, paired with new methods for uniquely identifying individuals in large data sets (Gamble et al., 2008), has advanced this goal significantly, though it relies on some simplifying assumptions. For example, upland survival estimates from this method represent the period of time from post-breeding to the subsequent pre-breeding capture points at drift fences, and therefore include sources of variability associated with breeding migrations. These limitations should be stated explicitly when parameterizing population models with these vital rates and addressed with the careful use and reporting of sensitivity and elasticity results (Biek et al., 2002). Ecological implications and future directions. Much recent work has emphasized the need to assess amphibian demography and population dynamics at multiple spatial and temporal scales when considering appropriate management or conservation strategies (Semlitsch, 2000; Cushman, 2006). Spatially, it is necessary to consider the breeding habitat (factors affecting breeding and larval success), the terrestrial habitat (factors affecting terrestrial survival), and the landscape or metapopulation scale (factors affecting dispersal and population connectivity; Compton et al., 2007). Temporally, it is necessary to consider variability in the different life stages, as well as interactions among these life stages and their effects on population-level changes (Vonesh and De la Cruz, 2002). In this study, with few exceptions, we did not observe dramatic variability in adult survival among six populations in proximity to each other. With the caveat that these populations were likely subjected to fairly homogeneous upland habitat conditions, this finding suggests that the distribution of variability in preadult stages will be important in determining the applicability of metapopulation principles to this system and to other pond-breeding ambystomatids in similar settings.