Estimating demographic parameters for loggerhead sea turtles using mark recapture data and a multistate model

Transcription

1 Popul Ecol (2011) 53: DOI /s x ORIGINAL ARTICLE Estimating demographic parameters for loggerhead sea turtles using mark recapture data and a multistate model Melissa Hedges Monk Jim Berkson Philippe Rivalan Received: 11 March 2009 / Accepted: 14 February 2010 / Published online: 23 April 2010 Ó The Society of Population Ecology and Springer 2010 Abstract The survival for adult loggerhead sea turtles from a saturation tagging study on Bald Head Island, NC, USA, was estimated using a multistate model with unobservable states to relax assumptions that are violated when survival is estimated from multistate models and produce more accurate estimates of survival, recapture, and breeding transition probabilities. The influence of time, trap dependence, and low site fidelity to the study nesting beach on survival and recapture were examined. The best model given the data included an imprecise site-fidelity effect on survival, constrained the reproductive cycle to 4 years, and contained a time effect on recapture rates. The estimate of annual survival for adult females was of 0.85, producing the highest estimate in the literature for loggerhead sea turtles. Multistate models should be applied to other nesting beach data for sea turtles to improve survival estimates and in turn the ability to model and manage populations. M. H. Monk Department of Fisheries and Wildlife Science, Virginia Tech, Blacksburg, VA 24061, USA J. Berkson National Marine Fisheries Service RTR Unit, Virginia Tech, 114 Cheatham Hall, Blacksburg, VA 24061, USA P. Rivalan Centre d Etudes Biologiques de Chizé, CNRS-UPR 1934, Villiers-en-Bois, Beauvoir-sur-Niort, France Present Address: M. H. Monk (&) Department of Oceanography and Coastal Studies, Louisiana State University, 2197 Energy, Coast, and Environment Building, Baton Rouge, LA 70803, USA mhedge1@tigers.lsu.edu Keywords Breeding cycle Caretta caretta Nesting-beach survey Population dynamics Survival Introduction Effective management of threatened and endangered species should begin with accurate estimates of life-history parameters, such as survival, birth, and mortality rates (Caswell 2000). Policymakers depend on the best available data to weigh management alternatives, which are only as robust as the data (Beissinger and Westphal 1998; Akçakaya and Sjögren-Gulve 2000). Whereas loggerhead turtles (Caretta caretta) are difficult to study due to their complex and elusive life history, the benefits gained from mark recapture studies are invaluable. A number of key demographic parameters can be estimated from mark recapture studies of nesting females, including survival, remigration intervals or breeding probabilities, and reproductive capacity (Nichols et al. 1992). These demographic parameters are all input parameters for population models used to manage the species, estimate recovery times, and promote conservation measures. Sensitivity analyses have shown that proportional changes in adult and, to a smaller extent, large juvenile survival rates have the largest effect on the population growth rate for marine turtle species (Crouse et al. 1987; Congdon et al. 1993; Heppell et al. 1996). Therefore, accurate and updated estimates of survival for adult loggerheads should be estimated whenever possible. Loggerhead females emerge from the ocean to lay multiple clutches of eggs during a breeding season and also exhibit site fidelity to a particular nesting beach within and between seasons (Bell and Richardson 1978; Limpus 1985). The ability or chance to capture a female multiple

2 166 Popul Ecol (2011) 53: times within a nesting season as well as site fidelity between seasons makes the reproductively active stage of loggerheads ideal for mark recapture studies. Despite the ease of capturing nesting loggerheads, the time and cost required to facilitate these studies over long periods is often a limiting factor. As a result, only three estimates of adult survival exist for loggerhead sea turtles nesting in the southeastern USA (Frazer 1983; TEWG 2000). Providing additional estimates of survival for the adult life stage has been identified as a critical research need (TEWG 2000; Heppell et al. 2003). The previous estimates of loggerhead survival have been obtained using single-state models, such as the Cormack Jolly Seber (CJS) model, which assume that at each sampling period (i.e., nesting season), every individual in the marked population has the same probability of capture at the study site (i.e., nesting beach) (Cormack 1964; Jolly 1965; Seber 1965). However, loggerheads exhibit reproductive skipping with observed breeding cycles ranging from 1 to 9 years and averaging years (Bell and Richardson 1978; Hughes 1982; Bjorndal et al. 1983; Limpus 1985; Dodd 1988; Schroeder et al. 2003). An individual that is not breeding a particular year is not available for capture on the nesting beach. Therefore, the capture probabilities for all individuals in the study population are not homogenous and depend on when the individual last nested (Balser 1981; Kendall and Nichols 1995; Prévot-Julliard et al. 1998). Models that do not account for heterogeneous capture probabilities negatively bias survival estimates. Loggerheads exhibit philopatry to an initial nesting beach and have been shown to return to a chosen nesting beach both within and between nesting seasons (Carr 1975; Miller et al. 2003; Schroeder et al. 2003). On average, the observed distance between nest sites within a nesting season for an individual female is 5 km or less, but the distance can be more variable both within and between seasons (Bell and Richardson 1978; Miller et al. 2003; Schroeder et al. 2003). Imprecise nest-site fidelity between years is one reason individuals may not be recaptured at the study site, especially in the nesting range of loggerheads along the eastern coast of the United States. Nesting-beach surveys are often constrained by the feasibility (time and cost) of monitoring a portion of a beach. Additionally, the boundary of a particular beach, e.g., a political or park boundary, limit the implementation of beach surveys. Loggerheads may exhibit site fidelity within an average distance, but individuals nesting at the boundary of the study area may nest on proximal beaches and will escape detection. Imprecise site fidelity is not the only reason individuals may not be recaptured. Other possibilities include mortality and nesting in the study area but escaping detection. A multistate mark recapture model offers a framework that can incorporate these characteristics of nesting loggerhead data (i.e., reproductive skipping and imprecise site fidelity) by allowing marked individuals to move between states (Kendall 2004). States in a multistate model are not limited to geographic locations but can be physiological (i.e., sex, weight, age), reproductive (i.e., breeder, nonbreeder), and completely unobservable (Nichols et al. 1992, 1994). Multistate models have been used to model unobservable seed/dormant states in plants (Kéry et al. 2005), breeding cycles in leatherback sea turtles (Rivalan et al. 2005) and fur seals (Beauplet et al. 2006), and reproductive cycles in bird species (Doligez et al. 2002; Véran et al. 2007; Hunter and Caswell 2009) but have never been applied to loggerhead sea turtles. The key development in multistate models for nesting sea turtle data was the ability to model an unobservable state, any life stage or portion thereof for which an individual s detection probability at the study site is zero (Lebreton et al. 1992; Fujiwara and Caswell 2002; Kendall and Nichols 2002; Lebreton and Pradel 2002). During the time an individual is unobservable, it is considered to have temporarily emigrated from the study population. This is appropriate for modeling reproductive skipping, as loggerheads physically leave the study area for foraging grounds between years of active reproduction. To observe the nonbreeding years, one would have to conduct comprehensive studies on the foraging grounds. A multistate model estimates the breeding transition probabilities given the individuals recapture histories. The probability of an individual transitioning to the breeder state is conditional on its state the previous year, i.e., how many years it has spent as a nonbreeder. Estimating the breeding transition probabilities provides more accurate estimates of the length of the reproductive cycle, which is variable on the individual level. By modeling reproductive skipping and site fidelity, the multistate model provides more accurate estimates of adult survival and unbiased breeding transition probabilities (Kendall and Bjorkland 2001; Rivalan et al. 2005). This study utilized the multistate model developed by Rivalan et al. (2005) to estimate survival rates, breeding transition probabilities, and recapture rates for loggerhead sea turtles nesting in the northern subpopulation along the southeastern coast of the USA. Methods Study site and data collection Data on nesting loggerhead sea turtles were collected on Bald Head Island, NC, USA, the southernmost barrier

3 Popul Ecol (2011) 53: island in North Carolina, located at the mouth of the Cape Fear River. From 1991 to 2006, a tagging study was conducted to identify nesting females and describe the activity of loggerheads on the island. Further details on the study site and data collection can be found in Hawkes et al. (2005). From 1991 to 2006, females were flipper-tagged on the trailing edge of each front flipper, and from 2002 to 2006, females were also tagged with passive integrated transponders (PIT tags) injected into the left shoulder. Estimates of tag loss were not available for this study before the use of more permanent PIT tags in Hawkes et al. (2005) analyzed the data set from 1980 to 2003, presenting descriptive statistics including nesting data and comparison statistics between apparent neophyte and remigrant nesting females. As part of our study, data were thoroughly reviewed and corrected for transcription errors in the electronic tagging database. Hawkes et al. (2005) reported that from 1991 to 2003, 477 individual loggerhead turtles had been tagged. We examined the original data sheets and found errors in data transcription to computer files. After review and after adding 3 more years of data, only 415 tagged individuals from 1991 to 2006 were logged in the new database. Nest and hatching data were also reviewed and corrected. Multistate model structure Capture histories were created for every tagged individual over the 16 years of available data, with one entry for each year. We followed a conditional Arnason Schwarz multistate model in which survival and capture probabilities were separately identifiable (Lebreton et al. 1992; Nichols et al. 1994; Schwarz and Arnason 1996). All models were run in the program M-SURGE using an identity link function (Choquet et al. 2005a), which transforms the mathematical parameters in the multistate model while keeping them within the biological range (0, 1). However, we used an identity link function that does not transform the data, because we modeled additive effects on more than two states (Choquet et al. 2005a). M-SURGE implements a specific algorithm to keep parameters within biological boundaries. Within the model, individuals could transition through a number of states, one observable breeder state, and a defined number of unobservable nonbreeder states (Fig. 1). Individuals were catchable only when they were on the beach during the breeder state. For the other unobservable states for years when nesting was skipped, each state was defined by how many years it had been since the individual last nested. Transitions between the breeding and skipped-breeding states, as well as among skippedbreeding states, were modeled as a first-order Markov process (Kendall et al. 1997; Pradel et al. 1997; Fujiwara and Caswell 2002; Kendall and Nichols 2002). Markovian temporary emigration conditions a female s transition probability on the previous year s breeding status (Viallefont et al. 1995; Kendall et al. 1997). The number of years spent in the nonbreeder state was constrained in the model structure by specifying the maximum length of the breeding cycle. Although skipping only 1 year between nesting is not the norm for loggerheads, it was observed on rare occasions during the study and included as a possibility for the length of the breeding cycle. The maximum length of the breeding cycle examined in this study was 5 years, which allowed individuals to skip up to 4 years before breeding again. The parameters estimated in the multistate model include breeding transition probabilities, capture probabilities, and survival. A transition probability (W) is the probability of transitioning from one state to the next. There is an observable breeder state denoted B and unobservable nonbreeder states denoted NB (Fig. 1). The numerical subscripts on NB indicate the number of years as a nonbreeder. Individuals were required to transition to the breeder state after reaching the last nonbreeder state Fig. 1 The four-stage model of the loggerhead breeding cycle. State B is the observable breeder state shown in the white circle. States NB 1, NB 2, and NB 3 are the unobservable nonbreeder states, representing the years of reproductive skipping and shown with the shaded circles. The transition probabilities (W) are the probabilities of an individual moving between states, with all possible transitions shown with an arrow. Superscripts in the transition probabilities represent movement from one state to the next. Transitions to nonbreeder states were calculated as the complements of the probability of transitions to the breeder state. The transition from the last possible nonbreeder state to the breeder state was constrained to 1.0 for all models. Survival probabilities (S) are equal for breeders and nonbreeders. Example: S W NB2!B is the probability of an individual surviving and transitioning to the breeding state after spending 2 years in the nonbreeder state

4 168 Popul Ecol (2011) 53: specified in the model. The transition W NBx!B, where x is the maximum years in the nonbreeder state, was constrained to one. The probabilities of transitioning between two nonbreeder states were calculated as the complements of the estimable probability of transitioning to the breeder state, e.g., W NB 1!NB 2 ¼ 1 W NB1!B : A more detailed description of the transition probabilities in matrix form can be found in Rivalan et al. (2005). Capture probability denoted by P is the probability an individual is recaptured at the study site while in the breeder state. The possible effects on capture probability include a time (year) effect, denoted P t, and a trap-dependence effect, denoted P m. The trap-dependence effect accounts for the heterogeneous recapture probabilities resulting from the temporary emigration from the study site during years as a nonbreeder. The assumption of homogeneous capture was relaxed by pooling the data to model immediate trap dependence (Lancia et al. 1994; Pollock et al. 2002). Individuals in the breeder state, regardless of being recaptured at time t, have a different capture probability at time t? 1 than individuals that were in any given nonbreeder state at time t (Pradel 1993). True trap dependence (i.e., trap happy, trap shy) was not considered an issue in this study, as tagged loggerheads do not show an affinity for or avoidance behavior of a nesting beach where it was originally tagged (Broderick and Godley 1999). A large proportion of individuals in this study were marked and never recaptured. If this phenomenon was ignored and individuals only captured once were included in the survival model, the estimated apparent survival would be negatively biased. We attribute the large number of individuals never recaptured to imprecise site fidelity of individuals, including first-time nesters that had not established a permanent nesting beach, and the boundaries of the study site. The study site is contiguous with suitable nesting habitat and also neighbors Oak Island, NC, which receives a comparable number of loggerhead nests but has no monitoring/tagging program. Loggerheads from this study have been resighted on nesting beaches as far south as Brevard County, FL, and as far north as Cape Lookout Point, NC. There have also been very few reports of dead recoveries carrying tags from our study, furthering our belief that the majority of individuals initially tagged on Bald Head Island remained alive in the larger population. To account for imprecise site fidelity, the methods described in Pradel et al. (1997) to account for transient individuals were used. Pradel et al. (1997) outlined two methods for estimating survival when transient individuals are detected, both of which estimate apparent resident survival for individuals recaptured at least once. The first method is an ad hoc approach of discarding the first capture of every individual and is appropriate when the capture rates over the duration of the study are high. In this study, capture rates were not consistently high, and therefore, the second, more formal, approach was applied. For the formal method, two artificial age groups are created from the capture histories. The age 1 group includes all individuals captured in the study and represents the time from first marking to the next year. The age 2 group includes only individuals that were recaptured at least once, discarding the first capture. Age classes are not reflective of an individual s true biological age, but rather of artificial groups, representing the number of years since the individual was first captured. Survival for age 1 is survival from the time at first marking to the next year and includes all individuals captured in the study, whereas survival for age 2 is conditioned on the first recapture and includes only individuals that were recaptured at least once. Survival estimate for age 2 may remain biased low, because individuals permanently emigrated from the study population cannot be discerned from individuals that exhibit imprecise nest-site fidelity or have died. The imprecise site fidelity effect in the model is denoted with a subscript s on the survival parameter S s (Pradel et al. 1997; Prévot-Julliard et al. 1998). Survival for s1 (S s1 ) is survival for age 1 individuals described above, or for those exhibiting imprecise site fidelity. Survival of s2 (S s2 ) is the survival estimate for individuals in the age 2 group and is the survival rate that is of biological relevance, representing the survival estimate for resident individuals. All individuals, whether in the observable or unobservable states, were assumed to have the same survival, as data were not available to estimate a separate survival for nonbreeding females on the foraging grounds (Kendall 2004). Determining the estimability of parameters is key to interpreting any biological meaning from the multistate model output (Catchpole and Morgan 1997). M-SURGE uses a numerical Catchpole Morgan Freeman approach to identify parameter redundancy (Choquet et al. 2005a). We also verified parameter estimability for the two state model with Appendix B from Schaub et al. (2004). Parameters known to be zero, such as transition probabilities between particular unobservable states and capture probabilities for unobservable states, were constrained to zero. Models with time-dependent transition probabilities were also not considered due to the unrealistic number of parameters they would introduce to the model and parameter estimability issues (Kendall and Nichols 2002; Schaub et al. 2004). Given the available data, convergence issues would arise in a model with many more parameters (Rivalan et al. 2005). The number of identifiable parameters for each model are presented with results in Table 1.

5 Popul Ecol (2011) 53: Table 1 Model selection statistics using the Quasi- Likelihood Akaike Information Criteria (QAIC), with the number of identifiable parameters in parentheses The five models with DQAIC values of less than 7.0 are in bold S the survival probability, P the capture probability, t time effect, m trap-dependence effect, s site fidelity effect,? an additive effect, * an interaction effect Capture probability model Survival probability model S s*t S s S t S s?t Breeding cycle constrained to 2 years P t (46) (17) (31) (46) P t?m (47) (19) (30) (47) P t*m (60) (23) (44) (60) Breeding cycle constrained to 3 years P t (46) (17) (24) (46) P t?m (48) (19) (29) (48) P t*m (61) (19) (43) (61) Breeding cycle constrained to 4 years P t (47) (17) (16) (47) P t?m (49) (18) (28) (49) P t*m (62) (18) (32) (62) Breeding cycle constrained to 5 years P t (48) (17) (16) (48) P t?m (47) (18) (29) (50) P t*m (63) (18) (32) (63) Candidate model parameterization The candidate models were parameterized by varying effects on survival and capture probabilities and changing the maximum length of the breeding cycle. Survival models included a time effect (year) (S t ), site-fidelity effect (S s ), interaction between time and site fidelity (S t*s ), and additive effects of time and site fidelity (S t?s ). The possible effects on capture included time (P t ), an interaction effect between time and trap dependence (P t*m ), and an additive effect of time and trap dependence (P t?m ). No models were tested with only a trap-dependence effect on recapture due to variability in sampling effort between seasons over the 16-year period (Hedges 2007). The suite of models simulated included each possible combination of effects on survival and recapture and with the breeding cycle constrained to 2, 3, 4, and 5 years (W cycle=2, W cycle=3, W cycle=4, W cycle=5 ). Goodness-of-fit (GOF) test No formal GOF test exists for a multistate model with unobservable states. As in Rivalan et al. (2005), the fit of the loggerhead data to a multistate model was tested with a modified single-state GOF test composed of two test components: test 2.CT and test 3.SR run in program U-CARE (Lebreton et al. 1992; Choquet et al. 2005b). Test 3.SR detected whether newly marked individuals had the same probably of recapture in successive years as those individuals marked in previous years (i.e., detection of individuals never recaptured). Test 2.CT determined that the probability of capture at time t - 1 was dependent on probability of capture at time t, interpreted as immediate trap dependence (Pradel 1993). Test 2.CT also detected nonrandom temporary emigration, a first-order Markov process, which would require data to be modeled in a multistate framework (Kendall et al. 1997; Pradel et al. 1997; Schaub et al. 2004). The final GOF and associated degrees of freedom (df) were calculated by subtracting the chi-squared values for test components of temporary emigration, the issue of site fidelity, and reproductive skipping from the full-model chi-squared test (Viallefont et al. 1995): GOF m ¼ v 2 total v2 Test2:CT v2 Test3:SR Ddev with df m ¼ df total df Test2:CT df Test3:SR 1 where Ddev was the change of deviance between the full model that accounted for reproductive skipping (denoted [S t*s, W cycle=2, P t*m ]) and the model that did not account for reproductive skipping (denoted [S t*s, P t*m ]). The variance inflation factor, ^c, was calculated as ^c ¼ GOF m =df m to account for overdispersion in the data (Burnham et al. 1987; Lebreton et al. 1992). Model selection Models were selected using the Quasi-Likelihood Akaike Information Criteria (QAIC), DQAIC (QAIC of any model subtracted from the QAIC of the top model) and model deviance with consideration given to biological relevance (Akaike 1973, 1983; Lebreton et al. 1992; Buckland et al. 1997; Burnham and Anderson 2002). Values were used to

6 170 Popul Ecol (2011) 53: analyze the relative support of the data for each model in relation to the model with the lowest QAIC (Burnham and Anderson 2002). Akaike weights were used to calculate normalized model weights, w, for any model i in a set of R models (Anderson and Burnham 1999): expð w i ¼ 1D 2 iþ P R r¼1 exp ð 1D 2 rþ where D i was the difference in model QAIC values. The model weight is a relative index for model i s likelihood against any other model in the set. Ratios of model likelihoods, w i/ w j, give a relative index for the strength of one model to another. Fig. 2 Number of observed loggerhead sea turtle nests deposited on Bald Head Island, NC. Points before 1992 may include an insignificant number of green sea turtle nests. Data from 1980 to 2003 were previously published in Hawkes et al. (2005) Results Nesting trend on Bald Head Island Annual nest numbers for loggerheads nesting on Bald Head Island are found in Fig. 2, including data from 1980 to 2003 first published in Hawkes et al. (2005). Annual nest numbers include all loggerhead nests identified on Bald Head Island regardless of female detection. The linear regression of the number of nests laid per year indicated a weak but significant negative trend through time over the period (R 2 = 0.29, F 1,25 = 10.21, P \ 0.001). Observed loggerhead nests on Bald Head Island are declining by -2.8 nests per year from 1980 to 2006, with a 95% confidence interval (CI) of (-5.2, -0.4). During the study period from 1991 to 2006, 415 female loggerheads were identified and tagged. A total of 67 marked individuals were recaptured at least once, and the other 348 individuals were never recaptured. Model selection The GOF test indicated a lack of fit of the data to the global model [S s*t, W cycle=2, P t*m ]. Test component 3.SR was significant (v 2 = 65.76, df = 12, P \ 0.001) and suggested individuals may not be faithful to the study site as a nesting beach. Test component 2.CT also tested significant (v 2 = 53.85, df = 13, P \ 0.001) and signified immediate trap dependence, which was biologically interpreted as nonrandom temporary emigration (Choquet et al. 2005b). The overdispersion coefficient, ^c ¼ 1:56 and was used in M-SURGE for adjusting standard deviations (SD) and calculating QAIC values. Model 1, parameterized as [S s, W cycle=4, P t ], had the lowest QAIC, with 17 identifiable parameters (Table 1). Model 2, [S s, W cycle=3, P t?m ], had the next lowest QAIC value, with 19 identifiable parameters. Models 3 and 4, [S s, W cycle=4, P t?m ] and [S s, W cycle=5, P t ], had the same model deviances, DQAIC values of 2, and contained 18 and 17 identifiable parameters, respectively (Table 2). No differentiation could be made among the top five models based upon model deviance or DQAIC values (Table 2). The model weights and relative likelihoods also lent little to the differentiation among models 1 5 (Table 2). Model 2 estimated the probability of surviving and breeding in two consecutive years at 22%, which did not accurately represent the pattern of reproductive skipping observed in the wild (Table 3) (Richardson et al. 1978; Hughes 1982; Frazer 1983). The high estimate of W B?B in model 2 was also a result of the additive trap-dependence effect on recapture (Rivalan et al. 2005). The trap-dependence effect introduced reproductive skipping into the capture estimates to compensate for the lack of skipping allowed when the breeding cycle was constrained to only 3 years. Therefore, model 2 was not deemed reasonable and was rejected from further consideration. Transition probability estimates for the top five models were the same for models 1, 3, 4, and 5 (Table 3). The probability of an individual nesting in two consecutive years was 3%. The probability of nesting after spending 1 year as a nonbreeder was 31%, and the probability of nesting after spending 2 years as a nonbreeder was 81%. In models 4 and 5, the breeding cycle was constrained to 5 years, but the model estimated the probability of breeding after spending 4 years as nonbreeder at 1.0. This indicated that all females had bred after 3 years as a nonbreeder, even though the model allowed them to remain as nonbreeders for 4 years, ruling out models 4 and 5 as the most appropriate given the data. Using the principle of parsimony, the model with the fewest number of parameters should be chosen over other top models as long as it is considered biologically reasonable (Burnham and Anderson 2002). Therefore, model 1 was selected over model 3 as the best model for the available data set.

7 Popul Ecol (2011) 53: Table 2 Comparison statistics for models with a Quasi-Likelihood Akaike Information Criteria (QAIC) value \7.0 Model mdev DQAIC QAIC weight Model likelihoods for model 1 Model 1 (S s, W cycle=4, P t ) x Model 2 (S s, W cycle=3, P t?m ) Model 3 (S s, W cycle=4, P t?m ) Model 4 (S s, W cycle=5, P t ) Model 5 (S s, W cycle=5, P t?m ) Support for model 1 is ratio of any model s weight to the model weight of model 1 calculated as (QAIC of model 1/QAIC of model N), where N = 2 5 W transition probability, cycle x the breeding cycle constrained to x number of years, mdev the model deviance, DQAIC difference in QAIC values between any given model and the model with the lowest QAIC, QAIC weight the relative contribution of each model to the set Table 3 Breeding transition probabilities for models 1 5 with 1 standard deviation (SD) are shown Transition Model 1 Model 2 Model 3 Models 4 and 5 Parameter estimates Model 1 estimated the apparent survival for those captured for the first time (i.e., influenced by the inclusion of individuals never recaptured) at 0.28 (CI: 0.19, 0.37) and for individuals recaptured at least once at 0.85 (CI: 0.78, 0.93) (Table 4). Models 3, 4, and 5 estimated identical adult survival rates as model 1 (Table 4). The survival estimated for model 2, which was deemed biologically inappropriate due to the breeding transition probabilities, was 0.83 (CI: 0.76, 0.90) (Table 4). Recapture probabilities for model 1 ranged from 0.20 to 1.0, discounting the recapture probability for the 1992 estimated at 0.0. The recapture probability during the second year of the tagging study was 0.0 because no individuals bred in both 1991 and The recapture rates were of less biological value than the survival estimates and breeding transition probabilities and will not be discussed further. Discussion Mean 1 SD Mean 1 SD Mean 1 SD Mean 1 SD W B?B W NB1!B W NB2!B W NB3!B The transitions from the last possible nonbreeder state to the breeder state were constrained to 1.0 and are not shown The ability to model key characteristic of nesting sea turtle data in a multistate model with unobservable states marks an extremely valuable advance in population modeling. Table 4 Estimates of survival for individuals exhibiting imprecise site fidelity (S s1 ) and individuals exhibiting site fidelity to the study site (S s2 ) Model S s1 S s2 Mean 1 SD Mean 1 SD Model 1 (S s, W cycle=4, P t ) Model 2 (S s, W cycle=3, P t?m ) Model 3 (S s, W cycle=4, P t?m ) Model 4 (S s, W cycle=5, P t ) Model 5 (S s, W cycle=5, P t?m ) S survival probability, P capture probability, t time effect, m trapdependence effect, s site-fidelity effect,? an additive effect, * an interaction effect This study provides a more accurate estimate of survival rates for loggerhead sea turtles than those from past singlestate models and also allows unbiased breeding transition probabilities to be directly incorporated into a population projection model (Perret et al. 2003). The survival estimate from the multistate model is a minimum survival estimate because we were not able to discern true transient individuals from those that permanently emigrated or died. This remains an issue with mark recapture studies, especially for species such as the loggerhead turtle with a complex life history. The resident survival estimated from this analysis (S = 0.85) was slightly higher than previous estimates from beaches in the southeastern USA, with the 95% CI encompassing previous estimates of survival. Frazer (1983) first estimated survival for adult females at using a life-history table from a mark recapture study on Little Cumberland Island, GA. Two other estimates from Wassaw Island, GA of 0.79 and Melbourne Beach, FL of 0.83 were estimated using CJS open population models (Cormack 1964; Jolly 1965; Seber 1965; NMFS-SEFSC 2001). The survival estimate calculated in this study may also be higher because the majority of data was collected after implementation of turtle excluder devices (TEDs) in

8 172 Popul Ecol (2011) 53: , which were implemented to reduce mortality. However, from 1990 to 2003, the escape openings of TEDs were not large enough to allow all large juvenile and adult loggerheads to escape, and this may also affect the estimated survival (Epperly and Teas 2002). We did not have the ability to discern or quantify sources of mortality in this study, and our estimate of survival represents a minimum survival estimate for the population. This study also provided estimates of breeding transition probabilities, which reflected the observed remigration patterns observed in the raw data and also that from Frazer (1983). With the breeding transition probabilities, the reproductive cycle can now be appropriately incorporated in a matrix population model. Multistate models not only provide improved survival estimates as output but will lead to improved population projection models. Analyses of the possible effects of alternative management actions will be affected if the estimated apparent survival rates do not adequately represent the true survival rates. The survival estimate from the multistate model should be incorporated into population projections because of the ability to incorporate reproductive skipping and the issue of site fidelity into the model structure, which are critical components of sea turtle life history. Recent analyses of nesting-beach data estimated adult survival for leatherbacks (Dermochelys coriacea) at 0.91 (Rivalan et al. 2005) and (Dutton et al. 2005) and hawksbills (Eretmochelys imbricata) at 0.94 (Kendall and Bjorkland 2001). All three studies accounted for nonrandom temporary emigration due to reproductive skipping and did not suffer from tag-loss issues. This indicates that the issue of tag loss in this study may be a cause of our lower survival estimate. However, for an actively hunted green turtle (Chelonia mydas) population, Campbell and Lagueux (2005) estimated annual adult female survival from a recovery model at 0.82 and did not account for tag loss. Troëng and Chaloupka (2007) analyzed the same green turtle data and estimated annual survival at 0.85 from two different models, a recovery model that accounted for tag loss and an open robust design model that did not. Our survival estimate for loggerheads is closer to that of the actively hunted green turtle population than other species. However, our 95% CI ( ) does encompass the range of estimates. It would be ideal to quantify estimates of anthropogenically caused mortality for loggerheads to discern why our estimates of survival are lower than those of other sea turtle species. Our data may not yet reflect the projected increase in juveniles surviving to the adult stage as a result of TEDs, circle hooks, and other conservation measures. Incorporating our estimate of adult survival into a population model will also help us understand how the range of possible survival rates given by the 95% CI effects population projections. Whereas the multistate model reduced bias in parameter estimates, there were still constraints in data collection. Mark recapture studies for sea turtles are often limited by time and space. This study spanned 16 years, but it still may not cover one full generation time. Spatial scale also became an issue due to the boundaries of the beach survey. Bald Head Island is connected via the beach to suitable nesting habitat to the north. Patrols in 2005 revealed that individual turtles carrying tags from Bald Head Island were in fact nesting outside the political boundaries of the study site, lowering the overall detection probability. If individuals nest outside the study site between years observed nesting within the study site, breeding frequencies would be underestimated. One limitation of the data from Bald Head Island is the lack of consistent and accurate records of tag loss. Records of tag loss would allow us to classify unmarked females as either apparent neophyte nesters or remigrant nesters with tag scars. Nesting outside the study area resulting in permanent emigration coupled with tag loss indicates that the survival estimate from this study remains an underestimate. We attempted to run the multistate model with data from PIT tags alone. However, the data were too sparse, and not enough years of data have been collected to observe full reproductive cycles of some of the females. Trends on Bald Head Island Bald Head Island actively protects sea turtle nests and has ordinances against beach-front lighting and beach furniture, but the loss of beach from erosion was a likely factor contributing to the decline of nest numbers over time. Due to the dynamic movements of barrier islands, there is less suitable beach habitat available. The decline may also be a factor of the variable remigration intervals, nesting outside the study area, as well as the individually variable clutch frequency observed in the raw data. Bald Head Island experienced unusually low loggerhead nest numbers in 2004, a phenomenon noted on all nesting beaches in the southeastern USA (M. Godfrey, personal communication). Nest numbers in 2005 and 2006 increased but still reflected the decreasing number of nests observed over time. However, total nest numbers for North Carolina from 1989 to 2006 showed no significant decline through time (M. Godfrey, personal communication). This lack of a trend indicates that individuals from Bald Head Island are most likely permanently emigrating to other nesting beaches and that the survival estimated here is not necessarily a result of a declining local population. From 1995 to 2006, raw data indicated that approximately 50% or more of the individuals captured each year were not previously observed on Bald Head Island (Hedges 2007). Due to the tag-loss issue and

9 Popul Ecol (2011) 53: capture probability of \1.0, we could not discern between individuals that were truly first-time nesters on Bald Head Island and those that had previously escaped capture. These estimates could be an indication of increased recruitment into adult life stages over the last 11 years. The mark recapture study on Bald Head Island will continue, which allows the multistate model to potentially be updated yearly. We recommend that all possible improvements be made to data collection, as this mark recapture study provides critical information on the northern subpopulation of loggerhead turtles. Acknowledgments We thank the Bald Head Island Conservancy for their commitment and funding the study on Bald Head Island as well all interns and volunteers involved in the project. Thank you to Dr. Nancy Thompson, Dr. Marcella Kelly, Dr. Matthew Godfrey, Dr. Joseph Powers and anonymous reviewers for comments on the manuscript. We are grateful to Dr. Bill Kendall for reviewing the manuscript and for conversations regarding the modeling content of the study. The Cooperative Marine Turtle Tagging Program has also been fundamental to maintaining a tagging database and providing information exchange among programs. Funding for this project was provided by the Bald Head Island Conservancy and NMFS RTR Unit at Virginia Tech. References Akaike H (1973) Information theory as an extension of the maximum likelihood principle. In: Petrov BN, Csaksi F (eds) Second international symposium on information theory. Akademiai Kiado, Budapest, Hungary, pp Akaike H (1983) Information measures and model selection. Int Stat Inst 44: Akçakaya HR, Sjögren-Gulve P (2000) Population viability analyses in conservation planning: an overview. Ecol Bull 48:9 21 Anderson DR, Burnham KP (1999) General strategies for the collection and analysis of ringing data. Bird Study 46: Balser JP (1981) Confidence interval estimation and tests for temporary migration in tag recapture studies. PhD Dissertation. Cornell University, Ithaca, NY Beauplet G, Barbraud C, Dabin W, Küssener C, Guinet C (2006) Agespecific survival and reproductive performances in fur seals: evidence of senescence and individual quality. Oikos 112: Beissinger SR, Westphal MI (1998) On the use of demographic models of population viability in endangered species management. J Wildl Manag 62: Bell R, Richardson JI (1978) An analysis of tag recoveries from loggerhead sea turtles (Caretta caretta) nesting on Little Cumberland Island, Georgia. Fla Mar Res Publ 33:1 166 Bjorndal KA, Meylan AB, Turner BJ (1983) Sea turtle nesting at Melbourne Beach, Florida. I. Size, growth, and reproductive biology. Biol Conserv 26:65 77 Broderick AC, Godley BJ (1999) Effect of tagging marine turtles on nesting behaviour and reproductive success. Anim Behav 58: Buckland ST, Burnham KP, Augustin NH (1997) Model selection: an integral part of inference. Biometrics 53: Burnham KP, Anderson DR (2002) Model selection and multimodel inference: a practical information-theoretic approach, 2nd edn. Springer, New York Burnham KP, Anderson DR, White GC, Brownie C, Pollock KH (1987) Design and analysis methods for fish survival experiments based on release-recapture. American Fisheries Society, Bethesda Campbell CL, Lagueux CJ (2005) Survival probability estimates for large juvenile and adult green turtles (Chelonia mydas) exposed to an artisanal marine turtle fishery in the western Caribbean. Herpetologica 61: Carr A (1975) The Ascension Island green turtle colony. Copeia 1975: Caswell H (2000) Matrix population models: construction, analysis, and interpretation, 2nd edn. Sinauer and Associates, Sunderland, MA Catchpole EA, Morgan BJT (1997) Detecting parameter redundancy. Biometrika 84: Choquet R, Reboulet A-M, Pradel R, Gimenez, Lebreton J-D (2005a) M-SURGE 1-7 User s Manual. CEFE-CNRS, Montpellier, France. Choquet R, Reboulet A-M, Lebreton J-D, Gimenez O, Pradel R (2005b) U-CARE 2.2 User s Manual. CEFE-CNRS, Montpellier, France. Congdon JD, Dunham AE, van Loben Sels RC (1993) Implications for conservation and management of long-lived organisms. Conserv Biol 7: Cormack RM (1964) Estimates of survival from the sighting of marked animals. Biometrika 51: Crouse DT, Crowder LB, Caswell H (1987) A stage-based population model for loggerhead sea turtles and implications for conservation. Ecology 68: Dodd CK (1988) Synopsis of the biological data on the loggerhead sea turtle Caretta caretta (Linnaeus 1758). USFWS Biol Rep 88:1 10 Doligez B, Colbert J, Pettifor RA, Rowcliffe M, Gustafsson L, Perrins CM, McCleery RH (2002) Costs of reproduction: assessing responses to brood size manipulation on life-history and behavioural traits using multi-state capture recapture models. J Appl Stat 29: Dutton DL, Dutton PH, Chaloupka M, Boulon RH (2005) Increase of a Caribbean leatherback turtle Dermochelys coriacea nesting population linked to long-term nest protection. Biol Conserv 126: Epperly SP, Teas WG (2002) Turtle excluder devices are the escape openings large enough? Fish Bull 100: Frazer NB (1983) Survivorship of adult female loggerhead sea turtles, Caretta caretta, nesting on Little Cumberland Island, Georgia, USA. Herpetologica 39: Fujiwara M, Caswell H (2002) Estimating population projection matrices from multi-stage mark-recapture data. Ecology 83: Hawkes LA, Broderick AC, Godfrey MH, Godley BJ (2005) Status of nesting loggerhead turtles Caretta caretta at Bald Head Island (North Carolina, USA) after 24 years of intensive monitoring and conservation. Oryx 39:65 72 Hedges ME (2007) Development and application of a multistate model to the northern subpopulation of loggerhead sea turtles. Master s Thesis. Virginia Tech, Blacksburg, VA Heppell SS, Crowder LB, Crouse DT (1996) Models to evaluate headstarting as a management tool for long-lived turtles. Ecol Appl 6: Heppell SS, Crowder LB, Crouse DT, Epperly SP, Frazer NB (2003) Population models for Atlantic loggerheads: past, present, and future. In: Bolten AB, Witherington BE (eds) Loggerhead sea turtles. Smithsonian Books, Washington, DC, pp Hughes GR (1982) Nesting cycles in sea turtles typical or atypical? In: Bjorndal KA (ed) Biology and conservation of sea turtles. Smithsonian Institution Press, Washington, DC, pp 81 89

10 174 Popul Ecol (2011) 53: Hunter CM, Caswell H (2009) Rank and redundancy of multistate mark-recapture models for seabird populations with unobservable states. In: Thomson DL, Cooch EG, Conroy MJ (eds) Modeling demographic processes in marked populations, vol 3. Springer, New York, pp Jolly GM (1965) Explicit estimates from capture recapture data with both death and immigration-stochastic model. Biometrika 52: Kendall WL (2004) Coping with unobservable and mis-classified states in capture recapture studies. Anim Biodivers Conserv 21: Kendall WL, Bjorkland RH (2001) Using open robust design models to estimate temporary emigration from capture recapture data. Biometrics 57: Kendall WL, Nichols JD (1995) On the use of secondary capture recapture samples to estimate temporary emigration and breeding proportions. J Appl Stat 22: Kendall WL, Nichols JD (2002) Estimating state-transition probabilities for unobservable states using capture recapture/resighting data. Ecology 83: Kendall WL, Nichols JD, Hines JE (1997) Estimating temporary emigration using capture recapture data with Pollock s robust design. Ecology 78: Kéry M, Gregg KB, Schaub M (2005) Demographic estimation methods for plants with unobservable life-states. Oikos 108: Lancia RA, Nichols JD, Pollock KH (1994) Estimating the number of animals in wildlife populations. In: Bookout TA (ed) Research and management techniques for wildlife and habitats, 5th edn. The Wildlife Society, Bethesda, MD, pp Lebreton J-D, Pradel R (2002) Multistate recapture models: modelling incomplete individual histories. J Appl Stat 29: Lebreton J-D, Burnham KP, Colbert J, Anderson DR (1992) Modeling survival and testing biological hypotheses using marked animals: a unified approach with case studies. Ecol Monogr 62: Limpus CJ (1985) A study of the loggerhead sea turtle, Caretta caretta, in Eastern Australia. PhD Dissertation. University of Queensland, St. Lucia, Queensland, Australia Miller JD, Limpus CJ, Godfrey MH (2003) Nest site selection, oviposition, eggs, development, hatching, and emergence of loggerhead turtles. In: Bolten AE, Witherington BE (eds) Loggerhead sea turtles. Smithsonian Books, Washington, DC, pp National Marine Fisheries Service Southeast Fisheries Science Center (NMFS-SEFSC) (2001) Stock assessments of loggerhead and leatherback sea turtles and an assessment of the impact of the pelagic longline fishery on the loggerhead and leatherback sea turtles of the Western North Atlantic. US Department of Commerce NOAA Technical Memorandum NMFS-SEFSC Nichols JD, Sauer JR, Pollock KH, Hestbeck JB (1992) Estimating transition probabilities for stage-based population projection matrices using capture recapture data. Ecology 73: Nichols JD, Hines JE, Pollock KH, Hinz R, Link WA (1994) Estimating breeding proportions and testing hypotheses about costs of reproduction with capture recapture data. Ecology 75: Perret N, Pradel R, Miaud C, Grolet O, Joly P (2003) Transience, dispersal and survival rates in newt patchy populations. J Anim Ecol 72: Pollock KH, Nichols JD, Simons TR, Farnsworth GL, Bailey LL, Sauer JR (2002) Large scale wildlife monitoring studies: statistical methods for design and analysis. Environmetrics 13: Pradel R (1993) Flexibility in survival analysis from recapture data: Handling trap-dependence. In: Lebreton J-D, North PM (eds) Marked individuals in the study of bird population. Birkhäuser Verlag, Basel, Switzerland, pp Pradel R, Hines JE, Lebreton J-D, Nichols JD (1997) Capture recapture survival models taking into account of transients. Biometrics 53:60 72 Prévot-Julliard A-C, Lebreton J-D, Pradel R (1998) Re-evaluation of adult survival of Black-headed Gulls (Larus ridibundus) in presence of recapture heterogeneity. Auk 115:85 95 Richardson JI, Richardson TH, Dix MW (1978) Remigration patterns of loggerhead sea turtles (Caretta caretta) nesting on Little Cumberland and Cumberland Islands, Georgia. Fla Mar Res Publ 33:39 44 Rivalan P, Prévot-Julliard A-C, Choquet R, Pradel R, Jacquemin B, Girondot M (2005) Trade-off between current reproductive effort and delay to next reproduction in the leatherback sea turtle. Oecologia 145: Schaub M, Gimenez O, Schmidt BR, Pradel R (2004) Estimating survival and temporary emigration in the multistate capture recapture framework. Ecology 85: Schroeder BA, Foley AM, Bagley DA (2003) Nesting patterns, reproductive migrations, and adult foraging areas of loggerhead turtles. In: Bolten AE, Witherington BE (eds) Loggerhead sea turtles. Smithsonian Books, Washington, DC, pp Schwarz CJ, Arnason AN (1996) A general methodology for the analysis of capture recapture experiments in open populations. Biometrics 52: Seber GAF (1965) A note on the multiple recapture census. Biometrika 52: Troëng S, Chaloupka M (2007) Variation in adult annual survival probability and remigration intervals of sea turtles. Mar Biol 151: Turtle Expert Working Group (TEWG) (2000) Assessment update for the Kemp s ridley and loggerhead sea turtle populations in the western North Atlantic. NOAA Technical Memorandum NMFS- SEFSC-444 Véran S, Gimenez O, Flint E, Kendall WL, Doherty PF Jr, Lebreton J-D (2007) Quantifying the impact of longline fisheries on adult survival in the black-footed albatross. J Appl Ecol 44: Viallefont A, Cooke F, Lebreton J-D (1995) Age-specific cost of firsttime breeding. Auk 112:67 76