, Volume 172, Issue 1, pp 129–140

Variability in temporary emigration rates of individually marked female Weddell seals prior to first reproduction


    • Department of EcologyMontana State University
  • Jay J. Rotella
    • Department of EcologyMontana State University
  • Robert A. Garrott
    • Department of EcologyMontana State University
Population ecology - Original research

DOI: 10.1007/s00442-012-2472-z

Cite this article as:
Stauffer, G.E., Rotella, J.J. & Garrott, R.A. Oecologia (2013) 172: 129. doi:10.1007/s00442-012-2472-z


In many species, temporary emigration (TE) is a phenomenon, often indicative of life-history characteristics such as dormancy, skipped reproduction, or partial migration, whereby certain individuals in a population are temporarily unobservable at a particular site. TE may be a flexible condition-dependent strategy that allows individuals to mitigate effects of adverse conditions. Consequently, TE rates ought to be highly variable, but sources of variations are poorly understood for most species. We used data from known-aged female Weddell seals (Leptonychotes weddellii) tagged in Erebus Bay, Antarctica, to investigate sources of variation in TE rates prior to reproduction and to evaluate possible implications for age-specific probability of first reproduction. TE rates were near 1 the year after birth, decreased to an average of 0.15 (\( \widehat{\text{SE}} \) = 0.01) by age 8, and were similar thereafter. TE rates varied substantially from year-to-year and were lower for seals that attended reproductive colonies the previous year than for seals that did not attend (e.g., \( \overline{{\hat{\psi }_{{i,{\text{age}}\,8}}^{\text{UU}} - \hat{\psi }_{{i,\,{\text{age}}\,8}}^{\text{PU}} }} \) = 0.22). Recruitment rates were marginally greater for seals that did attend than for seals that did not attend colonies the previous year. For Weddell seals specifically, our results suggest that (1) motivation to attend colonies varied temporally, (2) as seals grew older they had increased motivation to attend even before reproductive maturity, and (3) seals appear to follow various attendance strategies. More broadly, our results support the idea of TE as a variable, condition-dependent strategy, and highlight the utility of TE models for providing population and life-history insights for diverse taxa.


Capture–mark–recaptureColonial breedingCondition-dependentLeptonychotes weddelliiOpen robust designRecruitmentSocial learningTrade-offsUnobservable states


Using long-term capture–mark–recapture data to estimate and test hypotheses about demographic rates of wild populations (see reviews in Lebreton et al. 1992; Williams et al. 2002) has yielded insights about various ecological and demographic processes (e.g., Gaillard et al. 1997; Reid et al. 2003; Hadley et al. 2007b; Conn and Cooch 2009). One well-known assumption of capture–mark–recapture studies is that emigration from a study area is permanent; when such emigration occurs, estimates of apparent survival are the product of true survival and site-fidelity rates. However, in many species, certain individuals may make irregular and non-permanent moves away from a site; such movement is called temporary emigration (TE) in the ecological literature (e.g., Kendall et al. 1997; Converse et al. 2009). Recent literature has focused primarily on technical aspects of modeling TE rates and possible biases in other estimates resulting from non-random TE (e.g., Kendall and Nichols 2002; Kendall 2004; Schaub et al. 2004; Langtimm 2009; Bailey et al. 2010). However, from a capture–mark–recapture perspective, TE can be defined as a transition by individuals into any behavior that causes them to be unobservable until they transition out of that behavior. Improved understanding of variation in such transitions could provide important insights about life history, ecology, and population dynamics of diverse species.

A useful way to model TE is to define one or more unobservable states that encompass behaviors that cause individuals to be unobservable, and then to estimate transition probabilities into those states using multistate mark–recapture models (Kendall and Bjorkland 2001; Kendall and Nichols 2002; Lebreton et al. 2009). The classification or interpretation of behaviors as TE thus depends on the study system, the perspective of the observer, and the temporal scale at which observations are made. One example of TE is the absence from breeding areas of some immature individuals or adults skipping reproduction. Such behavior has been documented or explicitly modeled as TE in pinnipeds (e.g., Schwarz and Stobo 1997; Beauplet et al. 2005; Hadley et al. 2007b; de Bruyn et al. 2011), whales (e.g., Fujiwara and Caswell 2002; Bradford et al. 2006), colonial birds (e.g., Frederiksen and Bregnballe 2000; Converse et al. 2009), and pond-breeding amphibians (e.g., Bailey et al. 2004; Muths et al. 2006). In rare cases, it may be possible to observe individuals at all known populations centers, thus obviating the need to estimate or account for TE; such appears to be the case with Hawaiian monk seals (Monachus schauinslandi; Baker and Thompson 2007). Seasonal movements of individuals to breeding areas is analogous to migration (Dingle 1996), and TE by some individuals in some years is analogous to partial migration, (i.e., when there is a variable segment of the population that migrates; see White et al. 2007; Jahn et al. 2010). At shorter temporal scales, foraging trips away from colonies (Lunn et al. 1994) could be modeled as TE, provided transition periods are appropriately defined. Dispersal of young from a natal location to an adult breeding area or inter-annual dispersal of adults typically is permanent (Dobson and Jones 1985) and thus does not fit into a TE framework. However, inter-annual breeding dispersal between sub-populations of a metapopulation could be manifested as TE if observations are made at only one or a few sub-populations; in this case, TE would not represent reproductive skipping. Temporary emigration need not entail absence from a location. In small mammal studies, underground torpor or temporary movement off a trapping grid constitutes TE (e.g., Kendall et al. 1997), and in plants, annual underground dormancy rates can be estimated by modeling TE (e.g., Kéry et al. 2005).

In many populations, individual decisions whether to move (or to become dormant) or not in a given year may depend on current environmental conditions or status of individuals, and many ecologists are interested in understanding such variation in, and consequences of such movement choices (Ims and Hjermann 2001; Hoover 2003; White et al. 2007; Holyoak et al. 2008; Jahn et al. 2010). Unfortunately, inferences that can be made from capture-mark-recapture studies about movements away from a site generally are limited when animals are sampled from only that single study site (Nichols and Kaiser 1999). However, if absences are temporary and represent ecologically interesting phenomena such as skipped or delayed breeding (e.g., Muths et al. 2006; Converse et al. 2009), or dormancy (Kendall et al. 1997; Kéry et al. 2005), then estimates of TE can be useful and ecologically interesting, even when populations are sampled from only a single site.

For colonial breeding species, TE might be advantageous if it allows adults skipping reproduction or immature individuals to avoid potential costs to future survival resulting from lost foraging opportunities away from breeding areas, conflict with reproductive adults, or increased risk of predation. However, many immature individuals of some species, or adults skipping reproduction, sometimes do attend breeding sites, and such attendance can improve subsequent breeding success (Schjørring et al. 1999; Aubry et al. 2009). For example, close association with reproductive adults may facilitate the gathering of information or social skills (Danchin et al. 2004), or may synchronize ovulation and reproduction (McClintock 1978; Langvatn et al. 2004). If a major motivation for TE is greater access to resources, then individuals in the poorest condition should have the greatest motivation to avoid possible costs associated with attending colonies. Also, TE rates might increase during years when environmental conditions are poor if such conditions result in decreased ability or motivation to risk costs to survival (Frederiksen and Bregnballe 2000). For long-lived species, selection ought to minimize variability in survival at the expense of other demographic rates (Pfister 1998). Thus, temporary emigration might be an important and temporally variable demographic rate.

Temporary emigration from breeding sites might have either positive or negative ecological implications for individuals, but little is known for most taxa about sources of variation in this rate. A population of Weddell seals (Leptonychotes weddellii) breeding in Erebus Bay, Antarctica, is well suited for investigating TE because (1) the population lives in a temporally variable and relatively pristine environment that currently is little affected by anthropogenic perturbation (Ainley 2002); (2) an intensive mark–recapture study of this population has been conducted over multiple decades (Siniff et al. 1977; Cameron and Siniff 2004; Hadley et al. 2006); (3) TE is known to occur for both immature and mature seals at rates that likely vary among years as well as with animal age and reproductive state (Testa and Siniff 1987; Hadley et al. 2007b); and (4) female seals born in Erebus Bay exhibit remarkable site-fidelity, and surviving females eventually return to Erebus Bay to give birth to their first pup (Hadley et al. 2006; Cameron et al. 2007). Thus, we are confident that failure to observe seals in Erebus Bay only very rarely equates to reproduction elsewhere. We used 28 years (1980–2008) of data from this population to estimate rates of TE of female Weddell seals prior to first reproduction, to evaluate sources of variation, and to evaluate possible implications for age-specific rates of first reproduction (recruitment). Seals were unobservable when absent, or, perhaps rarely, when briefly present but never hauled out during any of the repeated surveys that were conducted. To make ecological interpretations, we therefore considered 2 behaviors: (1) colony attendance, which entails prolonged presence at breeding sites within a breeding season such that seals were readily detectable, and (2) TE, which represents complete absence from, or unobservable presence at the breeding colonies for at least 1 year between birth and first reproduction.

Materials and methods

Study area and population

Erebus Bay is located in southeastern McMurdo Sound in the western Ross Sea, Antarctica (Online Resource Fig. S1). Tidal action of fast ice against land and the movements of the Erebus Glacier Tongue pushing against the fast ice result in perennial cracks (Stirling 1969; Siniff et al. 1977) that provide reliable access for seals to the surface of the fast ice. Seal-pupping colonies form at 8–14 of these perennial cracks, and from late October to mid-November approximately 300–600 pups are born on the ice surface. Females usually give birth to a single pup (Gelatt et al. 2001), and mothers and pups remain in close association throughout the 30- to 45-day lactation period. Mothers rely primarily on stored body reserves for most of the lactation period (Eisert et al. 2005). Non-reproductive seals also routinely haul out in the study area and are readily detected. Males defend underwater territories, and mating typically takes place near the end of the pupping season each year (Stirling 1969; Siniff et al. 1977).

Data collection

Each year since 1969, Weddell seal pups in Erebus Bay were marked with 4 individually identifiable tags in the interdigital webbing of each hind flipper, and since 1979 nearly all new pups have been marked. Broken or missing tags on adults were subsequently replaced as necessary. Currently, the majority of the female population is marked, and most marked seals are of known ages. During each year from 1973 to the present, 5–8 resighting surveys of the study population were conducted every 3–6 days during early November through mid-December. Seals are highly approachable and tags usually could be read from within 0.5 m.

In this study, we used data from 5,450 female seals tagged as pups during 1980–2008, 1,300 of which were observed in ≥1 subsequent year, and 900 of which recruited into the study area’s breeding population (mean age at 1st reproduction = 7.5, SE = 0.05) by 2008. Two hundred thirty-three seals were not observed as prebreeders in Erebus Bay during the pupping season between the time they were born and when they first gave birth (mean age at 1st reproduction = 6.8 years, SE = 0.08), whereas 667 were observed ≥1 time during the same period (mean age at 1st reproduction = 7.7 years, SE = 0.06). For seals in the latter group, the mean age when they first were sighted after birth was 4.5 years (SE = 0.02) and the mean number of years observed as prebreeders was 2.1 (SE = 0.05). Most females attended colonies for ≤2 years prior to producing their first pup, and many attended for 0 years, but females clearly followed various attendance strategies (Fig. 1).
Fig. 1

Number of years that individually marked known-age female Weddell seals (Leptonychotes weddellii) were observed as prebreeders (after their birth year) in Erebus Bay, McMurdo Sound, Antarctica, during the 1980–2008 pupping seasons. Given that seals were seen prior to recruitment, the mean number of years observed as prebreeders was 2.1 (SE = 0.05)


We evaluated three possible sources of variation in TE and recruitment rates of female Weddell seals. We expected rates to vary with age, to vary depending on whether or not a seal attended a reproductive colony in the previous year, and to vary temporally depending on annual conditions in either the current year or the year a cohort was born.

We suspected that absence from breeding colonies early in life was partly motivated by limited food resources in Erebus Bay (Siniff et al. 1977; Testa et al. 1985) or by social conflict or competition between young seals and reproductive adults (Stirling 1969). Growth should be of primary importance to young seals, and given observed sighting patterns, we expected that TE rates would be very high for at least the first year or two of life. However, seals should have increasing motivation to attend the reproductive colonies as they approach reproductive size and maturity, because pupping colonies are foci of mating activity where female seals might observe underwater territorial defense by breeding males and interactions of males and females (Siniff et al. 1977). Seals might also benefit from observing interactions of mothers with pups. Given that many seals are observed ≥1 time prior to recruitment, we expected that TE rates would decrease well before females reached reproductive maturity.

Based on resighting patterns and the known life history of Weddell seals (Stirling 1969; Hadley et al. 2006), we also expected that TE rates would be state-dependent (i.e., follow a first-order Markov process) such that seals absent in year i would be more likely to be absent again in year i + 1 than seals that attended a colony in year i. Our hypothesis that attending colonies might facilitate beneficial information gathering by seals also led to the prediction that recruitment rates in year i + 1 should be greater for seals that attended breeding colonies in year i than for seals that did not attend. Alternatively, recruitment and TE rates might not be state-dependent (i.e., non-Markovian) if the benefits of TE compensate for loss of benefits from attending colonies, or if colony attendance is relatively unimportant.

Given the assumption of reduced foraging opportunity or increased intra-specific conflict when seals attend colonies, and given that this population of Weddell seals has been shown to buffer its demography from variability in survival (Rotella et al. 2012), we suspected that colony attendance might represent a trade-off of costs and benefits, the balance of which might vary with environment conditions. We thus predicted that TE and recruitment rates would vary annually across all ages and cohorts. In contrast, we predicted that TE and recruitment rates would vary by birth cohort if environmental conditions experienced by pups exerted a greater influence on phenotype than did later environmental conditions. Annual environmental conditions can influence demographic rates such as dispersal (Dugger et al. 2010), recruitment (Crespin et al. 2006; Hadley et al. 2006), or breeding success (Nevoux and Barbraud 2006), but environmental conditions experienced in the first year of life may also affect phenotypic quality and demographic rates later in life (Lindström 1999; Forchhammer et al. 2001; Cam et al. 2003; Garrott et al. 2012). Such effects can be of either short or long duration. Birth-year effects may be temporary if compensatory growth can overcome early disadvantages (Bjorndal et al. 2003). We predicted that differences in recruitment or TE rates among cohorts would persist for only a few years if seals are able to overcome poor early life conditions.

Although our primary focus was on TE rates, we also evaluated several sources of variation in apparent survival (hereafter survival) probabilities. We expected greater survival after age 3 than up to age 3 (Cameron and Siniff 2004; Hadley et al. 2006). Conditions at birth strongly influence early survival in several species of short-lived (Rödel et al. 2009) and long-lived animals (Gaillard et al. 1997; Forchhammer et al. 2001). We predicted a long-term cohort effect on survival if early conditions have greater influence on survival than more immediate conditions, and we also considered the possibility that cohort effects may only be temporary. In contrast, we predicted a year effect if current-year environmental conditions primarily influence survival. We did not have predictions about differences in survival rates depending on the TE status of seals, because generally such differences cannot be estimated (Kendall and Nichols 2002; Schaub et al. 2004) or testing for such differences is not possible (Bailey et al. 2010).

Data analysis

We applied open robust design multistate models (Kendall and Bjorkland 2001; Kendall 2006; Converse et al. 2009) to encounter histories for 5,450 female seals born in Erebus Bay during 1980–2008. Open robust design multistate models make use of multiple surveys per year to allow estimation of detection probabilities from within-season data, while properly accounting for staggered entry and exit of seals into the study area (further modeling details in Online Resource). One assumption is that individuals enter and exit the study area only once each season, i.e., there is no within-season TE; violations of this assumption result in negatively biased estimates of detection probability. Although we generally equated TE with absence from reproductive colonies, in some cases it might have been possible for seals to briefly visit colonies but never haul out long enough to be detected in one of our surveys. True absence and unobservable presence are indistinguishable from our data, but evidence from previous analysis suggests that prebreeders are highly detectable and that duration of attendance at colonies is usually long enough for seals to be available for detection during one or more surveys (Rotella et al. 2009). Because mothers with pups have a season-level detection rate of ~1.00 (Hadley et al. 2006; Rotella et al. 2009), we could reliably assign each seal 1 of 4 breeding state classifications each season (Fig. 2): pup (N); prebreeder (P, seal attending the study area and never documented with a pup); unobservable (U, prebreeder not attending the study area); and breeder (B, recruit, or first-time mother with pup). Temporary emigration was defined by transitions into U, including P to U (\( \psi_{{i,{\text{age}}}}^{\text{PU}} \)) as well as U back to U (\( \psi_{{i,{\text{age}}}}^{\text{UU}} \)), and recruitment as transitions from P or U into B (\( \psi_{{i,{\text{age}}}}^{\text{PB}} \) or \( \psi_{{i,{\text{age}}}}^{\text{UB}} \)). Because the probabilities of all possible transitions for any given state logically must sum to 1, one transition rate for each state was derived by subtraction (see Online Resource for further details). Our choice of which transition rates to estimate directly was made so that our competing models focused on testing hypotheses about two types of transitions: into an unobservable state (TE) and into a breeding state (recruitment). Our goal was to investigate variation in vital rates of prebreeders, so we right-censored information for seals after they recruited and only estimated rates for seals in states N, P, and U. Such censoring simplified the analysis by limiting the number of possible state transitions and was justified by the nearly certain detection of mothers.
Fig. 2

Breeding state classifications and possible transitions between states for individually marked, female Weddell seals studied in Erebus Bay during 1980–2008. States denote pups (newborn, N), older prebreeders either attending (prebreeder, P) or not attending (unobservable, U) a reproductive colony, and first-time mothers (breeder, B). Encounter histories were truncated after first reproduction, so no transitions out of state B were estimated. Black solid and dashed lines represent, respectively, temporary emigration and recruitment. Gray solid lines represent transitions not estimated directly in our open robust design multistate models

We used a 2-step approach to model \( S_{{i,{\text{age}}}}^{r} \) and \( \psi_{{i,{\text{age}}}}^{rz} \), where for a given year or cohort, i, and state, r, \( S_{{i,{\text{age}}}}^{r} \) is the probability of surviving and not permanently emigrating, and \( \psi_{i}^{rz} \) is the probability, given survival, that an individual makes a transition into state z. We used this sequential approach to model selection because it is efficient, and often is as effective as an all possible models approach when identifying the model with the lowest overall AICc (Doherty et al. 2010). We first modeled a suite of 3 a priori structures for temporal variation in \( S_{{i,{\text{age}}}}^{r} \) with \( \psi_{{i,{\text{age}}}}^{rz} \) modeled generally (age and time variation) and identified the model with the lowest value for the bias-corrected Akaike’s information criterion (AICc; Burnham and Anderson 2002). We then used this structure for \( S_{{i,{\text{age}}}}^{r} \) to evaluate competing hypotheses about \( \psi_{{i,{\text{age}}}}^{rz} \). The \( S_{{i,{\text{age}}}}^{r} \) suite included one model that allowed \( S_{{i,{\text{age}}}}^{r} \) to vary by year, where year effects were shared by all states and ages; this model evaluated the hypothesis that annual variations in environmental conditions influence all seals. Two competing models allowed \( S_{{i,{\text{age}}}}^{r} \) to vary by cohort (either permanently or temporarily for 3 years after birth); these models evaluated the hypotheses that environmental conditions at birth influenced survival rates and that adult survival is less related to current environmental conditions. In each of the three structures for \( S_{{i,{\text{age}}}}^{r} \) we allowed survival to vary by age, where age was modeled as a categorical variable with three nominal classes: 1, 2, and ≥3 (Hadley et al. 2006). Because hypotheses about state-dependent survival of females in an unobservable state cannot be differentiated on the basis of model selection (Bailey et al. 2010), we made the necessary assumption that \( S_{i}^{\text{U}} \) = \( S_{i}^{\text{P}} \). Previous analysis suggests that biases in survival estimates because of non-random TE were minimal for this population of Weddell seals when TE was modeled as time-invariant (Hadley et al. 2007b). However, when survival and state-dependent TE are both temporally variable, TE can cause negative bias in survival estimates at the end of a time-series (Langtimm 2009). Consequently, we avoided interpretation of survival estimates for the final four cohorts and focused primarily on possible ecological or biological reasons for state-dependent TE rather than on technical implications of TE for survival estimates. Also, because not all parameters are identifiable in fully time-dependent models, in models with annual variation we set survival equal for the final 2 years and did not interpret survival for the final years. We adjusted all survival estimates for tag loss as described in Hadley et al. (2006).

To address hypotheses about temporal variation in transition probabilities, we allowed rates to vary either as a function of year or birth cohort. Cohort effects were either permanent or, for TE rates only, terminated 3 years after birth. We also considered whether transition rates were Markovian or not. In all models, we estimated \( \psi_{i}^{{r{\text{U}}}} \) for 10 nominal age classes (1, 2–3, 4, 5, 6, 7, 8, 9, 10, ≥11 years old) and \( \psi_{i}^{{r{\text{B}}}} \) for 7 nominal age classes (5, 6, 7, 8, 9, 10, ≥11 years old), and we evaluated two functional forms for age-related variation in transition rates. Age was treated as either a categorical variable that allowed transition rates for each age to vary independently from each other or used as a continuous variable in a quadratic model that allowed rates to increase for young age classes and to then level off or decrease for older age classes. We fixed \( \psi_{i}^{\text{PU}} \) = 0 for the first age-class because all seals in this age-class necessarily were pups and transition thus originated from state N. Also, we fixed \( \psi_{i}^{{r{\text{B}}}} \) = 0 for all seals aged <5 years because no seals in our dataset recruited prior to age 5. We constrained year effects to be shared across all ages and states (Hadley et al. 2006); we constrained cohort effects similarly. Age denotes a female’s age at the end of an interval, i.e., \( \psi_{{i,{\text{age}}\,5}}^{rz} \) is the probability that a 4-year-old seal in state r in year i − 1 will transition state z when she is 5 years old in year i.

Based on apparent patterns from previous within-year analyses (Rotella et al. 2009), we felt justified in modeling only one structure for within-year parameters, which, for a given year, i, secondary occasion j, and state, r, were: probability that an individual is a new arrival to the study area (\( {\text{pent}}_{ij}^{r} \)); probability of detection (\( p_{ij}^{r} \)); and probability of staying in the study area until secondary occasion j + 1, given entry k occasions prior (\( \phi_{ijk}^{r} \)). We allowed \( {\text{pent}}_{ij}^{r} \) to follow a state-dependent (states P, B and N pooled) within-season linear trend for j > 1 (common for all seasons). Detection probability was also allowed to follow a year-constant and state-dependent (states N, P, and B) within-season linear trend. We allowed \( \phi_{ijk}^{r} \) to vary only by state (N, P, and B). All within-season parameters were fixed to zero for state U.

We performed analyses in program MARK (White and Burnham 1999) through the RMark package (Laake 2010) in program R (R Development Core Team 2012). Currently, there is no general goodness-of-fit test for the type of open robust design multistate models we used in our analysis, and the median \( \hat{c} \) procedure implemented into program MARK is not available for robust design data. In previous analyses of Weddell seal data from Erebus Bay with simpler models, Hadley et al. (2006) found only modest levels of overdispersion. Therefore, we used AICc rather than quasi-likelihood AICc (QAICc) as our model selection criterion. However, to assess the possible influence on overdispersion on model selection results, we examined how model rankings changed as we increased \( \hat{c} \) from 1.0 to 5.5, which is twice the highest level of overdispersion observed previously in these data (Hadley et al. 2007a).


The best-supported model in the survival suite included a permanent cohort effect on \( \hat{S}_{{i,{\text{age}}}}^{r} \) (Table 1). The top model retained its rank for all values of \( \hat{c} \) ≤ 5. Based on this model, the majority of female seals were present on the first survey each year, and the probability of detection at least once during the season was >0.91 for prebreeders and >0.99 for pups and mothers (see Online Resource for further details). As predicted, \( \hat{S} \) was greater for seals ≥2 years old (\( \bar{\hat{S}}_{\text{cohort}} \) = 0.93, \( \widehat{\text{SE}} \) = 0.01) than it was for pups (\( \bar{\hat{S}}_{\text{cohort}} \) = 0.64, \( \widehat{\text{SE}} \) = 0.08) or yearlings (\( \bar{\hat{S}}_{\text{cohort}} \) = 0.55, \( \widehat{\text{SE}} \) = 0.07), for which 95 % CIs overlapped broadly in each year (Fig. 3). Survival rates dropped sharply for the final few cohorts, possibly reflecting some negative bias at the end of the time series (Langtimm 2009).
Table 1

Model selection results for open robust design multistate capture–mark–recapture models representing hypotheses about variation in survival and transition probabilities for individually marked female Weddell seals (Leptonychotes weddellii) in Erebus Bay, Antarctica







Survival-probabilities suite




















Transition-probabilities suite


\( \psi_{{{\text{a7}} + {\text{year}}}}^{rB} ,\psi_{{{\text{a1}}0 + {\text{year}}}}^{rU} \), Markovian






\( \psi_{{{\text{A7}} + {\text{year}}}}^{rB} ,\psi_{{{\text{A1}}0 + {\text{year}}}}^{rU} \), Markovian






\( \psi_{{{\text{a7}} + {\text{cohort}}}}^{rB} ,\psi_{{{\text{a1}}0 + {\text{year}}}}^{rU} \), Markovian






The best parameterization for survival, identified in the survival-probabilities model suite, was used in the transition-probabilities suite, of which the top 3 models are shown here (see Table S2 for complete model selection results). Transitions represent inter-annual changes among states (U prebreeder not attending breeding colonies, P prebreeder attending breeding colonies, B a first-time mother in the breeding colonies). Age structures were as follows: a3 = 3 nominal age classes (1, 2, ≥3); a7 = 7 nominal age classes (5, 6, 7, 8, 9, 10, ≥11), and a10 = 10 nominal age classes (1, 2–3, 4, 5, 6, 7, 8, 9, 10, ≥11). a7 and a10 represent quadratic functional forms of the nominal age classes. AICc, ∆AICc, and wi (weight of evidence for each model) values represent within-suite values. Coh represents a short-term cohort effect ending after 3 years. Markovian transitions were 1st order

Fig. 3

Age- and cohort-specific apparent survival rates of individually marked prebreeder female Weddell seals in Erebus Bay, Antarctica, for 3 age classes and 24 cohorts (1980–2003). Estimates were corrected for tag loss, and error bars represent 95 % confidence intervals

Transition rates were very similar regardless of the parameterization of \( \hat{S}_{{i,{\text{age}}}}^{r} \). Therefore, we used the top model in the survival-probabilities suite to further model \( \psi_{{i,{\text{age}}}}^{rz} \). The best-supported model in the transition-probabilities suite received all the support from the data and included categorical age effects on \( \psi_{{i,{\text{age}}}}^{rz} \), year effects that were shared among states and age classes, and state-dependant transitions (Table 1). The top model retained its rank for all values of \( \hat{c} \) ≤ 3.0, and had >95 % of the model weight for values of \( \hat{c} \) ≤ 2.4. Consequently, we present results from only the top model (See Online Resource Table S3).

Our top model supported our predictions of age- and state-dependency in annual TE rates (Fig. 4). As expected, first-year TE rates were close to 1 (range = 0.91 [\( \widehat{\text{SE}} \) = 0.02] to 1.0 [\( \widehat{\text{SE}} \) < 0.01]; mean = 0.98 [\( \widehat{\text{SE}} \) < 0.01]), and estimates of TE rates decreased steadily with age. For seals that did attend the previous year, the average point estimates of TE rates were lowest at age 8 years (range = 0.01 [\( \widehat{\text{SE}} \) < 0.01] to 0.73 [\( \widehat{\text{SE}} \) = 0.04]; mean = 0.15 [\( \widehat{\text{SE}} \) = 0.01]), but estimates were similar across age classes for females >5 years old (Fig. 4). As predicted, TE rates were higher for females that had been temporary emigrants in the previous year (\( \hat{\beta }_{{{\text{Markovian}}\,{\text{effect}}}} \) = 1.4, \( \widehat{\text{SE}} \) = 0.10, 95 % CI = 1.2–1.6; e.g., \( \overline{{\hat{\psi }_{{i,\,{\text{age}}\,8}}^{\text{UU}} - \,\,\,\hat{\psi }_{{i,{\text{age}}\,8}}^{\text{PU}} }} \) = 0.22, \( \overline{{\widehat{\text{SE}}}} \) = 0.03). Temporary emigration rates also varied considerably from year-to-year, especially between 2000 and 2007. For example, \( \hat{\psi }_{{i,\,{\text{age}}\,8}}^{\text{PU}} \) was 0.088 (\( \widehat{\text{SE}} \) = 0.019) in 2003, 0.766 (\( \widehat{\text{SE}} \) = 0.046) in 2004, and 0.049 (\( \widehat{\text{SE}} \) = 0.014) in 2007. Such variability is consistent with our hypothesis that highly variable environment conditions should influence motivation or ability of prebreeders to attend colonies.
Fig. 4

a, b Temporary emigration rates and c, d recruitment rates of prebreeder female Weddell seals in Erebus Bay, Antarctica, during 1980–2008. Estimates were generated from the best-supported open robust design multistate model (with age- and state-dependent transitions and annual variation shared among ages), and are presented for seals that attended (\( \hat{\psi }_{i}^{\text{PU}} , \)\( \hat{\psi }_{i}^{\text{PB}} ; \)open symbols) or did not attend (\( \hat{\psi }_{i}^{\text{UU}} , \)\( \hat{\psi }_{i}^{\text{UB}} ; \)shaded symbols) reproductive colonies in Erebus Bay the previous year. States are denoted as U prebreeder not attending breeding colonies, P prebreeder attending breeding colonies, B first-time mother in the breeding colonies. a, cBoxplot summaries of age-specific a temporary emigration and c recruitment rates with medians (dark bars), means (×), whiskers bounding 2.5th and 97.5th percentiles of annual point estimates, and outliers (open and shaded circles). b, d Annual rates and 95 % confidence intervals for age-7 seals showing the temporal patterns shared in the model for each age class

Recruitment rates were also strongly age-dependent, varied by year, and depended on a female’s TE status in the previous year (Fig. 4). Recruitment rates were low at age 5 (e.g., for seals that attended the previous year the range of estimates was <0.01 [\( \widehat{\text{SE}} \) < 0.01] to 0.11 [\( \widehat{\text{SE}} \) = 0.03], and the mean was 0.06 [\( \widehat{\text{SE}} \) = 0.01]), increased gradually, and, depending on TE status in the previous year, peaked at either age 10 for seals that attended the previous year (range = 0.08 [\( \widehat{\text{SE}} \) = 0.03] to 0.72 [\( \widehat{\text{SE}} \) = 0.06]; mean = 0.52 [\( \widehat{\text{SE}} \) = 0.01]) or age 8 for seals that did not attend the previous year (range = 0.03 [\( \widehat{\text{SE}} \) = 0.01] to 0.56 [\( \widehat{\text{SE}} \) = 0.07]; mean = 0.34 [\( \widehat{\text{SE}} \) = 0.02]). As expected in a highly variable environment, annual variation in recruitment rates was substantial, and year-to-year variation was especially pronounced between 2000 and 2007. For example, \( \hat{\psi }_{{i,{\text{ age 8}}}}^{\text{PB}} \) was 0.66 (\( \widehat{\text{SE}} \) = 0.06) in 2002, 0.09 (\( \widehat{\text{SE}} \) = 0.03) in 2004, and 0.53 (\( \widehat{\text{SE}} \) = 0.05) in 2007. As predicted, recruitment rates were greater for seals that were present the previous year (e.g., \( \overline{{\hat{\psi }_{{i,{\text{age}}\,8}}^{\text{PB}} - \,\,\hat{\psi }_{{i,{\text{age}}\,8}}^{\text{UB}} }} \) = 0.13, \( \widehat{\text{SE}} \) = 0.03). However, 95 % confidence intervals for the effect of TE on recruitment included zero (\( \hat{\beta }_{{{\text{Markovian}}{\kern 1pt} {\text{effect}}}} \) = −0.11, \( \widehat{\text{SE}} \) = 0.11, 95 % CI = −0.32 to 0.10).


Using a 28-year time series of encounter histories of female Weddell seals, we were able to evaluate hypotheses and predictions about temporal variation and state-dependency in age-specific TE rates for a long-lived marine mammal. Our predictions about age-related and annual variability in TE and recruitment rates and about state dependent TE rates were supported by our results. Our data also provided some evidence that a female’s recruitment rate was related to her TE status in the previous year. Observed variability in survival rates among cohorts likely accounts for observed differences in the proportion of each female Weddell seal birth cohort that eventually recruits into the pup-producing portion of the population (Garrott et al. 2012).

When reviewing our survival estimates, it is important to remember that Markovian TE is known to sometimes cause negative biases in survival estimates at the end of time series, especially when survival rates are high and TE rates are variable (Langtimm 2009). Accordingly, we avoided making interpretations about or presenting survival estimates for the final few cohorts. However, given that the lowest survival estimate is for the final cohort we present (Fig. 3), it is possible that a slight negative bias remains for the final year or two of the time series we present. It is unlikely that such biases propagate back to earlier years (Langtimm 2009).

As expected, temporary emigration rates were high for young animals, decreased substantially prior to the age where recruitment was possible, and once seals made the transition to an observable state they were less likely than temporary emigrants to be unobservable the following year (Fig. 4). These results support the notion of a threshold body size or condition necessary for attending colonies (Laws 1956), and also provide some circumstantial evidence that, as they grow older, Weddell seals have motivation to attend colonies beyond that of immediate mating. One possible motivation might be to gather information that could improve subsequent reproductive success (Danchin et al. 2004). Current evidence of such benefits for prebreeders primarily comes from studies of colonial breeding seabirds (Schjørring et al. 1999; Aubry et al. 2009), but some reproductively mature birds (e.g., Doligez et al. 2002) and lekking mammals (e.g., Deutsch and Nefdt 1992) might also use information from conspecifics to assess quality of future nesting locations or to assess potential mates. Prebreeder Weddell seals might observe the interactions of mother-pup pairs during birth and lactation, and during the initial swimming bouts of pups. For example, mothers often help pups to haul out by reaming ice holes with their teeth. Inasmuch as the Weddell seal mating system is comparable to lekking (Stirling 1969; Siniff et al. 1977), prebreeders might also gain experience in assessing male quality when attending colonies in years prior to recruitment. Lastly, most surviving female seals born in Erebus Bay eventually recruit at one of the reproductive colonies in Erebus Bay (Stirling 1969; Cameron and Siniff 2004; Hadley et al. 2006), and prospecting within Erebus Bay may help prebreeders evaluate the quality of specific colonies.

Predator avoidance might also provide motivation to attend colonies, which, in striking contrast to the situation for many colonial breeding species, typically are safe locations inaccessible to important predators during the breeding season (Testa 1994). Thus, prebreeders in good condition may be able to forgo better foraging opportunities to attend colonies and avoid predation, and perhaps simultaneously acquire useful information. Necessary modeling assumptions precluded a test for differences in survival for seals that did or did not attend reproductive colonies. However, such a test is also irrelevant for the youngest age classes, which are most likely to have state-dependent survival rates, because they are almost entirely absent from Erebus Bay. Apparently, most young seals initially cannot afford to attend colonies regardless of any possible benefits from such attendance.

The state-dependent TE rates that we observed (Fig. 4), along with the observation that many seals apparently recruited without first attending colonies (Fig. 1), suggest that prolonged TE could be a strategy emphasizing improved body condition over potential benefits of information gathering at reproductive colonies. Emigration behavior and foraging efficiency, especially in adverse conditions, can be influenced by phenotype and genotype (Ims and Hjermann 2001; McMahon and Matter 2006; Lescroël et al. 2010). Benefits of improved body condition from prolonged TE could include earlier onset of reproduction than would otherwise be possible or the ability to produce larger pups, which are more likely to survive (McMahon et al. 2003; Proffitt et al. 2008). It is possible that attendance strategies influence choices of recruitment locations, interactions with pups, or long-term reproductive trajectories. Accumulating data on parturition and weaning masses of first-time mothers and their pups should prove useful for evaluating possible trade-offs of alternative TE strategies followed by prebreeders. For at least one species (black-legged kittiwake; Rissa tridactyla), an information-gathering strategy rather than a strategy of prolonged TE resulted in the greatest long-term reproductive success (Aubry et al. 2009). Postulating benefits of TE rests on the reasonable assumptions that TE allows higher-quality foraging than is possible in the reproductive areas and that such increased opportunity can translate into improved body condition. Validating these assumptions will require currently unavailable data on actual movements and body conditions of seals.

We observed considerable annual variation in recruitment and TE rates, which suggests that perhaps one or both types of demographic rates are influenced by variations in annual environmental conditions. In particular, our results indicate a period of increased TE rates and roughly corresponding decreases in recruitment rates after 2000 (Fig. 4b, d), with a possible consequent increase in the average recruitment age (7.1, SE = 0.6 before 2004; 8.3, SE = 09 afterward). During this period, an unusually large iceberg that broke off the Ross Ice Shelf became grounded in the Ross Sea northeast of Ross Island and dramatically altered sea-ice dynamics in McMurdo Sound and beyond, and also directly or indirectly influenced primary productivity and population dynamics of top-trophic species (Arrigo et al. 2002; Arrigo and van Dijken 2003; Kooyman et al. 2007; Remy et al. 2008). Recruitment rates of Weddell seal rebounded sharply and TE rates dropped after the iceberg broke up and moved away from McMurdo Sound in 2006 (Fig. 4). It remains to be seen whether the average recruitment age will decrease. Although the mechanisms by which the iceberg might have influenced Weddell seal demographic rates remain somewhat unclear, the timing of shifts in recruitment and TE rates that we observed suggests that shifts likely are related to changes in ice conditions resulting from the grounding of this large iceberg. Sea-ice dynamics possibly influence TE rates by directly influencing food availability, by shifting spatial distribution of the best foraging areas, or by altering access to Erebus Bay and other alternative breeding locations. Primary production in the southwest Ross Sea tends to be greatest in years with minimal sea-ice extent (Arrigo and van Dijken 2003), and although lags in the presumed trophic cascades relevant to Weddell seal foraging ecology are not well described, increased productivity eventually should drive increased foraging success for Weddell seals, which in turn could motivate greater colony attendance as growth or maintenance needs are met. In contrast, in years with extensive fast-ice, the prey-rich marginal ice zone (Brierley et al. 2002) is distant from Erebus Bay, and colony attendance could necessitate more time away from high-quality foraging areas.

Consistent with life-history theory (Pfister 1998), Weddell seal population growth rate is most sensitive to changes in adult survival, and adult survival rate has low temporal variability (Rotella et al. 2012). In a highly variable environment, the costs and benefits of attending colonies should also be variable, and we suggest that Weddell seal populations and other species might vary their TE rates in accordance with environmental conditions to buffer the effects of environmental variability, perhaps including annual population density (Rotella et al. 2009), on survival. The Southern Ocean environment is highly variable, and reproductive effort or success consequently varies considerably for colonial pinnipeds and seabirds (e.g., Nevoux and Barbraud 2005; Forcada et al. 2008). Prebreeder elephant seals (Mirounga leonina), subantarctic fur seals (Arctocephalus tropicalis), and Antarctic fur seals (A. gazella) typically are absent from reproductive colonies for the first few years of life (Hindell 1991; Lunn et al. 1994; Beauplet et al. 2005), and TE rates for subantarctic fur seals vary among cohorts and are state-dependent (Beauplet et al. 2005, 2006). Lake et al. (2008) estimated highly variable rates of detection for Weddell seals in eastern Antarctica, which may in part reflect variability in TE rates. In the northern hemisphere, colonial-breeding phocids such as northern elephant seals (M. angustirostris) and grey seals (Halichoerus grypus) have juveniles and non-reproductive adults that typically, but not always, are absent from the reproductive colonies (Sydeman et al. 1991; Pomeroy et al. 1994). We suspect that TE rates in many of the above populations might be highly variable among years.

It is possible that our estimates of TE rates reflect a slight negative bias. The probability that a seal is observed, given survival and no permanent emigration, is (1 − \( \psi^{{r{\text{U}}}} \))p*, where 1 − \( \psi_{{}}^{\text{rU}} \) is the probability of not being a temporary emigrant and p* is the probability of detection at least once, given attendance in the study area. One assumption of open robust design multistate models is that no TE occurs within seasons, i.e., once individuals enter the study area, they do not temporarily leave and then return in the same season. If this assumption was violated, e.g., if seals made within-season foraging trips, p* was under-estimated, and consequently, \( \psi_{{}}^{\text{rU}} \) was also under-estimated. Currently, we cannot evaluate the extent of such bias, but given that our estimates of p* were ≥0.91 (≥0.92 in most years) and very precise (see Online Resource for further details), we suspect that any negative bias in \( \hat{p}* \), and consequently in \( \hat{\psi }_{{}}^{\text{rU}} \), was minimal.

Studies that explicitly estimate TE rates and provide ecological interpretations are uncommon in the ecological literature, but we believe that an understanding of variability in TE rates could provide life-history insights for diverse taxa, especially in cases where actual movement of individuals in a population cannot be followed. For example, partial migration is a common and often conditional phenomenon in many species, especially fish, birds, and mammals (e.g., White et al. 2007; Brodersen et al. 2008; Boyle 2008), and can be manifested as TE if individuals are not observable in certain periods. Similarly, individuals skipping reproduction in certain years are often, but not always, unobservable (Frederiksen and Bregnballe 2000; Rivalan et al. 2005; Muths et al. 2006; Converse et al. 2009), and TE models can be used to evaluate hypotheses about reproductive costs (Kendall et al. 1997; Hadley et al. 2007b; Converse et al. 2009). Temporary emigration models can also provide interesting insights about environmental and individual influences on patterns of torpor in small mammals (e.g., Kendall et al. 1997) or dormancy in herbaceous perennial plants (e.g., Kéry et al. 2005). Lastly, information about TE allows a fuller understanding of change in population size than is possible with estimates of only survival and reproduction. Based on our work and other examples cited here, we regard estimation of TE rates not as a nuisance but as an opportunity to gain useful insights about ecological and demographic processes of populations.


This work was supported by the National Science Foundation, Office of Polar Programs (grant no. ANT-0635739 to R.A. Garrott, J.J. Rotella, and D.B. Siniff, and previous grants to D.B. Siniff and J.W. Testa). Animal handling protocol was approved by Montana State University’s Institutional Animal Care and Use Committee (Protocol #41-05) and complied with the Marine Mammal Permit Act of the USA and the multinational Antarctic Conservation Act. Raytheon Polar Services Corporation, Petroleum Helicopters International, and the New York National Guard provided logistical support for field work in Antarctica. We thank D.B. Siniff for numerous conversations about Weddell seal ecology and helpful advice throughout this work, W.L. Kendall for advice on analysis and critical review of previous manuscript drafts, and J.L. Laake for help with RMark coding. We appreciate the dozens of field assistants that supported this project through the years. Two anonymous reviewers provided helpful comments on an earlier manuscript draft.

Supplementary material

442_2012_2472_MOESM1_ESM.pdf (869 kb)
Supplementary material 1 (PDF 868 kb)

Copyright information

© Springer-Verlag 2012