Population Ecology

, Volume 55, Issue 3, pp 405–415

Birth-year and current-year influences on survival and recruitment rates of female Weddell seals

Authors

    • Department of EcologyMontana State University
    • The Pennsylvania State University
  • Jay J. Rotella
    • Department of EcologyMontana State University
  • Robert A. Garrott
    • Department of EcologyMontana State University
Original article

DOI: 10.1007/s10144-013-0379-0

Cite this article as:
Stauffer, G.E., Rotella, J.J. & Garrott, R.A. Popul Ecol (2013) 55: 405. doi:10.1007/s10144-013-0379-0

Abstract

In long-lived species, juvenile survival typically is lower and more variable than adult survival, and modeling such variation is important for understanding population dynamics. Variability in juvenile survival can be related to birth- or current-year influences, and the birth-year influences can be transient, persistent, or intermediate in duration. We used multi-state models and data collected from 5,459 known-aged prebreeder female Weddell seals (Leptonychotes weddellii Lesson) tagged in Erebus Bay, Antarctica from 1980–2007 to evaluate the duration of potential birth-year influences on survival rates and the importance of birth- and current-year influences on survival and recruitment rates. Survival rates differed for each birth cohort and were positively related to current-year winter sea-ice conditions. The estimated duration of birth-cohort effects on survival was intermediate (6 years) rather than transient (2 years) or permanent. Estimated survivorship from birth to 6 years of age varied among cohorts from 0.13 (SE = 0.04) to 0.42 (SE = 0.06), and averaged 0.25 (SE = 0.02). Recruitment rates (probability of transitioning from prebreeder to breeder state) varied annually but apparently were not related to birth-year conditions. Our results provide evidence that birth- and current-year conditions act in combination to influence survival. Although for many long-lived species the influences of either birth- or current-year conditions on survival are well-studied, we suggest that modeling survival rates as a function of birth- and current-year influences simultaneously could lead to better understanding of survival and improved stochastic models to project population dynamics.

Keywords

Capture-mark-recaptureEnvironmental variabilityMarine mammalMcMurdo SoundPinnipedSouthern Ocean

Introduction

Population growth rates of long-lived species often are most sensitive to changes in adult survival rates, but variability in juvenile survival rates may be responsible for much of the actual variation in population growth rates (Gaillard et al. 1998). This is because in long-lived species, evolution of life-history strategies that minimize or buffer variability in adult survival are expected (Pfister 1998; Gaillard et al. 2000). In contrast, juvenile survival rates often are highly variable, perhaps in part because juveniles typically are less developed and less experienced than adults and are thus more vulnerable to mortality risks, especially during harsh conditions (McMahon et al. 2000; Forcada et al. 2008).

Variability in juvenile survival rates can arise in diverse ways. Attributes of individuals or environmental conditions experienced around the time of birth or hatching can have either transient or persistent influence on survival rates. When conditions experienced by young during development influence early survival, but have no further influences on fitness, they can be viewed as filters that simply allow fewer or more individuals experiencing those conditions to survive to maturity (e.g., Gaillard et al. 1997; Drummond et al. 2011). In contrast, early conditions may set the stage for persistent fitness differences among individuals. For example, poorly provisioned young may experience reduced fecundity later in life or persistently face greater mortality risk relative to well-provisioned young. Grafen (1988) called this phenomenon a “lead spoon” effect, and one may likewise imagine a “silver spoon” effect where individuals experience exceptionally good conditions early in life, with consequent persistent increased fitness relative to other individuals. Persistent differences in various fitness components, including survival, have been shown for diverse taxa (Lindström 1999; Monaghan 2008). The duration of birth-year influences could also be intermediate in duration, rather than permanent.

Various attributes of individuals, their parents, or their rearing environment have been linked to transient or persistent intra-cohort variation in fitness components. For example, maternal attributes such as age and mass at parturition were related to transient and persistent intra-cohort variations in survival, onset of breeding, and lifetime reproductive success (LRS; Hadley et al. 2007a; Rödel and von Holst 2009). Influences of maternal attributes might be especially important during harsh conditions when mothers mitigate their own mortality risk by passing survival costs on to their offspring (Martin and Festa-Bianchet 2010). In southern elephant seals (Mirounga leonine Gill), the influence of weaning mass and birth-year environmental conditions on survival was transient, as only first-year survival was related to these covariates (McMahon and Burton 2005). Similarly, in subantarctic fur seals (Arctocephalus tropicalis Gray), first-year survival, but not subsequent survival rates were related to pup growth rate during lactation and environmental conditions in the 6 months after weaning (Beauplet et al. 2005). Subordinate nestling blue-footed boobies (Sula nebouxii Milne-Edwards) had elevated stress hormone levels and suffered greater prefledging mortality than did singletons or dominant nestlings, but again no differences were detected among the groups in subsequent rates of survival or reproduction (Drummond et al. 2011). In contrast, hatching order of kittiwakes (Rissa tridactyla Linnaeus) was related to persistent intra-clutch differences in survival and recruitment probabilities (Cam et al. 2003). Similarly, oystercatchers (Haematopus ostralegus Linnaeus) raised in high-quality habitat had greater juvenile and adult prebreeder survival, were more likely themselves to settle in high-quality habitat as breeders, and had greater LRS than did individuals raised in low-quality habitat (van de Pol et al. 2006).

Transient or persistent effects can be related to individual attributes as in the above examples, but also can be shared by entire birth cohorts when they are related to environmental conditions experienced early in life by entire cohorts. For example, food availability during development influence growth rates of all individual snakes in a cohort (Madsen and Shine 2000). In red-billed choughs (Pyrrhocorax pyrrhocorax Linnaeus), there was a transient relationship between survival and environmental conditions experienced during fledging, but a more persistent influence on reproductive success later in life (Reid et al. 2003). In roe deer (Capreolus capreolus Linnaeus), birth-year environmental variation resulted in persistent differences among cohorts in survival rates and breeding abilities in one population whereas in another population differences among cohorts were manifested only as a transient effect influencing how many members of each cohort survived the first year or two of life (Gaillard et al. 1997).

Conditions experienced at birth can influence survival during subsequent years, but conditions throughout life should also exert influence, particularly for slow-maturing species experiencing substantial annual variation in environmental conditions. It might therefore be important to consider influences of birth year and current year on variability in juvenile survival. Current-year environmental conditions might influence survival of all extant cohorts similarly, but inter-cohort variation could persist because of conditions experienced during the birth year. Regardless of subsequent environmental conditions, the “filter hypothesis” predicts that age-specific survival rates will differ among cohorts at young ages but then be similar later in life, whereas the spoon hypothesis predicts persistent differences in age-specific survival rates. Even if conditions experienced at birth have little influence beyond the birth year, different cohorts could still have different survivorship to older ages simply because they experience different series of environmental conditions; however, annual survival rates for older age classes might be similar because all older animals are influenced by the same current-year conditions.

To address questions about the importance and persistence of birth-year influences and of current-year conditions on survival in long-lived species requires long-term data from populations in variable environments. We used 27 years of data collected from a population of Weddell seals (Leptonychotes weddellii Lesson) in Erebus Bay, Antarctica to evaluate the following: (1, 2) the magnitude and duration of inter-cohort variation in survival rates; (3) potential associations between birth-year environmental covariates and inter-cohort variation in survival either early in life, later in life, or both; (4) potential associations between current-year covariates and variation in subsequent annual survival shared by surviving members of all cohorts; and (5) relative support for inter-cohort or inter-annual variations in recruitment rates. We restricted analysis to data from females prior to first reproduction (i.e., prebreeders) as this allowed us to focus our attention on survival during the life-history period when approximately 80 % of females typically disappear from a cohort (Garrott et al. 2012). This population of seals lives in a polar environment with substantial annual variation in ice conditions; maternal influences on survival and subsequent reproduction (Hastings and Testa 1998; Hadley et al. 2007a; Proffitt et al. 2010) as well as differences in survival among cohorts (Cameron and Siniff 2004) have previously been documented. Further, the proportion of each female cohort that eventually produces a pup varies fourfold, which likely results from either transient or persistent effects of early conditions on survival (Garrott et al. 2012).

Methods

Study area and population

Erebus Bay is located in McMurdo Sound in the southwestern Ross Sea, Antarctica, where sea-ice condition are highly variable (e.g., summer sea-ice extent in the Ross Sea varies fivefold from year-to year). In Erebus Bay, tidal action and glacial pressures create and maintain 8–14 sea-ice cracks in the same areas each year, and reproductive colonies of seals form each austral spring around these cracks. Weddell seals typically give birth to a single pup on the sea-ice surface during late-October to early-December following a gestation period of approximately 9 months. Mothers and pups remain on the ice surface nearly continually for about a week post-parturition, after which they swim for brief periods each day. Lactation lasts 5–6 weeks, is supported primarily from resources that mothers accumulated during the previous summer and winter, and is terminated abruptly when mothers abandon pups. Many other seals are present in the colonies for ≥1 years prior to reproductive maturity (prebreeders) or in years where they skip reproduction (skip-breeders). Most seals in the colonies spend a considerable portion of each day on the surface of the ice and consequently are highly detectable and approachable.

Food resources for seals probably are limited in Erebus Bay, and adults and pups are believed to move out into food-rich foraging areas in the Southwest Ross Sea following the pupping season (Smith 1965; Testa et al. 1985; Burns et al. 1999). Females born in Erebus Bay typically are not observed for ≥1 years following birth, but surviving females are highly philopatric and most prebreeders return to Erebus Bay for ≥1 year prior to recruiting into the reproductive population (Hadley et al. 2006). The mean age of recruitment is 7.6 years, and females typically produce one pup every 1.5–2.2 years thereafter (Hadley et al. 2006, 2007b).

Data collection

Each year since 1969, Weddell seal pups born in Erebus Bay were marked with individually identifiable and paired plastic livestock tags attached to the interdigital webbing of each hind flipper. The proportion of pups tagged each year varied in the early years of the study, but since 1982, virtually all pups born in the study area have been tagged. Subsequently, broken or missing tags were replaced as necessary. Resighting surveys of the study population were conducted every 3–6 days (5–8 surveys/year) during early November through mid-December of each year. Our dataset included observations from 5,459 female seals born between 1980 and 2007, 862 (16 %) of which went on to produce a pup by 2007.

Covariates of survival

The environmental covariates of prebreeders survival that we considered were (1) winter sea-ice extent (SIEw) for the Ross Sea sector measured just prior to the pupping season in September of year t, (2) summer sea ice-extent (SIEs) measured in February of year t + 1 shortly after the end of the pupping season in year t, (3) Southern Oscillation Index (SOIw) averaged for the winter (July–September) prior to pupping in year t, (4) Southern Oscillation Index (SOIs) averaged for the summer (December–February) following the pupping of year t, and (5) Antarctic Dipole (ADP) averaged for the summer (December–February) following the pupping of year t. Microbial ice communities constitute a substantial portion an annual primary productivity in the Ross Sea (Arrigo and Thomas 2004), and winter sea-ice is important in the life cycle of krill (Loeb et al. 1997). Increased krill abundance may benefit Antarctic silverfish (Pleurogramma antarcticum), a primary food source for Weddell seals (Burns et al. 1998). Additionally, Weddell seals are better able to exploit resources under ice than are other Antarctic predators, and extensive winter sea-ice may reduce competition for prey. Therefore, we reasoned that mothers could build body reserves best in years with extensive winter-sea-ice, and we predicted a positive relationship between birth-year SIEw and survival of prebreeders. Similarly, we predicted that current-year SIEw would be positively related to survival. In contrast, extensive summer sea-ice reduces the amount of open water available for phytoplankton blooms in the Ross Sea and apparently reduces foraging success of female Weddell seals during the summer and is negatively related to subsequent pup weaning mass (Proffitt et al. 2007a, b). However, extensive summer sea-ice might also provide protection from predation for young seals. Therefore, we predicted that survival rates for prebreeders would be negatively related to birth-year SIEs, but positively related to current-year SIEs. Although sea-ice extent probably is an important factor in Weddell seal biology, large-scale, integrative indices may capture more environmental variation and thus better explain variation in vital rates than do local covariates (Stenseth et al. 2003). The Southern Oscillation Index (SOI) and the ADP are two such measures available for the Ross Sea. SOI is a climatic index based on sea-surface pressure differences in the Southern Ocean and commonly used as a measure of El Niño/Southern Oscillation (ENSO) strength. SOI is positively correlated with sea-ice extent in the Ross Sea (Kwok and Comiso 2002), and positive SOI phases have been linked to increases in seal pup weaning mass, population size, and reproductive rates (Testa et al. 1991; Proffitt et al. 2007a; Rotella et al. 2009). Therefore, we predicted increased survival for cohorts produced following positive SOI phases in either the winter or summer of the birth year or current year. The ADP is a climate mode reflecting high-latitude characteristics of ENSO forcing (Yuan 2004). For Erebus Bay seals, the ADP the summer following birth is positively related to eventual recruitment into the reproductive population (Garrott et al. 2012), and so we predicted that survival would be positively related to ADP during the summer months following birth. We recognize that population responses to environmental conditions may involve nonlinearities and time lags, but as a logical starting point we chose to restrict our analysis to linear and direct environmental effects. We used standardized values (mean = 0, SD = 1) of each of the environmental covariates in all analyses. SIE data were obtained from ftp://sidads.colorado.edu/pub/DATASETS/nsidc0192_seaice_trends_climo/ice-extent/nasateam/ and SOI data from http://www.bom.gov.au/climate/current/soihtm1.shtml. Covariate values varied substantially among years (Fig. 1) and pair-wise correlations among covariates ranged from −0.16 to 0.42.
https://static-content.springer.com/image/art%3A10.1007%2Fs10144-013-0379-0/MediaObjects/10144_2013_379_Fig1_HTML.gif
Fig. 1

Covariates considered in multistate models estimating survival rates of Weddell seals prior to first reproduction. Environmental covariate values are expressed as standardized values (mean = 0, SD = 1), and cohort size was log-transformed (natural log) prior to modeling. Environmental covariates are Sea-ice extent (SIE), Southern Oscillation Index (SOI), and Antarctic Dipole (ADP)

Besides environmental covariates, we also considered birth-year and current-year cohort size as covariates that might explain variation among prebreeder survival rates, where cohort-size is the count of pups born in Erebus Bay. It is likely that many environmental factors influence the ability of female Weddell seals to produce a pup in any given year, and pupping rates and cohort size should be greatest when overall environmental conditions are favorable for pup production. In Weddell seals, birth-year cohort size is positively related to the probability that a female seal will eventually recruit into the reproductive population (Garrott et al. 2012). Therefore, we viewed cohort size as an integrated index of overall environmental favorability, and we predicted that cohort size in either the birth year or the current year would be positively related to survival of prebreeders. The population of seals available to produce pups is finite, and we expected that the effect of incremental increases in cohort size would diminish as environmental conditions improved. Therefore, we use the natural log of cohort size in our analysis rather than actual cohort size.

Data analysis

We used multistate capture-mark-recapture models (e.g., Brownie et al. 1993; Nichols and Kendall 1995) to evaluate influences of environmental conditions on survival rates of female Weddell seals prior to first reproduction. Mothers with pups in this population are highly approachable and reliably detected each breeding season (Hadley et al. 2006; Rotella et al. 2009). Consequently, we could readily determine when females produced pups and could therefore reliably assign each seal into either a prebreeder (P, present in the study area and never documented with a pup) or breeder (B, mother with pup) state each year it was observed. The transition from P to B between year t and year t + 1 defined recruitment, and multistate models allow estimation of this recruitment probability each year. For the hypotheses evaluated here, we were interested primarily in survival rates prior to recruitment and so did not include encounters of females after they recruited into the breeding population. In conducting our modeling, we used age classes for prebreeders previously shown to be appropriate for this population (Hadley et al. 2006). Accordingly, survival was modeled for 3 age classes: age 1 = survival from birth to age 1 year, age 2 = survival from age 2 to age 3 years, and age 3+ = subsequent annual survival, and detection probabilities were modeled for 4 age classes: ages 1, 2, 3–6 years, and greater than 6 years old. Recruitment probabilities were modeled for 8 age classes: ages 4, 5, 6, 7, 8, 9, 10, and 11+ years old. For example, recruitment at age 7 denoted the probability of making a transition from state P at age 6 to state B at age 7 years old. We used a 2-step model-selection process, first identifying an appropriate parameterization for survival probabilities, then modifying the top survival model to evaluate two different structures for recruitment (Doherty et al. 2010).

We considered three types of survival models: models that included only influences of birth-year covariates, models that included only influences of current-year covariates, and models that combined influences of both the birth-year and the current year. We also evaluated models without environmental covariates that instead allowed survival to differ for each cohort (cohort effect) or each year. All covariates were modeled so as to influence log-odds of survival in a linear fashion. To avoid overparameterization and confounding of birth-and current-year effects, we did not include any interaction terms. For example, the logit-linear model for apparent survival that included an effect of birth-year and an effect of current-year winter sea-ice extent was logit(S) = β0 + β1 × age0 + β2 × age1 + β3 × cohort1 + ··· + βpenultimate × cohortlast + βultimate × ySIEw. In combined models, we allowed only one current-year covariate or a year effect and one birth-year covariate or a cohort effect, and we considered all possible combinations of such models. To evaluate duration of birth-year influences, we considered combined models that allowed birth-year effects to persist indefinitely or to last only for the first 2 years of life, when survival rates are substantially lower than later in life (Hadley et al. 2006). For example, when effects were persistent, the log-odds of survival differed for each cohort by a constant amount. When effects were transient, the difference in log-odds lasted for only the first two age classes. We recognized that influences of birth-year conditions might be intermediate in duration, so after a priori analyses were completed, we also explored several models where the duration of birth-year influences was allowed to vary from 2–8 years. All exploratory models were modifications of our top-ranked model. Previous analysis indicated that an appropriate model structure for detection and recruitment probabilities allowed variation among age classes and differences among years shared by each age class (Hadley et al. 2006), so we used this structure throughout when evaluating survival models. Once we identified our best survival model, we evaluated one additional model where recruitment rates were allowed to differ by cohort rather than year.

We performed analyses in program MARK (White and Burnham 1999) through the RMark package (Laake 2010) in program R (R Development Core Team 2012). To evaluate goodness-of-fit, we used the median \( \hat{c} \) procedure implemented in program MARK to estimate an overdispersion parameter, c, for our most general model. This model did not include cohort size or any environmental covariates and instead allowed survival to differ by cohort and year. We subsequently used \( \hat{c} \) to inflate the variances of our estimates and to calculate a bias-corrected and quasi-likelihood form of Akaike’s information criterion (QAICc; Burnham and Anderson 2002), which we then used to rank models.

When individuals are typically absent for the first few years of life, their unavailability for capture can result in negative bias in survival rates for younger age classes near the end of the time series. In general, if data are collected from additional years when surviving, but previously unobservable individuals become available for capture, the bias disappears (Langtimm 2009). Because of this potential for negative bias, we avoided making inferences for cohorts born in the final 4 years of our time series, i.e., cohorts born in 2003–2006.

Results

We estimated overdispersion as 1.56 and accordingly used QAICc to rank models and \( \hat{c} \) = 1.56 to inflate variances of estimates. Detection probabilities (p) were highly variable among years and very low for young seals (\( \hat{\bar{p}}_{{{\text{age}}1}} \) = 0.03, SE ≤ 0.01), but increased substantially for older seals (\( \hat{\bar{p}}_{{{\text{age}}7}} \) = 0.54, SE = 0.03). These estimates largely reflect temporary emigration, as true detection rates for this population is nearly 1 (Rotella et al. 2009). Model selection results provided clear evidence of temporal variation in survival rates; models that allowed survival to vary by age but not over time received no support (Table 1). Models that combined birth-year and current-year influences were better-supported than models that included only birth-year or only current-year influences, and models that predicted survival as a function of named birth-year environmental covariates or cohort size were not as well supported as models that simply allowed unique survival rates for each cohort (cohort effect). A persistent cohort effect was supported over a transient cohort effect, but exploratory analysis indicated that the duration of birth-year influences on juvenile survival was intermediate (6 years; Table 2). A model that allowed age-specific recruitment probabilities to vary by current-year was better-supported than a model that allowed variation to vary by birth cohort (Table 2). Thus, the data provided no evidence for permanent effects on recruitment probability but do indicate multi-year effects of early conditions on survival.
Table 1

Multistate model selection results representing structures of variation in survival rates for female Weddell seals in Erebus Bay, Antarctica

Modela

K

QAICc

∆QAICc

wi

 (136) cohort + ySIEw

88

12,447.0

0.00

0.181

 (142) coh + ySIEw

88

12,448.1

1.14

0.103

 (133) cohort + yADP

88

12,448.5

1.47

0.087

 (137) cohort + ySOIw

88

12,448.7

1.66

0.085

 (66) cohort

87

12,449.1

2.15

0.079

 (132) coh + ycohsize

88

12,449.3

2.27

0.058

 (143) coh + ySOIw

88

12,449.4

2.36

0.056

 (140) coh + ySIEw

88

12,450.1

3.07

0.039

 (141) coh + ySOIw

88

12,450.4

3.36

0.034

 (139) coh + yADP

88

12,450.6

3.64

0.029

 (138) coh + ycohsize

88

12,450.8

3.82

0.027

 (135) cohort + ySOIs

88

12,450.9

3.93

0.025

 (1) no temporal variation

62

12,522.0

75.0

0.000

Results are shown for all models that had ∆QAICc <4, and for a model with no temporal variation. In all models, detection probabilities were modelled for 4 nominal age classes (1, 2, 3–6, ≥7 years old) and with time variation shared among all age classes. Similarly, in all models recruitment probabilities (transitions from a non-reproductive to a reproductive state) were modelled with annual variation shared among 8 nominal age classes (4, 5, 6, 7, 8, 9, 10, 11+ years). In all models, survival was modelled for 3 nominal age classes (1, 2, ≥3). K represents number of parameters in the model and wi represents model weight

aCategorical covariates are: cohort permanent effect of birth year, coh temporary effect of birth year; Current-year covariates are: ySIEw effect of current-year winter sea-ice extent (SIE), yADP effect of summer Antarctic Dipole, ySOIs effect of summer Southern Oscillation Index (SOI), ySOIw effect of winter SOI, ycohsize effect of cohort size. Birth-year covariates are denoted similarly, but without preceding y, e.g., cohsize effect of birth-year cohort size

Table 2

Comparison of multistate models where (1) duration of a birth-year influence on survival of female Weddell seals prior to first reproduction was allowed to vary from 2 years to >8 years, and (2) age-specific recruitment probabilities were modelled as a function of age and birth-year or age and current-year

Modela

K

QAICc

∆QAICc

wi

Duration of birth-year influence

 6 years

88

12,439.3

0.00

0.831

 8 years

88

12,443.3

3.96

0.114

 4 years

88

12,446.2

6.85

0.027

 More than 8 yearsb

88

12,447.0

7.67

0.018

 2 yearsc

88

12,448.1

8.81

0.010

Source of variation in recruitment probability

 Current yearb

88

12,447.0

0.00

1.000

 Birth year

88

12,591.2

144.16

0.000

K represents number of parameters in the model and wi represents model weight. QAICc is given for comparison with Table 1

aModels are either identical to, or modifications of model 136 in Table 1, which also included effects on recruitment of 8 nominal age classes (4, 5, 6, 7, 8, 9, 10, 11+ years) and additive effects on survival of 3 nominal age classes (1, 2, ≥3 years old) and current-year sea-ice extent

bModel is identical to model 136 in Table 1

cModel is identical to model 142 in Table 1

The best-supported model included a persistent categorical cohort effect and, as expected, a positive influence of current-year SIEw (\( \hat{\beta }_{\text{ySIEw}} \) = 0.19, SE = 0.09). The second ranked model (∆QAICc = 1.14) was similar to the top-ranked model except that it included a transient rather than persistent cohort effect. Three other models within 2 QAICc units (Table 1) included persistent cohort effects and positive influences of either current-year ADP (\( \hat{\beta }_{\text{yADP}} \) = 0.16, SE = 0.09), current-year cohort size (\( \hat{\beta }_{\text{ycohsize}} \) = 0.30, SE = 0.41), or current-year SOIw (\( \hat{\beta }_{\text{ySOIw}} \)= 0.15, SE = 0.09). In each of these 3 models, the sign of the relationship of the current-year covariate with juvenile survival was in the predicted direction, but 95 % CIs for the estimated coefficient relating the covariate to survival overlapped zero. The best-supported model combining a birth-year covariate (rather than a cohort effect) and a current-year covariate (∆QAICc = 7.32) contained a persistent influence of birth-year cohort size that, as predicted, was positive, but imprecise (\( \hat{\beta }_{\text{cohsize}} \) = 0.23, SE = 0.23) as well as a positive influence of current-year SIEw (\( \hat{\beta }_{\text{ySIEw}} \) = 0.34, SE = 0.05). Estimates of first-year survival rates from the top-ranked model ranged from 0.45 (SE = 0.10) for a weak cohort (1980) in a hypothetical year with unfavorable (minimal extent) winter sea-ice conditions to 0.76 (SE = 0.07) for a strong cohort (1999) in a year with favorable (maximal extent) winter sea-ice conditions (Fig. 2a). Although the top model allowed age-specific survival rates to vary by cohort with no constraints on the pattern of variation, cohort-specific estimates of age-specific rates of survival were actually quite similar for most cohorts; only two of 23 1st-year survival estimates were <0.54 and only four were >0.69 (Fig. 2b).
https://static-content.springer.com/image/art%3A10.1007%2Fs10144-013-0379-0/MediaObjects/10144_2013_379_Fig2_HTML.gif
Fig. 2

Predicted survival rates a for “weak”, “average”, and “strong” cohorts when sea-ice extant (SIE) is minimal, average, or maximal, and estimated survival rates b from best-supported multistate model for 1st, 2nd, and 3rd-year survival for each cohort. Error bars indicate 95 % confidence intervals

The data provided strong support for models in which birth- and current-year conditions both are related to survival rates of prebreeders and reasonable support for models that included a persistent rather than a transient influence of birth-year conditions. In exploratory analysis, we modified our top model to evaluate whether birth-year effects on survival might be of some intermediate duration. The best-supported exploratory model indicated that the duration for the influence of birth-year on survival rates was 6 years (Table 2), and this model had substantially more support than a model that allowed the influence of birth year to persist for only 2 years (∆QAICc = 8.8) or indefinitely (∆QAICc = 7.7). The second-best-supported model (∆QAICc = 4.0) allowed the influence of birth year to persist for 8 years.

Survivorship trajectories from birth to age 6 years were influenced by initial differences in survival rates among cohorts as well as differences arising from subsequent sea-ice conditions. Initial differences appeared to have the greatest influence, as indicated by a comparison of trajectories for (1) hypothetical “strong” and “weak” cohorts that subsequently experienced consecutive years of either favorable or unfavorable sea-ice conditions, and (2) actual cohorts that experienced actual sea-ice conditions (Fig. 3). Differences between strong and weak cohorts were slightly greater than differences attributable to subsequent favorable or unfavorable sea-ice conditions (Fig. 3a, b). Also, differences among cohorts tended to persist despite actual differences in sea-ice conditions experienced by cohorts (i.e., trajectories only occasionally crossed; Fig. 3d). The proportion of each actual cohort that was estimated to survive to age 6 varied from a low of 0.13 (SE = 0.04) to a high of 0.42 (SE = 0.06) and averaged 0.25 (SE = 0.02).
https://static-content.springer.com/image/art%3A10.1007%2Fs10144-013-0379-0/MediaObjects/10144_2013_379_Fig3_HTML.gif
Fig. 3

Predicted survivorship trajectories comparing a a “strong” cohort and b a “weak” cohort that each subsequently experienced consecutive years with either favorable or unfavorable winter sea-ice extent (SIE); and survivorship trajectories based on actual estimated survival rates for c a “strong” and “weak” cohort, and d survivorship trajectories for all cohorts born during 1980 through 2003. Predictions were generated by multiplying together age-specific survival estimates from our best-supported multistate model, which allowed different estimates for each cohort and included a positive influence of current-year winter SIE. Error bars (95 % confidence intervals) represent uncertainties estimated by using the delta method (Powell 2007)

We compared our selected survival model, which estimated age-specific recruitment probabilities as a function of the current-year, with an analogous model that instead estimated recruitment probabilities as a function of the birth-year. The model with inter-annual variations in recruitment probabilities (our best model) was far better supported (∆QAICc = 144.2) than the model with inter-cohort variations in recruitment probabilities (Table 2). Age-specific recruitment probability was greatest for prebreeders of age 7 years and was highly variable among years; the average estimate for recruitment probability for this age class was 0.43 (SE = 0.03), but estimates varied from a low of 0.07 (SE = 0.03) to a high of 0.66 (SE = 0.07).

Discussion

We evaluated birth-year and current-year influences on survival and recruitment rates of female prebreeder Weddell seals. Here we highlight three main results from our analysis. First, survival rates prior to first reproduction were influenced by (1) birth-year conditions that were unique to each cohort, but were not identified by any of our measured covariates; and (2) current-year conditions experienced by surviving members of all extant cohorts and best modeled using the extent of winter sea-ice. Second, the influence of birth-year conditions on subsequent survival persisted for about 6 years but did not appear to be permanent. We do not believe this selection of an intermediate duration is a consequence of sparse data resulting from censoring seals after they recruited. Of the seals in our sample that recruited, the average age of recruitment was 7.5 years, but more than half spent >6 years as prebreeders, and an appreciable number spent >8 years as prebreeders (Fig. 4). Third, age-specific recruitment probabilities were influenced primarily by current-year conditions experienced by surviving members of all extant cohorts rather than by persistent influences of conditions experienced by pups or moms during or prior to the birth year. Thus, the influence of birth-year conditions on survival of Weddell seals did not appear to be transient such as those documented for other Southern Ocean pinnipeds (McMahon et al. 2000; Hall et al. 2001; Beauplet et al. 2005; McMahon and Burton 2005). However, there also was little evidence of permanent effects on survival resulting from environmental conditions experienced at birth. Our results thus strongly support the suggestion by Garrott et al. (2012) that differences in survival rates among cohorts over the first several years of life, rather than difference in eventual age-specific recruitment rates, primarily determine what proportion of females in each cohort eventually produces pups. Although the duration of a cohort effect clearly was greater than 2 years, most of the variation in survivorship curves for different cohorts did occur in the first year or two of life, and most, though not all, trajectories maintained their position relative to other trajectories over time (i.e., trajectories rarely crossed; Fig. 3d).
https://static-content.springer.com/image/art%3A10.1007%2Fs10144-013-0379-0/MediaObjects/10144_2013_379_Fig4_HTML.gif
Fig. 4

For 862 female Weddell seals, the number of year they spent as prebreeders prior to being observed with their first pup

Persistent consequences of maternal or individual attributions or natal location have been documented in Weddell seals and other species (e.g., Cam et al. 2003; Reid et al. 2003; van de Pol et al. 2006; Hadley et al. 2007a, 2008; Proffitt et al. 2008a), and persistent differences in survival and reproduction among cohorts also have been documented (e.g., Gaillard et al. 1997; Reid et al. 2003). We cannot dismiss the possibility that, given recruitment, subsequent reproductive output by Weddell seals could be related to birth-year conditions experienced by all members of a cohort. However, we suggest that the eventual fitness of each cohort might be affected more by persistent influences of birth-year conditions on survival than by persistent influences of birth-year conditions on reproductive success.

We propose two plausible hypotheses to explain why annual survival rates for prebreeders from different birth cohorts might be similar beyond 6–8 years of age. First, once seals reach a certain threshold body size or condition, past inequalities might no longer matter, or might be evident only in senescence (Metcalfe and Monaghan 2001). Second, mortality might filter out the “frailest” members of each cohort (Cam et al. 2002), and tend to leave as survivors more of the higher-quality individuals that are less affected by unfavorable environmental conditions. The observations that age-specific recruitment probabilities do not vary among cohorts also is consistent with such a frailty hypothesis. Heterogeneous frailty has not yet been investigated in this population, but variation in reproductive quality among individuals has been proposed (Hadley et al. 2008; Proffitt et al. 2008b) and will be the subject of future investigations.

Our results suggest that current-year conditions and birth-year conditions both influenced survival. Although several environmental covariates were included in well-supported models, winter sea-ice extent was the best-supported covariate. Extensive winter ice is positively associated with krill abundances (Loeb et al. 1997), but current understanding of trophic connections between fish-eating seals and krill is limited. Based on the idea that the Ross Sea ecosystem is structured from the top down (Ainley et al. 2006), Garrott et al. (2012) suggested that Weddell seals, given their ability to dive deep and forage below extensive areas of ice, might face less competition from other top predators for food resources in years with extensive winter sea-ice extent. Evidence from several avian species of fish predators does indicate that species that need open-water areas for foraging suffer in years with extensive sea ice (Barbraud et al. 2000; Wilson et al. 2001; Jenouvrier et al. 2006). However, data are limited regarding other potential competitors such as cetaceans, other seal species, or Antarctic toothfish (Dissostichus mawsoni Norman), the latter which is also a potential prey item for Weddell seals (Ainley and Siniff 2009).

Winter sea-ice extent was out best-supported covariate, and captured some, but clearly not all of the annual variation in adult survival (Fig. 5). A useful alternative to the fixed effects approach we took would be to model random effects of cohort and year. While logistically challenging (Choquet and Gimenez 2012), this approach would facilitate a better accounting of the process variation captured by covariates, by measuring the reduction of the standard deviation of the random effects.
https://static-content.springer.com/image/art%3A10.1007%2Fs10144-013-0379-0/MediaObjects/10144_2013_379_Fig5_HTML.gif
Fig. 5

Comparison of adult survival rates from a model showing annual variation and a model showing variation among cohorts but with no additional annual variation (a), and for one example cohort (b), comparison of estimates from a model with annual variation, a model with different rates for each cohort (only 1 cohort shown here), but no additional annual variation, and our best model (model 136 in Table 1) with different rates for each cohort (only 1 cohort shown here) and additional variation related to current-year winter sea-ice extent

Population ecologists commonly use stochastic models to project population dynamics. We suggest that in some cases, especially when environmental covariates are used to predict survival, it might be important to account for persistent influences of birth-year conditions on survival as well as current-year conditions experienced by cohorts. Although we found evidence that birth-year influences on survival for prebreeders did not persist past about 6 years of age, it is possible that conditions experienced near the time of birth might influence later reproductive performance or patterns of survival (e.g., Metcalfe and Monaghan 2001; Reid et al. 2003). Birth-year conditions experienced by cohorts (or individuals) also might influence what set of future conditions is optimal for a given cohort of individual. For example, an individual born in harsh conditions might be better adapted for future poor conditions, relative to other individuals. In contrast, such an individual might be burdened with a “lead spoon” that exacerbates negative effects of poor conditions in the future (Monaghan 2008). By gaining better knowledge of environmental influences on Ross Sea productivity, and of the nature, timing, and strength of trophic links between primary producers and Weddell seals, we should be better positioned to predict influences of possible future environmental conditions on population dynamics of Weddell seals. Furthermore, improved understanding of the roles of birth-year and subsequent environmental conditions in demography of other long-lived species ought to be useful for developing predictive models of population dynamics, especially in the face of projected changes in climate patterns.

Acknowledgments

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). Raytheon Polar Services Corporation, Petroleum Helicopters International, and the New York National Guard provided logistical support for field work in Antarctica. W. L. Kendall, T. E. McMahon, and D. B. Siniff and 2 anonymous reviewers provided useful comments on previous manuscript drafts. Many field assistants assisted with data collection.

Copyright information

© The Society of Population Ecology and Springer Japan 2013